adding math to bob

batch bill  summarizer sends a batch api call to gpt

.github/workflows/build_systems.yml

@@ -23,6 +23,6 @@ jobs:
     steps:
       - uses: actions/checkout@v4
       - name: Build default package
-        run: "nixos-rebuild build --accept-flake-config --flake ./#${{ matrix.system }}"
+        run: "nixos-rebuild build --flake ./#${{ matrix.system }}"
       - name: copy to nix-cache
         run: nix copy --accept-flake-config --to unix:///host-nix/var/nix/daemon-socket/socket .#nixosConfigurations.${{ matrix.system }}.config.system.build.toplevel

.github/workflows/fix_eval_warnings.yml

@@ -0,0 +1,30 @@
+name: fix_eval_warnings
+on:
+  workflow_run:
+    workflows: ["build_systems"]
+    types: [completed]
+
+jobs:
+  check-warnings:
+    if: >-
+      github.event.workflow_run.conclusion != 'cancelled' &&
+      github.event.workflow_run.head_branch == 'main' &&
+      (github.event.workflow_run.event == 'push' || github.event.workflow_run.event == 'schedule')
+    runs-on: self-hosted
+    permissions:
+      contents: write
+      pull-requests: write
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Fix eval warnings
+        env:
+          GH_TOKEN: ${{ secrets.GH_TOKEN_FOR_UPDATES }}
+        run: >-
+          nix develop .#devShells.x86_64-linux.default -c
+          python -m python.eval_warnings.main
+          --run-id "${{ github.event.workflow_run.id }}"
+          --repo "${{ github.repository }}"
+          --ollama-url "${{ secrets.OLLAMA_URL }}"
+          --run-url "${{ github.event.workflow_run.html_url }}"

.github/workflows/merge_flake_lock_update.yml

@@ -6,18 +6,24 @@ on:
 
 jobs:
   merge:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
 
     permissions:
       contents: write
       pull-requests: write
 
     steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
       - name: merge_flake_lock_update
-        run: >-
+        run: |
-          nix develop .#devShells.x86_64-linux.default -c
+          pr_number=$(gh pr list --state open --author RichieCahill --label flake_lock_update --json number --jq '.[0].number')
-          python -m python.gitea_flake_lock merge
+          echo "pr_number=$pr_number" >> $GITHUB_ENV
-          --repo "${{ github.repository }}"
+          if [ -n "$pr_number" ]; then
+            gh pr merge "$pr_number" --rebase
+          else
+            echo "No open PR found with label flake_lock_update"
+          fi
         env:
-          GITEA_TOKEN: ${{ secrets.GITEA_TOKEN }}
+          GITHUB_TOKEN: ${{ secrets.GH_TOKEN_FOR_UPDATES }}
-          GITEA_URL: https://gitea.tmmworkshop.com

.github/workflows/pytest.yml

@@ -1,13 +1,13 @@
 name: pytest
 
 on:
-  workflow_dispatch:
   push:
     branches:
       - main
   pull_request:
     branches:
       - main
+  merge_group:
 
 jobs:
   pytest:

.github/workflows/update-flake-lock.yml

@@ -6,21 +6,18 @@ on:
 
 jobs:
   lockfile:
-    runs-on: self-hosted
+    runs-on: ubuntu-latest
-    permissions:
-      actions: write
-      contents: write
-      pull-requests: write
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
+      - name: Install Nix
+        uses: DeterminateSystems/nix-installer-action@main
       - name: Update flake.lock
-        run: nix flake update
+        uses: DeterminateSystems/update-flake-lock@main
-      - name: Create or update flake.lock PR
+        with:
-        env:
+          token: ${{ secrets.GH_TOKEN_FOR_UPDATES }}
-          GITEA_TOKEN: ${{ secrets.GITEA_TOKEN }}
+          pr-title: "Update flake.lock"
-          GITEA_URL: https://gitea.tmmworkshop.com
+          pr-labels: |
-        run: >-
+            dependencies
-          nix develop .#devShells.x86_64-linux.default -c
+            automated
-          python -m python.gitea_flake_lock update
+            flake_lock_update
-          --repo "${{ github.repository }}"

.gitignore

@@ -170,5 +170,6 @@ test.*
 frontend/dist/
 frontend/node_modules/
 
-# data from testing llms
+# data dir for training, validation, and testing
-data/*
+data/
+config.toml

.vscode/settings.json

@@ -40,6 +40,7 @@
     "cgroupdriver",
     "charliermarsh",
     "Checkpointing",
+    "cloudflared",
     "codellama",
     "codezombiech",
     "compactmode",
@@ -203,7 +204,6 @@
     "peerconnection",
     "PESKYFOX",
     "PGID",
-    "pgvector",
     "pipewire",
     "pkgs",
     "plugdev",

AGENTS.md

@@ -0,0 +1,12 @@
+## Dev environment tips
+
+- use treefmt to format all files
+- make python code ruff compliant
+- use pytest to test python code
+- always use the minimum amount of complexity
+- if judgment calls are easy to reverse make them. if not ask me first
+- Match existing code style.
+- Use builtin helpers getenv() over os.environ.get.
+- Prefer single-purpose functions over “do everything” helpers.
+- Avoid compatibility branches like PG_USER and POSTGRESQL_URL unless requested.
+- Keep helpers only if reused or they simplify the code otherwise inline.

common/global/default.nix

@@ -23,10 +23,7 @@
   boot = {
     tmp.useTmpfs = true;
     kernelPackages = lib.mkDefault pkgs.linuxPackages_6_12;
-    zfs = {
+    zfs.package = lib.mkDefault pkgs.zfs_2_4;
-      package = lib.mkDefault pkgs.zfs_2_4;
-      forceImportRoot = lib.mkDefault false;
-    };
   };
 
   hardware.enableRedistributableFirmware = true;
@@ -40,17 +37,10 @@
 
   nixpkgs = {
     overlays = builtins.attrValues outputs.overlays;
-    config = {
+    config.allowUnfree = true;
-      allowUnfree = true;
-      permittedInsecurePackages = [
-        "openssl-1.1.1w" # This is for discord-canary
-      ];
-    };
   };
 
   services = {
-    dbus.implementation = "dbus";
-
     # firmware update
     fwupd.enable = true;
 

common/global/nix.nix

@@ -34,7 +34,6 @@ in
       warn-dirty = false;
       flake-registry = ""; # disable global flake registries
       connect-timeout = 10;
-      download-buffer-size = 536870912;
       fallback = true;
     };
 

common/optional/monitoring-agent.nix

@@ -1,256 +0,0 @@
-{
-  config,
-  lib,
-  pkgs,
-  ...
-}:
-let
-  monitoringInterface = "ztwfunumly";
-  nodeTextfileDir = "/var/lib/prometheus-node-exporter-textfile";
-
-  mkProcessNameTemplate =
-    perPid: template: if perPid then "${template}:{{.PID}}:{{.StartTime}}" else template;
-
-  mkProcessMatchers = perPid: [
-    {
-      name = mkProcessNameTemplate perPid "{{.Username}}:{{.Matches.Module}}";
-      cmdline = [ "^/nix/store[^ ]*/bin/python[^ ]* -m (?P<Module>[^ ]+)" ];
-    }
-    {
-      name = mkProcessNameTemplate perPid "{{.Username}}:{{.Matches.Wrapped}}";
-      cmdline = [
-        "^/nix/store[^ ]*/bin/python[^ ]* /nix/store[^ ]*/bin/\\.?(?P<Wrapped>[^ /]+?)(?:-wrapped)?(?:\\s|$)"
-      ];
-    }
-    {
-      name = mkProcessNameTemplate perPid "{{.Username}}:{{.Matches.Wrapped}}";
-      cmdline = [
-        "^/nix/store[^ ]*/bin/node /nix/store[^ ]*-(?P<Wrapped>[A-Za-z0-9._+-]+)-[0-9][^ /]*/"
-      ];
-    }
-    {
-      name = mkProcessNameTemplate perPid "{{.Username}}:{{.Matches.Wrapped}}";
-      cmdline = [ "^/nix/store[^ ]*/(?:bin/|lib/[^ ]*/)?\\.?(?P<Wrapped>[^ /]+?)(?:-wrapped)?(?:\\s|$)" ];
-    }
-    {
-      name = mkProcessNameTemplate perPid "{{.Username}}:{{.ExeBase}}";
-      cmdline = [ ".+" ];
-    }
-  ];
-
-  perPidConfig = pkgs.writeText "process-exporter-per-pid.yaml" (
-    builtins.toJSON {
-      process_names = mkProcessMatchers true;
-    }
-  );
-
-  zpoolLatencyScript = pkgs.writeShellScript "zpool-latency-exporter" ''
-        set -euo pipefail
-
-        out_dir=${lib.escapeShellArg nodeTextfileDir}
-        host=${lib.escapeShellArg config.networking.hostName}
-        tmp_file="$(mktemp "$out_dir/zpool.prom.XXXXXX")"
-        trap 'rm -f "$tmp_file"' EXIT
-
-        pools="$(zpool list -H -o name | paste -sd, -)"
-
-        cat >"$tmp_file" <<'EOF'
-    # HELP zpool_iostat_total_wait_read_ns Average total read wait time reported by zpool iostat.
-    # TYPE zpool_iostat_total_wait_read_ns gauge
-    # HELP zpool_iostat_total_wait_write_ns Average total write wait time reported by zpool iostat.
-    # TYPE zpool_iostat_total_wait_write_ns gauge
-    # HELP zpool_iostat_disk_wait_read_ns Average disk read wait time reported by zpool iostat.
-    # TYPE zpool_iostat_disk_wait_read_ns gauge
-    # HELP zpool_iostat_disk_wait_write_ns Average disk write wait time reported by zpool iostat.
-    # TYPE zpool_iostat_disk_wait_write_ns gauge
-    # HELP zpool_iostat_syncq_wait_read_ns Average synchronous queue read wait time reported by zpool iostat.
-    # TYPE zpool_iostat_syncq_wait_read_ns gauge
-    # HELP zpool_iostat_syncq_wait_write_ns Average synchronous queue write wait time reported by zpool iostat.
-    # TYPE zpool_iostat_syncq_wait_write_ns gauge
-    # HELP zpool_iostat_asyncq_wait_read_ns Average asynchronous queue read wait time reported by zpool iostat.
-    # TYPE zpool_iostat_asyncq_wait_read_ns gauge
-    # HELP zpool_iostat_asyncq_wait_write_ns Average asynchronous queue write wait time reported by zpool iostat.
-    # TYPE zpool_iostat_asyncq_wait_write_ns gauge
-    EOF
-
-        zpool iostat -Hplvy -y 1 1 | awk -F '\t' -v host="$host" -v pools="$pools" '
-          function esc(str, out) {
-            out = str
-            gsub(/\\/, "\\\\", out)
-            gsub(/"/, "\\\"", out)
-            return out
-          }
-
-          function emit(metric, pool, vdev, value) {
-            if (value == "" || value == "-") {
-              return
-            }
-
-            printf "%s{host=\"%s\",pool=\"%s\",vdev=\"%s\"} %s\n",
-              metric,
-              esc(host),
-              esc(pool),
-              esc(vdev),
-              value
-          }
-
-          BEGIN {
-            split(pools, pool_names, ",")
-            for (idx in pool_names) {
-              if (pool_names[idx] != "") {
-                known_pools[pool_names[idx]] = 1
-              }
-            }
-          }
-
-          NF == 0 {
-            next
-          }
-
-          {
-            row_name = $1
-
-            if (row_name in known_pools) {
-              current_pool = row_name
-              current_vdev = "_pool"
-            } else if (current_pool == "") {
-              next
-            } else {
-              current_vdev = row_name
-            }
-
-            emit("zpool_iostat_total_wait_read_ns", current_pool, current_vdev, $8)
-            emit("zpool_iostat_total_wait_write_ns", current_pool, current_vdev, $9)
-            emit("zpool_iostat_disk_wait_read_ns", current_pool, current_vdev, $10)
-            emit("zpool_iostat_disk_wait_write_ns", current_pool, current_vdev, $11)
-            emit("zpool_iostat_syncq_wait_read_ns", current_pool, current_vdev, $12)
-            emit("zpool_iostat_syncq_wait_write_ns", current_pool, current_vdev, $13)
-            emit("zpool_iostat_asyncq_wait_read_ns", current_pool, current_vdev, $14)
-            emit("zpool_iostat_asyncq_wait_write_ns", current_pool, current_vdev, $15)
-          }
-        ' >>"$tmp_file"
-
-        mv "$tmp_file" "$out_dir/zpool.prom"
-        trap - EXIT
-  '';
-in
-{
-  networking.firewall.interfaces.${monitoringInterface}.allowedTCPPorts = [
-    9100
-    9134
-    9256
-    9257
-    9633
-  ];
-
-  services.prometheus.exporters = {
-    node = {
-      enable = true;
-      enabledCollectors = [
-        "pressure"
-        "processes"
-        "systemd"
-      ];
-      extraFlags = [ "--collector.textfile.directory=${nodeTextfileDir}" ];
-    };
-
-    process = {
-      enable = true;
-      user = "root";
-      group = "root";
-      settings.process_names = mkProcessMatchers false;
-      extraFlags = [
-        "-gather-smaps=false"
-        "-remove-empty-groups=true"
-        "-threads=false"
-      ];
-    };
-
-    smartctl.enable = true;
-    zfs.enable = true;
-  };
-
-  programs.atop = {
-    enable = true;
-    atopService.enable = true;
-    atopRotateTimer.enable = true;
-    atopacctService.enable = true;
-    settings.interval = 30;
-  };
-
-  systemd = {
-    services = {
-      prometheus-process-pid-exporter = {
-        description = "Prometheus process exporter with per-PID naming";
-        wantedBy = [ "multi-user.target" ];
-        after = [ "network.target" ];
-        serviceConfig = {
-          ExecStart = ''
-            ${pkgs.prometheus-process-exporter}/bin/process-exporter \
-              --web.listen-address 0.0.0.0:9257 \
-              --config.path ${perPidConfig} \
-              -children=false \
-              -gather-smaps=false \
-              -remove-empty-groups=true \
-              -threads=false
-          '';
-          User = "root";
-          Group = "root";
-          Restart = "always";
-          WorkingDirectory = "/tmp";
-          CapabilityBoundingSet = [ "" ];
-          DeviceAllow = [ "" ];
-          LockPersonality = true;
-          MemoryDenyWriteExecute = true;
-          NoNewPrivileges = true;
-          PrivateDevices = true;
-          PrivateTmp = true;
-          ProtectClock = true;
-          ProtectControlGroups = true;
-          ProtectHome = true;
-          ProtectHostname = true;
-          ProtectKernelLogs = true;
-          ProtectKernelModules = true;
-          ProtectKernelTunables = true;
-          ProtectSystem = "strict";
-          RemoveIPC = true;
-          RestrictAddressFamilies = [
-            "AF_INET"
-            "AF_INET6"
-          ];
-          RestrictNamespaces = true;
-          RestrictRealtime = true;
-          RestrictSUIDSGID = true;
-          SystemCallArchitectures = "native";
-          UMask = "0077";
-        };
-      };
-
-      zpool-latency-exporter = {
-        description = "Exports ZFS latency metrics for node_exporter textfile collection";
-        after = [ "zfs-import.target" ];
-        requires = [ "zfs-import.target" ];
-        path = [
-          config.boot.zfs.package
-          pkgs.coreutils
-          pkgs.gawk
-        ];
-        serviceConfig = {
-          Type = "oneshot";
-          ExecStart = zpoolLatencyScript;
-        };
-      };
-    };
-
-    timers.zpool-latency-exporter = {
-      wantedBy = [ "timers.target" ];
-      timerConfig = {
-        OnBootSec = "2m";
-        OnUnitActiveSec = "60s";
-        Unit = "zpool-latency-exporter.service";
-      };
-    };
-
-    tmpfiles.rules = [ "d ${nodeTextfileDir} 0755 root root - -" ];
-  };
-}

common/optional/syncthing_base.nix

@@ -12,7 +12,7 @@
       brain.id = "SSCGIPI-IV3VYKB-TRNIJE3-COV4T2H-CDBER7F-I2CGHYA-NWOEUDU-3T5QAAN"; # cspell:disable-line
       ipad.id = "KI76T3X-SFUGV2L-VSNYTKR-TSIUV5L-SHWD3HE-GQRGRCN-GY4UFMD-CW6Z6AX"; # cspell:disable-line
       jeeves.id = "ICRHXZW-ECYJCUZ-I4CZ64R-3XRK7CG-LL2HAAK-FGOHD22-BQA4AI6-5OAL6AG"; # cspell:disable-line
-      phone.id = "JPVQKQW-CFXOJXT-Q5G5F3H-QIDHDRE-GKHPTQB-GXZUQSP-U7FR7F7-INP3AAH"; # cspell:disable-line
+      phone.id = "TBRULKD-7DZPGGZ-F6LLB7J-MSO54AY-7KLPBIN-QOFK6PX-W2HBEWI-PHM2CQI"; # cspell:disable-line
       rhapsody-in-green.id = "ASL3KC4-3XEN6PA-7BQBRKE-A7JXLI6-DJT43BY-Q4WPOER-7UALUAZ-VTPQ6Q4"; # cspell:disable-line
     };
   };

common/optional/update.nix

@@ -4,7 +4,7 @@
     flags = [ "--accept-flake-config" ];
     randomizedDelaySec = "1h";
     persistent = true;
-    flake = "git+https://gitea.tmmworkshop.com/richie/dotfiles?ref=main";
+    flake = "github:RichieCahill/dotfiles";
     allowReboot = true;
     dates = "Sat *-*-* 06:00:00";
   };

docs/zfs_failed_root_import_recovry.md

@@ -1,76 +0,0 @@
-# ZFS failed root import recovery
-
-## Fast path
-
-If the machine fails to boot because ZFS refuses to import `root_pool`:
-
-### GRUB
-
-1. At the bootloader menu, select the normal NixOS entry.
-2. Press `e`.
-3. Find the line that starts with `linux`.
-4. Append this to the end of that line:
-
-```text
-zfs_force=1
-```
-
-5. Boot once with `Ctrl+x` or `F10`.
-
-### systemd-boot
-
-1. At the bootloader menu, highlight the normal NixOS entry.
-2. Press `e`.
-3. Append this to the end of the options line:
-
-```text
-zfs_force=1
-```
-
-4. Press `Enter` to boot once.
-
-## After boot
-
-Run:
-
-```bash
-sudo zpool status
-sudo zpool import
-journalctl -b | rg "ZFS|zfs|import|root_pool"
-```
-
-## Expected result
-
-`sudo zpool status` should show `root_pool` as `ONLINE`.
-
-## Reboot test
-
-Run:
-
-```bash
-sudo reboot
-```
-
-Do not add `zfs_force=1` the second time.
-
-## If it still fails
-
-Boot once more with:
-
-```text
-zfs_force=1
-```
-
-Then run:
-
-```bash
-sudo zpool status -v
-sudo zpool history | tail -n 50
-journalctl -b | rg "ZFS|zfs|import|root_pool"
-```
-
-## Notes
-
-- Root pool name is `root_pool`.
-- This is a one-time recovery path after disk moves, controller changes, dirty exports, or interrupted imports.
-- Some hosts also need the LUKS unlock USB key inserted before boot.

flake.lock

@@ -8,11 +8,11 @@
       },
       "locked": {
         "dir": "pkgs/firefox-addons",
-        "lastModified": 1781150628,
+        "lastModified": 1773979456,
-        "narHash": "sha256-b4mp8l3qWuSCyYYo9HSngDtcB3PpecYiOXjULrjwwlw=",
+        "narHash": "sha256-9kBMJ5IvxqNlkkj/swmE8uK1Sc7TL/LIRUI958m7uBM=",
         "owner": "rycee",
         "repo": "nur-expressions",
-        "rev": "753319310f4673a2dabbfab87482187b40bf9bac",
+        "rev": "81e28f47ac18d9e89513929c77e711e657b64851",
         "type": "gitlab"
       },
       "original": {
@@ -29,11 +29,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1781189114,
+        "lastModified": 1774007980,
-        "narHash": "sha256-5inaamLgUMWy+MOBE9ChF9QAF1o/74LFuHkI0W/9rqc=",
+        "narHash": "sha256-FOnZjElEI8pqqCvB6K/1JRHTE8o4rer8driivTpq2uo=",
         "owner": "nix-community",
         "repo": "home-manager",
-        "rev": "486595d2cf49cfcd649b58a284fa11ac0e34da22",
+        "rev": "9670de2921812bc4e0452f6e3efd8c859696c183",
         "type": "github"
       },
       "original": {
@@ -43,15 +43,12 @@
       }
     },
     "nixos-hardware": {
-      "inputs": {
-        "nixpkgs": "nixpkgs"
-      },
       "locked": {
-        "lastModified": 1781168557,
+        "lastModified": 1774018263,
-        "narHash": "sha256-LOnLQ2tpYF9gqIDDr3+j3DbpJJr/QCH6zPRT2GzEUOE=",
+        "narHash": "sha256-HHYEwK1A22aSaxv2ibhMMkKvrDGKGlA/qObG4smrSqc=",
         "owner": "nixos",
         "repo": "nixos-hardware",
-        "rev": "6358ff76821101c178e3ab4919a62799bfe3652e",
+        "rev": "2d4b4717b2534fad5c715968c1cece04a172b365",
         "type": "github"
       },
       "original": {
@@ -63,24 +60,27 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1767892417,
+        "lastModified": 1773821835,
-        "narHash": "sha256-8bW3q88CEg2u4hSP66Vf4lpbLonHz7hqDNBMcCY7E9U=",
+        "narHash": "sha256-TJ3lSQtW0E2JrznGVm8hOQGVpXjJyXY2guAxku2O9A4=",
-        "rev": "3497aa5c9457a9d88d71fa93a4a8368816fbeeba",
+        "owner": "nixos",
-        "type": "tarball",
+        "repo": "nixpkgs",
-        "url": "https://releases.nixos.org/nixos/unstable/nixos-26.05pre924538.3497aa5c9457/nixexprs.tar.xz"
+        "rev": "b40629efe5d6ec48dd1efba650c797ddbd39ace0",
+        "type": "github"
       },
       "original": {
-        "type": "tarball",
+        "owner": "nixos",
-        "url": "https://channels.nixos.org/nixos-unstable/nixexprs.tar.xz"
+        "ref": "nixos-unstable",
+        "repo": "nixpkgs",
+        "type": "github"
       }
     },
     "nixpkgs-master": {
       "locked": {
-        "lastModified": 1781229721,
+        "lastModified": 1774051532,
-        "narHash": "sha256-ORvqDbb/LYxiJljGIejapjkc/kJbVote2N1WSb9W45I=",
+        "narHash": "sha256-d3CGMweyYIcPuTj5BKq+1Lx4zwlgL31nVtN647tOZKo=",
         "owner": "nixos",
         "repo": "nixpkgs",
-        "rev": "173d0ad7a974f8543a9ab01d2271b2e290341b33",
+        "rev": "8620c0b5cc8fbe76502442181be1d0514bc3a1b7",
         "type": "github"
       },
       "original": {
@@ -106,28 +106,12 @@
         "type": "github"
       }
     },
-    "nixpkgs_2": {
-      "locked": {
-        "lastModified": 1781074563,
-        "narHash": "sha256-md8WlXOlfnIeHeOScMTTHFyf2d6iaTwPl2apR5EQ3P4=",
-        "owner": "nixos",
-        "repo": "nixpkgs",
-        "rev": "9ae611a455b90cf061d8f332b977e387bda8e1ca",
-        "type": "github"
-      },
-      "original": {
-        "owner": "nixos",
-        "ref": "nixos-unstable",
-        "repo": "nixpkgs",
-        "type": "github"
-      }
-    },
     "root": {
       "inputs": {
         "firefox-addons": "firefox-addons",
         "home-manager": "home-manager",
         "nixos-hardware": "nixos-hardware",
-        "nixpkgs": "nixpkgs_2",
+        "nixpkgs": "nixpkgs",
         "nixpkgs-master": "nixpkgs-master",
         "nixpkgs-stable": "nixpkgs-stable",
         "sops-nix": "sops-nix",
@@ -141,11 +125,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1780547341,
+        "lastModified": 1773889674,
-        "narHash": "sha256-Gq8KNx5A7hBB3uGJaj6eQfLDIz5YdLu92gqBcvHvoUo=",
+        "narHash": "sha256-+ycaiVAk3MEshJTg35cBTUa0MizGiS+bgpYw/f8ohkg=",
         "owner": "Mic92",
         "repo": "sops-nix",
-        "rev": "9ed65852b6257fbeae4355bc24ecfea307ca759a",
+        "rev": "29b6519f3e0780452bca0ac0be4584f04ac16cc5",
         "type": "github"
       },
       "original": {

frontend/.gitignore

@@ -0,0 +1,24 @@
+# Logs
+logs
+*.log
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
+lerna-debug.log*
+
+node_modules
+dist
+dist-ssr
+*.local
+
+# Editor directories and files
+.vscode/*
+!.vscode/extensions.json
+.idea
+.DS_Store
+*.suo
+*.ntvs*
+*.njsproj
+*.sln
+*.sw?

overlays/default.nix

@@ -24,6 +24,7 @@
         fastapi
         fastapi-cli
         httpx
+        huggingface-hub
         mypy
         orjson
         polars

pyproject.toml

@@ -12,6 +12,7 @@ dependencies = [
     "alembic",
     "apprise",
     "apscheduler",
+    "huggingface-hub",
     "httpx",
     "python-multipart",
     "polars",
@@ -26,7 +27,11 @@ dependencies = [
 [project.scripts]
 database = "python.database_cli:app"
 van-inventory = "python.van_inventory.main:serve"
-whisper-transcribe = "python.tools.whisper.transcribe:main"
+prompt-bench = "python.prompt_bench.main:cli"
+prompt-bench-download = "python.prompt_bench.downloader:cli"
+finetune = "python.prompt_bench.finetune:cli"
+finetune-container = "python.prompt_bench.finetune_container:cli"
+build-finetune-dataset = "python.prompt_bench.build_finetune_dataset:cli"
 
 [dependency-groups]
 dev = [
@@ -51,7 +56,6 @@ lint.ignore = [
     "COM812", # (TEMP) conflicts when used with the formatter
     "ISC001", # (TEMP) conflicts when used with the formatter
     "S603",   # (PERM) This is known to cause a false positive
-    "S607",   # (PERM) This is becoming a consistent annoyance
 ]
 
 [tool.ruff.lint.per-file-ignores]
@@ -80,10 +84,20 @@ lint.ignore = [
 "python/congress_tracker/**" = [
     "TC003", # (perm) this creates issues because sqlalchemy uses these at runtime
 ]
-
+"python/eval_warnings/**" = [
+    "S607", # (perm) gh and git are expected on PATH in the runner environment
+]
+"python/prompt_bench/**" = [
+    "FBT002",  # (perm) typer requires boolean defaults for --flag/--no-flag options
+    "PLR0913", # (perm) typer CLIs naturally have many parameters
+    "S607",    # (perm) docker and nvidia-smi are expected on PATH
+]
 "python/alembic/**" = [
     "INP001", # (perm) this creates LSP issues for alembic
 ]
+"python/signal_bot/**" = [
+    "D107", # (perm) class docstrings cover __init__
+]
 
 [tool.ruff.lint.pydocstyle]
 convention = "google"

python/alembic/data_science_dev/versions/2026_03_24-adding_failedingestion_2f43120e3ffc.py

@@ -0,0 +1,50 @@
+"""adding FailedIngestion.
+
+Revision ID: 2f43120e3ffc
+Revises: f99be864fe69
+Create Date: 2026-03-24 23:46:17.277897
+
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+import sqlalchemy as sa
+from alembic import op
+
+from python.orm import DataScienceDevBase
+
+if TYPE_CHECKING:
+    from collections.abc import Sequence
+
+# revision identifiers, used by Alembic.
+revision: str = "2f43120e3ffc"
+down_revision: str | None = "f99be864fe69"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+schema = DataScienceDevBase.schema_name
+
+
+def upgrade() -> None:
+    """Upgrade."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.create_table(
+        "failed_ingestion",
+        sa.Column("raw_line", sa.Text(), nullable=False),
+        sa.Column("error", sa.Text(), nullable=False),
+        sa.Column("id", sa.Integer(), nullable=False),
+        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_failed_ingestion")),
+        schema=schema,
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    """Downgrade."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_table("failed_ingestion", schema=schema)
+    # ### end Alembic commands ###

python/alembic/data_science_dev/versions/2026_03_25-attach_partitions_to_posts.py

@@ -0,0 +1,72 @@
+"""Attach all partition tables to the posts parent table.
+
+Alembic autogenerate creates partition tables as standalone tables but does not
+emit the ALTER TABLE ... ATTACH PARTITION statements needed for PostgreSQL to
+route inserts to the correct partition.
+
+Revision ID: a1b2c3d4e5f6
+Revises: 605b1794838f
+Create Date: 2026-03-25 10:00:00.000000
+
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+from alembic import op
+from sqlalchemy import text
+
+from python.orm import DataScienceDevBase
+from python.orm.data_science_dev.posts.partitions import (
+    PARTITION_END_YEAR,
+    PARTITION_START_YEAR,
+    iso_weeks_in_year,
+    week_bounds,
+)
+
+if TYPE_CHECKING:
+    from collections.abc import Sequence
+
+# revision identifiers, used by Alembic.
+revision: str = "a1b2c3d4e5f6"
+down_revision: str | None = "605b1794838f"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+schema = DataScienceDevBase.schema_name
+
+ALREADY_ATTACHED_QUERY = text("""
+    SELECT inhrelid::regclass::text
+    FROM pg_inherits
+    WHERE inhparent = :parent::regclass
+""")
+
+
+def upgrade() -> None:
+    """Attach all weekly partition tables to the posts parent table."""
+    connection = op.get_bind()
+    already_attached = {row[0] for row in connection.execute(ALREADY_ATTACHED_QUERY, {"parent": f"{schema}.posts"})}
+
+    for year in range(PARTITION_START_YEAR, PARTITION_END_YEAR + 1):
+        for week in range(1, iso_weeks_in_year(year) + 1):
+            table_name = f"posts_{year}_{week:02d}"
+            qualified_name = f"{schema}.{table_name}"
+            if qualified_name in already_attached:
+                continue
+            start, end = week_bounds(year, week)
+            start_str = start.strftime("%Y-%m-%d %H:%M:%S")
+            end_str = end.strftime("%Y-%m-%d %H:%M:%S")
+            op.execute(
+                f"ALTER TABLE {schema}.posts "
+                f"ATTACH PARTITION {qualified_name} "
+                f"FOR VALUES FROM ('{start_str}') TO ('{end_str}')"
+            )
+
+
+def downgrade() -> None:
+    """Detach all weekly partition tables from the posts parent table."""
+    for year in range(PARTITION_START_YEAR, PARTITION_END_YEAR + 1):
+        for week in range(1, iso_weeks_in_year(year) + 1):
+            table_name = f"posts_{year}_{week:02d}"
+            op.execute(f"ALTER TABLE {schema}.posts DETACH PARTITION {schema}.{table_name}")

python/alembic/data_science_dev/versions/2026_03_27-adding_congress_data_83bfc8af92d8.py

@@ -0,0 +1,153 @@
+"""adding congress data.
+
+Revision ID: 83bfc8af92d8
+Revises: a1b2c3d4e5f6
+Create Date: 2026-03-27 10:43:02.324510
+
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+import sqlalchemy as sa
+from alembic import op
+
+from python.orm import DataScienceDevBase
+
+if TYPE_CHECKING:
+    from collections.abc import Sequence
+
+# revision identifiers, used by Alembic.
+revision: str = "83bfc8af92d8"
+down_revision: str | None = "a1b2c3d4e5f6"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+schema = DataScienceDevBase.schema_name
+
+
+def upgrade() -> None:
+    """Upgrade."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.create_table(
+        "bill",
+        sa.Column("congress", sa.Integer(), nullable=False),
+        sa.Column("bill_type", sa.String(), nullable=False),
+        sa.Column("number", sa.Integer(), nullable=False),
+        sa.Column("title", sa.String(), nullable=True),
+        sa.Column("title_short", sa.String(), nullable=True),
+        sa.Column("official_title", sa.String(), nullable=True),
+        sa.Column("status", sa.String(), nullable=True),
+        sa.Column("status_at", sa.Date(), nullable=True),
+        sa.Column("sponsor_bioguide_id", sa.String(), nullable=True),
+        sa.Column("subjects_top_term", sa.String(), nullable=True),
+        sa.Column("id", sa.Integer(), nullable=False),
+        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_bill")),
+        sa.UniqueConstraint("congress", "bill_type", "number", name="uq_bill_congress_type_number"),
+        schema=schema,
+    )
+    op.create_index("ix_bill_congress", "bill", ["congress"], unique=False, schema=schema)
+    op.create_table(
+        "legislator",
+        sa.Column("bioguide_id", sa.Text(), nullable=False),
+        sa.Column("thomas_id", sa.String(), nullable=True),
+        sa.Column("lis_id", sa.String(), nullable=True),
+        sa.Column("govtrack_id", sa.Integer(), nullable=True),
+        sa.Column("opensecrets_id", sa.String(), nullable=True),
+        sa.Column("fec_ids", sa.String(), nullable=True),
+        sa.Column("first_name", sa.String(), nullable=False),
+        sa.Column("last_name", sa.String(), nullable=False),
+        sa.Column("official_full_name", sa.String(), nullable=True),
+        sa.Column("nickname", sa.String(), nullable=True),
+        sa.Column("birthday", sa.Date(), nullable=True),
+        sa.Column("gender", sa.String(), nullable=True),
+        sa.Column("current_party", sa.String(), nullable=True),
+        sa.Column("current_state", sa.String(), nullable=True),
+        sa.Column("current_district", sa.Integer(), nullable=True),
+        sa.Column("current_chamber", sa.String(), nullable=True),
+        sa.Column("id", sa.Integer(), nullable=False),
+        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_legislator")),
+        schema=schema,
+    )
+    op.create_index(op.f("ix_legislator_bioguide_id"), "legislator", ["bioguide_id"], unique=True, schema=schema)
+    op.create_table(
+        "bill_text",
+        sa.Column("bill_id", sa.Integer(), nullable=False),
+        sa.Column("version_code", sa.String(), nullable=False),
+        sa.Column("version_name", sa.String(), nullable=True),
+        sa.Column("text_content", sa.String(), nullable=True),
+        sa.Column("date", sa.Date(), nullable=True),
+        sa.Column("id", sa.Integer(), nullable=False),
+        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.ForeignKeyConstraint(
+            ["bill_id"], [f"{schema}.bill.id"], name=op.f("fk_bill_text_bill_id_bill"), ondelete="CASCADE"
+        ),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_bill_text")),
+        sa.UniqueConstraint("bill_id", "version_code", name="uq_bill_text_bill_id_version_code"),
+        schema=schema,
+    )
+    op.create_table(
+        "vote",
+        sa.Column("congress", sa.Integer(), nullable=False),
+        sa.Column("chamber", sa.String(), nullable=False),
+        sa.Column("session", sa.Integer(), nullable=False),
+        sa.Column("number", sa.Integer(), nullable=False),
+        sa.Column("vote_type", sa.String(), nullable=True),
+        sa.Column("question", sa.String(), nullable=True),
+        sa.Column("result", sa.String(), nullable=True),
+        sa.Column("result_text", sa.String(), nullable=True),
+        sa.Column("vote_date", sa.Date(), nullable=False),
+        sa.Column("yea_count", sa.Integer(), nullable=True),
+        sa.Column("nay_count", sa.Integer(), nullable=True),
+        sa.Column("not_voting_count", sa.Integer(), nullable=True),
+        sa.Column("present_count", sa.Integer(), nullable=True),
+        sa.Column("bill_id", sa.Integer(), nullable=True),
+        sa.Column("id", sa.Integer(), nullable=False),
+        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.ForeignKeyConstraint(["bill_id"], [f"{schema}.bill.id"], name=op.f("fk_vote_bill_id_bill")),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_vote")),
+        sa.UniqueConstraint("congress", "chamber", "session", "number", name="uq_vote_congress_chamber_session_number"),
+        schema=schema,
+    )
+    op.create_index("ix_vote_congress_chamber", "vote", ["congress", "chamber"], unique=False, schema=schema)
+    op.create_index("ix_vote_date", "vote", ["vote_date"], unique=False, schema=schema)
+    op.create_table(
+        "vote_record",
+        sa.Column("vote_id", sa.Integer(), nullable=False),
+        sa.Column("legislator_id", sa.Integer(), nullable=False),
+        sa.Column("position", sa.String(), nullable=False),
+        sa.ForeignKeyConstraint(
+            ["legislator_id"],
+            [f"{schema}.legislator.id"],
+            name=op.f("fk_vote_record_legislator_id_legislator"),
+            ondelete="CASCADE",
+        ),
+        sa.ForeignKeyConstraint(
+            ["vote_id"], [f"{schema}.vote.id"], name=op.f("fk_vote_record_vote_id_vote"), ondelete="CASCADE"
+        ),
+        sa.PrimaryKeyConstraint("vote_id", "legislator_id", name=op.f("pk_vote_record")),
+        schema=schema,
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    """Downgrade."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_table("vote_record", schema=schema)
+    op.drop_index("ix_vote_date", table_name="vote", schema=schema)
+    op.drop_index("ix_vote_congress_chamber", table_name="vote", schema=schema)
+    op.drop_table("vote", schema=schema)
+    op.drop_table("bill_text", schema=schema)
+    op.drop_index(op.f("ix_legislator_bioguide_id"), table_name="legislator", schema=schema)
+    op.drop_table("legislator", schema=schema)
+    op.drop_index("ix_bill_congress", table_name="bill", schema=schema)
+    op.drop_table("bill", schema=schema)
+    # ### end Alembic commands ###

python/alembic/data_science_dev/versions/2026_03_29-adding_legislatorsocialmedia_5cd7eee3549d.py

@@ -0,0 +1,58 @@
+"""adding LegislatorSocialMedia.
+
+Revision ID: 5cd7eee3549d
+Revises: 83bfc8af92d8
+Create Date: 2026-03-29 11:53:44.224799
+
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+import sqlalchemy as sa
+from alembic import op
+
+from python.orm import DataScienceDevBase
+
+if TYPE_CHECKING:
+    from collections.abc import Sequence
+
+# revision identifiers, used by Alembic.
+revision: str = "5cd7eee3549d"
+down_revision: str | None = "83bfc8af92d8"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+schema = DataScienceDevBase.schema_name
+
+
+def upgrade() -> None:
+    """Upgrade."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.create_table(
+        "legislator_social_media",
+        sa.Column("legislator_id", sa.Integer(), nullable=False),
+        sa.Column("platform", sa.String(), nullable=False),
+        sa.Column("account_name", sa.String(), nullable=False),
+        sa.Column("url", sa.String(), nullable=True),
+        sa.Column("source", sa.String(), nullable=False),
+        sa.Column("id", sa.Integer(), nullable=False),
+        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.ForeignKeyConstraint(
+            ["legislator_id"],
+            [f"{schema}.legislator.id"],
+            name=op.f("fk_legislator_social_media_legislator_id_legislator"),
+        ),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_legislator_social_media")),
+        schema=schema,
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    """Downgrade."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_table("legislator_social_media", schema=schema)
+    # ### end Alembic commands ###

python/alembic/richie/versions/2026_06_03-adding_audiobook_libreary_metadata_d7864d1ffc17.py

@@ -1,93 +0,0 @@
-"""adding audiobook libreary metadata.
-
-Revision ID: d7864d1ffc17
-Revises: c8a794340928
-Create Date: 2026-06-03 20:24:09.200837
-
-"""
-
-from __future__ import annotations
-
-from typing import TYPE_CHECKING
-
-import sqlalchemy as sa
-from alembic import op
-
-from python.orm import RichieBase
-
-if TYPE_CHECKING:
-    from collections.abc import Sequence
-
-# revision identifiers, used by Alembic.
-revision: str = "d7864d1ffc17"
-down_revision: str | None = "c8a794340928"
-branch_labels: str | Sequence[str] | None = None
-depends_on: str | Sequence[str] | None = None
-
-schema = RichieBase.schema_name
-
-
-def upgrade() -> None:
-    """Upgrade."""
-    # ### commands auto generated by Alembic - please adjust! ###
-    op.create_table(
-        "audiobook_author",
-        sa.Column("name", sa.String(), nullable=False),
-        sa.Column("id", sa.Integer(), nullable=False),
-        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
-        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
-        sa.PrimaryKeyConstraint("id", name=op.f("pk_audiobook_author")),
-        sa.UniqueConstraint("name", name=op.f("uq_audiobook_author_name")),
-        schema=schema,
-    )
-    op.create_table(
-        "audiobook_series",
-        sa.Column("name", sa.String(), nullable=False),
-        sa.Column("author_id", sa.Integer(), nullable=False),
-        sa.Column("id", sa.Integer(), nullable=False),
-        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
-        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
-        sa.ForeignKeyConstraint(
-            ["author_id"],
-            [f"{schema}.audiobook_author.id"],
-            name=op.f("fk_audiobook_series_author_id_audiobook_author"),
-            ondelete="CASCADE",
-        ),
-        sa.PrimaryKeyConstraint("id", name=op.f("pk_audiobook_series")),
-        sa.UniqueConstraint("author_id", "name", name=op.f("uq_audiobook_series_author_id")),
-        schema=schema,
-    )
-    op.create_table(
-        "audiobook",
-        sa.Column("title", sa.String(), nullable=False),
-        sa.Column("author_id", sa.Integer(), nullable=False),
-        sa.Column("series_id", sa.Integer(), nullable=True),
-        sa.Column("series_index", sa.Integer(), nullable=False),
-        sa.Column("id", sa.Integer(), nullable=False),
-        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
-        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
-        sa.ForeignKeyConstraint(
-            ["author_id"],
-            [f"{schema}.audiobook_author.id"],
-            name=op.f("fk_audiobook_author_id_audiobook_author"),
-            ondelete="CASCADE",
-        ),
-        sa.ForeignKeyConstraint(
-            ["series_id"],
-            [f"{schema}.audiobook_series.id"],
-            name=op.f("fk_audiobook_series_id_audiobook_series"),
-            ondelete="SET NULL",
-        ),
-        sa.PrimaryKeyConstraint("id", name=op.f("pk_audiobook")),
-        schema=schema,
-    )
-    # ### end Alembic commands ###
-
-
-def downgrade() -> None:
-    """Downgrade."""
-    # ### commands auto generated by Alembic - please adjust! ###
-    op.drop_table("audiobook", schema=schema)
-    op.drop_table("audiobook_series", schema=schema)
-    op.drop_table("audiobook_author", schema=schema)
-    # ### end Alembic commands ###

python/alembic/richie/versions/2026_06_10-updated_series_index_to_float_and_added__b3c60cc5beb5.py

@@ -1,63 +0,0 @@
-"""updated series_index to float and added UniqueConstraint to audiobook and audiobook_author.
-
-Revision ID: b3c60cc5beb5
-Revises: d7864d1ffc17
-Create Date: 2026-06-10 20:02:43.073725
-
-"""
-
-from __future__ import annotations
-
-from typing import TYPE_CHECKING
-
-import sqlalchemy as sa
-from alembic import op
-
-from python.orm import RichieBase
-
-if TYPE_CHECKING:
-    from collections.abc import Sequence
-
-# revision identifiers, used by Alembic.
-revision: str = "b3c60cc5beb5"
-down_revision: str | None = "d7864d1ffc17"
-branch_labels: str | Sequence[str] | None = None
-depends_on: str | Sequence[str] | None = None
-
-schema = RichieBase.schema_name
-
-
-def upgrade() -> None:
-    """Upgrade."""
-    # ### commands auto generated by Alembic - please adjust! ###
-    op.alter_column(
-        "audiobook",
-        "series_index",
-        existing_type=sa.INTEGER(),
-        type_=sa.Float(),
-        existing_nullable=False,
-        schema=schema,
-    )
-    op.create_unique_constraint(
-        op.f("uq_audiobook_author_id"),
-        "audiobook",
-        ["author_id", "series_id", "title"],
-        schema=schema,
-        postgresql_nulls_not_distinct=True,
-    )
-    # ### end Alembic commands ###
-
-
-def downgrade() -> None:
-    """Downgrade."""
-    # ### commands auto generated by Alembic - please adjust! ###
-    op.drop_constraint(op.f("uq_audiobook_author_id"), "audiobook", schema=schema, type_="unique")
-    op.alter_column(
-        "audiobook",
-        "series_index",
-        existing_type=sa.Float(),
-        type_=sa.INTEGER(),
-        existing_nullable=False,
-        schema=schema,
-    )
-    # ### end Alembic commands ###

python/alembic/signal_bot/versions/2026_03_17-seprating_signal_bot_database_6eaf696e07a5.py

@@ -0,0 +1,100 @@
+"""seprating signal_bot database.
+
+Revision ID: 6eaf696e07a5
+Revises:
+Create Date: 2026-03-17 21:35:37.612672
+
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+from python.orm import SignalBotBase
+
+if TYPE_CHECKING:
+    from collections.abc import Sequence
+
+# revision identifiers, used by Alembic.
+revision: str = "6eaf696e07a5"
+down_revision: str | None = None
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+schema = SignalBotBase.schema_name
+
+
+def upgrade() -> None:
+    """Upgrade."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.create_table(
+        "dead_letter_message",
+        sa.Column("source", sa.String(), nullable=False),
+        sa.Column("message", sa.Text(), nullable=False),
+        sa.Column("received_at", sa.DateTime(timezone=True), nullable=False),
+        sa.Column(
+            "status", postgresql.ENUM("UNPROCESSED", "PROCESSED", name="message_status", schema=schema), nullable=False
+        ),
+        sa.Column("id", sa.Integer(), nullable=False),
+        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_dead_letter_message")),
+        schema=schema,
+    )
+    op.create_table(
+        "role",
+        sa.Column("name", sa.String(length=50), nullable=False),
+        sa.Column("id", sa.SmallInteger(), nullable=False),
+        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_role")),
+        sa.UniqueConstraint("name", name=op.f("uq_role_name")),
+        schema=schema,
+    )
+    op.create_table(
+        "signal_device",
+        sa.Column("phone_number", sa.String(length=50), nullable=False),
+        sa.Column("safety_number", sa.String(), nullable=True),
+        sa.Column(
+            "trust_level",
+            postgresql.ENUM("VERIFIED", "UNVERIFIED", "BLOCKED", name="trust_level", schema=schema),
+            nullable=False,
+        ),
+        sa.Column("last_seen", sa.DateTime(timezone=True), nullable=False),
+        sa.Column("id", sa.Integer(), nullable=False),
+        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_signal_device")),
+        sa.UniqueConstraint("phone_number", name=op.f("uq_signal_device_phone_number")),
+        schema=schema,
+    )
+    op.create_table(
+        "device_role",
+        sa.Column("device_id", sa.Integer(), nullable=False),
+        sa.Column("role_id", sa.SmallInteger(), nullable=False),
+        sa.Column("id", sa.Integer(), nullable=False),
+        sa.Column("created", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.Column("updated", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=False),
+        sa.ForeignKeyConstraint(
+            ["device_id"], [f"{schema}.signal_device.id"], name=op.f("fk_device_role_device_id_signal_device")
+        ),
+        sa.ForeignKeyConstraint(["role_id"], [f"{schema}.role.id"], name=op.f("fk_device_role_role_id_role")),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_device_role")),
+        sa.UniqueConstraint("device_id", "role_id", name="uq_device_role_device_role"),
+        schema=schema,
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    """Downgrade."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_table("device_role", schema=schema)
+    op.drop_table("signal_device", schema=schema)
+    op.drop_table("role", schema=schema)
+    op.drop_table("dead_letter_message", schema=schema)
+    # ### end Alembic commands ###

python/alembic/signal_bot/versions/2026_03_18-test_66bdd532bcab.py

@@ -0,0 +1,72 @@
+"""test.
+
+Revision ID: 66bdd532bcab
+Revises: 6eaf696e07a5
+Create Date: 2026-03-18 19:21:14.561568
+
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+from python.orm import SignalBotBase
+
+if TYPE_CHECKING:
+    from collections.abc import Sequence
+
+# revision identifiers, used by Alembic.
+revision: str = "66bdd532bcab"
+down_revision: str | None = "6eaf696e07a5"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+schema = SignalBotBase.schema_name
+
+
+def upgrade() -> None:
+    """Upgrade."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.alter_column(
+        "dead_letter_message",
+        "status",
+        existing_type=postgresql.ENUM("UNPROCESSED", "PROCESSED", name="message_status", schema=schema),
+        type_=sa.Enum("UNPROCESSED", "PROCESSED", name="message_status", native_enum=False),
+        existing_nullable=False,
+        schema=schema,
+    )
+    op.alter_column(
+        "signal_device",
+        "trust_level",
+        existing_type=postgresql.ENUM("VERIFIED", "UNVERIFIED", "BLOCKED", name="trust_level", schema=schema),
+        type_=sa.Enum("VERIFIED", "UNVERIFIED", "BLOCKED", name="trust_level", native_enum=False),
+        existing_nullable=False,
+        schema=schema,
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    """Downgrade."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.alter_column(
+        "signal_device",
+        "trust_level",
+        existing_type=sa.Enum("VERIFIED", "UNVERIFIED", "BLOCKED", name="trust_level", native_enum=False),
+        type_=postgresql.ENUM("VERIFIED", "UNVERIFIED", "BLOCKED", name="trust_level", schema=schema),
+        existing_nullable=False,
+        schema=schema,
+    )
+    op.alter_column(
+        "dead_letter_message",
+        "status",
+        existing_type=sa.Enum("UNPROCESSED", "PROCESSED", name="message_status", native_enum=False),
+        type_=postgresql.ENUM("UNPROCESSED", "PROCESSED", name="message_status", schema=schema),
+        existing_nullable=False,
+        schema=schema,
+    )
+    # ### end Alembic commands ###

python/data_science/__init__.py

@@ -0,0 +1,3 @@
+"""Data science CLI tools."""
+
+from __future__ import annotations

python/data_science/ingest_congress.py

@@ -0,0 +1,613 @@
+"""Ingestion pipeline for loading congress data from unitedstates/congress JSON files.
+
+Loads legislators, bills, votes, vote records, and bill text into the data_science_dev database.
+Expects the parent directory to contain congress-tracker/ and congress-legislators/ as siblings.
+
+Usage:
+    ingest-congress /path/to/parent/
+    ingest-congress /path/to/parent/ --congress 118
+    ingest-congress /path/to/parent/ --congress 118 --only bills
+"""
+
+from __future__ import annotations
+
+import logging
+from pathlib import Path  # noqa: TC003 needed at runtime for typer CLI argument
+from typing import TYPE_CHECKING, Annotated
+
+import orjson
+import typer
+import yaml
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
+from python.common import configure_logger
+from python.orm.common import get_postgres_engine
+from python.orm.data_science_dev.congress import Bill, BillText, Legislator, LegislatorSocialMedia, Vote, VoteRecord
+
+if TYPE_CHECKING:
+    from collections.abc import Iterator
+
+    from sqlalchemy.engine import Engine
+
+logger = logging.getLogger(__name__)
+
+BATCH_SIZE = 10_000
+
+app = typer.Typer(help="Ingest unitedstates/congress data into data_science_dev.")
+
+
+@app.command()
+def main(
+    parent_dir: Annotated[
+        Path,
+        typer.Argument(help="Parent directory containing congress-tracker/ and congress-legislators/"),
+    ],
+    congress: Annotated[int | None, typer.Option(help="Only ingest a specific congress number")] = None,
+    only: Annotated[
+        str | None,
+        typer.Option(help="Only run a specific step: legislators, social-media, bills, votes, bill-text"),
+    ] = None,
+) -> None:
+    """Ingest congress data from unitedstates/congress JSON files."""
+    configure_logger(level="INFO")
+
+    data_dir = parent_dir / "congress-tracker/congress/data/"
+    legislators_dir = parent_dir / "congress-legislators"
+
+    if not data_dir.is_dir():
+        typer.echo(f"Expected congress-tracker/ directory: {data_dir}", err=True)
+        raise typer.Exit(code=1)
+
+    if not legislators_dir.is_dir():
+        typer.echo(f"Expected congress-legislators/ directory: {legislators_dir}", err=True)
+        raise typer.Exit(code=1)
+
+    engine = get_postgres_engine(name="DATA_SCIENCE_DEV")
+
+    congress_dirs = _resolve_congress_dirs(data_dir, congress)
+    if not congress_dirs:
+        typer.echo("No congress directories found.", err=True)
+        raise typer.Exit(code=1)
+
+    logger.info("Found %d congress directories to process", len(congress_dirs))
+
+    steps: dict[str, tuple] = {
+        "legislators": (ingest_legislators, (engine, legislators_dir)),
+        "legislators-social-media": (ingest_social_media, (engine, legislators_dir)),
+        "bills": (ingest_bills, (engine, congress_dirs)),
+        "votes": (ingest_votes, (engine, congress_dirs)),
+        "bill-text": (ingest_bill_text, (engine, congress_dirs)),
+    }
+
+    if only:
+        if only not in steps:
+            typer.echo(f"Unknown step: {only}. Choose from: {', '.join(steps)}", err=True)
+            raise typer.Exit(code=1)
+        steps = {only: steps[only]}
+
+    for step_name, (step_func, step_args) in steps.items():
+        logger.info("=== Starting step: %s ===", step_name)
+        step_func(*step_args)
+        logger.info("=== Finished step: %s ===", step_name)
+
+    logger.info("ingest-congress done")
+
+
+def _resolve_congress_dirs(data_dir: Path, congress: int | None) -> list[Path]:
+    """Find congress number directories under data_dir."""
+    if congress is not None:
+        target = data_dir / str(congress)
+        return [target] if target.is_dir() else []
+    return sorted(path for path in data_dir.iterdir() if path.is_dir() and path.name.isdigit())
+
+
+def _flush_batch(session: Session, batch: list[object], label: str) -> int:
+    """Add a batch of ORM objects to the session and commit. Returns count added."""
+    if not batch:
+        return 0
+    session.add_all(batch)
+    session.commit()
+    count = len(batch)
+    logger.info("Committed %d %s", count, label)
+    batch.clear()
+    return count
+
+
+# ---------------------------------------------------------------------------
+# Legislators — loaded from congress-legislators YAML files
+# ---------------------------------------------------------------------------
+
+
+def ingest_legislators(engine: Engine, legislators_dir: Path) -> None:
+    """Load legislators from congress-legislators YAML files."""
+    legislators_data = _load_legislators_yaml(legislators_dir)
+    logger.info("Loaded %d legislators from YAML files", len(legislators_data))
+
+    with Session(engine) as session:
+        existing_legislators = {
+            legislator.bioguide_id: legislator for legislator in session.scalars(select(Legislator)).all()
+        }
+        logger.info("Found %d existing legislators in DB", len(existing_legislators))
+
+        total_inserted = 0
+        total_updated = 0
+        for entry in legislators_data:
+            bioguide_id = entry.get("id", {}).get("bioguide")
+            if not bioguide_id:
+                continue
+
+            fields = _parse_legislator(entry)
+            if existing := existing_legislators.get(bioguide_id):
+                changed = False
+                for field, value in fields.items():
+                    if value is not None and getattr(existing, field) != value:
+                        setattr(existing, field, value)
+                        changed = True
+                if changed:
+                    total_updated += 1
+            else:
+                session.add(Legislator(bioguide_id=bioguide_id, **fields))
+                total_inserted += 1
+
+        session.commit()
+    logger.info("Inserted %d new legislators, updated %d existing", total_inserted, total_updated)
+
+
+def _load_legislators_yaml(legislators_dir: Path) -> list[dict]:
+    """Load and combine legislators-current.yaml and legislators-historical.yaml."""
+    legislators: list[dict] = []
+    for filename in ("legislators-current.yaml", "legislators-historical.yaml"):
+        path = legislators_dir / filename
+        if not path.exists():
+            logger.warning("Legislators file not found: %s", path)
+            continue
+        with path.open() as file:
+            data = yaml.safe_load(file)
+            if isinstance(data, list):
+                legislators.extend(data)
+    return legislators
+
+
+def _parse_legislator(entry: dict) -> dict:
+    """Extract Legislator fields from a congress-legislators YAML entry."""
+    ids = entry.get("id", {})
+    name = entry.get("name", {})
+    bio = entry.get("bio", {})
+    terms = entry.get("terms", [])
+    latest_term = terms[-1] if terms else {}
+
+    fec_ids = ids.get("fec")
+    fec_ids_joined = ",".join(fec_ids) if isinstance(fec_ids, list) else fec_ids
+
+    chamber = latest_term.get("type")
+    chamber_normalized = {"rep": "House", "sen": "Senate"}.get(chamber, chamber)
+
+    return {
+        "thomas_id": ids.get("thomas"),
+        "lis_id": ids.get("lis"),
+        "govtrack_id": ids.get("govtrack"),
+        "opensecrets_id": ids.get("opensecrets"),
+        "fec_ids": fec_ids_joined,
+        "first_name": name.get("first"),
+        "last_name": name.get("last"),
+        "official_full_name": name.get("official_full"),
+        "nickname": name.get("nickname"),
+        "birthday": bio.get("birthday"),
+        "gender": bio.get("gender"),
+        "current_party": latest_term.get("party"),
+        "current_state": latest_term.get("state"),
+        "current_district": latest_term.get("district"),
+        "current_chamber": chamber_normalized,
+    }
+
+
+# ---------------------------------------------------------------------------
+# Social Media — loaded from legislators-social-media.yaml
+# ---------------------------------------------------------------------------
+
+SOCIAL_MEDIA_PLATFORMS = {
+    "twitter": "https://twitter.com/{account}",
+    "facebook": "https://facebook.com/{account}",
+    "youtube": "https://youtube.com/{account}",
+    "instagram": "https://instagram.com/{account}",
+    "mastodon": None,
+}
+
+
+def ingest_social_media(engine: Engine, legislators_dir: Path) -> None:
+    """Load social media accounts from legislators-social-media.yaml."""
+    social_media_path = legislators_dir / "legislators-social-media.yaml"
+    if not social_media_path.exists():
+        logger.warning("Social media file not found: %s", social_media_path)
+        return
+
+    with social_media_path.open() as file:
+        social_media_data = yaml.safe_load(file)
+
+    if not isinstance(social_media_data, list):
+        logger.warning("Unexpected format in %s", social_media_path)
+        return
+
+    logger.info("Loaded %d entries from legislators-social-media.yaml", len(social_media_data))
+
+    with Session(engine) as session:
+        legislator_map = _build_legislator_map(session)
+        existing_accounts = {
+            (account.legislator_id, account.platform)
+            for account in session.scalars(select(LegislatorSocialMedia)).all()
+        }
+        logger.info("Found %d existing social media accounts in DB", len(existing_accounts))
+
+        total_inserted = 0
+        total_updated = 0
+        for entry in social_media_data:
+            bioguide_id = entry.get("id", {}).get("bioguide")
+            if not bioguide_id:
+                continue
+
+            legislator_id = legislator_map.get(bioguide_id)
+            if legislator_id is None:
+                continue
+
+            social = entry.get("social", {})
+            for platform, url_template in SOCIAL_MEDIA_PLATFORMS.items():
+                account_name = social.get(platform)
+                if not account_name:
+                    continue
+
+                url = url_template.format(account=account_name) if url_template else None
+
+                if (legislator_id, platform) in existing_accounts:
+                    total_updated += 1
+                else:
+                    session.add(
+                        LegislatorSocialMedia(
+                            legislator_id=legislator_id,
+                            platform=platform,
+                            account_name=str(account_name),
+                            url=url,
+                            source="https://github.com/unitedstates/congress-legislators",
+                        )
+                    )
+                    existing_accounts.add((legislator_id, platform))
+                    total_inserted += 1
+
+        session.commit()
+    logger.info("Inserted %d new social media accounts, updated %d existing", total_inserted, total_updated)
+
+
+def _iter_voters(position_group: object) -> Iterator[dict]:
+    """Yield voter dicts from a vote position group (handles list, single dict, or string)."""
+    if isinstance(position_group, dict):
+        yield position_group
+    elif isinstance(position_group, list):
+        for voter in position_group:
+            if isinstance(voter, dict):
+                yield voter
+
+
+# ---------------------------------------------------------------------------
+# Bills
+# ---------------------------------------------------------------------------
+
+
+def ingest_bills(engine: Engine, congress_dirs: list[Path]) -> None:
+    """Load bill data.json files."""
+    with Session(engine) as session:
+        existing_bills = {(bill.congress, bill.bill_type, bill.number) for bill in session.scalars(select(Bill)).all()}
+        logger.info("Found %d existing bills in DB", len(existing_bills))
+
+        total_inserted = 0
+        batch: list[Bill] = []
+        for congress_dir in congress_dirs:
+            bills_dir = congress_dir / "bills"
+            if not bills_dir.is_dir():
+                continue
+            logger.info("Scanning bills from %s", congress_dir.name)
+            for bill_file in bills_dir.rglob("data.json"):
+                data = _read_json(bill_file)
+                if data is None:
+                    continue
+                bill = _parse_bill(data, existing_bills)
+                if bill is not None:
+                    batch.append(bill)
+                    if len(batch) >= BATCH_SIZE:
+                        total_inserted += _flush_batch(session, batch, "bills")
+
+        total_inserted += _flush_batch(session, batch, "bills")
+    logger.info("Inserted %d new bills total", total_inserted)
+
+
+def _parse_bill(data: dict, existing_bills: set[tuple[int, str, int]]) -> Bill | None:
+    """Parse a bill data.json dict into a Bill ORM object, skipping existing."""
+    raw_congress = data.get("congress")
+    bill_type = data.get("bill_type")
+    raw_number = data.get("number")
+    if raw_congress is None or bill_type is None or raw_number is None:
+        return None
+    congress = int(raw_congress)
+    number = int(raw_number)
+    if (congress, bill_type, number) in existing_bills:
+        return None
+
+    sponsor_bioguide = None
+    sponsor = data.get("sponsor")
+    if sponsor:
+        sponsor_bioguide = sponsor.get("bioguide_id")
+
+    return Bill(
+        congress=congress,
+        bill_type=bill_type,
+        number=number,
+        title=data.get("short_title") or data.get("official_title"),
+        title_short=data.get("short_title"),
+        official_title=data.get("official_title"),
+        status=data.get("status"),
+        status_at=data.get("status_at"),
+        sponsor_bioguide_id=sponsor_bioguide,
+        subjects_top_term=data.get("subjects_top_term"),
+    )
+
+
+# ---------------------------------------------------------------------------
+# Votes (and vote records)
+# ---------------------------------------------------------------------------
+
+
+def ingest_votes(engine: Engine, congress_dirs: list[Path]) -> None:
+    """Load vote data.json files with their vote records."""
+    with Session(engine) as session:
+        legislator_map = _build_legislator_map(session)
+        logger.info("Loaded %d legislators into lookup map", len(legislator_map))
+        bill_map = _build_bill_map(session)
+        logger.info("Loaded %d bills into lookup map", len(bill_map))
+        existing_votes = {
+            (vote.congress, vote.chamber, vote.session, vote.number) for vote in session.scalars(select(Vote)).all()
+        }
+        logger.info("Found %d existing votes in DB", len(existing_votes))
+
+        total_inserted = 0
+        batch: list[Vote] = []
+        for congress_dir in congress_dirs:
+            votes_dir = congress_dir / "votes"
+            if not votes_dir.is_dir():
+                continue
+            logger.info("Scanning votes from %s", congress_dir.name)
+            for vote_file in votes_dir.rglob("data.json"):
+                data = _read_json(vote_file)
+                if data is None:
+                    continue
+                vote = _parse_vote(data, legislator_map, bill_map, existing_votes)
+                if vote is not None:
+                    batch.append(vote)
+                    if len(batch) >= BATCH_SIZE:
+                        total_inserted += _flush_batch(session, batch, "votes")
+
+        total_inserted += _flush_batch(session, batch, "votes")
+    logger.info("Inserted %d new votes total", total_inserted)
+
+
+def _build_legislator_map(session: Session) -> dict[str, int]:
+    """Build a mapping of bioguide_id -> legislator.id."""
+    return {legislator.bioguide_id: legislator.id for legislator in session.scalars(select(Legislator)).all()}
+
+
+def _build_bill_map(session: Session) -> dict[tuple[int, str, int], int]:
+    """Build a mapping of (congress, bill_type, number) -> bill.id."""
+    return {(bill.congress, bill.bill_type, bill.number): bill.id for bill in session.scalars(select(Bill)).all()}
+
+
+def _parse_vote(
+    data: dict,
+    legislator_map: dict[str, int],
+    bill_map: dict[tuple[int, str, int], int],
+    existing_votes: set[tuple[int, str, int, int]],
+) -> Vote | None:
+    """Parse a vote data.json dict into a Vote ORM object with records."""
+    raw_congress = data.get("congress")
+    chamber = data.get("chamber")
+    raw_number = data.get("number")
+    vote_date = data.get("date")
+    if raw_congress is None or chamber is None or raw_number is None or vote_date is None:
+        return None
+
+    raw_session = data.get("session")
+    if raw_session is None:
+        return None
+
+    congress = int(raw_congress)
+    number = int(raw_number)
+    session_number = int(raw_session)
+
+    # Normalize chamber from "h"/"s" to "House"/"Senate"
+    chamber_normalized = {"h": "House", "s": "Senate"}.get(chamber, chamber)
+
+    if (congress, chamber_normalized, session_number, number) in existing_votes:
+        return None
+
+    # Resolve linked bill
+    bill_id = None
+    bill_ref = data.get("bill")
+    if bill_ref:
+        bill_key = (
+            int(bill_ref.get("congress", congress)),
+            bill_ref.get("type"),
+            int(bill_ref.get("number", 0)),
+        )
+        bill_id = bill_map.get(bill_key)
+
+    raw_votes = data.get("votes", {})
+    vote_counts = _count_votes(raw_votes)
+    vote_records = _build_vote_records(raw_votes, legislator_map)
+
+    return Vote(
+        congress=congress,
+        chamber=chamber_normalized,
+        session=session_number,
+        number=number,
+        vote_type=data.get("type"),
+        question=data.get("question"),
+        result=data.get("result"),
+        result_text=data.get("result_text"),
+        vote_date=vote_date[:10] if isinstance(vote_date, str) else vote_date,
+        bill_id=bill_id,
+        vote_records=vote_records,
+        **vote_counts,
+    )
+
+
+def _count_votes(raw_votes: dict) -> dict[str, int]:
+    """Count voters per position category, correctly handling dict and list formats."""
+    yea_count = 0
+    nay_count = 0
+    not_voting_count = 0
+    present_count = 0
+
+    for position, position_group in raw_votes.items():
+        voter_count = sum(1 for _ in _iter_voters(position_group))
+        if position in ("Yea", "Aye"):
+            yea_count += voter_count
+        elif position in ("Nay", "No"):
+            nay_count += voter_count
+        elif position == "Not Voting":
+            not_voting_count += voter_count
+        elif position == "Present":
+            present_count += voter_count
+
+    return {
+        "yea_count": yea_count,
+        "nay_count": nay_count,
+        "not_voting_count": not_voting_count,
+        "present_count": present_count,
+    }
+
+
+def _build_vote_records(raw_votes: dict, legislator_map: dict[str, int]) -> list[VoteRecord]:
+    """Build VoteRecord objects from raw vote data."""
+    records: list[VoteRecord] = []
+    for position, position_group in raw_votes.items():
+        for voter in _iter_voters(position_group):
+            bioguide_id = voter.get("id")
+            if not bioguide_id:
+                continue
+            legislator_id = legislator_map.get(bioguide_id)
+            if legislator_id is None:
+                continue
+            records.append(
+                VoteRecord(
+                    legislator_id=legislator_id,
+                    position=position,
+                )
+            )
+    return records
+
+
+# ---------------------------------------------------------------------------
+# Bill Text
+# ---------------------------------------------------------------------------
+
+
+def ingest_bill_text(engine: Engine, congress_dirs: list[Path]) -> None:
+    """Load bill text from text-versions directories."""
+    with Session(engine) as session:
+        bill_map = _build_bill_map(session)
+        logger.info("Loaded %d bills into lookup map", len(bill_map))
+        existing_bill_texts = {
+            (bill_text.bill_id, bill_text.version_code) for bill_text in session.scalars(select(BillText)).all()
+        }
+        logger.info("Found %d existing bill text versions in DB", len(existing_bill_texts))
+
+        total_inserted = 0
+        batch: list[BillText] = []
+        for congress_dir in congress_dirs:
+            logger.info("Scanning bill texts from %s", congress_dir.name)
+            for bill_text in _iter_bill_texts(congress_dir, bill_map, existing_bill_texts):
+                batch.append(bill_text)
+                if len(batch) >= BATCH_SIZE:
+                    total_inserted += _flush_batch(session, batch, "bill texts")
+
+        total_inserted += _flush_batch(session, batch, "bill texts")
+    logger.info("Inserted %d new bill text versions total", total_inserted)
+
+
+def _iter_bill_texts(
+    congress_dir: Path,
+    bill_map: dict[tuple[int, str, int], int],
+    existing_bill_texts: set[tuple[int, str]],
+) -> Iterator[BillText]:
+    """Yield BillText objects for a single congress directory, skipping existing."""
+    bills_dir = congress_dir / "bills"
+    if not bills_dir.is_dir():
+        return
+
+    for bill_dir in bills_dir.rglob("text-versions"):
+        if not bill_dir.is_dir():
+            continue
+        bill_key = _bill_key_from_dir(bill_dir.parent, congress_dir)
+        if bill_key is None:
+            continue
+        bill_id = bill_map.get(bill_key)
+        if bill_id is None:
+            continue
+
+        for version_dir in sorted(bill_dir.iterdir()):
+            if not version_dir.is_dir():
+                continue
+            if (bill_id, version_dir.name) in existing_bill_texts:
+                continue
+            text_content = _read_bill_text(version_dir)
+            version_data = _read_json(version_dir / "data.json")
+            yield BillText(
+                bill_id=bill_id,
+                version_code=version_dir.name,
+                version_name=version_data.get("version_name") if version_data else None,
+                date=version_data.get("issued_on") if version_data else None,
+                text_content=text_content,
+            )
+
+
+def _bill_key_from_dir(bill_dir: Path, congress_dir: Path) -> tuple[int, str, int] | None:
+    """Extract (congress, bill_type, number) from directory structure."""
+    congress = int(congress_dir.name)
+    bill_type = bill_dir.parent.name
+    name = bill_dir.name
+    # Directory name is like "hr3590" — strip the type prefix to get the number
+    number_str = name[len(bill_type) :]
+    if not number_str.isdigit():
+        return None
+    return (congress, bill_type, int(number_str))
+
+
+def _read_bill_text(version_dir: Path) -> str | None:
+    """Read bill text from a version directory, preferring .txt over .xml."""
+    for extension in ("txt", "htm", "html", "xml"):
+        candidates = list(version_dir.glob(f"document.{extension}"))
+        if not candidates:
+            candidates = list(version_dir.glob(f"*.{extension}"))
+        if candidates:
+            try:
+                return candidates[0].read_text(encoding="utf-8")
+            except Exception:
+                logger.exception("Failed to read %s", candidates[0])
+    return None
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def _read_json(path: Path) -> dict | None:
+    """Read and parse a JSON file, returning None on failure."""
+    try:
+        return orjson.loads(path.read_bytes())
+    except FileNotFoundError:
+        return None
+    except Exception:
+        logger.exception("Failed to parse %s", path)
+        return None
+
+
+if __name__ == "__main__":
+    app()

python/data_science/ingest_posts.py

@@ -0,0 +1,247 @@
+"""Ingestion pipeline for loading JSONL post files into the weekly-partitioned posts table.
+
+Usage:
+    ingest-posts /path/to/files/
+    ingest-posts /path/to/single_file.jsonl
+    ingest-posts /data/dir/ --workers 4 --batch-size 5000
+"""
+
+from __future__ import annotations
+
+import logging
+from datetime import UTC, datetime
+from pathlib import Path  # noqa: TC003 this is needed for typer
+from typing import TYPE_CHECKING, Annotated
+
+import orjson
+import psycopg
+import typer
+
+from python.common import configure_logger
+from python.orm.common import get_connection_info
+from python.parallelize import parallelize_process
+
+if TYPE_CHECKING:
+    from collections.abc import Iterator
+
+logger = logging.getLogger(__name__)
+
+
+app = typer.Typer(help="Ingest JSONL post files into the partitioned posts table.")
+
+
+@app.command()
+def main(
+    path: Annotated[Path, typer.Argument(help="Directory containing JSONL files, or a single JSONL file")],
+    batch_size: Annotated[int, typer.Option(help="Rows per INSERT batch")] = 10000,
+    workers: Annotated[int, typer.Option(help="Parallel workers for multi-file ingestion")] = 4,
+    pattern: Annotated[str, typer.Option(help="Glob pattern for JSONL files")] = "*.jsonl",
+) -> None:
+    """Ingest JSONL post files into the weekly-partitioned posts table."""
+    configure_logger(level="INFO")
+
+    logger.info("starting ingest-posts")
+    logger.info("path=%s batch_size=%d workers=%d pattern=%s", path, batch_size, workers, pattern)
+    if path.is_file():
+        ingest_file(path, batch_size=batch_size)
+    elif path.is_dir():
+        ingest_directory(path, batch_size=batch_size, max_workers=workers, pattern=pattern)
+    else:
+        typer.echo(f"Path does not exist: {path}", err=True)
+        raise typer.Exit(code=1)
+
+    logger.info("ingest-posts done")
+
+
+def ingest_directory(
+    directory: Path,
+    *,
+    batch_size: int,
+    max_workers: int,
+    pattern: str = "*.jsonl",
+) -> None:
+    """Ingest all JSONL files in a directory using parallel workers."""
+    files = sorted(directory.glob(pattern))
+    if not files:
+        logger.warning("No JSONL files found in %s", directory)
+        return
+
+    logger.info("Found %d JSONL files to ingest", len(files))
+
+    kwargs_list = [{"path": fp, "batch_size": batch_size} for fp in files]
+    parallelize_process(ingest_file, kwargs_list, max_workers=max_workers)
+
+
+SCHEMA = "main"
+
+COLUMNS = (
+    "post_id",
+    "user_id",
+    "instance",
+    "date",
+    "text",
+    "langs",
+    "like_count",
+    "reply_count",
+    "repost_count",
+    "reply_to",
+    "replied_author",
+    "thread_root",
+    "thread_root_author",
+    "repost_from",
+    "reposted_author",
+    "quotes",
+    "quoted_author",
+    "labels",
+    "sent_label",
+    "sent_score",
+)
+
+INSERT_FROM_STAGING = f"""
+    INSERT INTO {SCHEMA}.posts ({", ".join(COLUMNS)})
+    SELECT {", ".join(COLUMNS)} FROM pg_temp.staging
+    ON CONFLICT (post_id, date) DO NOTHING
+"""  # noqa: S608
+
+FAILED_INSERT = f"""
+    INSERT INTO {SCHEMA}.failed_ingestion (raw_line, error)
+    VALUES (%(raw_line)s, %(error)s)
+"""  # noqa: S608
+
+
+def get_psycopg_connection() -> psycopg.Connection:
+    """Create a raw psycopg3 connection from environment variables."""
+    database, host, port, username, password = get_connection_info("DATA_SCIENCE_DEV")
+    return psycopg.connect(
+        dbname=database,
+        host=host,
+        port=int(port),
+        user=username,
+        password=password,
+        autocommit=False,
+    )
+
+
+def ingest_file(path: Path, *, batch_size: int) -> None:
+    """Ingest a single JSONL file into the posts table."""
+    log_trigger = max(100_000 // batch_size, 1)
+    failed_lines: list[dict] = []
+    try:
+        with get_psycopg_connection() as connection:
+            for index, batch in enumerate(read_jsonl_batches(path, batch_size, failed_lines), 1):
+                ingest_batch(connection, batch)
+                if index % log_trigger == 0:
+                    logger.info("Ingested %d batches (%d rows) from %s", index, index * batch_size, path)
+
+            if failed_lines:
+                logger.warning("Recording %d malformed lines from %s", len(failed_lines), path.name)
+                with connection.cursor() as cursor:
+                    cursor.executemany(FAILED_INSERT, failed_lines)
+                connection.commit()
+    except Exception:
+        logger.exception("Failed to ingest file: %s", path)
+        raise
+
+
+def ingest_batch(connection: psycopg.Connection, batch: list[dict]) -> None:
+    """COPY batch into a temp staging table, then INSERT ... ON CONFLICT into posts."""
+    if not batch:
+        return
+
+    try:
+        with connection.cursor() as cursor:
+            cursor.execute(f"""
+                CREATE TEMP TABLE IF NOT EXISTS staging
+                (LIKE {SCHEMA}.posts INCLUDING DEFAULTS)
+                ON COMMIT DELETE ROWS
+            """)
+            cursor.execute("TRUNCATE pg_temp.staging")
+
+            with cursor.copy(f"COPY pg_temp.staging ({', '.join(COLUMNS)}) FROM STDIN") as copy:
+                for row in batch:
+                    copy.write_row(tuple(row.get(column) for column in COLUMNS))
+
+            cursor.execute(INSERT_FROM_STAGING)
+        connection.commit()
+    except Exception as error:
+        connection.rollback()
+
+        if len(batch) == 1:
+            logger.exception("Skipping bad row post_id=%s", batch[0].get("post_id"))
+            with connection.cursor() as cursor:
+                cursor.execute(
+                    FAILED_INSERT,
+                    {
+                        "raw_line": orjson.dumps(batch[0], default=str).decode(),
+                        "error": str(error),
+                    },
+                )
+            connection.commit()
+            return
+
+        midpoint = len(batch) // 2
+        ingest_batch(connection, batch[:midpoint])
+        ingest_batch(connection, batch[midpoint:])
+
+
+def read_jsonl_batches(file_path: Path, batch_size: int, failed_lines: list[dict]) -> Iterator[list[dict]]:
+    """Stream a JSONL file and yield batches of transformed rows."""
+    batch: list[dict] = []
+    with file_path.open("r", encoding="utf-8") as handle:
+        for raw_line in handle:
+            line = raw_line.strip()
+            if not line:
+                continue
+            batch.extend(parse_line(line, file_path, failed_lines))
+            if len(batch) >= batch_size:
+                yield batch
+                batch = []
+    if batch:
+        yield batch
+
+
+def parse_line(line: str, file_path: Path, failed_lines: list[dict]) -> Iterator[dict]:
+    """Parse a JSONL line, handling concatenated JSON objects."""
+    try:
+        yield transform_row(orjson.loads(line))
+    except orjson.JSONDecodeError:
+        if "}{" not in line:
+            logger.warning("Skipping malformed line in %s: %s", file_path.name, line[:120])
+            failed_lines.append({"raw_line": line, "error": "malformed JSON"})
+            return
+        fragments = line.replace("}{", "}\n{").split("\n")
+        for fragment in fragments:
+            try:
+                yield transform_row(orjson.loads(fragment))
+            except (orjson.JSONDecodeError, KeyError, ValueError) as error:
+                logger.warning("Skipping malformed fragment in %s: %s", file_path.name, fragment[:120])
+                failed_lines.append({"raw_line": fragment, "error": str(error)})
+    except Exception as error:
+        logger.exception("Skipping bad row in %s: %s", file_path.name, line[:120])
+        failed_lines.append({"raw_line": line, "error": str(error)})
+
+
+def transform_row(raw: dict) -> dict:
+    """Transform a raw JSONL row into a dict matching the Posts table columns."""
+    raw["date"] = parse_date(raw["date"])
+    if raw.get("langs") is not None:
+        raw["langs"] = orjson.dumps(raw["langs"])
+    if raw.get("text") is not None:
+        raw["text"] = raw["text"].replace("\x00", "")
+    return raw
+
+
+def parse_date(raw_date: int) -> datetime:
+    """Parse compact YYYYMMDDHHmm integer into a naive datetime (input is UTC by spec)."""
+    return datetime(
+        raw_date // 100000000,
+        (raw_date // 1000000) % 100,
+        (raw_date // 10000) % 100,
+        (raw_date // 100) % 100,
+        raw_date % 100,
+        tzinfo=UTC,
+    )
+
+
+if __name__ == "__main__":
+    app()

python/database_cli.py

@@ -83,6 +83,20 @@ DATABASES: dict[str, DatabaseConfig] = {
         base_class_name="VanInventoryBase",
         models_module="python.orm.van_inventory.models",
     ),
+    "signal_bot": DatabaseConfig(
+        env_prefix="SIGNALBOT",
+        version_location="python/alembic/signal_bot/versions",
+        base_module="python.orm.signal_bot.base",
+        base_class_name="SignalBotBase",
+        models_module="python.orm.signal_bot.models",
+    ),
+    "data_science_dev": DatabaseConfig(
+        env_prefix="DATA_SCIENCE_DEV",
+        version_location="python/alembic/data_science_dev/versions",
+        base_module="python.orm.data_science_dev.base",
+        base_class_name="DataScienceDevBase",
+        models_module="python.orm.data_science_dev.models",
+    ),
 }
 
 

python/gitea.py

@@ -1,347 +0,0 @@
-"""Small Gitea API client for repository automation."""
-
-from __future__ import annotations
-
-from dataclasses import dataclass
-from typing import Self
-from urllib.parse import quote
-
-import httpx
-
-DEFAULT_PAGE_SIZE = 100
-EXPECTED_NO_CONTENT = 204
-EXPECTED_CREATED = 201
-EXPECTED_OK = 200
-
-
-@dataclass(frozen=True)
-class CreatedIssue:
-    """Issue data returned by Gitea."""
-
-    number: int | None
-    html_url: str | None
-    title: str
-
-
-@dataclass(frozen=True)
-class PullRequest:
-    """Pull request data returned by Gitea."""
-
-    number: int
-    title: str
-    html_url: str | None
-    labels: tuple[str, ...]
-    head_branch: str | None
-    base_branch: str | None
-
-
-@dataclass(frozen=True)
-class WorkflowJob:
-    """Workflow job data returned by Gitea Actions."""
-
-    id: int
-    name: str
-    run_id: int | None
-    status: str | None
-    conclusion: str | None
-
-
-class GiteaError(RuntimeError):
-    """Raised when Gitea rejects an API request."""
-
-
-def split_repo_name(repo: str) -> tuple[str, str]:
-    """Split an owner/repo string into its parts."""
-    owner, separator, repo_name = repo.partition("/")
-    if not separator or not owner or not repo_name:
-        msg = f"Invalid repository name: {repo}"
-        raise ValueError(msg)
-    return owner, repo_name
-
-
-class GiteaClient:
-    """HTTP client for the subset of Gitea APIs used in this repository."""
-
-    def __init__(
-        self,
-        *,
-        base_url: str,
-        token: str,
-        timeout: int = 30,
-        transport: httpx.BaseTransport | None = None,
-    ) -> None:
-        """Initialize the Gitea client."""
-        self._client = httpx.Client(
-            base_url=base_url.rstrip("/"),
-            timeout=timeout,
-            headers={"Authorization": f"token {token}"},
-            transport=transport,
-        )
-
-    def create_issue(
-        self,
-        *,
-        owner: str,
-        repo: str,
-        title: str,
-        body: str,
-        labels: list[int] | None = None,
-    ) -> CreatedIssue:
-        """Create a Gitea issue."""
-        payload: dict[str, object] = {"title": title, "body": body, "labels": labels or []}
-        response = self._request(
-            "POST",
-            f"/api/v1/repos/{owner}/{repo}/issues",
-            expected_statuses={EXPECTED_CREATED},
-            json=payload,
-        )
-        data = response.json()
-        return CreatedIssue(
-            number=_optional_int(data.get("number")),
-            html_url=_optional_str(data.get("html_url")),
-            title=str(data.get("title", title)),
-        )
-
-    def resolve_label_ids(self, *, owner: str, repo: str, labels: list[str]) -> list[int]:
-        """Resolve label names to Gitea label IDs."""
-        if not labels:
-            return []
-
-        available_labels: dict[str, int] = {}
-        page = 1
-        while True:
-            response = self._request(
-                "GET",
-                f"/api/v1/repos/{owner}/{repo}/labels",
-                params={"page": page, "limit": DEFAULT_PAGE_SIZE},
-            )
-            batch = response.json()
-            if not batch:
-                break
-            for label in batch:
-                label_name = str(label.get("name", ""))
-                label_id = _optional_int(label.get("id"))
-                if label_name and label_id is not None:
-                    available_labels[label_name] = label_id
-            if len(batch) < DEFAULT_PAGE_SIZE:
-                break
-            page += 1
-
-        missing = [label for label in labels if label not in available_labels]
-        if missing:
-            missing_names = ", ".join(sorted(missing))
-            msg = f"Missing Gitea labels: {missing_names}"
-            raise GiteaError(msg)
-
-        return [available_labels[label] for label in labels]
-
-    def list_open_pull_requests(
-        self,
-        *,
-        owner: str,
-        repo: str,
-        labels: list[str] | None = None,
-        head: str | None = None,
-    ) -> list[PullRequest]:
-        """List open pull requests for a repository."""
-        expected_labels = set(labels or [])
-        pull_requests: list[PullRequest] = []
-        page = 1
-        while True:
-            response = self._request(
-                "GET",
-                f"/api/v1/repos/{owner}/{repo}/pulls",
-                params={"state": "open", "page": page, "limit": DEFAULT_PAGE_SIZE},
-            )
-            batch = response.json()
-            if not batch:
-                break
-
-            for item in batch:
-                pull_request = _pull_request_from_api(item)
-                if head and pull_request.head_branch != head:
-                    continue
-                if expected_labels and not expected_labels.issubset(set(pull_request.labels)):
-                    continue
-                pull_requests.append(pull_request)
-
-            if len(batch) < DEFAULT_PAGE_SIZE:
-                break
-            page += 1
-
-        return pull_requests
-
-    def create_pull_request(
-        self,
-        *,
-        owner: str,
-        repo: str,
-        title: str,
-        body: str,
-        head: str,
-        base: str,
-        labels: list[str] | None = None,
-    ) -> PullRequest:
-        """Create a pull request."""
-        payload: dict[str, object] = {
-            "title": title,
-            "body": body,
-            "head": head,
-            "base": base,
-        }
-        if labels:
-            payload["labels"] = self.resolve_label_ids(owner=owner, repo=repo, labels=labels)
-
-        response = self._request(
-            "POST",
-            f"/api/v1/repos/{owner}/{repo}/pulls",
-            expected_statuses={EXPECTED_CREATED},
-            json=payload,
-        )
-        return _pull_request_from_api(response.json())
-
-    def merge_pull_request(
-        self,
-        *,
-        owner: str,
-        repo: str,
-        number: int,
-        merge_method: str = "rebase",
-        head_commit_id: str | None = None,
-        delete_branch_after_merge: bool = False,
-    ) -> None:
-        """Merge a pull request."""
-        payload: dict[str, object] = {
-            "Do": merge_method,
-            "delete_branch_after_merge": delete_branch_after_merge,
-        }
-        if head_commit_id:
-            payload["head_commit_id"] = head_commit_id
-
-        self._request(
-            "POST",
-            f"/api/v1/repos/{owner}/{repo}/pulls/{number}/merge",
-            json=payload,
-        )
-
-    def dispatch_workflow(self, *, owner: str, repo: str, workflow_id: str, ref: str) -> None:
-        """Trigger a workflow_dispatch run."""
-        workflow_path = quote(workflow_id, safe="")
-        self._request(
-            "POST",
-            f"/api/v1/repos/{owner}/{repo}/actions/workflows/{workflow_path}/dispatches",
-            expected_statuses={EXPECTED_OK, EXPECTED_NO_CONTENT},
-            json={"ref": ref},
-        )
-
-    def list_run_jobs(self, *, owner: str, repo: str, run_id: str | int) -> list[WorkflowJob]:
-        """List workflow jobs for a specific run."""
-        jobs: list[WorkflowJob] = []
-        page = 1
-        while True:
-            response = self._request(
-                "GET",
-                f"/api/v1/repos/{owner}/{repo}/actions/jobs",
-                params={"page": page, "limit": DEFAULT_PAGE_SIZE},
-            )
-            payload = response.json()
-            batch = payload.get("jobs", [])
-            if not batch:
-                break
-
-            for item in batch:
-                if str(item.get("run_id")) != str(run_id):
-                    continue
-                jobs.append(_workflow_job_from_api(item))
-
-            if len(batch) < DEFAULT_PAGE_SIZE:
-                break
-            page += 1
-
-        return jobs
-
-    def download_job_logs(self, *, owner: str, repo: str, job_id: int) -> str:
-        """Download logs for a workflow job."""
-        response = self._request(
-            "GET",
-            f"/api/v1/repos/{owner}/{repo}/actions/jobs/{job_id}/logs",
-        )
-        return response.text
-
-    def close(self) -> None:
-        """Close the underlying HTTP client."""
-        self._client.close()
-
-    def __enter__(self) -> Self:
-        """Enter the context manager."""
-        return self
-
-    def __exit__(self, *args: object) -> None:
-        """Close the HTTP client."""
-        self.close()
-
-    def _request(
-        self,
-        method: str,
-        path: str,
-        *,
-        expected_statuses: set[int] | None = None,
-        **kwargs: object,
-    ) -> httpx.Response:
-        """Send an HTTP request and validate the response status."""
-        response = self._client.request(method, path, **kwargs)
-        statuses = expected_statuses or {EXPECTED_OK}
-        if response.status_code not in statuses:
-            msg = f"Gitea request failed ({response.status_code}): {response.text}"
-            raise GiteaError(msg)
-        return response
-
-
-def _pull_request_from_api(data: dict[str, object]) -> PullRequest:
-    """Convert Gitea API pull-request data into a dataclass."""
-    number = _optional_int(data.get("number")) or _optional_int(data.get("index"))
-    if number is None:
-        msg = "Gitea pull request payload is missing a number"
-        raise GiteaError(msg)
-
-    labels = tuple(str(label.get("name", "")) for label in data.get("labels", []))
-    head = data.get("head", {})
-    base = data.get("base", {})
-    return PullRequest(
-        number=number,
-        title=str(data.get("title", "")),
-        html_url=_optional_str(data.get("html_url")),
-        labels=tuple(label for label in labels if label),
-        head_branch=_optional_str(head.get("ref")) or _optional_str(data.get("head_branch")),
-        base_branch=_optional_str(base.get("ref")) or _optional_str(data.get("base_branch")),
-    )
-
-
-def _workflow_job_from_api(data: dict[str, object]) -> WorkflowJob:
-    """Convert Gitea API workflow-job data into a dataclass."""
-    job_id = _optional_int(data.get("id"))
-    if job_id is None:
-        msg = "Gitea workflow job payload is missing an ID"
-        raise GiteaError(msg)
-
-    return WorkflowJob(
-        id=job_id,
-        name=str(data.get("name", "")),
-        run_id=_optional_int(data.get("run_id")),
-        status=_optional_str(data.get("status")),
-        conclusion=_optional_str(data.get("conclusion")),
-    )
-
-
-def _optional_int(value: object) -> int | None:
-    """Convert an API value to an integer when present."""
-    if value is None:
-        return None
-    return int(value)
-
-
-def _optional_str(value: object) -> str | None:
-    """Convert an API value to a string when present."""
-    if value is None:
-        return None
-    return str(value)

python/gitea_flake_lock.py

@@ -1,148 +0,0 @@
-"""Automation helpers for flake.lock pull requests on Gitea."""
-
-from __future__ import annotations
-
-import subprocess
-from os import getenv
-from typing import Annotated
-
-import typer
-
-from python.gitea import GiteaClient, PullRequest, split_repo_name
-
-DEFAULT_BASE_BRANCH = "main"
-DEFAULT_BRANCH = "automation/update-flake-lock"
-DEFAULT_GITEA_URL = "https://gitea.tmmworkshop.com"
-PR_LABELS = ["dependencies", "automated", "flake_lock_update"]
-PR_CHECK_WORKFLOWS = ["build_systems.yml", "treefmt.yml", "pytest.yml"]
-PR_TITLE = "Update flake.lock"
-PR_BODY = "Automated flake.lock update."
-
-app = typer.Typer(add_completion=False)
-
-
-def run_cmd(cmd: list[str], *, check: bool = True) -> subprocess.CompletedProcess[str]:
-    """Run a subprocess command."""
-    return subprocess.run(cmd, capture_output=True, text=True, check=check)
-
-
-def ensure_flake_lock_pull_request(
-    client: GiteaClient,
-    *,
-    owner: str,
-    repo: str,
-    branch: str,
-    base: str,
-) -> PullRequest:
-    """Return an existing flake.lock PR for the branch or create one."""
-    pull_requests = client.list_open_pull_requests(owner=owner, repo=repo, head=branch)
-    if pull_requests:
-        return pull_requests[0]
-
-    return client.create_pull_request(
-        owner=owner,
-        repo=repo,
-        title=PR_TITLE,
-        body=PR_BODY,
-        head=branch,
-        base=base,
-        labels=PR_LABELS,
-    )
-
-
-def find_flake_lock_pull_request(client: GiteaClient, *, owner: str, repo: str) -> PullRequest | None:
-    """Find the first open flake.lock pull request."""
-    pull_requests = client.list_open_pull_requests(owner=owner, repo=repo, labels=["flake_lock_update"])
-    if not pull_requests:
-        return None
-    return pull_requests[0]
-
-
-def dispatch_pull_request_checks(client: GiteaClient, *, owner: str, repo: str, branch: str) -> None:
-    """Dispatch the workflows that normally run for pull requests."""
-    for workflow in PR_CHECK_WORKFLOWS:
-        client.dispatch_workflow(owner=owner, repo=repo, workflow_id=workflow, ref=branch)
-
-
-def has_worktree_changes() -> bool:
-    """Return whether `flake.lock` has worktree changes."""
-    result = run_cmd(["git", "diff", "--quiet", "--", "flake.lock"], check=False)
-    return result.returncode != 0
-
-
-def commit_flake_lock_update(*, branch: str) -> None:
-    """Commit the updated lock file to the automation branch."""
-    run_cmd(["git", "config", "user.name", "gitea-actions[bot]"])
-    run_cmd(["git", "config", "user.email", "gitea-actions@tmmworkshop.com"])
-    run_cmd(["git", "checkout", "-B", branch])
-    run_cmd(["git", "add", "flake.lock"])
-    run_cmd(["git", "commit", "-m", "chore: update flake.lock"])
-
-
-def push_branch(*, branch: str) -> None:
-    """Push the automation branch to origin."""
-    run_cmd(["git", "push", "origin", f"HEAD:{branch}", "--force"])
-
-
-def _required_gitea_token() -> str:
-    """Read the required Gitea token from the environment."""
-    token = getenv("GITEA_TOKEN")
-    if token:
-        return token
-
-    msg = "GITEA_TOKEN environment variable is required"
-    raise RuntimeError(msg)
-
-
-@app.command()
-def update(
-    repo: Annotated[str, typer.Option("--repo", help="Gitea repository in owner/repo form")],
-    base: Annotated[str, typer.Option("--base", help="Base branch")] = DEFAULT_BASE_BRANCH,
-    branch: Annotated[str, typer.Option("--branch", help="Automation branch")] = DEFAULT_BRANCH,
-) -> None:
-    """Commit flake.lock changes and ensure a pull request exists."""
-    if not has_worktree_changes():
-        typer.echo("No flake.lock changes detected")
-        return
-
-    commit_flake_lock_update(branch=branch)
-    push_branch(branch=branch)
-
-    owner, repo_name = split_repo_name(repo)
-    with GiteaClient(
-        base_url=getenv("GITEA_URL", DEFAULT_GITEA_URL),
-        token=_required_gitea_token(),
-    ) as client:
-        pull_request = ensure_flake_lock_pull_request(
-            client,
-            owner=owner,
-            repo=repo_name,
-            branch=branch,
-            base=base,
-        )
-        # We can remove this if Gitea fixes the following issue:
-        # https://github.com/go-gitea/gitea/issues/33963
-        dispatch_pull_request_checks(client, owner=owner, repo=repo_name, branch=branch)
-    typer.echo(pull_request.html_url or f"Pull request #{pull_request.number}")
-
-
-@app.command()
-def merge(
-    repo: Annotated[str, typer.Option("--repo", help="Gitea repository in owner/repo form")],
-) -> None:
-    """Merge the first open flake.lock pull request."""
-    owner, repo_name = split_repo_name(repo)
-    with GiteaClient(
-        base_url=getenv("GITEA_URL", DEFAULT_GITEA_URL),
-        token=_required_gitea_token(),
-    ) as client:
-        pull_request = find_flake_lock_pull_request(client, owner=owner, repo=repo_name)
-        if not pull_request:
-            typer.echo("No open PR found with label flake_lock_update")
-            return
-        client.merge_pull_request(owner=owner, repo=repo_name, number=pull_request.number, merge_method="rebase")
-    typer.echo(f"Merged PR #{pull_request.number}")
-
-
-if __name__ == "__main__":
-    app()

python/orm/__init__.py

@@ -1,9 +1,13 @@
 """ORM package exports."""
 
+from python.orm.data_science_dev.base import DataScienceDevBase
 from python.orm.richie.base import RichieBase
+from python.orm.signal_bot.base import SignalBotBase
 from python.orm.van_inventory.base import VanInventoryBase
 
 __all__ = [
+    "DataScienceDevBase",
     "RichieBase",
+    "SignalBotBase",
     "VanInventoryBase",
 ]

python/orm/data_science_dev/__init__.py

@@ -0,0 +1,11 @@
+"""Data science dev database ORM exports."""
+
+from __future__ import annotations
+
+from python.orm.data_science_dev.base import DataScienceDevBase, DataScienceDevTableBase, DataScienceDevTableBaseBig
+
+__all__ = [
+    "DataScienceDevBase",
+    "DataScienceDevTableBase",
+    "DataScienceDevTableBaseBig",
+]

python/orm/data_science_dev/base.py

@@ -0,0 +1,52 @@
+"""Data science dev database ORM base."""
+
+from __future__ import annotations
+
+from datetime import datetime
+
+from sqlalchemy import BigInteger, DateTime, MetaData, func
+from sqlalchemy.ext.declarative import AbstractConcreteBase
+from sqlalchemy.orm import DeclarativeBase, Mapped, mapped_column
+
+from python.orm.common import NAMING_CONVENTION
+
+
+class DataScienceDevBase(DeclarativeBase):
+    """Base class for data_science_dev database ORM models."""
+
+    schema_name = "main"
+
+    metadata = MetaData(
+        schema=schema_name,
+        naming_convention=NAMING_CONVENTION,
+    )
+
+
+class _TableMixin:
+    """Shared timestamp columns for all table bases."""
+
+    created: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True),
+        server_default=func.now(),
+    )
+    updated: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True),
+        server_default=func.now(),
+        onupdate=func.now(),
+    )
+
+
+class DataScienceDevTableBase(_TableMixin, AbstractConcreteBase, DataScienceDevBase):
+    """Table with Integer primary key."""
+
+    __abstract__ = True
+
+    id: Mapped[int] = mapped_column(primary_key=True)
+
+
+class DataScienceDevTableBaseBig(_TableMixin, AbstractConcreteBase, DataScienceDevBase):
+    """Table with BigInteger primary key."""
+
+    __abstract__ = True
+
+    id: Mapped[int] = mapped_column(BigInteger, primary_key=True)

python/orm/data_science_dev/congress/__init__.py

@@ -0,0 +1,14 @@
+"""init."""
+
+from python.orm.data_science_dev.congress.bill import Bill, BillText
+from python.orm.data_science_dev.congress.legislator import Legislator, LegislatorSocialMedia
+from python.orm.data_science_dev.congress.vote import Vote, VoteRecord
+
+__all__ = [
+    "Bill",
+    "BillText",
+    "Legislator",
+    "LegislatorSocialMedia",
+    "Vote",
+    "VoteRecord",
+]

python/orm/data_science_dev/congress/bill.py

@@ -0,0 +1,66 @@
+"""Bill model - legislation introduced in Congress."""
+
+from __future__ import annotations
+
+from datetime import date
+from typing import TYPE_CHECKING
+
+from sqlalchemy import ForeignKey, Index, UniqueConstraint
+from sqlalchemy.orm import Mapped, mapped_column, relationship
+
+from python.orm.data_science_dev.base import DataScienceDevTableBase
+
+if TYPE_CHECKING:
+    from python.orm.data_science_dev.congress.vote import Vote
+
+
+class Bill(DataScienceDevTableBase):
+    """Legislation with congress number, type, titles, status, and sponsor."""
+
+    __tablename__ = "bill"
+
+    congress: Mapped[int]
+    bill_type: Mapped[str]
+    number: Mapped[int]
+
+    title: Mapped[str | None]
+    title_short: Mapped[str | None]
+    official_title: Mapped[str | None]
+
+    status: Mapped[str | None]
+    status_at: Mapped[date | None]
+
+    sponsor_bioguide_id: Mapped[str | None]
+
+    subjects_top_term: Mapped[str | None]
+
+    votes: Mapped[list[Vote]] = relationship(
+        "Vote",
+        back_populates="bill",
+    )
+    bill_texts: Mapped[list[BillText]] = relationship(
+        "BillText",
+        back_populates="bill",
+        cascade="all, delete-orphan",
+    )
+
+    __table_args__ = (
+        UniqueConstraint("congress", "bill_type", "number", name="uq_bill_congress_type_number"),
+        Index("ix_bill_congress", "congress"),
+    )
+
+
+class BillText(DataScienceDevTableBase):
+    """Stores different text versions of a bill (introduced, enrolled, etc.)."""
+
+    __tablename__ = "bill_text"
+
+    bill_id: Mapped[int] = mapped_column(ForeignKey("main.bill.id", ondelete="CASCADE"))
+    version_code: Mapped[str]
+    version_name: Mapped[str | None]
+    text_content: Mapped[str | None]
+    date: Mapped[date | None]
+
+    bill: Mapped[Bill] = relationship("Bill", back_populates="bill_texts")
+
+    __table_args__ = (UniqueConstraint("bill_id", "version_code", name="uq_bill_text_bill_id_version_code"),)

python/orm/data_science_dev/congress/legislator.py

@@ -0,0 +1,66 @@
+"""Legislator model - members of Congress."""
+
+from __future__ import annotations
+
+from datetime import date
+from typing import TYPE_CHECKING
+
+from sqlalchemy import ForeignKey, Text
+from sqlalchemy.orm import Mapped, mapped_column, relationship
+
+from python.orm.data_science_dev.base import DataScienceDevTableBase
+
+if TYPE_CHECKING:
+    from python.orm.data_science_dev.congress.vote import VoteRecord
+
+
+class Legislator(DataScienceDevTableBase):
+    """Members of Congress with identification and current term info."""
+
+    __tablename__ = "legislator"
+
+    bioguide_id: Mapped[str] = mapped_column(Text, unique=True, index=True)
+
+    thomas_id: Mapped[str | None]
+    lis_id: Mapped[str | None]
+    govtrack_id: Mapped[int | None]
+    opensecrets_id: Mapped[str | None]
+    fec_ids: Mapped[str | None]
+
+    first_name: Mapped[str]
+    last_name: Mapped[str]
+    official_full_name: Mapped[str | None]
+    nickname: Mapped[str | None]
+
+    birthday: Mapped[date | None]
+    gender: Mapped[str | None]
+
+    current_party: Mapped[str | None]
+    current_state: Mapped[str | None]
+    current_district: Mapped[int | None]
+    current_chamber: Mapped[str | None]
+
+    social_media_accounts: Mapped[list[LegislatorSocialMedia]] = relationship(
+        "LegislatorSocialMedia",
+        back_populates="legislator",
+        cascade="all, delete-orphan",
+    )
+    vote_records: Mapped[list[VoteRecord]] = relationship(
+        "VoteRecord",
+        back_populates="legislator",
+        cascade="all, delete-orphan",
+    )
+
+
+class LegislatorSocialMedia(DataScienceDevTableBase):
+    """Social media account linked to a legislator."""
+
+    __tablename__ = "legislator_social_media"
+
+    legislator_id: Mapped[int] = mapped_column(ForeignKey("main.legislator.id"))
+    platform: Mapped[str]
+    account_name: Mapped[str]
+    url: Mapped[str | None]
+    source: Mapped[str]
+
+    legislator: Mapped[Legislator] = relationship(back_populates="social_media_accounts")

python/orm/data_science_dev/congress/vote.py

@@ -0,0 +1,79 @@
+"""Vote model - roll call votes in Congress."""
+
+from __future__ import annotations
+
+from datetime import date
+from typing import TYPE_CHECKING
+
+from sqlalchemy import ForeignKey, Index, UniqueConstraint
+from sqlalchemy.orm import Mapped, mapped_column, relationship
+
+from python.orm.data_science_dev.base import DataScienceDevBase, DataScienceDevTableBase
+
+if TYPE_CHECKING:
+    from python.orm.data_science_dev.congress.bill import Bill
+    from python.orm.data_science_dev.congress.legislator import Legislator
+    from python.orm.data_science_dev.congress.vote import Vote
+
+
+class VoteRecord(DataScienceDevBase):
+    """Links a vote to a legislator with their position (Yea, Nay, etc.)."""
+
+    __tablename__ = "vote_record"
+
+    vote_id: Mapped[int] = mapped_column(
+        ForeignKey("main.vote.id", ondelete="CASCADE"),
+        primary_key=True,
+    )
+    legislator_id: Mapped[int] = mapped_column(
+        ForeignKey("main.legislator.id", ondelete="CASCADE"),
+        primary_key=True,
+    )
+    position: Mapped[str]
+
+    vote: Mapped[Vote] = relationship("Vote", back_populates="vote_records")
+    legislator: Mapped[Legislator] = relationship("Legislator", back_populates="vote_records")
+
+
+class Vote(DataScienceDevTableBase):
+    """Roll call votes with counts and optional bill linkage."""
+
+    __tablename__ = "vote"
+
+    congress: Mapped[int]
+    chamber: Mapped[str]
+    session: Mapped[int]
+    number: Mapped[int]
+
+    vote_type: Mapped[str | None]
+    question: Mapped[str | None]
+    result: Mapped[str | None]
+    result_text: Mapped[str | None]
+
+    vote_date: Mapped[date]
+
+    yea_count: Mapped[int | None]
+    nay_count: Mapped[int | None]
+    not_voting_count: Mapped[int | None]
+    present_count: Mapped[int | None]
+
+    bill_id: Mapped[int | None] = mapped_column(ForeignKey("main.bill.id"))
+
+    bill: Mapped[Bill | None] = relationship("Bill", back_populates="votes")
+    vote_records: Mapped[list[VoteRecord]] = relationship(
+        "VoteRecord",
+        back_populates="vote",
+        cascade="all, delete-orphan",
+    )
+
+    __table_args__ = (
+        UniqueConstraint(
+            "congress",
+            "chamber",
+            "session",
+            "number",
+            name="uq_vote_congress_chamber_session_number",
+        ),
+        Index("ix_vote_date", "vote_date"),
+        Index("ix_vote_congress_chamber", "congress", "chamber"),
+    )

python/orm/data_science_dev/models.py

@@ -0,0 +1,16 @@
+"""Data science dev database ORM models."""
+
+from __future__ import annotations
+
+from python.orm.data_science_dev.congress import Bill, BillText, Legislator, Vote, VoteRecord
+from python.orm.data_science_dev.posts import partitions  # noqa: F401 — registers partition classes in metadata
+from python.orm.data_science_dev.posts.tables import Posts
+
+__all__ = [
+    "Bill",
+    "BillText",
+    "Legislator",
+    "Posts",
+    "Vote",
+    "VoteRecord",
+]

python/orm/data_science_dev/posts/__init__.py

@@ -0,0 +1,11 @@
+"""Posts module — weekly-partitioned posts table and partition ORM models."""
+
+from __future__ import annotations
+
+from python.orm.data_science_dev.posts.failed_ingestion import FailedIngestion
+from python.orm.data_science_dev.posts.tables import Posts
+
+__all__ = [
+    "FailedIngestion",
+    "Posts",
+]

python/orm/data_science_dev/posts/columns.py

@@ -0,0 +1,33 @@
+"""Shared column definitions for the posts partitioned table family."""
+
+from __future__ import annotations
+
+from datetime import datetime
+
+from sqlalchemy import BigInteger, SmallInteger, Text
+from sqlalchemy.orm import Mapped, mapped_column
+
+
+class PostsColumns:
+    """Mixin providing all posts columns. Used by both the parent table and partitions."""
+
+    post_id: Mapped[int] = mapped_column(BigInteger, primary_key=True)
+    user_id: Mapped[int] = mapped_column(BigInteger)
+    instance: Mapped[str]
+    date: Mapped[datetime] = mapped_column(primary_key=True)
+    text: Mapped[str] = mapped_column(Text)
+    langs: Mapped[str | None]
+    like_count: Mapped[int]
+    reply_count: Mapped[int]
+    repost_count: Mapped[int]
+    reply_to: Mapped[int | None] = mapped_column(BigInteger)
+    replied_author: Mapped[int | None] = mapped_column(BigInteger)
+    thread_root: Mapped[int | None] = mapped_column(BigInteger)
+    thread_root_author: Mapped[int | None] = mapped_column(BigInteger)
+    repost_from: Mapped[int | None] = mapped_column(BigInteger)
+    reposted_author: Mapped[int | None] = mapped_column(BigInteger)
+    quotes: Mapped[int | None] = mapped_column(BigInteger)
+    quoted_author: Mapped[int | None] = mapped_column(BigInteger)
+    labels: Mapped[str | None]
+    sent_label: Mapped[int | None] = mapped_column(SmallInteger)
+    sent_score: Mapped[float | None]

python/orm/data_science_dev/posts/failed_ingestion.py

@@ -0,0 +1,17 @@
+"""Table for storing JSONL lines that failed during post ingestion."""
+
+from __future__ import annotations
+
+from sqlalchemy import Text
+from sqlalchemy.orm import Mapped, mapped_column
+
+from python.orm.data_science_dev.base import DataScienceDevTableBase
+
+
+class FailedIngestion(DataScienceDevTableBase):
+    """Stores raw JSONL lines and their error messages when ingestion fails."""
+
+    __tablename__ = "failed_ingestion"
+
+    raw_line: Mapped[str] = mapped_column(Text)
+    error: Mapped[str] = mapped_column(Text)

python/orm/data_science_dev/posts/partitions.py

@@ -0,0 +1,71 @@
+"""Dynamically generated ORM classes for each weekly partition of the posts table.
+
+Each class maps to a PostgreSQL partition table (e.g. posts_2024_01).
+These are real ORM models tracked by Alembic autogenerate.
+
+Uses ISO week numbering (datetime.isocalendar().week). ISO years can have
+52 or 53 weeks, and week boundaries are always Monday to Monday.
+"""
+
+from __future__ import annotations
+
+import sys
+from datetime import UTC, datetime
+
+from python.orm.data_science_dev.base import DataScienceDevBase
+from python.orm.data_science_dev.posts.columns import PostsColumns
+
+PARTITION_START_YEAR = 2023
+PARTITION_END_YEAR = 2026
+
+_current_module = sys.modules[__name__]
+
+
+def iso_weeks_in_year(year: int) -> int:
+    """Return the number of ISO weeks in a given year (52 or 53)."""
+    dec_28 = datetime(year, 12, 28, tzinfo=UTC)
+    return dec_28.isocalendar().week
+
+
+def week_bounds(year: int, week: int) -> tuple[datetime, datetime]:
+    """Return (start, end) datetimes for an ISO week.
+
+    Start = Monday 00:00:00 UTC of the given ISO week.
+    End   = Monday 00:00:00 UTC of the following ISO week.
+    """
+    start = datetime.fromisocalendar(year, week, 1).replace(tzinfo=UTC)
+    if week < iso_weeks_in_year(year):
+        end = datetime.fromisocalendar(year, week + 1, 1).replace(tzinfo=UTC)
+    else:
+        end = datetime.fromisocalendar(year + 1, 1, 1).replace(tzinfo=UTC)
+    return start, end
+
+
+def _build_partition_classes() -> dict[str, type]:
+    """Generate one ORM class per ISO week partition."""
+    classes: dict[str, type] = {}
+
+    for year in range(PARTITION_START_YEAR, PARTITION_END_YEAR + 1):
+        for week in range(1, iso_weeks_in_year(year) + 1):
+            class_name = f"PostsWeek{year}W{week:02d}"
+            table_name = f"posts_{year}_{week:02d}"
+
+            partition_class = type(
+                class_name,
+                (PostsColumns, DataScienceDevBase),
+                {
+                    "__tablename__": table_name,
+                    "__table_args__": ({"implicit_returning": False},),
+                },
+            )
+
+            classes[class_name] = partition_class
+
+    return classes
+
+
+# Generate all partition classes and register them on this module
+_partition_classes = _build_partition_classes()
+for _name, _cls in _partition_classes.items():
+    setattr(_current_module, _name, _cls)
+__all__ = list(_partition_classes.keys())

python/orm/data_science_dev/posts/tables.py

@@ -0,0 +1,13 @@
+"""Posts parent table with PostgreSQL weekly range partitioning on date column."""
+
+from __future__ import annotations
+
+from python.orm.data_science_dev.base import DataScienceDevBase
+from python.orm.data_science_dev.posts.columns import PostsColumns
+
+
+class Posts(PostsColumns, DataScienceDevBase):
+    """Parent partitioned table for posts, partitioned by week on `date`."""
+
+    __tablename__ = "posts"
+    __table_args__ = ({"postgresql_partition_by": "RANGE (date)"},)

python/orm/richie/__init__.py

@@ -2,7 +2,6 @@
 
 from __future__ import annotations
 
-from python.orm.richie.audiobook import Audiobook, AudiobookAuthor, AudiobookSeries
 from python.orm.richie.base import RichieBase, TableBase, TableBaseBig, TableBaseSmall
 from python.orm.richie.contact import (
     Contact,
@@ -13,9 +12,6 @@ from python.orm.richie.contact import (
 )
 
 __all__ = [
-    "Audiobook",
-    "AudiobookAuthor",
-    "AudiobookSeries",
     "Contact",
     "ContactNeed",
     "ContactRelationship",

python/orm/richie/audiobook.py

@@ -1,55 +0,0 @@
-"""Audiobook catalog models."""
-
-from __future__ import annotations
-
-from sqlalchemy import ForeignKey, String, UniqueConstraint
-from sqlalchemy.orm import Mapped, mapped_column, relationship
-
-from python.orm.richie.base import TableBase
-
-
-class AudiobookAuthor(TableBase):
-    """Canonical audiobook author."""
-
-    __tablename__ = "audiobook_author"
-    __table_args__ = (UniqueConstraint("name"),)
-
-    name: Mapped[str] = mapped_column(String, unique=True)
-
-    books: Mapped[list[Audiobook]] = relationship("Audiobook", back_populates="author")
-    series: Mapped[list[AudiobookSeries]] = relationship("AudiobookSeries", back_populates="author")
-
-
-class AudiobookSeries(TableBase):
-    """Canonical audiobook series."""
-
-    __tablename__ = "audiobook_series"
-    __table_args__ = (UniqueConstraint("author_id", "name"),)
-
-    name: Mapped[str] = mapped_column(String)
-    author_id: Mapped[int] = mapped_column(ForeignKey("main.audiobook_author.id", ondelete="CASCADE"))
-
-    author: Mapped[AudiobookAuthor] = relationship("AudiobookAuthor", back_populates="series")
-    books: Mapped[list[Audiobook]] = relationship("Audiobook", back_populates="series")
-
-
-class Audiobook(TableBase):
-    """Canonical audiobook title."""
-
-    __tablename__ = "audiobook"
-    __table_args__ = (
-        UniqueConstraint(
-            "author_id",
-            "series_id",
-            "title",
-            postgresql_nulls_not_distinct=True,
-        ),
-    )
-
-    title: Mapped[str] = mapped_column(String)
-    author_id: Mapped[int] = mapped_column(ForeignKey("main.audiobook_author.id", ondelete="CASCADE"))
-    series_id: Mapped[int | None] = mapped_column(ForeignKey("main.audiobook_series.id", ondelete="SET NULL"))
-    series_index: Mapped[float] = mapped_column(default=0.0)
-
-    author: Mapped[AudiobookAuthor] = relationship("AudiobookAuthor", back_populates="books")
-    series: Mapped[AudiobookSeries | None] = relationship("AudiobookSeries", back_populates="books")

python/orm/signal_bot/__init__.py

@@ -0,0 +1,16 @@
+"""Signal bot database ORM exports."""
+
+from __future__ import annotations
+
+from python.orm.signal_bot.base import SignalBotBase, SignalBotTableBase, SignalBotTableBaseSmall
+from python.orm.signal_bot.models import DeadLetterMessage, DeviceRole, RoleRecord, SignalDevice
+
+__all__ = [
+    "DeadLetterMessage",
+    "DeviceRole",
+    "RoleRecord",
+    "SignalBotBase",
+    "SignalBotTableBase",
+    "SignalBotTableBaseSmall",
+    "SignalDevice",
+]

python/orm/signal_bot/base.py

@@ -0,0 +1,52 @@
+"""Signal bot database ORM base."""
+
+from __future__ import annotations
+
+from datetime import datetime
+
+from sqlalchemy import DateTime, MetaData, SmallInteger, func
+from sqlalchemy.ext.declarative import AbstractConcreteBase
+from sqlalchemy.orm import DeclarativeBase, Mapped, mapped_column
+
+from python.orm.common import NAMING_CONVENTION
+
+
+class SignalBotBase(DeclarativeBase):
+    """Base class for signal_bot database ORM models."""
+
+    schema_name = "main"
+
+    metadata = MetaData(
+        schema=schema_name,
+        naming_convention=NAMING_CONVENTION,
+    )
+
+
+class _TableMixin:
+    """Shared timestamp columns for all table bases."""
+
+    created: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True),
+        server_default=func.now(),
+    )
+    updated: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True),
+        server_default=func.now(),
+        onupdate=func.now(),
+    )
+
+
+class SignalBotTableBaseSmall(_TableMixin, AbstractConcreteBase, SignalBotBase):
+    """Table with SmallInteger primary key."""
+
+    __abstract__ = True
+
+    id: Mapped[int] = mapped_column(SmallInteger, primary_key=True)
+
+
+class SignalBotTableBase(_TableMixin, AbstractConcreteBase, SignalBotBase):
+    """Table with Integer primary key."""
+
+    __abstract__ = True
+
+    id: Mapped[int] = mapped_column(primary_key=True)

python/orm/signal_bot/models.py

@@ -0,0 +1,62 @@
+"""Signal bot device, role, and dead letter ORM models."""
+
+from __future__ import annotations
+
+from datetime import datetime
+
+from sqlalchemy import DateTime, Enum, ForeignKey, SmallInteger, String, Text, UniqueConstraint
+from sqlalchemy.orm import Mapped, mapped_column, relationship
+
+from python.orm.signal_bot.base import SignalBotTableBase, SignalBotTableBaseSmall
+from python.signal_bot.models import MessageStatus, TrustLevel
+
+
+class RoleRecord(SignalBotTableBaseSmall):
+    """Lookup table for RBAC roles, keyed by smallint."""
+
+    __tablename__ = "role"
+
+    name: Mapped[str] = mapped_column(String(50), unique=True)
+
+
+class DeviceRole(SignalBotTableBase):
+    """Association between a device and a role."""
+
+    __tablename__ = "device_role"
+    __table_args__ = (
+        UniqueConstraint("device_id", "role_id", name="uq_device_role_device_role"),
+        {"schema": "main"},
+    )
+
+    device_id: Mapped[int] = mapped_column(ForeignKey("main.signal_device.id"))
+    role_id: Mapped[int] = mapped_column(SmallInteger, ForeignKey("main.role.id"))
+
+
+class SignalDevice(SignalBotTableBase):
+    """A Signal device tracked by phone number and safety number."""
+
+    __tablename__ = "signal_device"
+
+    phone_number: Mapped[str] = mapped_column(String(50), unique=True)
+    safety_number: Mapped[str | None]
+    trust_level: Mapped[TrustLevel] = mapped_column(
+        Enum(TrustLevel, name="trust_level", create_constraint=False, native_enum=False),
+        default=TrustLevel.UNVERIFIED,
+    )
+    last_seen: Mapped[datetime] = mapped_column(DateTime(timezone=True))
+
+    roles: Mapped[list[RoleRecord]] = relationship(secondary=DeviceRole.__table__)
+
+
+class DeadLetterMessage(SignalBotTableBase):
+    """A Signal message that failed processing and was sent to the dead letter queue."""
+
+    __tablename__ = "dead_letter_message"
+
+    source: Mapped[str]
+    message: Mapped[str] = mapped_column(Text)
+    received_at: Mapped[datetime] = mapped_column(DateTime(timezone=True))
+    status: Mapped[MessageStatus] = mapped_column(
+        Enum(MessageStatus, name="message_status", create_constraint=False, native_enum=False),
+        default=MessageStatus.UNPROCESSED,
+    )

python/prompt_bench/Dockerfile.finetune

@@ -0,0 +1,25 @@
+# Unsloth fine-tuning container for Qwen 3.5 4B on RTX 3090.
+#
+# Build:
+#   docker build -f python/prompt_bench/Dockerfile.finetune -t bill-finetune .
+#
+# Run:
+#   docker run --rm --device=nvidia.com/gpu=all --ipc=host \
+#     -v $(pwd)/output:/workspace/output \
+#     -v $(pwd)/output/finetune_dataset.jsonl:/workspace/dataset.jsonl:ro \
+#     -v /zfs/models/hf:/models \
+#     bill-finetune \
+#     --dataset /workspace/dataset.jsonl \
+#     --output-dir /workspace/output/qwen-bill-summarizer
+
+FROM ghcr.io/unslothai/unsloth:latest
+
+RUN pip install --no-cache-dir typer
+
+WORKDIR /workspace
+COPY python/prompt_bench/finetune.py python/prompt_bench/finetune.py
+COPY python/prompt_bench/summarization_prompts.py python/prompt_bench/summarization_prompts.py
+COPY python/prompt_bench/__init__.py python/prompt_bench/__init__.py
+COPY python/__init__.py python/__init__.py
+
+ENTRYPOINT ["python", "-m", "python.prompt_bench.finetune"]

python/prompt_bench/__init__.py

@@ -0,0 +1 @@
+"""Prompt benchmarking system for evaluating LLMs via vLLM."""

python/prompt_bench/batch_bill_summarizer.py

@@ -0,0 +1,233 @@
+"""Submit an OpenAI Batch API bill-summarization job over compressed text.
+
+Reads the first N bills from a CSV with a `text_content` column, compresses
+each via `bill_token_compression.compress_bill_text`, builds a JSONL file of
+summarization requests, and submits it as an asynchronous Batch API job
+against `/v1/chat/completions`. Also writes a CSV of per-bill pre/post-
+compression token counts.
+"""
+
+from __future__ import annotations
+
+import csv
+import json
+import logging
+import re
+import sys
+from os import getenv
+from pathlib import Path
+from typing import Annotated
+
+import httpx
+import typer
+from tiktoken import Encoding, get_encoding
+
+from python.prompt_bench.bill_token_compression import compress_bill_text
+from python.prompt_bench.summarization_prompts import SUMMARIZATION_SYSTEM_PROMPT, SUMMARIZATION_USER_TEMPLATE
+
+logger = logging.getLogger(__name__)
+
+OPENAI_API_BASE = "https://api.openai.com/v1"
+
+
+def load_bills(csv_path: Path, count: int = 0) -> list[tuple[str, str]]:
+    """Return (bill_id, text_content) tuples with non-empty text.
+
+    If `count` is 0 or negative, all rows are returned.
+    """
+    csv.field_size_limit(sys.maxsize)
+    bills: list[tuple[str, str]] = []
+    with csv_path.open(newline="", encoding="utf-8") as handle:
+        reader = csv.DictReader(handle)
+        for row in reader:
+            text_content = (row.get("text_content") or "").strip()
+            if not text_content:
+                continue
+            bill_id = row.get("bill_id") or row.get("id") or f"row-{len(bills)}"
+            version_code = row.get("version_code") or ""
+            unique_id = f"{bill_id}-{version_code}" if version_code else bill_id
+            bills.append((unique_id, text_content))
+            if count > 0 and len(bills) >= count:
+                break
+    return bills
+
+
+def safe_filename(value: str) -> str:
+    """Make a string safe for use as a filename or batch custom_id."""
+    return re.sub(r"[^A-Za-z0-9._-]+", "_", value).strip("_") or "unnamed"
+
+
+def build_request(custom_id: str, model: str, bill_text: str) -> dict:
+    """Build one OpenAI batch request line."""
+    return {
+        "custom_id": custom_id,
+        "method": "POST",
+        "url": "/v1/chat/completions",
+        "body": {
+            "model": model,
+            "messages": [
+                {"role": "system", "content": SUMMARIZATION_SYSTEM_PROMPT},
+                {"role": "user", "content": SUMMARIZATION_USER_TEMPLATE.format(text_content=bill_text)},
+            ],
+        },
+    }
+
+
+def write_jsonl(path: Path, lines: list[dict]) -> None:
+    """Write a list of dicts as JSONL."""
+    with path.open("w", encoding="utf-8") as handle:
+        for line in lines:
+            handle.write(json.dumps(line, ensure_ascii=False))
+            handle.write("\n")
+
+
+def upload_file(client: httpx.Client, path: Path) -> str:
+    """Upload a JSONL file to the OpenAI Files API and return its file id."""
+    with path.open("rb") as handle:
+        response = client.post(
+            f"{OPENAI_API_BASE}/files",
+            files={"file": (path.name, handle, "application/jsonl")},
+            data={"purpose": "batch"},
+        )
+    response.raise_for_status()
+    return response.json()["id"]
+
+
+def prepare_requests(
+    bills: list[tuple[str, str]],
+    *,
+    model: str,
+    encoder: Encoding,
+) -> tuple[list[dict], list[dict]]:
+    """Build (request_lines, token_rows) from bills.
+
+    Each bill is compressed before being turned into a request line.
+    Each `token_rows` entry has chars + token counts for one bill so the caller
+    can write a per-bill CSV.
+    """
+    request_lines: list[dict] = []
+    token_rows: list[dict] = []
+    for bill_id, text_content in bills:
+        raw_token_count = len(encoder.encode(text_content))
+        compressed_text = compress_bill_text(text_content)
+        compressed_token_count = len(encoder.encode(compressed_text))
+        token_rows.append(
+            {
+                "bill_id": bill_id,
+                "raw_chars": len(text_content),
+                "compressed_chars": len(compressed_text),
+                "raw_tokens": raw_token_count,
+                "compressed_tokens": compressed_token_count,
+                "token_ratio": (compressed_token_count / raw_token_count) if raw_token_count else None,
+            },
+        )
+        safe_id = safe_filename(bill_id)
+        request_lines.append(build_request(safe_id, model, compressed_text))
+    return request_lines, token_rows
+
+
+def write_token_csv(path: Path, token_rows: list[dict]) -> tuple[int, int]:
+    """Write per-bill token counts to CSV. Returns (raw_total, compressed_total)."""
+    with path.open("w", newline="", encoding="utf-8") as handle:
+        writer = csv.DictWriter(
+            handle,
+            fieldnames=["bill_id", "raw_chars", "compressed_chars", "raw_tokens", "compressed_tokens", "token_ratio"],
+        )
+        writer.writeheader()
+        writer.writerows(token_rows)
+    raw_total = sum(row["raw_tokens"] for row in token_rows)
+    compressed_total = sum(row["compressed_tokens"] for row in token_rows)
+    return raw_total, compressed_total
+
+
+def create_batch(client: httpx.Client, input_file_id: str, description: str) -> dict:
+    """Create a batch job and return its full response payload."""
+    response = client.post(
+        f"{OPENAI_API_BASE}/batches",
+        json={
+            "input_file_id": input_file_id,
+            "endpoint": "/v1/chat/completions",
+            "completion_window": "24h",
+            "metadata": {"description": description},
+        },
+    )
+    response.raise_for_status()
+    return response.json()
+
+
+def main(
+    csv_path: Annotated[Path, typer.Option("--csv", help="Bills CSV path")] = Path("bills.csv"),
+    output_dir: Annotated[Path, typer.Option("--output-dir", help="Where to write JSONL + metadata")] = Path(
+        "output/openai_batch",
+    ),
+    model: Annotated[str, typer.Option(help="OpenAI model id")] = "gpt-5-mini",
+    count: Annotated[int, typer.Option(help="Max bills to process, 0 = all")] = 0,
+    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
+) -> None:
+    """Submit an OpenAI Batch job of compressed bill summaries."""
+    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
+
+    api_key = getenv("CLOSEDAI_TOKEN") or getenv("OPENAI_API_KEY")
+    if not api_key:
+        message = "Neither CLOSEDAI_TOKEN nor OPENAI_API_KEY is set"
+        raise typer.BadParameter(message)
+    if not csv_path.is_file():
+        message = f"CSV not found: {csv_path}"
+        raise typer.BadParameter(message)
+
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    logger.info("Loading %d bills from %s", count, csv_path)
+    bills = load_bills(csv_path, count)
+    if len(bills) < count:
+        logger.warning("Only %d bills available (requested %d)", len(bills), count)
+
+    encoder = get_encoding("o200k_base")
+    request_lines, token_rows = prepare_requests(bills, model=model, encoder=encoder)
+
+    token_csv_path = output_dir / "token_counts.csv"
+    raw_tokens_total, compressed_tokens_total = write_token_csv(token_csv_path, token_rows)
+    logger.info(
+        "Token counts: raw=%d compressed=%d ratio=%.3f -> %s",
+        raw_tokens_total,
+        compressed_tokens_total,
+        (compressed_tokens_total / raw_tokens_total) if raw_tokens_total else 0.0,
+        token_csv_path,
+    )
+
+    jsonl_path = output_dir / "requests.jsonl"
+    write_jsonl(jsonl_path, request_lines)
+    logger.info("Wrote %s (%d bills)", jsonl_path, len(request_lines))
+
+    headers = {"Authorization": f"Bearer {api_key}"}
+    with httpx.Client(headers=headers, timeout=httpx.Timeout(300.0)) as client:
+        logger.info("Uploading JSONL")
+        file_id = upload_file(client, jsonl_path)
+        logger.info("Uploaded: %s", file_id)
+
+        logger.info("Creating batch")
+        batch = create_batch(client, file_id, f"compressed bill summaries x{len(request_lines)} ({model})")
+        logger.info("Batch created: %s", batch["id"])
+
+    metadata = {
+        "model": model,
+        "count": len(bills),
+        "jsonl": str(jsonl_path),
+        "input_file_id": file_id,
+        "batch_id": batch["id"],
+        "raw_tokens_total": raw_tokens_total,
+        "compressed_tokens_total": compressed_tokens_total,
+        "batch": batch,
+    }
+    metadata_path = output_dir / "batch.json"
+    metadata_path.write_text(json.dumps(metadata, indent=2))
+    logger.info("Wrote metadata to %s", metadata_path)
+
+
+def cli() -> None:
+    """Typer entry point."""
+    typer.run(main)
+
+
+if __name__ == "__main__":
+    cli()

python/prompt_bench/bill_token_compression.py

@@ -0,0 +1,162 @@
+"""Lossless-ish text compression for Congressional bill text."""
+
+from __future__ import annotations
+
+import re
+
+STATES = (
+    "Alabama",
+    "Alaska",
+    "Arizona",
+    "Arkansas",
+    "California",
+    "Colorado",
+    "Connecticut",
+    "Delaware",
+    "Florida",
+    "Georgia",
+    "Hawaii",
+    "Idaho",
+    "Illinois",
+    "Indiana",
+    "Iowa",
+    "Kansas",
+    "Kentucky",
+    "Louisiana",
+    "Maine",
+    "Maryland",
+    "Massachusetts",
+    "Michigan",
+    "Minnesota",
+    "Mississippi",
+    "Missouri",
+    "Montana",
+    "Nebraska",
+    "Nevada",
+    "New Hampshire",
+    "New Jersey",
+    "New Mexico",
+    "New York",
+    "North Carolina",
+    "North Dakota",
+    "Ohio",
+    "Oklahoma",
+    "Oregon",
+    "Pennsylvania",
+    "Rhode Island",
+    "South Carolina",
+    "South Dakota",
+    "Tennessee",
+    "Texas",
+    "Utah",
+    "Vermont",
+    "Virginia",
+    "Washington",
+    "West Virginia",
+    "Wisconsin",
+    "Wyoming",
+    "Puerto Rico",
+    "Guam",
+    "American Samoa",
+    "District of Columbia",
+    "US Virgin Islands",
+)
+STATE_PATTERNS = [(re.compile(re.escape(state), re.IGNORECASE), state) for state in STATES]
+
+
+def normalize_state_names(text: str) -> str:
+    """Replace any casing of state names with title case."""
+    for pattern, replacement in STATE_PATTERNS:
+        text = pattern.sub(replacement, text)
+    return text
+
+
+def strip_number_commas(text: str) -> str:
+    """Remove commas from numeric thousands separators."""
+    return re.sub(r"(\d{1,3}(?:,\d{3})+)", lambda match: match.group().replace(",", ""), text)
+
+
+def strip_horizontal_rules(text: str) -> str:
+    """Remove ASCII horizontal-rule lines built from underscores, dashes, equals, or asterisks."""
+    return re.sub(r"^\s*[_\-=\*]{3,}\s*$", "", text, flags=re.MULTILINE)
+
+
+def collapse_double_dashes(text: str) -> str:
+    """Replace ``--`` em-dash stand-ins with a single space so they don't tokenize oddly."""
+    return text.replace("--", " ")
+
+
+def collapse_inline_whitespace(text: str) -> str:
+    """Collapse runs of horizontal whitespace (spaces, tabs) into a single space, leaving newlines intact."""
+    return re.sub(r"[^\S\n]+", " ", text)
+
+
+def collapse_blank_lines(text: str) -> str:
+    """Collapse three-or-more consecutive newlines down to a blank-line separator."""
+    return re.sub(r"\n{3,}", "\n\n", text)
+
+
+def trim_line_edges(text: str) -> str:
+    """Strip spaces immediately before and after newline characters on every line."""
+    text = re.sub(r" +\n", "\n", text)
+    return re.sub(r"\n +", "\n", text)
+
+
+def shorten_section_markers(text: str) -> str:
+    """Rewrite ``Sec. 12.`` style section headings as the more compact ``SEC 12``."""
+    return re.sub(r"(?i)sec\.\s*(\d+[a-zA-Z]?)\.", r"SEC \1", text)
+
+
+def unwrap_parens(text: str) -> str:
+    """Strip parentheses around short alphanumeric labels like ``(a)`` or ``(12)``."""
+    return re.sub(r"\(([a-zA-Z0-9]+)\)", r"\1", text)
+
+
+def strip_typeset_quotes(text: str) -> str:
+    """Remove the `` and '' typeset quote markers used in the GPO bill format."""
+    return text.replace("``", "").replace("''", "")
+
+
+def normalize_usc_acronym(text: str) -> str:
+    """Collapse ``U.S.C.`` to ``USC`` to save tokens on the common citation."""
+    return text.replace("U.S.C.", "USC")
+
+
+def normalize_us_acronym(text: str) -> str:
+    """Normalize the various ``U.S.``/``U. S.`` spellings to the bare ``US`` form."""
+    for acronym in ("U. S.", "u. s.", "U.S. ", "u.s. "):
+        text = text.replace(acronym, "US ")
+    return text
+
+
+def collapse_ellipses(text: str) -> str:
+    """Collapse runs of two-or-more periods (``...``, ``....``) down to a single period."""
+    return re.sub(r"\.{2,}", ".", text)
+
+
+COMPRESSION_STEPS = (
+    strip_horizontal_rules,
+    collapse_double_dashes,
+    collapse_inline_whitespace,
+    collapse_blank_lines,
+    trim_line_edges,
+    shorten_section_markers,
+    unwrap_parens,
+    strip_typeset_quotes,
+    normalize_usc_acronym,
+    normalize_us_acronym,
+    strip_number_commas,
+    collapse_ellipses,
+    normalize_state_names,
+)
+
+
+def compress_bill_text(text: str) -> str:
+    """Apply lossless-ish whitespace and boilerplate compression to bill text.
+
+    Runs every transform in :data:`COMPRESSION_STEPS` in order, then strips
+    leading/trailing whitespace from the final result.
+    """
+    for step in COMPRESSION_STEPS:
+        text = step(text)
+    return text.strip()

python/prompt_bench/compresion_test.py

@@ -0,0 +1,236 @@
+"""Run two interactive OpenAI chat-completion sweeps over bill text.
+
+Reads the first N bills from a CSV with a `text_content` column and sends two
+sweeps through `/v1/chat/completions` concurrently — one with the raw bill
+text, one with the compressed bill text. Each request's prompt is saved to
+disk alongside the OpenAI response id so the prompts and responses can be
+correlated later.
+"""
+
+from __future__ import annotations
+
+import csv
+import json
+import logging
+import re
+import sys
+import time
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from os import getenv
+from pathlib import Path
+from typing import Annotated
+
+import httpx
+import typer
+
+from python.prompt_bench.bill_token_compression import compress_bill_text
+from python.prompt_bench.summarization_prompts import SUMMARIZATION_SYSTEM_PROMPT, SUMMARIZATION_USER_TEMPLATE
+
+logger = logging.getLogger(__name__)
+
+OPENAI_API_BASE = "https://api.openai.com/v1"
+DEFAULT_MODEL = "gpt-5.4-mini"
+DEFAULT_COUNT = 100
+SEED = 42
+
+
+def load_bills(csv_path: Path, count: int) -> list[tuple[str, str]]:
+    """Return up to `count` (bill_id, text_content) tuples with non-empty text."""
+    csv.field_size_limit(sys.maxsize)
+    bills: list[tuple[str, str]] = []
+    with csv_path.open(newline="", encoding="utf-8") as handle:
+        reader = csv.DictReader(handle)
+        for row in reader:
+            text_content = (row.get("text_content") or "").strip()
+            if not text_content:
+                continue
+            bill_id = row.get("bill_id") or row.get("id") or f"row-{len(bills)}"
+            version_code = row.get("version_code") or ""
+            unique_id = f"{bill_id}-{version_code}" if version_code else bill_id
+            bills.append((unique_id, text_content))
+            if len(bills) >= count:
+                break
+    return bills
+
+
+def build_messages(bill_text: str) -> list[dict]:
+    """Return the system + user message pair for a bill."""
+    return [
+        {"role": "system", "content": SUMMARIZATION_SYSTEM_PROMPT},
+        {"role": "user", "content": SUMMARIZATION_USER_TEMPLATE.format(text_content=bill_text)},
+    ]
+
+
+def safe_filename(value: str) -> str:
+    """Make a string safe for use as a filename."""
+    return re.sub(r"[^A-Za-z0-9._-]+", "_", value).strip("_") or "unnamed"
+
+
+def run_one_request(
+    client: httpx.Client,
+    *,
+    bill_id: str,
+    label: str,
+    bill_text: str,
+    model: str,
+    output_path: Path,
+) -> tuple[bool, float, str | None]:
+    """Send one chat-completion request and persist prompt + response.
+
+    Returns (success, elapsed_seconds, response_id).
+    """
+    messages = build_messages(bill_text)
+    payload = {
+        "model": model,
+        "messages": messages,
+        "seed": SEED,
+    }
+    start = time.monotonic()
+    record: dict = {
+        "bill_id": bill_id,
+        "label": label,
+        "model": model,
+        "seed": SEED,
+        "input_chars": len(bill_text),
+        "messages": messages,
+    }
+    try:
+        response = client.post(f"{OPENAI_API_BASE}/chat/completions", json=payload)
+        response.raise_for_status()
+        body = response.json()
+    except httpx.HTTPStatusError as error:
+        elapsed = time.monotonic() - start
+        record["error"] = {
+            "status_code": error.response.status_code,
+            "body": error.response.text,
+            "elapsed_seconds": elapsed,
+        }
+        output_path.write_text(json.dumps(record, ensure_ascii=False, indent=2))
+        logger.exception("HTTP error for %s/%s after %.2fs", label, bill_id, elapsed)
+        return False, elapsed, None
+    except Exception as error:
+        elapsed = time.monotonic() - start
+        record["error"] = {"message": str(error), "elapsed_seconds": elapsed}
+        output_path.write_text(json.dumps(record, ensure_ascii=False, indent=2))
+        logger.exception("Failed: %s/%s after %.2fs", label, bill_id, elapsed)
+        return False, elapsed, None
+
+    elapsed = time.monotonic() - start
+    response_id = body.get("id")
+    record["response_id"] = response_id
+    record["elapsed_seconds"] = elapsed
+    record["usage"] = body.get("usage")
+    record["response"] = body
+    output_path.write_text(json.dumps(record, ensure_ascii=False, indent=2))
+    logger.info("Done: %s/%s id=%s in %.2fs", label, bill_id, response_id, elapsed)
+    return True, elapsed, response_id
+
+
+def main(
+    csv_path: Annotated[Path, typer.Option("--csv", help="Bills CSV path")] = Path("bills.csv"),
+    output_dir: Annotated[Path, typer.Option("--output-dir", help="Where to write per-request JSON")] = Path(
+        "output/openai_runs",
+    ),
+    model: Annotated[str, typer.Option(help="OpenAI model id")] = DEFAULT_MODEL,
+    count: Annotated[int, typer.Option(help="Number of bills per set")] = DEFAULT_COUNT,
+    concurrency: Annotated[int, typer.Option(help="Concurrent in-flight requests")] = 16,
+    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
+) -> None:
+    """Run two interactive OpenAI sweeps (compressed + uncompressed) over bill text."""
+    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
+
+    api_key = getenv("CLOSEDAI_TOKEN") or getenv("OPENAI_API_KEY")
+    if not api_key:
+        message = "Neither CLOSEDAI_TOKEN nor OPENAI_API_KEY is set"
+        raise typer.BadParameter(message)
+    if not csv_path.is_file():
+        message = f"CSV not found: {csv_path}"
+        raise typer.BadParameter(message)
+
+    compressed_dir = output_dir / "compressed"
+    uncompressed_dir = output_dir / "uncompressed"
+    compressed_dir.mkdir(parents=True, exist_ok=True)
+    uncompressed_dir.mkdir(parents=True, exist_ok=True)
+
+    logger.info("Loading %d bills from %s", count, csv_path)
+    bills = load_bills(csv_path, count)
+    if len(bills) < count:
+        logger.warning("Only %d bills available (requested %d)", len(bills), count)
+
+    tasks: list[tuple[str, str, str, Path]] = []
+    for bill_id, text_content in bills:
+        filename = f"{safe_filename(bill_id)}.json"
+        tasks.append((bill_id, "compressed", compress_bill_text(text_content), compressed_dir / filename))
+        tasks.append((bill_id, "uncompressed", text_content, uncompressed_dir / filename))
+
+    logger.info("Submitting %d requests at concurrency=%d", len(tasks), concurrency)
+
+    headers = {"Authorization": f"Bearer {api_key}"}
+    completed = 0
+    failed = 0
+    index: list[dict] = []
+    wall_start = time.monotonic()
+    with (
+        httpx.Client(headers=headers, timeout=httpx.Timeout(300.0)) as client,
+        ThreadPoolExecutor(
+            max_workers=concurrency,
+        ) as executor,
+    ):
+        future_to_task = {
+            executor.submit(
+                run_one_request,
+                client,
+                bill_id=bill_id,
+                label=label,
+                bill_text=bill_text,
+                model=model,
+                output_path=output_path,
+            ): (bill_id, label, output_path)
+            for bill_id, label, bill_text, output_path in tasks
+        }
+        for future in as_completed(future_to_task):
+            bill_id, label, output_path = future_to_task[future]
+            success, elapsed, response_id = future.result()
+            if success:
+                completed += 1
+            else:
+                failed += 1
+            index.append(
+                {
+                    "bill_id": bill_id,
+                    "label": label,
+                    "response_id": response_id,
+                    "elapsed_seconds": elapsed,
+                    "success": success,
+                    "path": str(output_path),
+                },
+            )
+    wall_elapsed = time.monotonic() - wall_start
+
+    summary = {
+        "model": model,
+        "count": len(bills),
+        "completed": completed,
+        "failed": failed,
+        "wall_seconds": wall_elapsed,
+        "concurrency": concurrency,
+        "results": index,
+    }
+    summary_path = output_dir / "summary.json"
+    summary_path.write_text(json.dumps(summary, indent=2))
+    logger.info(
+        "Done: completed=%d failed=%d wall=%.1fs summary=%s",
+        completed,
+        failed,
+        wall_elapsed,
+        summary_path,
+    )
+
+
+def cli() -> None:
+    """Typer entry point."""
+    typer.run(main)
+
+
+if __name__ == "__main__":
+    cli()

python/prompt_bench/containers/__init__.py

@@ -0,0 +1 @@
+"""Prompt benchmarking system for evaluating LLMs via vLLM."""

python/prompt_bench/containers/finetune.py

@@ -0,0 +1,165 @@
+"""Docker container lifecycle management for Unsloth fine-tuning."""
+
+from __future__ import annotations
+
+import logging
+import subprocess
+from pathlib import Path
+from typing import Annotated
+
+import typer
+
+from python.prompt_bench.containers.lib import check_gpu_free
+
+logger = logging.getLogger(__name__)
+
+CONTAINER_NAME = "bill-finetune"
+FINETUNE_IMAGE = "bill-finetune:latest"
+DOCKERFILE_PATH = "/home/richie/dotfiles/python/prompt_bench/Dockerfile.finetune"
+DEFAULT_HF_CACHE = Path("/zfs/models/hf")
+
+
+def build_image() -> None:
+    """Build the fine-tuning Docker image."""
+    logger.info("Building fine-tuning image: %s", FINETUNE_IMAGE)
+    result = subprocess.run(
+        ["docker", "build", "-f", DOCKERFILE_PATH, "-t", FINETUNE_IMAGE, "."],
+        text=True,
+        check=False,
+    )
+    if result.returncode != 0:
+        message = "Failed to build fine-tuning image"
+        raise RuntimeError(message)
+    logger.info("Image built: %s", FINETUNE_IMAGE)
+
+
+def start_finetune(
+    *,
+    dataset_path: Path,
+    output_dir: Path,
+    hf_cache: Path = DEFAULT_HF_CACHE,
+) -> None:
+    """Run the fine-tuning container.
+
+    Args:
+        dataset_path: Host path to the fine-tuning JSONL dataset.
+        output_dir: Host path where the trained model will be saved.
+        hf_cache: Host path to HuggingFace model cache (bind-mounted to avoid re-downloading).
+        validation_split: Fraction of data held out for validation.
+    """
+    dataset_path = dataset_path.resolve()
+    output_dir = output_dir.resolve()
+
+    if not dataset_path.is_file():
+        message = f"Dataset not found: {dataset_path}"
+        raise FileNotFoundError(message)
+
+    output_dir.mkdir(parents=True, exist_ok=True)
+    stop_finetune()
+
+    hf_cache = hf_cache.resolve()
+    hf_cache.mkdir(parents=True, exist_ok=True)
+
+    command = [
+        "docker",
+        "run",
+        "--name",
+        CONTAINER_NAME,
+        "--device=nvidia.com/gpu=all",
+        "--ipc=host",
+        "-v",
+        f"{hf_cache}:/root/.cache/huggingface",
+        "-v",
+        f"{output_dir}:/workspace/output/qwen-bill-summarizer",
+        "-v",
+        f"{dataset_path}:/workspace/dataset.jsonl:ro",
+        FINETUNE_IMAGE,
+        "--dataset",
+        "/workspace/dataset.jsonl",
+        "--output-dir",
+        "/workspace/output/qwen-bill-summarizer",
+    ]
+
+    logger.info("Starting fine-tuning container")
+    logger.info("  Dataset:    %s", dataset_path)
+    logger.info("  Output:     %s", output_dir)
+
+    result = subprocess.run(command, text=True, check=False)
+    if result.returncode != 0:
+        message = f"Fine-tuning container exited with code {result.returncode}"
+        raise RuntimeError(message)
+    logger.info("Fine-tuning complete. Model saved to %s", output_dir)
+
+
+def stop_finetune() -> None:
+    """Stop and remove the fine-tuning container."""
+    logger.info("Stopping fine-tuning container")
+    subprocess.run(["docker", "stop", CONTAINER_NAME], capture_output=True, check=False)
+    subprocess.run(["docker", "rm", "-f", CONTAINER_NAME], capture_output=True, check=False)
+
+
+def logs_finetune() -> str | None:
+    """Return recent logs from the fine-tuning container, or None if not running."""
+    result = subprocess.run(
+        ["docker", "logs", "--tail", "50", CONTAINER_NAME],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    if result.returncode != 0:
+        return None
+    return result.stdout + result.stderr
+
+
+app = typer.Typer(help="Fine-tuning container management.")
+
+
+@app.command()
+def build() -> None:
+    """Build the fine-tuning Docker image."""
+    build_image()
+
+
+@app.command()
+def run(
+    dataset: Annotated[Path, typer.Option(help="Fine-tuning JSONL")] = Path(
+        "/home/richie/dotfiles/data/finetune_dataset.jsonl"
+    ),
+    output_dir: Annotated[Path, typer.Option(help="Where to save the trained model")] = Path(
+        "/home/richie/dotfiles/data/output/qwen-bill-summarizer",
+    ),
+    hf_cache: Annotated[Path, typer.Option(help="Host path to HuggingFace model cache")] = DEFAULT_HF_CACHE,
+    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
+) -> None:
+    """Run fine-tuning inside a Docker container."""
+    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
+    check_gpu_free()
+    start_finetune(
+        dataset_path=dataset,
+        output_dir=output_dir,
+        hf_cache=hf_cache,
+    )
+
+@app.command()
+def stop() -> None:
+    """Stop and remove the fine-tuning container."""
+    stop_finetune()
+
+
+@app.command()
+def logs() -> None:
+    """Show recent logs from the fine-tuning container."""
+    output = logs_finetune()
+    if output is None:
+        typer.echo("No running fine-tuning container found.")
+        raise typer.Exit(code=1)
+    typer.echo(output)
+
+
+def cli() -> None:
+    """Typer entry point."""
+    app()
+
+
+if __name__ == "__main__":
+    cli()

python/prompt_bench/containers/lib.py

@@ -0,0 +1,23 @@
+from __future__ import annotations
+
+import logging
+import subprocess
+
+logger = logging.getLogger(__name__)
+
+
+def check_gpu_free() -> None:
+    """Warn if GPU-heavy processes (e.g. Ollama) are running."""
+    result = subprocess.run(
+        ["nvidia-smi", "--query-compute-apps=pid,process_name", "--format=csv,noheader"],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    if result.returncode != 0:
+        logger.warning("Could not query GPU processes: %s", result.stderr.strip())
+        return
+    processes = result.stdout.strip()
+    if processes:
+        logger.warning("GPU processes detected:\n%s", processes)
+        logger.warning("Consider stopping Ollama (sudo systemctl stop ollama) before benchmarking")

python/prompt_bench/containers/vllm.py

@@ -0,0 +1,70 @@
+"""Docker container lifecycle management for vLLM."""
+
+from __future__ import annotations
+
+import logging
+import subprocess
+
+logger = logging.getLogger(__name__)
+
+CONTAINER_NAME = "vllm-bench"
+VLLM_IMAGE = "vllm/vllm-openai:v0.19.0"
+
+
+def start_vllm(
+    *,
+    model: str,
+    port: int,
+    model_dir: str,
+    gpu_memory_utilization: float,
+) -> None:
+    """Start a vLLM container serving the given model.
+
+    Args:
+        model: HuggingFace model directory name (relative to model_dir).
+        port: Host port to bind.
+        model_dir: Host path containing HuggingFace model directories.
+        gpu_memory_utilization: Fraction of GPU memory to use (0-1).
+    """
+    command = [
+        "docker",
+        "run",
+        "-d",
+        "--name",
+        CONTAINER_NAME,
+        "--device=nvidia.com/gpu=all",
+        "--ipc=host",
+        "-v",
+        f"{model_dir}:/models",
+        "-p",
+        f"{port}:8000",
+        VLLM_IMAGE,
+        "--model",
+        f"/models/{model}",
+        "--served-model-name",
+        model,
+        "--gpu-memory-utilization",
+        str(gpu_memory_utilization),
+        "--max-model-len",
+        "4096",
+    ]
+    logger.info("Starting vLLM container with model: %s", model)
+    stop_vllm()
+    result = subprocess.run(command, capture_output=True, text=True, check=False)
+    if result.returncode != 0:
+        msg = f"Failed to start vLLM container: {result.stderr.strip()}"
+        raise RuntimeError(msg)
+    logger.info("vLLM container started: %s", result.stdout.strip()[:12])
+
+
+def stop_vllm() -> None:
+    """Stop and remove the vLLM benchmark container."""
+    logger.info("Stopping vLLM container")
+    subprocess.run(["docker", "stop", CONTAINER_NAME], capture_output=True, check=False)
+    subprocess.run(["docker", "rm", "-f", CONTAINER_NAME], capture_output=True, check=False)
+    subprocess.run(
+        ["docker", "network", "disconnect", "-f", "bridge", CONTAINER_NAME],
+        capture_output=True,
+        check=False,
+    )
+    logger.info("vLLM container stopped and removed")

python/prompt_bench/downloader.py

@@ -0,0 +1,75 @@
+"""HuggingFace model downloader."""
+
+from __future__ import annotations
+
+import logging
+from pathlib import Path
+from typing import Annotated
+
+import typer
+from huggingface_hub import snapshot_download
+
+from python.prompt_bench.models import BenchmarkConfig
+
+logger = logging.getLogger(__name__)
+
+
+def local_model_path(repo: str, model_dir: str) -> Path:
+    """Return the local directory path for a HuggingFace repo."""
+    return Path(model_dir) / repo
+
+
+def is_model_present(repo: str, model_dir: str) -> bool:
+    """Check if a model has already been downloaded."""
+    path = local_model_path(repo, model_dir)
+    return path.exists() and any(path.iterdir())
+
+
+def download_model(repo: str, model_dir: str) -> Path:
+    """Download a HuggingFace model to the local model directory.
+
+    Skips the download if the model directory already exists and contains files.
+    """
+    local_path = local_model_path(repo, model_dir)
+
+    if is_model_present(repo, model_dir):
+        logger.info("Model already exists: %s", local_path)
+        return local_path
+
+    logger.info("Downloading model: %s -> %s", repo, local_path)
+    snapshot_download(
+        repo_id=repo,
+        local_dir=str(local_path),
+    )
+    logger.info("Download complete: %s", repo)
+    return local_path
+
+
+def download_all(config: BenchmarkConfig) -> None:
+    """Download every model listed in the config, top to bottom."""
+    for repo in config.models:
+        download_model(repo, config.model_dir)
+
+
+def main(
+    config: Annotated[Path, typer.Option(help="Path to TOML config file")] = Path("bench.toml"),
+    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
+) -> None:
+    """Download all models listed in the benchmark config."""
+    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
+
+    if not config.is_file():
+        message = f"Config file does not exist: {config}"
+        raise typer.BadParameter(message)
+
+    benchmark_config = BenchmarkConfig.from_toml(config)
+    download_all(benchmark_config)
+
+
+def cli() -> None:
+    """Typer entry point."""
+    typer.run(main)
+
+
+if __name__ == "__main__":
+    cli()

python/prompt_bench/finetune.py

@@ -0,0 +1,214 @@
+"""Fine-tune Qwen 3.5 4B on bill summarization data using Unsloth.
+
+Loads a ChatML-style JSONL dataset (system/user/assistant messages),
+applies QLoRA with 4-bit quantization, and saves the merged model
+in HuggingFace format. Designed for a single RTX 3090 (24GB).
+
+Usage:
+    python -m python.prompt_bench.finetune \
+        --dataset output/finetune_dataset.jsonl \
+        --output-dir output/qwen-bill-summarizer
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Annotated
+
+import tomllib
+import typer
+from unsloth import FastLanguageModel
+from datasets import Dataset
+from transformers import TrainingArguments
+from trl import SFTTrainer
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class LoraConfig:
+    """LoRA adapter hyperparameters."""
+
+    rank: int
+    alpha: int
+    dropout: float
+    targets: list[str]
+
+
+@dataclass
+class TrainingConfig:
+    """Training loop hyperparameters."""
+
+    learning_rate: float
+    epochs: int
+    batch_size: int
+    gradient_accumulation: int
+    max_seq_length: int
+    warmup_ratio: float
+    weight_decay: float
+    logging_steps: int
+    save_steps: int
+
+
+@dataclass
+class FinetuneConfig:
+    """Top-level finetune configuration."""
+
+    base_model: str
+    lora: LoraConfig
+    training: TrainingConfig
+
+    @classmethod
+    def from_toml(cls, config_path: Path) -> FinetuneConfig:
+        """Load finetune config from a TOML file."""
+        raw = tomllib.loads(config_path.read_text())["finetune"]
+        return cls(
+            base_model=raw["base_model"],
+            lora=LoraConfig(**raw["lora"]),
+            training=TrainingConfig(**raw["training"]),
+        )
+
+
+def _messages_to_chatml(messages: list[dict]) -> str:
+    r"""Convert a message list to Qwen ChatML format.
+
+    Produces:
+        <|im_start|>system\n...\n<|im_end|>
+        <|im_start|>user\n...\n<|im_end|>
+        <|im_start|>assistant\n...\n<|im_end|>
+    """
+    parts = []
+    for message in messages:
+        role = message["role"]
+        content = message["content"]
+        parts.append(f"<|im_start|>{role}\n{content}<|im_end|>")
+    return "\n".join(parts)
+
+
+def load_dataset_from_jsonl(path: Path) -> Dataset:
+    """Load a ChatML JSONL file into a HuggingFace Dataset.
+
+    Each line must have {"messages": [{"role": ..., "content": ...}, ...]}.
+    Pre-formats into a `text` column with the Qwen ChatML template applied,
+    which SFTTrainer consumes directly.
+    """
+    records = []
+    with path.open(encoding="utf-8") as handle:
+        for raw_line in handle:
+            stripped = raw_line.strip()
+            if stripped:
+                entry = json.loads(stripped)
+                records.append({"text": _messages_to_chatml(entry["messages"])})
+    logger.info("Loaded %d examples from %s", len(records), path)
+    return Dataset.from_list(records)
+
+
+def main(
+    dataset_path: Annotated[Path, typer.Option("--dataset", help="Fine-tuning JSONL")] = Path(
+        "output/finetune_dataset.jsonl",
+    ),
+    validation_split: Annotated[float, typer.Option("--val-split", help="Fraction held out for validation")] = 0.1,
+    output_dir: Annotated[Path, typer.Option("--output-dir", help="Where to save the merged model")] = Path(
+        "output/qwen-bill-summarizer",
+    ),
+    config_path: Annotated[
+        Path,
+        typer.Option("--config", help="TOML config file"),
+    ] = Path(__file__).parent / "config.toml",
+    save_gguf: Annotated[bool, typer.Option("--save-gguf/--no-save-gguf", help="Also save GGUF")] = False,
+) -> None:
+    """Fine-tune Qwen 3.5 4B on bill summarization with Unsloth + QLoRA."""
+    logging.basicConfig(level="INFO", format="%(asctime)s %(levelname)s %(name)s: %(message)s")
+
+    if not dataset_path.is_file():
+        message = f"Dataset not found: {dataset_path}"
+        raise typer.BadParameter(message)
+
+    config = FinetuneConfig.from_toml(config_path)
+
+    logger.info("Loading base model: %s", config.base_model)
+    model, tokenizer = FastLanguageModel.from_pretrained(
+        model_name=config.base_model,
+        max_seq_length=config.training.max_seq_length,
+        load_in_4bit=True,
+        dtype=None,
+    )
+
+    logger.info("Applying LoRA (rank=%d, alpha=%d)", config.lora.rank, config.lora.alpha)
+    model = FastLanguageModel.get_peft_model(
+        model,
+        r=config.lora.rank,
+        lora_alpha=config.lora.alpha,
+        lora_dropout=config.lora.dropout,
+        target_modules=config.lora.targets,
+        bias="none",
+        use_gradient_checkpointing="unsloth",
+        random_state=42,
+    )
+
+    full_dataset = load_dataset_from_jsonl(dataset_path)
+    split = full_dataset.train_test_split(test_size=validation_split, seed=42)
+    train_dataset = split["train"]
+    validation_dataset = split["test"]
+    logger.info("Split: %d train, %d validation", len(train_dataset), len(validation_dataset))
+    training_args = TrainingArguments(
+        output_dir=str(output_dir / "checkpoints"),
+        num_train_epochs=config.training.epochs,
+        per_device_train_batch_size=config.training.batch_size,
+        gradient_accumulation_steps=config.training.gradient_accumulation,
+        learning_rate=config.training.learning_rate,
+        warmup_ratio=config.training.warmup_ratio,
+        weight_decay=config.training.weight_decay,
+        lr_scheduler_type="cosine",
+        logging_steps=config.training.logging_steps,
+        save_steps=config.training.save_steps,
+        save_total_limit=3,
+        eval_strategy="steps",
+        eval_steps=config.training.save_steps,
+        load_best_model_at_end=True,
+        bf16=True,
+        optim="adamw_8bit",
+        seed=42,
+        report_to="none",
+    )
+
+    trainer = SFTTrainer(
+        model=model,
+        tokenizer=tokenizer,
+        train_dataset=train_dataset,
+        eval_dataset=validation_dataset,
+        args=training_args,
+        max_seq_length=config.training.max_seq_length,
+        packing=True,
+    )
+
+    logger.info(
+        "Starting training: %d train, %d val, %d epochs",
+        len(train_dataset),
+        len(validation_dataset),
+        config.training.epochs,
+    )
+    trainer.train()
+
+    merged_path = str(output_dir / "merged")
+    logger.info("Saving merged model to %s", merged_path)
+    model.save_pretrained_merged(merged_path, tokenizer, save_method="merged_16bit")
+
+    if save_gguf:
+        gguf_path = str(output_dir / "gguf")
+        logger.info("Saving GGUF to %s", gguf_path)
+        model.save_pretrained_gguf(gguf_path, tokenizer, quantization_method="q4_k_m")
+
+    logger.info("Done! Model saved to %s", output_dir)
+
+
+def cli() -> None:
+    """Typer entry point."""
+    typer.run(main)
+
+
+if __name__ == "__main__":
+    cli()

python/prompt_bench/main.py

@@ -0,0 +1,215 @@
+"""CLI entry point for the prompt benchmarking system."""
+
+from __future__ import annotations
+
+import json
+import logging
+import time
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from pathlib import Path
+from typing import Annotated
+
+import typer
+
+from python.prompt_bench.containers.lib import check_gpu_free
+from python.prompt_bench.containers.vllm import start_vllm, stop_vllm
+from python.prompt_bench.downloader import is_model_present
+from python.prompt_bench.models import BenchmarkConfig
+from python.prompt_bench.vllm_client import VLLMClient
+
+logger = logging.getLogger(__name__)
+
+
+def discover_prompts(input_dir: Path) -> list[Path]:
+    """Find all .txt files in the input directory."""
+    prompts = list(input_dir.glob("*.txt"))
+    if not prompts:
+        message = f"No .txt files found in {input_dir}"
+        raise FileNotFoundError(message)
+    return prompts
+
+
+def _run_prompt(
+    client: VLLMClient,
+    prompt_path: Path,
+    *,
+    repo: str,
+    model_dir_name: str,
+    model_output: Path,
+    temperature: float,
+) -> tuple[bool, float]:
+    """Run a single prompt. Returns (success, elapsed_seconds)."""
+    filename = prompt_path.name
+    output_path = model_output / filename
+    start = time.monotonic()
+    try:
+        prompt_text = prompt_path.read_text()
+        response = client.complete(prompt_text, model_dir_name, temperature=temperature)
+        output_path.write_text(response)
+        elapsed = time.monotonic() - start
+        logger.info("Completed: %s / %s in %.2fs", repo, filename, elapsed)
+    except Exception:
+        elapsed = time.monotonic() - start
+        error_path = model_output / f"{filename}.error"
+        logger.exception("Failed: %s / %s after %.2fs", repo, filename, elapsed)
+        error_path.write_text(f"Error processing {filename}")
+        return False, elapsed
+    return True, elapsed
+
+
+def benchmark_model(
+    client: VLLMClient,
+    prompts: list[Path],
+    *,
+    repo: str,
+    model_dir_name: str,
+    model_output: Path,
+    temperature: float,
+    concurrency: int,
+) -> tuple[int, int]:
+    """Run all prompts against a single model in parallel.
+
+    vLLM batches concurrent requests internally, so submitting many at once is
+    significantly faster than running them serially.
+    """
+    pending = [prompt for prompt in prompts if not (model_output / prompt.name).exists()]
+    skipped = len(prompts) - len(pending)
+    if skipped:
+        logger.info("Skipping %d prompts with existing output for %s", skipped, repo)
+
+    if not pending:
+        logger.info("Nothing to do for %s", repo)
+        return 0, 0
+
+    completed = 0
+    failed = 0
+    latencies: list[float] = []
+
+    wall_start = time.monotonic()
+    with ThreadPoolExecutor(max_workers=concurrency) as executor:
+        futures = [
+            executor.submit(
+                _run_prompt,
+                client,
+                prompt_path,
+                repo=repo,
+                model_dir_name=model_dir_name,
+                model_output=model_output,
+                temperature=temperature,
+            )
+            for prompt_path in pending
+        ]
+        for future in as_completed(futures):
+            success, elapsed = future.result()
+            latencies.append(elapsed)
+            if success:
+                completed += 1
+            else:
+                failed += 1
+    wall_elapsed = time.monotonic() - wall_start
+
+    attempted = completed + failed
+    avg_latency = sum(latencies) / attempted
+    throughput = attempted / wall_elapsed if wall_elapsed > 0 else 0.0
+    timing = {
+        "repo": repo,
+        "wall_seconds": wall_elapsed,
+        "attempted": attempted,
+        "completed": completed,
+        "failed": failed,
+        "avg_latency_seconds": avg_latency,
+        "throughput_prompts_per_second": throughput,
+        "concurrency": concurrency,
+    }
+    timing_path = model_output / "_timing.json"
+    timing_path.write_text(json.dumps(timing, indent=2))
+
+    return completed, failed
+
+
+def run_benchmark(
+    config: BenchmarkConfig,
+    input_dir: Path,
+    output_dir: Path,
+) -> None:
+    """Execute the benchmark across all models and prompts."""
+    prompts = discover_prompts(input_dir)
+    logger.info("Found %d prompts in %s", len(prompts), input_dir)
+
+    check_gpu_free()
+
+    total_completed = 0
+    total_failed = 0
+
+    for repo in config.models:
+        if not is_model_present(repo, config.model_dir):
+            logger.warning("Skipping (not downloaded): %s", repo)
+            continue
+
+        model_output = output_dir / repo
+        model_output.mkdir(parents=True, exist_ok=True)
+
+        logger.info("=== Benchmarking model: %s ===", repo)
+
+        stop_vllm()
+        try:
+            start_vllm(
+                model=repo,
+                port=config.port,
+                model_dir=config.model_dir,
+                gpu_memory_utilization=config.gpu_memory_utilization,
+            )
+        except RuntimeError:
+            logger.exception("Failed to start vLLM for %s, skipping", repo)
+            continue
+        logger.info("vLLM started for %s", repo)
+        try:
+            with VLLMClient(port=config.port, timeout=config.timeout) as client:
+                client.wait_ready(max_wait=config.vllm_startup_timeout)
+                completed, failed = benchmark_model(
+                    client,
+                    prompts,
+                    repo=repo,
+                    model_dir_name=repo,
+                    model_output=model_output,
+                    temperature=config.temperature,
+                    concurrency=config.concurrency,
+                )
+                total_completed += completed
+                total_failed += failed
+        finally:
+            stop_vllm()
+
+    logger.info("=== Benchmark complete ===")
+    logger.info("Completed: %d | Failed: %d", total_completed, total_failed)
+
+
+def main(
+    input_dir: Annotated[Path, typer.Argument(help="Directory containing input .txt prompt files")],
+    config: Annotated[Path, typer.Option(help="Path to TOML config file")] = Path("bench.toml"),
+    output_dir: Annotated[Path, typer.Option(help="Output directory for results")] = Path("output"),
+    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
+) -> None:
+    """Run prompts through multiple LLMs via vLLM and save results."""
+    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
+
+    if not input_dir.is_dir():
+        message = f"Input directory does not exist: {input_dir}"
+        raise typer.BadParameter(message)
+    if not config.is_file():
+        message = f"Config file does not exist: {config}"
+        raise typer.BadParameter(message)
+
+    benchmark_config = BenchmarkConfig.from_toml(config)
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    run_benchmark(benchmark_config, input_dir, output_dir)
+
+
+def cli() -> None:
+    """Typer entry point."""
+    typer.run(main)
+
+
+if __name__ == "__main__":
+    cli()

python/prompt_bench/models.py

@@ -0,0 +1,30 @@
+"""Pydantic models for benchmark configuration."""
+
+from __future__ import annotations
+
+import tomllib
+from typing import TYPE_CHECKING
+
+from pydantic import BaseModel
+
+if TYPE_CHECKING:
+    from pathlib import Path
+
+
+class BenchmarkConfig(BaseModel):
+    """Top-level benchmark configuration loaded from TOML."""
+
+    models: list[str]
+    model_dir: str = "/zfs/models/hf"
+    port: int = 8000
+    gpu_memory_utilization: float = 0.90
+    temperature: float = 0.0
+    timeout: int = 300
+    concurrency: int = 4
+    vllm_startup_timeout: int = 900
+
+    @classmethod
+    def from_toml(cls, config_path: Path) -> BenchmarkConfig:
+        """Load benchmark config from a TOML file."""
+        raw = tomllib.loads(config_path.read_text())["bench"]
+        return cls(**raw)

python/prompt_bench/summarization_prompts.py

@@ -0,0 +1,34 @@
+SUMMARIZATION_SYSTEM_PROMPT = """You are a legislative analyst extracting policy substance from Congressional bill text.
+
+Your job is to compress a bill into a dense, neutral structured summary that captures every distinct policy action — including secondary effects that might be buried in subsections.
+
+EXTRACTION RULES:
+- IGNORE: whereas clauses, congressional findings that are purely political statements, recitals, preambles, citations of existing law by number alone, and procedural boilerplate.
+- FOCUS ON: operative verbs — what the bill SHALL do, PROHIBIT, REQUIRE, AUTHORIZE, AMEND, APPROPRIATE, or ESTABLISH.
+- SURFACE ALL THREADS: If the bill touches multiple policy areas, list each thread separately. Do not collapse them.
+- BE CONCRETE: Name the affected population, the mechanism, and the direction (expands/restricts/maintains).
+- STAY NEUTRAL: No political framing. Describe what the text does, not what its sponsors claim it does.
+
+OUTPUT FORMAT — plain structured text, not JSON:
+
+OPERATIVE ACTIONS:
+[Numbered list of what the bill actually does, one action per line, max 20 words each]
+
+AFFECTED POPULATIONS:
+[Who gains something, who loses something, or whose behavior is regulated]
+
+MECHANISMS:
+[How it works: new funding, mandate, prohibition, amendment to existing statute, grant program, study commission, etc.]
+
+POLICY THREADS:
+[List each distinct policy domain this bill touches, even minor ones. Use plain language, not domain codes.]
+
+SYMBOLIC/PROCEDURAL ONLY:
+[Yes or No — is this bill primarily a resolution, designation, or awareness declaration with no operative effect?]
+
+LENGTH TARGET: 150-250 words total. Be ruthless about cutting. Density over completeness."""
+
+SUMMARIZATION_USER_TEMPLATE = """Summarize the following Congressional bill according to your instructions.
+
+BILL TEXT:
+{text_content}"""

python/prompt_bench/tools/build_finetune_dataset.py

@@ -0,0 +1,114 @@
+"""Build a fine-tuning JSONL dataset from batch request + output files.
+
+Joins the original request JSONL (system + user messages) with the batch
+output JSONL (assistant completions) by custom_id to produce a ChatML-style
+messages JSONL suitable for fine-tuning.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+from pathlib import Path
+from typing import Annotated
+
+import typer
+
+logger = logging.getLogger(__name__)
+
+HTTP_OK = 200
+
+
+def load_requests(path: Path) -> dict[str, list[dict]]:
+    """Parse request JSONL into {custom_id: messages}."""
+    results: dict[str, list[dict]] = {}
+    with path.open(encoding="utf-8") as handle:
+        for raw_line in handle:
+            stripped = raw_line.strip()
+            if not stripped:
+                continue
+            record = json.loads(stripped)
+            custom_id = record["custom_id"]
+            messages = record["body"]["messages"]
+            results[custom_id] = messages
+    return results
+
+
+def load_completions(path: Path) -> dict[str, str]:
+    """Parse batch output JSONL into {custom_id: assistant_content}."""
+    results: dict[str, str] = {}
+    with path.open(encoding="utf-8") as handle:
+        for line_number, raw_line in enumerate(handle, 1):
+            stripped = raw_line.strip()
+            if not stripped:
+                continue
+            record = json.loads(stripped)
+            custom_id = record["custom_id"]
+            response = record.get("response", {})
+            if response.get("status_code") != HTTP_OK:
+                logger.warning("Skipping %s (line %d): status %s", custom_id, line_number, response.get("status_code"))
+                continue
+            body = response.get("body", {})
+            choices = body.get("choices", [])
+            if not choices:
+                logger.warning("Skipping %s (line %d): no choices", custom_id, line_number)
+                continue
+            content = choices[0].get("message", {}).get("content", "")
+            if not content:
+                logger.warning("Skipping %s (line %d): empty content", custom_id, line_number)
+                continue
+            results[custom_id] = content
+    return results
+
+
+def main(
+    requests_path: Annotated[Path, typer.Option("--requests", help="Batch request JSONL")] = Path(
+        "output/openai_batch/requests.jsonl",
+    ),
+    batch_output: Annotated[Path, typer.Option("--batch-output", help="Batch output JSONL")] = Path(
+        "batch_69d84558d91c819091d53f08d78f9fd6_output.jsonl",
+    ),
+    output_path: Annotated[Path, typer.Option("--output", help="Fine-tuning JSONL output")] = Path(
+        "output/finetune_dataset.jsonl",
+    ),
+    log_level: Annotated[str, typer.Option(help="Log level")] = "INFO",
+) -> None:
+    """Build fine-tuning dataset by joining request and output JSONL files."""
+    logging.basicConfig(level=log_level, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
+
+    logger.info("Loading requests from %s", requests_path)
+    requests = load_requests(requests_path)
+    logger.info("Loaded %d requests", len(requests))
+
+    logger.info("Loading completions from %s", batch_output)
+    completions = load_completions(batch_output)
+    logger.info("Loaded %d completions", len(completions))
+
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    matched = 0
+    skipped = 0
+
+    with output_path.open("w", encoding="utf-8") as handle:
+        for custom_id, messages in requests.items():
+            assistant_content = completions.get(custom_id)
+            if assistant_content is None:
+                skipped += 1
+                continue
+
+            example = {
+                "messages": [*messages, {"role": "assistant", "content": assistant_content}],
+            }
+            handle.write(json.dumps(example, ensure_ascii=False))
+            handle.write("\n")
+            matched += 1
+
+    logger.info("Wrote %d examples to %s (skipped %d unmatched)", matched, output_path, skipped)
+
+
+def cli() -> None:
+    """Typer entry point."""
+    typer.run(main)
+
+
+if __name__ == "__main__":
+    cli()

python/prompt_bench/tools/count_tokens.py

@@ -0,0 +1,97 @@
+"""Sum token usage across compressed and uncompressed run directories."""
+
+from __future__ import annotations
+
+import json
+import logging
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Annotated
+
+import typer
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class UsageTotals:
+    """Aggregate usage counters for a directory of run records."""
+
+    files: int = 0
+    errors: int = 0
+    prompt_tokens: int = 0
+    cached_tokens: int = 0
+    completion_tokens: int = 0
+    reasoning_tokens: int = 0
+    total_tokens: int = 0
+    per_file: list[tuple[str, int, int, int]] = field(default_factory=list)
+
+
+def tally_directory(directory: Path) -> UsageTotals:
+    """Return aggregated usage stats for every JSON record in a directory."""
+    totals = UsageTotals()
+    decoder = json.JSONDecoder()
+    for path in sorted(directory.glob("*.json")):
+        text = path.read_text().lstrip()
+        record, _ = decoder.raw_decode(text)
+        totals.files += 1
+        usage = record.get("usage")
+        if not usage:
+            totals.errors += 1
+            continue
+        prompt_tokens = usage.get("prompt_tokens", 0)
+        completion_tokens = usage.get("completion_tokens", 0)
+        total_tokens = usage.get("total_tokens", 0)
+        cached_tokens = (usage.get("prompt_tokens_details") or {}).get("cached_tokens", 0)
+        reasoning_tokens = (usage.get("completion_tokens_details") or {}).get("reasoning_tokens", 0)
+        totals.prompt_tokens += prompt_tokens
+        totals.completion_tokens += completion_tokens
+        totals.total_tokens += total_tokens
+        totals.cached_tokens += cached_tokens
+        totals.reasoning_tokens += reasoning_tokens
+        totals.per_file.append((path.name, prompt_tokens, completion_tokens, total_tokens))
+    return totals
+
+
+def log_totals(label: str, totals: UsageTotals) -> None:
+    """Log a one-block summary for a directory."""
+    counted = totals.files - totals.errors
+    average_total = totals.total_tokens / counted if counted else 0
+    logger.info("[%s]", label)
+    logger.info("  files          : %d (with usage: %d, errors: %d)", totals.files, counted, totals.errors)
+    logger.info("  prompt tokens  : %d", totals.prompt_tokens)
+    logger.info("  cached tokens  : %d", totals.cached_tokens)
+    logger.info("  completion tok : %d", totals.completion_tokens)
+    logger.info("  reasoning tok  : %d", totals.reasoning_tokens)
+    logger.info("  total tokens   : %d", totals.total_tokens)
+    logger.info("  avg total/file : %.1f", average_total)
+
+
+def main(
+    runs_dir: Annotated[Path, typer.Option("--runs-dir")] = Path("output/openai_runs_temp_1"),
+    log_level: Annotated[str, typer.Option("--log-level")] = "INFO",
+) -> None:
+    """Print token usage totals for the compressed and uncompressed run directories."""
+    logging.basicConfig(level=log_level, format="%(message)s")
+
+    grand = UsageTotals()
+    for label in ("compressed", "uncompressed"):
+        directory = runs_dir / label
+        if not directory.is_dir():
+            logger.warning("%s: directory not found at %s", label, directory)
+            continue
+        totals = tally_directory(directory)
+        log_totals(label, totals)
+        grand.files += totals.files
+        grand.errors += totals.errors
+        grand.prompt_tokens += totals.prompt_tokens
+        grand.cached_tokens += totals.cached_tokens
+        grand.completion_tokens += totals.completion_tokens
+        grand.reasoning_tokens += totals.reasoning_tokens
+        grand.total_tokens += totals.total_tokens
+
+    log_totals("grand total", grand)
+
+
+if __name__ == "__main__":
+    typer.run(main)

python/prompt_bench/vllm_client.py

@@ -0,0 +1,68 @@
+"""OpenAI-compatible client for vLLM's API."""
+
+from __future__ import annotations
+
+import logging
+import time
+from typing import Self
+
+import httpx
+
+logger = logging.getLogger(__name__)
+
+READY_POLL_INTERVAL = 2.0
+
+
+class VLLMClient:
+    """Talk to a vLLM server via its OpenAI-compatible API.
+
+    Args:
+        host: vLLM host.
+        port: vLLM port.
+        timeout: Per-request timeout in seconds.
+    """
+
+    def __init__(self, *, host: str = "localhost", port: int = 8000, timeout: int = 300) -> None:
+        """Create a client connected to a vLLM server."""
+        self._client = httpx.Client(base_url=f"http://{host}:{port}", timeout=timeout)
+
+    def wait_ready(self, max_wait: int) -> None:
+        """Poll /v1/models until the server is ready or timeout."""
+        deadline = time.monotonic() + max_wait
+        while time.monotonic() < deadline:
+            try:
+                response = self._client.get("/v1/models")
+                if response.is_success:
+                    logger.info("vLLM server is ready")
+                    return
+            except httpx.TransportError:
+                pass
+            time.sleep(READY_POLL_INTERVAL)
+        msg = f"vLLM server not ready after {max_wait}s"
+        raise TimeoutError(msg)
+
+    def complete(self, prompt: str, model: str, *, temperature: float = 0.0, max_tokens: int = 4096) -> str:
+        """Send a prompt to /v1/completions and return the response text."""
+        payload = {
+            "model": model,
+            "prompt": prompt,
+            "temperature": temperature,
+            "max_tokens": max_tokens,
+        }
+        logger.info("Sending prompt to %s (%d chars)", model, len(prompt))
+        response = self._client.post("/v1/completions", json=payload)
+        response.raise_for_status()
+        data = response.json()
+        return data["choices"][0]["text"]
+
+    def close(self) -> None:
+        """Close the HTTP client."""
+        self._client.close()
+
+    def __enter__(self) -> Self:
+        """Enter the context manager."""
+        return self
+
+    def __exit__(self, *args: object) -> None:
+        """Close the HTTP client on exit."""
+        self.close()

python/signal_bot/__init__.py

@@ -0,0 +1 @@
+"""Signal command and control bot."""

python/signal_bot/commands/__init__.py

@@ -0,0 +1 @@
+"""Signal bot commands."""

python/signal_bot/commands/inventory.py

@@ -0,0 +1,137 @@
+"""Van inventory command — parse receipts and item lists via LLM, push to API."""
+
+from __future__ import annotations
+
+import json
+import logging
+from typing import TYPE_CHECKING, Any
+
+import httpx
+
+from python.signal_bot.models import InventoryItem, InventoryUpdate
+
+if TYPE_CHECKING:
+    from python.signal_bot.llm_client import LLMClient
+    from python.signal_bot.models import SignalMessage
+    from python.signal_bot.signal_client import SignalClient
+
+logger = logging.getLogger(__name__)
+
+SYSTEM_PROMPT = """\
+You are an inventory assistant. Extract items from the input and return ONLY
+a JSON array. Each element must have these fields:
+  - "name": item name (string)
+  - "quantity": numeric count or amount (default 1)
+  - "unit": unit of measure (e.g. "each", "lb", "oz", "gallon", "bag", "box")
+  - "category": category like "food", "tools", "supplies", etc.
+  - "notes": any extra detail (empty string if none)
+
+Example output:
+[{"name": "water bottles", "quantity": 6, "unit": "gallon", "category": "supplies", "notes": "1 gallon each"}]
+
+Return ONLY the JSON array, no other text.\
+"""
+
+IMAGE_PROMPT = "Extract all items from this receipt or inventory photo."
+TEXT_PROMPT = "Extract all items from this inventory list."
+
+
+def parse_llm_response(raw: str) -> list[InventoryItem]:
+    """Parse the LLM JSON response into InventoryItem list."""
+    text = raw.strip()
+    # Strip markdown code fences if present
+    if text.startswith("```"):
+        lines = text.split("\n")
+        lines = [line for line in lines if not line.startswith("```")]
+        text = "\n".join(lines)
+
+    items_data: list[dict[str, Any]] = json.loads(text)
+    return [InventoryItem.model_validate(item) for item in items_data]
+
+
+def _upsert_item(api_url: str, item: InventoryItem) -> None:
+    """Create or update an item via the van_inventory API.
+
+    Fetches existing items, and if one with the same name exists,
+    patches its quantity (summing). Otherwise creates a new item.
+    """
+    base = api_url.rstrip("/")
+    response = httpx.get(f"{base}/api/items", timeout=10)
+    response.raise_for_status()
+    existing: list[dict[str, Any]] = response.json()
+
+    match = next((e for e in existing if e["name"].lower() == item.name.lower()), None)
+
+    if match:
+        new_qty = match["quantity"] + item.quantity
+        patch = {"quantity": new_qty}
+        if item.category:
+            patch["category"] = item.category
+        response = httpx.patch(f"{base}/api/items/{match['id']}", json=patch, timeout=10)
+        response.raise_for_status()
+        return
+    payload = {
+        "name": item.name,
+        "quantity": item.quantity,
+        "unit": item.unit,
+        "category": item.category or None,
+    }
+    response = httpx.post(f"{base}/api/items", json=payload, timeout=10)
+    response.raise_for_status()
+
+
+def handle_inventory_update(
+    message: SignalMessage,
+    signal: SignalClient,
+    llm: LLMClient,
+    api_url: str,
+) -> InventoryUpdate:
+    """Process an inventory update from a Signal message.
+
+    Accepts either an image (receipt photo) or text list.
+    Uses the LLM to extract structured items, then pushes to the van_inventory API.
+    """
+    try:
+        logger.info(f"Processing inventory update from {message.source}")
+        if message.attachments:
+            image_data = signal.get_attachment(message.attachments[0])
+            raw_response = llm.chat(
+                IMAGE_PROMPT,
+                image_data=image_data,
+                system=SYSTEM_PROMPT,
+            )
+            source_type = "receipt_photo"
+        elif message.message.strip():
+            raw_response = llm.chat(
+                f"{TEXT_PROMPT}\n\n{message.message}",
+                system=SYSTEM_PROMPT,
+            )
+            source_type = "text_list"
+        else:
+            signal.reply(message, "Send a photo of a receipt or a text list of items to update inventory.")
+            return InventoryUpdate()
+
+        logger.info(f"{raw_response=}")
+
+        new_items = parse_llm_response(raw_response)
+
+        logger.info(f"{new_items=}")
+
+        for item in new_items:
+            _upsert_item(api_url, item)
+
+        summary = _format_summary(new_items)
+        signal.reply(message, f"Inventory updated with {len(new_items)} item(s):\n{summary}")
+
+        return InventoryUpdate(items=new_items, raw_response=raw_response, source_type=source_type)
+
+    except Exception:
+        logger.exception("Failed to process inventory update")
+        signal.reply(message, "Failed to process inventory update. Check logs for details.")
+        return InventoryUpdate()
+
+
+def _format_summary(items: list[InventoryItem]) -> str:
+    """Format items into a readable summary."""
+    lines = [f"  - {item.name} x{item.quantity} {item.unit} [{item.category}]" for item in items]
+    return "\n".join(lines)

python/signal_bot/commands/location.py

@@ -0,0 +1,64 @@
+"""Location command for the Signal bot."""
+
+from __future__ import annotations
+
+import logging
+from typing import TYPE_CHECKING, Any
+
+import httpx
+
+if TYPE_CHECKING:
+    from python.signal_bot.models import SignalMessage
+    from python.signal_bot.signal_client import SignalClient
+
+logger = logging.getLogger(__name__)
+
+
+def _get_entity_state(ha_url: str, ha_token: str, entity_id: str) -> dict[str, Any]:
+    """Fetch an entity's state from Home Assistant."""
+    entity_url = f"{ha_url}/api/states/{entity_id}"
+    logger.debug(f"Fetching {entity_url=}")
+    response = httpx.get(
+        entity_url,
+        headers={"Authorization": f"Bearer {ha_token}"},
+        timeout=30,
+    )
+    response.raise_for_status()
+    return response.json()
+
+
+def _format_location(latitude: str, longitude: str) -> str:
+    """Render a friendly location response."""
+    return f"Van location: {latitude}, {longitude}\nhttps://maps.google.com/?q={latitude},{longitude}"
+
+
+def handle_location_request(
+    message: SignalMessage,
+    signal: SignalClient,
+    ha_url: str | None,
+    ha_token: str | None,
+) -> None:
+    """Reply with van location from Home Assistant."""
+    if ha_url is None or ha_token is None:
+        signal.reply(message, "Location command is not configured (missing HA_URL or HA_TOKEN).")
+        return
+
+    lat_payload = None
+    lon_payload = None
+    try:
+        lat_payload = _get_entity_state(ha_url, ha_token, "sensor.van_last_known_latitude")
+        lon_payload = _get_entity_state(ha_url, ha_token, "sensor.van_last_known_longitude")
+    except httpx.HTTPError:
+        logger.exception("Couldn't fetch van location from Home Assistant right now.")
+        logger.debug(f"{ha_url=} {lat_payload=} {lon_payload=}")
+        signal.reply(message, "Couldn't fetch van location from Home Assistant right now.")
+        return
+
+    latitude = lat_payload.get("state", "")
+    longitude = lon_payload.get("state", "")
+
+    if not latitude or not longitude or latitude == "unavailable" or longitude == "unavailable":
+        signal.reply(message, "Van location is unavailable in Home Assistant right now.")
+        return
+
+    signal.reply(message, _format_location(latitude, longitude))

python/signal_bot/device_registry.py

@@ -0,0 +1,284 @@
+"""Device registry — tracks verified/unverified devices by safety number."""
+
+from __future__ import annotations
+
+import logging
+from datetime import datetime, timedelta
+from typing import TYPE_CHECKING, NamedTuple
+
+from sqlalchemy import delete, select
+from sqlalchemy.orm import Session
+
+from python.common import utcnow
+from python.orm.signal_bot.models import RoleRecord, SignalDevice
+from python.signal_bot.models import Role, TrustLevel
+
+if TYPE_CHECKING:
+    from sqlalchemy.engine import Engine
+
+    from python.signal_bot.signal_client import SignalClient
+
+logger = logging.getLogger(__name__)
+
+_BLOCKED_TTL = timedelta(minutes=60)
+_DEFAULT_TTL = timedelta(minutes=5)
+
+
+class _CacheEntry(NamedTuple):
+    expires: datetime
+    trust_level: TrustLevel
+    has_safety_number: bool
+    safety_number: str | None
+    roles: list[Role]
+
+
+class DeviceRegistry:
+    """Manage device trust based on Signal safety numbers.
+
+    Devices start as UNVERIFIED. An admin verifies them over SSH by calling
+    ``verify(phone_number)`` which marks the device VERIFIED and also tells
+    signal-cli to trust the identity.
+
+    Only VERIFIED devices may execute commands.
+    """
+
+    def __init__(self, signal_client: SignalClient, engine: Engine) -> None:
+        self.signal_client = signal_client
+        self.engine = engine
+        self._contact_cache: dict[str, _CacheEntry] = {}
+
+    def is_verified(self, phone_number: str) -> bool:
+        """Check if a phone number is verified."""
+        if entry := self._cached(phone_number):
+            return entry.trust_level == TrustLevel.VERIFIED
+        device = self._load_device(phone_number)
+        return device is not None and device.trust_level == TrustLevel.VERIFIED
+
+    def record_contact(self, phone_number: str, safety_number: str | None = None) -> None:
+        """Record seeing a device. Creates entry if new, updates last_seen."""
+        now = utcnow()
+
+        entry = self._cached(phone_number)
+        if entry and entry.safety_number == safety_number:
+            return
+
+        with Session(self.engine) as session:
+            device = session.scalars(
+                select(SignalDevice).where(SignalDevice.phone_number == phone_number)
+            ).one_or_none()
+
+            if device:
+                if device.safety_number != safety_number and device.trust_level != TrustLevel.BLOCKED:
+                    logger.warning(f"Safety number changed for {phone_number}, resetting to UNVERIFIED")
+                    device.safety_number = safety_number
+                    device.trust_level = TrustLevel.UNVERIFIED
+                device.last_seen = now
+            else:
+                device = SignalDevice(
+                    phone_number=phone_number,
+                    safety_number=safety_number,
+                    trust_level=TrustLevel.UNVERIFIED,
+                    last_seen=now,
+                )
+                session.add(device)
+                logger.info(f"New device registered: {phone_number}")
+
+            session.commit()
+            self._update_cache(phone_number, device)
+
+    def has_safety_number(self, phone_number: str) -> bool:
+        """Check if a device has a safety number on file."""
+        if entry := self._cached(phone_number):
+            return entry.has_safety_number
+        device = self._load_device(phone_number)
+        return device is not None and device.safety_number is not None
+
+    def verify(self, phone_number: str) -> bool:
+        """Mark a device as verified. Called by admin over SSH.
+
+        Returns True if the device was found and verified.
+        """
+        with Session(self.engine) as session:
+            device = session.scalars(
+                select(SignalDevice).where(SignalDevice.phone_number == phone_number)
+            ).one_or_none()
+
+            if not device:
+                logger.warning(f"Cannot verify unknown device: {phone_number}")
+                return False
+
+            device.trust_level = TrustLevel.VERIFIED
+            self.signal_client.trust_identity(phone_number, trust_all_known_keys=True)
+            session.commit()
+            self._update_cache(phone_number, device)
+            logger.info(f"Device verified: {phone_number}")
+            return True
+
+    def block(self, phone_number: str) -> bool:
+        """Block a device."""
+        return self._set_trust(phone_number, TrustLevel.BLOCKED, "Device blocked")
+
+    def unverify(self, phone_number: str) -> bool:
+        """Reset a device to unverified."""
+        return self._set_trust(phone_number, TrustLevel.UNVERIFIED)
+
+    # -- role management ------------------------------------------------------
+
+    def get_roles(self, phone_number: str) -> list[Role]:
+        """Return the roles for a device, defaulting to empty."""
+        if entry := self._cached(phone_number):
+            return entry.roles
+        device = self._load_device(phone_number)
+        return _extract_roles(device) if device else []
+
+    def has_role(self, phone_number: str, role: Role) -> bool:
+        """Check if a device has a specific role or is admin."""
+        roles = self.get_roles(phone_number)
+        return Role.ADMIN in roles or role in roles
+
+    def grant_role(self, phone_number: str, role: Role) -> bool:
+        """Add a role to a device. Called by admin over SSH."""
+        with Session(self.engine) as session:
+            device = session.scalars(
+                select(SignalDevice).where(SignalDevice.phone_number == phone_number)
+            ).one_or_none()
+
+            if not device:
+                logger.warning(f"Cannot grant role for unknown device: {phone_number}")
+                return False
+
+            if any(record.name == role for record in device.roles):
+                return True
+
+            role_record = session.scalars(select(RoleRecord).where(RoleRecord.name == role)).one_or_none()
+
+            if not role_record:
+                logger.warning(f"Unknown role: {role}")
+                return False
+
+            device.roles.append(role_record)
+            session.commit()
+            self._update_cache(phone_number, device)
+            logger.info(f"Device {phone_number} granted role {role}")
+            return True
+
+    def revoke_role(self, phone_number: str, role: Role) -> bool:
+        """Remove a role from a device. Called by admin over SSH."""
+        with Session(self.engine) as session:
+            device = session.scalars(
+                select(SignalDevice).where(SignalDevice.phone_number == phone_number)
+            ).one_or_none()
+
+            if not device:
+                logger.warning(f"Cannot revoke role for unknown device: {phone_number}")
+                return False
+
+            device.roles = [record for record in device.roles if record.name != role]
+            session.commit()
+            self._update_cache(phone_number, device)
+            logger.info(f"Device {phone_number} revoked role {role}")
+            return True
+
+    def set_roles(self, phone_number: str, roles: list[Role]) -> bool:
+        """Replace all roles for a device. Called by admin over SSH."""
+        with Session(self.engine) as session:
+            device = session.scalars(
+                select(SignalDevice).where(SignalDevice.phone_number == phone_number)
+            ).one_or_none()
+
+            if not device:
+                logger.warning(f"Cannot set roles for unknown device: {phone_number}")
+                return False
+
+            role_names = [str(role) for role in roles]
+            records = session.scalars(select(RoleRecord).where(RoleRecord.name.in_(role_names))).all()
+            device.roles = records
+            session.commit()
+            self._update_cache(phone_number, device)
+            logger.info(f"Device {phone_number} roles set to {role_names}")
+            return True
+
+    # -- queries --------------------------------------------------------------
+
+    def list_devices(self) -> list[SignalDevice]:
+        """Return all known devices."""
+        with Session(self.engine) as session:
+            return list(session.scalars(select(SignalDevice)).all())
+
+    def sync_identities(self) -> None:
+        """Pull identity list from signal-cli and record any new ones."""
+        identities = self.signal_client.get_identities()
+        for identity in identities:
+            number = identity.get("number", "")
+            safety = identity.get("safety_number", identity.get("fingerprint", ""))
+            if number:
+                self.record_contact(number, safety)
+
+    # -- internals ------------------------------------------------------------
+
+    def _cached(self, phone_number: str) -> _CacheEntry | None:
+        """Return the cache entry if it exists and hasn't expired."""
+        entry = self._contact_cache.get(phone_number)
+        if entry and utcnow() < entry.expires:
+            return entry
+        return None
+
+    def _load_device(self, phone_number: str) -> SignalDevice | None:
+        """Fetch a device by phone number (with joined roles)."""
+        with Session(self.engine) as session:
+            return session.scalars(select(SignalDevice).where(SignalDevice.phone_number == phone_number)).one_or_none()
+
+    def _update_cache(self, phone_number: str, device: SignalDevice) -> None:
+        """Refresh the cache entry for a device."""
+        ttl = _BLOCKED_TTL if device.trust_level == TrustLevel.BLOCKED else _DEFAULT_TTL
+        self._contact_cache[phone_number] = _CacheEntry(
+            expires=utcnow() + ttl,
+            trust_level=device.trust_level,
+            has_safety_number=device.safety_number is not None,
+            safety_number=device.safety_number,
+            roles=_extract_roles(device),
+        )
+
+    def _set_trust(self, phone_number: str, level: str, log_msg: str | None = None) -> bool:
+        """Update the trust level for a device."""
+        with Session(self.engine) as session:
+            device = session.scalars(
+                select(SignalDevice).where(SignalDevice.phone_number == phone_number)
+            ).one_or_none()
+
+            if not device:
+                return False
+
+            device.trust_level = level
+            session.commit()
+            self._update_cache(phone_number, device)
+            if log_msg:
+                logger.info(f"{log_msg}: {phone_number}")
+            return True
+
+
+def _extract_roles(device: SignalDevice) -> list[Role]:
+    """Convert a device's RoleRecord objects to a list of Role enums."""
+    return [Role(record.name) for record in device.roles]
+
+
+def sync_roles(engine: Engine) -> None:
+    """Sync the Role enum to the role table, adding new and removing stale entries."""
+    expected = {role.value for role in Role}
+
+    with Session(engine) as session:
+        existing = set(session.scalars(select(RoleRecord.name)).all())
+
+        to_add = expected - existing
+        to_remove = existing - expected
+
+        for name in to_add:
+            session.add(RoleRecord(name=name))
+            logger.info(f"Role added: {name}")
+
+        if to_remove:
+            session.execute(delete(RoleRecord).where(RoleRecord.name.in_(to_remove)))
+            for name in to_remove:
+                logger.info(f"Role removed: {name}")
+
+        session.commit()

python/signal_bot/llm_client.py

@@ -0,0 +1,80 @@
+"""Flexible LLM client for ollama backends."""
+
+from __future__ import annotations
+
+import base64
+import logging
+from typing import Any, Self
+
+import httpx
+
+logger = logging.getLogger(__name__)
+
+
+class LLMClient:
+    """Talk to an ollama instance.
+
+    Args:
+        model: Ollama model name.
+        host: Ollama host.
+        port: Ollama port.
+        temperature: Sampling temperature.
+    """
+
+    def __init__(
+        self,
+        *,
+        model: str,
+        host: str,
+        port: int = 11434,
+        temperature: float = 0.1,
+        timeout: int = 300,
+    ) -> None:
+        self.model = model
+        self.temperature = temperature
+        self._client = httpx.Client(base_url=f"http://{host}:{port}", timeout=timeout)
+
+    def chat(self, prompt: str, image_data: bytes | None = None, system: str | None = None) -> str:
+        """Send a text prompt and return the response."""
+        messages: list[dict[str, Any]] = []
+        if system:
+            messages.append({"role": "system", "content": system})
+
+        user_msg = {"role": "user", "content": prompt}
+        if image_data:
+            user_msg["images"] = [base64.b64encode(image_data).decode()]
+
+        messages.append(user_msg)
+        return self._generate(messages)
+
+    def _generate(self, messages: list[dict[str, Any]]) -> str:
+        """Call the ollama chat API."""
+        payload = {
+            "model": self.model,
+            "messages": messages,
+            "stream": False,
+            "options": {"temperature": self.temperature},
+        }
+        logger.info(f"LLM request to {self.model}")
+        response = self._client.post("/api/chat", json=payload)
+        response.raise_for_status()
+        data = response.json()
+        return data["message"]["content"]
+
+    def list_models(self) -> list[str]:
+        """List available models on the ollama instance."""
+        response = self._client.get("/api/tags")
+        response.raise_for_status()
+        return [m["name"] for m in response.json().get("models", [])]
+
+    def __enter__(self) -> Self:
+        """Enter the context manager."""
+        return self
+
+    def __exit__(self, *args: object) -> None:
+        """Close the HTTP client on exit."""
+        self.close()
+
+    def close(self) -> None:
+        """Close the HTTP client."""
+        self._client.close()

python/signal_bot/main.py

@@ -0,0 +1,239 @@
+"""Signal command and control bot — main entry point."""
+
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass
+from os import getenv
+from typing import TYPE_CHECKING, Annotated
+
+if TYPE_CHECKING:
+    from collections.abc import Callable
+
+import typer
+from alembic.command import upgrade
+from sqlalchemy.orm import Session
+from tenacity import before_sleep_log, retry, stop_after_attempt, wait_exponential
+
+from python.common import configure_logger, utcnow
+from python.database_cli import DATABASES
+from python.orm.common import get_postgres_engine
+from python.orm.signal_bot.models import DeadLetterMessage
+from python.signal_bot.commands.inventory import handle_inventory_update
+from python.signal_bot.commands.location import handle_location_request
+from python.signal_bot.device_registry import DeviceRegistry, sync_roles
+from python.signal_bot.llm_client import LLMClient
+from python.signal_bot.models import BotConfig, MessageStatus, Role, SignalMessage
+from python.signal_bot.signal_client import SignalClient
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass(frozen=True, slots=True)
+class Command:
+    """A registered bot command."""
+
+    action: Callable[[SignalMessage, str], None]
+    help_text: str
+    role: Role | None  # None = no role required (always allowed)
+
+
+class Bot:
+    """Holds shared resources and dispatches incoming messages to command handlers."""
+
+    def __init__(
+        self,
+        signal: SignalClient,
+        llm: LLMClient,
+        registry: DeviceRegistry,
+        config: BotConfig,
+    ) -> None:
+        self.signal = signal
+        self.llm = llm
+        self.registry = registry
+        self.config = config
+        self.commands: dict[str, Command] = {
+            "help": Command(action=self._help, help_text="show this help message", role=None),
+            "status": Command(action=self._status, help_text="show bot status", role=Role.STATUS),
+            "inventory": Command(
+                action=self._inventory,
+                help_text="update van inventory from a text list or receipt photo",
+                role=Role.INVENTORY,
+            ),
+            "location": Command(
+                action=self._location,
+                help_text="get current van location",
+                role=Role.LOCATION,
+            ),
+        }
+
+    # -- actions --------------------------------------------------------------
+
+    def _help(self, message: SignalMessage, _cmd: str) -> None:
+        """Return help text filtered to the sender's roles."""
+        self.signal.reply(message, self._build_help(self.registry.get_roles(message.source)))
+
+    def _status(self, message: SignalMessage, _cmd: str) -> None:
+        """Return the status of the bot."""
+        models = self.llm.list_models()
+        model_list = ", ".join(models[:10])
+        device_count = len(self.registry.list_devices())
+        self.signal.reply(
+            message,
+            f"Bot online.\nLLM: {self.llm.model}\nAvailable models: {model_list}\nKnown devices: {device_count}",
+        )
+
+    def _inventory(self, message: SignalMessage, _cmd: str) -> None:
+        """Process an inventory update."""
+        handle_inventory_update(message, self.signal, self.llm, self.config.inventory_api_url)
+
+    def _location(self, message: SignalMessage, _cmd: str) -> None:
+        """Reply with current van location."""
+        handle_location_request(message, self.signal, self.config.ha_url, self.config.ha_token)
+
+    # -- dispatch -------------------------------------------------------------
+
+    def _build_help(self, roles: list[Role]) -> str:
+        """Build help text showing only the commands the user can access."""
+        is_admin = Role.ADMIN in roles
+        lines = ["Available commands:"]
+        for name, cmd in self.commands.items():
+            if cmd.role is None or is_admin or cmd.role in roles:
+                lines.append(f"  {name:20s} — {cmd.help_text}")
+        return "\n".join(lines)
+
+    def dispatch(self, message: SignalMessage) -> None:
+        """Route an incoming message to the right command handler."""
+        source = message.source
+
+        if not self.registry.is_verified(source):
+            logger.info(f"Device {source} not verified, ignoring message")
+            return
+
+        if not self.registry.has_safety_number(source) and self.registry.has_role(source, Role.ADMIN):
+            logger.warning(f"Admin device {source} missing safety number, ignoring message")
+            return
+
+        text = message.message.strip()
+        parts = text.split()
+
+        if not parts and not message.attachments:
+            return
+
+        cmd = parts[0].lower() if parts else ""
+
+        logger.info(f"f{source=} running {cmd=} with {message=}")
+
+        command = self.commands.get(cmd)
+        if command is None:
+            if message.attachments:
+                command = self.commands["inventory"]
+                cmd = "inventory"
+            else:
+                return
+
+        if command.role is not None and not self.registry.has_role(source, command.role):
+            logger.warning(f"Device {source} denied access to {cmd!r}")
+            self.signal.reply(message, f"Permission denied: you do not have the '{command.role}' role.")
+            return
+
+        command.action(message, cmd)
+
+    def process_message(self, message: SignalMessage) -> None:
+        """Process a single message, sending it to the dead letter queue after repeated failures."""
+        max_attempts = self.config.max_message_attempts
+        for attempt in range(1, max_attempts + 1):
+            try:
+                safety_number = self.signal.get_safety_number(message.source)
+                self.registry.record_contact(message.source, safety_number)
+                self.dispatch(message)
+            except Exception:
+                logger.exception(f"Failed to process message (attempt {attempt}/{max_attempts})")
+            else:
+                return
+
+        logger.error(f"Message from {message.source} failed {max_attempts} times, sending to dead letter queue")
+        with Session(self.config.engine) as session:
+            session.add(
+                DeadLetterMessage(
+                    source=message.source,
+                    message=message.message,
+                    received_at=utcnow(),
+                    status=MessageStatus.UNPROCESSED,
+                )
+            )
+            session.commit()
+
+    def run(self) -> None:
+        """Listen for messages via WebSocket, reconnecting on failure."""
+        logger.info("Bot started — listening via WebSocket")
+
+        @retry(
+            stop=stop_after_attempt(self.config.max_retries),
+            wait=wait_exponential(multiplier=self.config.reconnect_delay, max=self.config.max_reconnect_delay),
+            before_sleep=before_sleep_log(logger, logging.WARNING),
+            reraise=True,
+        )
+        def _listen() -> None:
+            for message in self.signal.listen():
+                logger.info(f"Message from {message.source}: {message.message[:80]}")
+                self.process_message(message)
+
+        try:
+            _listen()
+        except Exception:
+            logger.critical("Max retries exceeded, shutting down")
+            raise
+
+
+def main(
+    log_level: Annotated[str, typer.Option()] = "DEBUG",
+    llm_timeout: Annotated[int, typer.Option()] = 600,
+) -> None:
+    """Run the Signal command and control bot."""
+    configure_logger(log_level)
+    signal_api_url = getenv("SIGNAL_API_URL")
+    phone_number = getenv("SIGNAL_PHONE_NUMBER")
+    inventory_api_url = getenv("INVENTORY_API_URL")
+
+    if signal_api_url is None:
+        error = "SIGNAL_API_URL environment variable not set"
+        raise ValueError(error)
+    if phone_number is None:
+        error = "SIGNAL_PHONE_NUMBER environment variable not set"
+        raise ValueError(error)
+    if inventory_api_url is None:
+        error = "INVENTORY_API_URL environment variable not set"
+        raise ValueError(error)
+
+    signal_bot_config = DATABASES["signal_bot"].alembic_config()
+    upgrade(signal_bot_config, "head")
+    engine = get_postgres_engine(name="SIGNALBOT")
+    sync_roles(engine)
+    config = BotConfig(
+        signal_api_url=signal_api_url,
+        phone_number=phone_number,
+        inventory_api_url=inventory_api_url,
+        ha_url=getenv("HA_URL"),
+        ha_token=getenv("HA_TOKEN"),
+        engine=engine,
+    )
+
+    llm_host = getenv("LLM_HOST")
+    llm_model = getenv("LLM_MODEL", "qwen3-vl:32b")
+    llm_port = int(getenv("LLM_PORT", "11434"))
+    if llm_host is None:
+        error = "LLM_HOST environment variable not set"
+        raise ValueError(error)
+
+    with (
+        SignalClient(config.signal_api_url, config.phone_number) as signal,
+        LLMClient(model=llm_model, host=llm_host, port=llm_port, timeout=llm_timeout) as llm,
+    ):
+        registry = DeviceRegistry(signal, engine)
+        bot = Bot(signal, llm, registry, config)
+        bot.run()
+
+
+if __name__ == "__main__":
+    typer.run(main)

python/signal_bot/models.py

@@ -0,0 +1,97 @@
+"""Models for the Signal command and control bot."""
+
+from __future__ import annotations
+
+from datetime import datetime  # noqa: TC003 - pydantic needs this at runtime
+from enum import StrEnum
+from typing import Any
+
+from pydantic import BaseModel, ConfigDict
+from sqlalchemy.engine import Engine  # noqa: TC002 - pydantic needs this at runtime
+
+
+class TrustLevel(StrEnum):
+    """Device trust level."""
+
+    VERIFIED = "verified"
+    UNVERIFIED = "unverified"
+    BLOCKED = "blocked"
+
+
+class Role(StrEnum):
+    """RBAC roles — one per command, plus admin which grants all."""
+
+    ADMIN = "admin"
+    STATUS = "status"
+    INVENTORY = "inventory"
+    LOCATION = "location"
+
+
+class MessageStatus(StrEnum):
+    """Dead letter queue message status."""
+
+    UNPROCESSED = "unprocessed"
+    PROCESSED = "processed"
+
+
+class Device(BaseModel):
+    """A registered device tracked by safety number."""
+
+    phone_number: str
+    safety_number: str
+    trust_level: TrustLevel = TrustLevel.UNVERIFIED
+    first_seen: datetime
+    last_seen: datetime
+
+
+class SignalMessage(BaseModel):
+    """An incoming Signal message."""
+
+    source: str
+    timestamp: int
+    message: str = ""
+    attachments: list[str] = []
+    group_id: str | None = None
+    is_receipt: bool = False
+
+
+class SignalEnvelope(BaseModel):
+    """Raw envelope from signal-cli-rest-api."""
+
+    envelope: dict[str, Any]
+    account: str | None = None
+
+
+class InventoryItem(BaseModel):
+    """An item in the van inventory."""
+
+    name: str
+    quantity: float = 1
+    unit: str = "each"
+    category: str = ""
+    notes: str = ""
+
+
+class InventoryUpdate(BaseModel):
+    """Result of processing an inventory update."""
+
+    items: list[InventoryItem] = []
+    raw_response: str = ""
+    source_type: str = ""  # "receipt_photo" or "text_list"
+
+
+class BotConfig(BaseModel):
+    """Top-level bot configuration."""
+
+    model_config = ConfigDict(arbitrary_types_allowed=True)
+
+    signal_api_url: str
+    phone_number: str
+    inventory_api_url: str
+    ha_url: str | None = None
+    ha_token: str | None = None
+    engine: Engine
+    reconnect_delay: int = 5
+    max_reconnect_delay: int = 300
+    max_retries: int = 10
+    max_message_attempts: int = 3

python/signal_bot/signal_client.py

@@ -0,0 +1,141 @@
+"""Client for the signal-cli-rest-api."""
+
+from __future__ import annotations
+
+import json
+import logging
+from typing import TYPE_CHECKING, Any, Self
+
+import httpx
+import websockets.sync.client
+
+if TYPE_CHECKING:
+    from collections.abc import Generator
+
+from python.signal_bot.models import SignalMessage
+
+logger = logging.getLogger(__name__)
+
+
+def _parse_envelope(envelope: dict[str, Any]) -> SignalMessage | None:
+    """Parse a signal-cli envelope into a SignalMessage, or None if not a data message."""
+    data_message = envelope.get("dataMessage")
+    if not data_message:
+        return None
+
+    attachment_ids = [att["id"] for att in data_message.get("attachments", []) if "id" in att]
+
+    group_info = data_message.get("groupInfo")
+    group_id = group_info.get("groupId") if group_info else None
+
+    return SignalMessage(
+        source=envelope.get("source", ""),
+        timestamp=envelope.get("timestamp", 0),
+        message=data_message.get("message", "") or "",
+        attachments=attachment_ids,
+        group_id=group_id,
+    )
+
+
+class SignalClient:
+    """Communicate with signal-cli-rest-api.
+
+    Args:
+        base_url: URL of the signal-cli-rest-api (e.g. http://localhost:8989).
+        phone_number: The registered phone number to send/receive as.
+    """
+
+    def __init__(self, base_url: str, phone_number: str) -> None:
+        self.base_url = base_url.rstrip("/")
+        self.phone_number = phone_number
+        self._client = httpx.Client(base_url=self.base_url, timeout=30)
+
+    def _ws_url(self) -> str:
+        """Build the WebSocket URL from the base HTTP URL."""
+        url = self.base_url.replace("http://", "ws://").replace("https://", "wss://")
+        return f"{url}/v1/receive/{self.phone_number}"
+
+    def listen(self) -> Generator[SignalMessage]:
+        """Connect via WebSocket and yield messages as they arrive."""
+        ws_url = self._ws_url()
+        logger.info(f"Connecting to WebSocket: {ws_url}")
+
+        with websockets.sync.client.connect(ws_url) as ws:
+            for raw in ws:
+                try:
+                    data = json.loads(raw)
+                    envelope = data.get("envelope", {})
+                    message = _parse_envelope(envelope)
+                    if message:
+                        yield message
+                except json.JSONDecodeError:
+                    logger.warning(f"Non-JSON WebSocket frame: {raw[:200]}")
+
+    def send(self, recipient: str, message: str) -> None:
+        """Send a text message."""
+        payload = {
+            "message": message,
+            "number": self.phone_number,
+            "recipients": [recipient],
+        }
+        response = self._client.post("/v2/send", json=payload)
+        response.raise_for_status()
+
+    def send_to_group(self, group_id: str, message: str) -> None:
+        """Send a message to a group."""
+        payload = {
+            "message": message,
+            "number": self.phone_number,
+            "recipients": [group_id],
+        }
+        response = self._client.post("/v2/send", json=payload)
+        response.raise_for_status()
+
+    def get_attachment(self, attachment_id: str) -> bytes:
+        """Download an attachment by ID."""
+        response = self._client.get(f"/v1/attachments/{attachment_id}")
+        response.raise_for_status()
+        return response.content
+
+    def get_identities(self) -> list[dict[str, Any]]:
+        """List known identities and their trust levels."""
+        response = self._client.get(f"/v1/identities/{self.phone_number}")
+        response.raise_for_status()
+        return response.json()
+
+    def get_safety_number(self, phone_number: str) -> str | None:
+        """Look up the safety number for a contact from signal-cli's local store."""
+        for identity in self.get_identities():
+            if identity.get("number") == phone_number:
+                return identity.get("safety_number", identity.get("fingerprint", ""))
+        return None
+
+    def trust_identity(self, number_to_trust: str, *, trust_all_known_keys: bool = False) -> None:
+        """Trust an identity (verify safety number)."""
+        payload: dict[str, Any] = {}
+        if trust_all_known_keys:
+            payload["trust_all_known_keys"] = True
+        response = self._client.put(
+            f"/v1/identities/{self.phone_number}/trust/{number_to_trust}",
+            json=payload,
+        )
+        response.raise_for_status()
+
+    def reply(self, message: SignalMessage, text: str) -> None:
+        """Reply to a message, routing to group or individual."""
+        if message.group_id:
+            self.send_to_group(message.group_id, text)
+        else:
+            self.send(message.source, text)
+
+    def __enter__(self) -> Self:
+        """Enter the context manager."""
+        return self
+
+    def __exit__(self, *args: object) -> None:
+        """Close the HTTP client on exit."""
+        self.close()
+
+    def close(self) -> None:
+        """Close the HTTP client."""
+        self._client.close()

python/tools/audiobook/__init__.py

@@ -1 +0,0 @@
-"""Audiobook tools."""

python/tools/audiobook/audible_convert.py

@@ -1,471 +0,0 @@
-"""Convert Audible AAX downloads into Audiobookshelf-friendly M4B files."""
-
-from __future__ import annotations
-
-import json
-import logging
-import re
-import shutil
-import subprocess
-from concurrent.futures import ThreadPoolExecutor
-from dataclasses import asdict, dataclass
-from os import getenv
-from pathlib import Path  # noqa: TC003 This is required for the typer CLI
-from typing import TYPE_CHECKING, Annotated, Any
-from uuid import uuid7
-
-import typer
-
-from python.common import configure_logger
-from python.orm.common import get_postgres_engine
-from python.tools.audiobook.metadata_agent import (
-    AgentConfig,
-    StandardBookMetadata,
-    standard_book_metadata,
-    write_agent_log,
-)
-
-if TYPE_CHECKING:
-    from sqlalchemy.engine import Engine
-
-logger = logging.getLogger(__name__)
-
-SENSITIVE_COMMAND_ARGUMENTS = {"-activation_bytes"}
-BOOK_RANGE_PATTERN = re.compile(r"(?:^|-)books?-(?P<start>[1-9]\d*)-(?P<end>[1-9]\d*)(?:-|$)")
-
-
-@dataclass(frozen=True)
-class ConversionConfig:
-    """Runtime settings for one conversion command."""
-
-    resolved_output: Path
-    ollama_api_key: str
-    agent_config: AgentConfig
-    engine: Engine
-    activation_bytes: str | None
-    dry_run: bool
-    overwrite: bool
-    work_directory_name: str = ".audible_convert"
-    dry_run_directory_name: str = "dry-run"
-    temp_directory_name: str = "tmp"
-    log_directory_name: str = "logs"
-    review_directory_name: str = "review"
-
-
-@dataclass(frozen=True)
-class ConcurrentConversionResult:
-    """Result from running ffmpeg and metadata resolution together."""
-
-    metadata: StandardBookMetadata | None
-    conversion_error: Exception | None
-    metadata_error: Exception | None
-
-
-class CommandExecutionError(RuntimeError):
-    """Command failed without exposing sensitive arguments."""
-
-    def __init__(self, arguments: list[str], returncode: int) -> None:
-        """Create a redacted command failure."""
-        self.arguments = tuple(arguments)
-        self.returncode = returncode
-        command = " ".join(redact_command_arguments(arguments))
-        super().__init__(f"Command failed with exit code {returncode}: {command}")
-
-
-def main(
-    input_directory: Annotated[Path, typer.Argument(help="Directory audible-cli downloads AAX files into.")],
-    output_directory: Annotated[Path, typer.Argument(help="Audiobook output directory.")],
-    *,
-    dry_run: Annotated[
-        bool,
-        typer.Option("--dry-run", help="Print planned output files and write marker files without converting."),
-    ] = False,
-    overwrite: Annotated[bool, typer.Option("--overwrite", help="Overwrite existing M4B files.")] = False,
-) -> None:
-    """Convert AAX files from a download directory into M4B files."""
-    configure_logger()
-    resolved_input = input_directory.resolve(strict=True)
-    resolved_output = output_directory.resolve()
-    if not dry_run:
-        resolved_output.mkdir(parents=True, exist_ok=True)
-
-    ollama_api_key = getenv("OLLAMA_API_KEY")
-    if not ollama_api_key:
-        msg = "OLLAMA_API_KEY is required for audiobook metadata resolution"
-        raise RuntimeError(msg)
-
-    config = ConversionConfig(
-        resolved_output=resolved_output,
-        ollama_api_key=ollama_api_key,
-        agent_config=AgentConfig(),
-        engine=get_postgres_engine(name="RICHIE"),
-        activation_bytes=getenv("AUDIBLE_ACTIVATION_BYTES"),
-        dry_run=dry_run,
-        overwrite=overwrite,
-    )
-
-    aax_files = sorted(resolved_input.glob("*.aax"))
-    if not aax_files:
-        logger.info("No AAX files found in %s", resolved_input)
-        return
-    for aax_file in aax_files:
-        logger.info("Converting %s", aax_file)
-        convert_aax_file_with_agent(aax_file, config)
-
-
-def run_command(arguments: list[str], *, capture: bool = False) -> subprocess.CompletedProcess[str]:
-    """Run a command and return the completed process.
-
-    Args:
-        arguments: Command and arguments to run.
-        capture: Whether to capture stdout and stderr.
-
-    Returns:
-        The completed process.
-    """
-    logger.debug("%s", " ".join(redact_command_arguments(arguments)))
-    try:
-        return subprocess.run(arguments, check=True, capture_output=capture, text=True)
-    except subprocess.CalledProcessError as error:
-        raise CommandExecutionError(arguments, error.returncode) from error
-
-
-def redact_command_arguments(arguments: list[str]) -> list[str]:
-    """Return command arguments with sensitive values redacted."""
-    redacted = []
-    redact_next = False
-    for argument in arguments:
-        if redact_next:
-            redacted.append("<redacted>")
-            redact_next = False
-            continue
-
-        redacted.append(argument)
-        redact_next = argument in SENSITIVE_COMMAND_ARGUMENTS
-    return redacted
-
-
-def read_metadata(aax_file: Path) -> dict[str, str]:
-    """Read ffprobe format tags from an AAX file.
-
-    Args:
-        aax_file: AAX file to inspect.
-
-    Returns:
-        Lower-cased metadata tag names mapped to their values.
-    """
-    completed = run_command(
-        [
-            "ffprobe",
-            "-v",
-            "quiet",
-            "-print_format",
-            "json",
-            "-show_format",
-            str(aax_file),
-        ],
-        capture=True,
-    )
-    ffprobe_data: dict[str, Any] = json.loads(completed.stdout)
-    tags = ffprobe_data.get("format", {}).get("tags", {})
-    return {str(key).lower(): str(value) for key, value in tags.items()}
-
-
-def output_stem(metadata: StandardBookMetadata) -> str:
-    """Build the output stem for a book.
-
-    Args:
-        metadata: Book metadata.
-
-    Returns:
-        Output stem in author-series_01-title form.
-    """
-    index_slug = series_index_slug(metadata.series_index, metadata.title)
-    return f"{metadata.author}-{metadata.series}_{index_slug}-{metadata.title}"
-
-
-def series_index_slug(series_index: float, title: str = "") -> str:
-    """Return a filename-safe series index."""
-    if title_range := title_series_range_slug(series_index, title):
-        return title_range
-    index = float(series_index)
-    if index.is_integer():
-        return f"{int(index):02}"
-    return f"{int(index):02}.5"
-
-
-def title_series_range_slug(series_index: float, title: str) -> str | None:
-    """Return a series range slug found in an omnibus title."""
-    index = float(series_index)
-    if not index.is_integer():
-        return None
-    first_index = int(index)
-    for match in BOOK_RANGE_PATTERN.finditer(title):
-        start = int(match.group("start"))
-        end = int(match.group("end"))
-        if start == first_index and end > start:
-            return f"{start:02}-{end:02}"
-    return None
-
-
-def metadata_output_path(output_directory: Path, metadata: StandardBookMetadata) -> Path:
-    """Build the final M4B path from resolved metadata."""
-    stem = output_stem(metadata)
-    return output_directory / stem / f"{stem}.m4b"
-
-
-def convert_aax_file(
-    aax_file: Path,
-    destination: Path,
-    activation_bytes: str | None,
-    *,
-    overwrite: bool,
-) -> None:
-    """Convert an AAX file into an M4B file.
-
-    Args:
-        aax_file: Source AAX file.
-        destination: Destination M4B file.
-        activation_bytes: Optional Audible activation bytes for ffmpeg.
-        overwrite: Whether to overwrite an existing M4B.
-    """
-    if destination.exists() and not overwrite:
-        logger.info("Skipping existing file %s", destination)
-        return
-
-    destination.parent.mkdir(parents=True, exist_ok=True)
-    arguments = ["ffmpeg", "-hide_banner", "-y" if overwrite else "-n"]
-    if activation_bytes:
-        arguments.extend(["-activation_bytes", activation_bytes])
-    arguments.extend(["-i", str(aax_file), "-map_metadata", "0", "-c", "copy", str(destination)])
-    run_command(arguments)
-
-
-def write_review_file(
-    *,
-    destination: Path | None,
-    ffprobe_metadata: dict[str, str],
-    log_file: Path,
-    metadata: StandardBookMetadata | None,
-    reason: str,
-    review_file: Path,
-    source: Path,
-    temp_file: Path | None,
-) -> None:
-    """Write a manual review file for an unresolved conversion."""
-    review_file.parent.mkdir(parents=True, exist_ok=True)
-    payload = {
-        "destination": str(destination) if destination else None,
-        "ffprobe_metadata": ffprobe_metadata,
-        "metadata": asdict(metadata) if metadata else None,
-        "reason": reason,
-        "source": str(source),
-        "temp_file": str(temp_file) if temp_file else None,
-    }
-    review_file.write_text(json.dumps(payload, indent=2, sort_keys=True), encoding="utf-8")
-    write_agent_log(log_file, "review_written", path=str(review_file), reason=reason)
-
-
-def cleanup_temp_output(temp_file: Path) -> None:
-    """Remove a run's temporary output directory."""
-    shutil.rmtree(temp_file.parent, ignore_errors=True)
-
-
-def dry_run_aax_file_with_agent(
-    aax_file: Path,
-    ffprobe_metadata: dict[str, str],
-    engine: Engine,
-    config: ConversionConfig,
-    log_file: Path,
-    review_file: Path,
-) -> None:
-    """Resolve and print the planned output path without converting."""
-    metadata = standard_book_metadata(
-        aax_file.name,
-        ffprobe_metadata,
-        engine,
-        log_file,
-        config.ollama_api_key,
-        config.agent_config,
-    )
-    destination = None if metadata.needs_review else metadata_output_path(config.resolved_output, metadata)
-    if metadata.needs_review:
-        write_review_file(
-            destination=destination,
-            ffprobe_metadata=ffprobe_metadata,
-            log_file=log_file,
-            metadata=metadata,
-            reason="metadata_needs_review",
-            review_file=review_file,
-            source=aax_file,
-            temp_file=None,
-        )
-        typer.echo(f"{aax_file} -> REVIEW {review_file}")
-    else:
-        stem = output_stem(metadata)
-        dry_run_file = (
-            config.resolved_output / config.work_directory_name / config.dry_run_directory_name / stem / f"{stem}.m4b"
-        )
-        dry_run_file.parent.mkdir(parents=True, exist_ok=True)
-        dry_run_file.write_text(f"{destination}\n", encoding="utf-8")
-        write_agent_log(
-            log_file,
-            "dry_run_file_written",
-            destination=str(destination),
-            path=str(dry_run_file),
-        )
-        typer.echo(f"{aax_file} -> {destination}")
-
-
-def convert_temp_file_and_resolve_metadata(
-    aax_file: Path,
-    temp_file: Path,
-    ffprobe_metadata: dict[str, str],
-    config: ConversionConfig,
-    log_file: Path,
-) -> ConcurrentConversionResult:
-    """Run ffmpeg and metadata resolution in parallel."""
-    conversion_error: Exception | None = None
-    metadata_error: Exception | None = None
-    metadata: StandardBookMetadata | None = None
-
-    with ThreadPoolExecutor(max_workers=2) as executor:
-        conversion_future = executor.submit(
-            convert_aax_file,
-            aax_file,
-            temp_file,
-            config.activation_bytes,
-            overwrite=True,
-        )
-        metadata_future = executor.submit(
-            standard_book_metadata,
-            aax_file.name,
-            ffprobe_metadata,
-            config.engine,
-            log_file,
-            config.ollama_api_key,
-            config.agent_config,
-        )
-
-        conversion_error = conversion_future.exception()
-        if conversion_error is None:
-            conversion_future.result()
-
-        metadata_error = metadata_future.exception()
-        if metadata_error is None:
-            metadata = metadata_future.result()
-
-    return ConcurrentConversionResult(
-        metadata=metadata,
-        conversion_error=conversion_error,
-        metadata_error=metadata_error,
-    )
-
-
-def convert_aax_file_with_agent(aax_file: Path, config: ConversionConfig) -> None:
-    """Convert one AAX file using the metadata agent for the final path."""
-    run_id = uuid7().hex
-    log_file = config.resolved_output / config.work_directory_name / config.log_directory_name / f"{run_id}.jsonl"
-    review_file = config.resolved_output / config.work_directory_name / config.review_directory_name / f"{run_id}.json"
-    write_agent_log(log_file, "conversion_start", source=str(aax_file), dry_run=config.dry_run)
-    try:
-        ffprobe_metadata = read_metadata(aax_file)
-    except Exception as error:
-        logger.exception("ffprobe failed")
-        write_review_file(
-            destination=None,
-            ffprobe_metadata={},
-            log_file=log_file,
-            metadata=None,
-            reason=f"ffprobe_failed: {error}",
-            review_file=review_file,
-            source=aax_file,
-            temp_file=None,
-        )
-        return
-
-    if config.dry_run:
-        dry_run_aax_file_with_agent(
-            aax_file,
-            ffprobe_metadata,
-            config.engine,
-            config,
-            log_file,
-            review_file,
-        )
-        return
-
-    temp_file = (
-        config.resolved_output / config.work_directory_name / config.temp_directory_name / run_id / "converted.m4b"
-    )
-    temp_file.parent.mkdir(parents=True, exist_ok=True)
-
-    result = convert_temp_file_and_resolve_metadata(aax_file, temp_file, ffprobe_metadata, config, log_file)
-
-    if result.conversion_error:
-        reason = f"ffmpeg_failed: {result.conversion_error}"
-        write_review_file(
-            destination=None,
-            ffprobe_metadata=ffprobe_metadata,
-            log_file=log_file,
-            metadata=result.metadata,
-            reason=reason,
-            review_file=review_file,
-            source=aax_file,
-            temp_file=temp_file if temp_file.exists() else None,
-        )
-        return
-
-    if result.metadata_error:
-        write_review_file(
-            destination=None,
-            ffprobe_metadata=ffprobe_metadata,
-            log_file=log_file,
-            metadata=None,
-            reason=f"metadata_failed: {result.metadata_error}",
-            review_file=review_file,
-            source=aax_file,
-            temp_file=temp_file,
-        )
-        return
-
-    if result.metadata is None or result.metadata.needs_review:
-        write_review_file(
-            destination=None,
-            ffprobe_metadata=ffprobe_metadata,
-            log_file=log_file,
-            metadata=result.metadata,
-            reason="metadata_needs_review",
-            review_file=review_file,
-            source=aax_file,
-            temp_file=temp_file,
-        )
-        return
-
-    destination = metadata_output_path(config.resolved_output, result.metadata)
-    if destination.exists() and not config.overwrite:
-        write_agent_log(log_file, "destination_exists", destination=str(destination))
-        cleanup_temp_output(temp_file)
-        return
-
-    destination.parent.mkdir(parents=True, exist_ok=True)
-    try:
-        temp_file.replace(destination)
-    except Exception as error:  # noqa: BLE001
-        write_review_file(
-            destination=destination,
-            ffprobe_metadata=ffprobe_metadata,
-            log_file=log_file,
-            metadata=result.metadata,
-            reason=f"rename_failed: {error}",
-            review_file=review_file,
-            source=aax_file,
-            temp_file=temp_file if temp_file.exists() else None,
-        )
-    else:
-        cleanup_temp_output(temp_file)
-        write_agent_log(log_file, "conversion_complete", destination=str(destination))
-
-
-if __name__ == "__main__":
-    typer.run(main)

python/tools/audiobook/catalog.py

@@ -1,176 +0,0 @@
-"""Import audiobook catalog authors and series from CSV files."""
-
-from __future__ import annotations
-
-import csv
-import logging
-from pathlib import Path  # noqa: TC003 This is required for the typer CLI
-from typing import Annotated
-
-import typer
-from sqlalchemy import select
-from sqlalchemy.orm import Session
-
-from python.common import configure_logger
-from python.orm.common import get_postgres_engine
-from python.orm.richie import AudiobookAuthor, AudiobookSeries
-
-logger = logging.getLogger(__name__)
-
-AUTHOR_NAME_COLUMN = "author_name"
-ID_COLUMN = "id"
-NAME_COLUMN = "name"
-
-
-class CatalogImportError(ValueError):
-    """CSV catalog import failed validation."""
-
-
-def main(
-    authors_csv: Annotated[Path, typer.Argument(help="CSV with name and optional id.")],
-    series_csv: Annotated[Path, typer.Argument(help="CSV with name, author_name, and optional id.")],
-) -> None:
-    """Upsert audiobook authors and series from CSV files."""
-    configure_logger()
-    try:
-        engine = get_postgres_engine(name="RICHIE")
-        with Session(engine) as session:
-            author_count = upsert_authors_from_csv(session, authors_csv)
-            series_count = upsert_series_from_csv(session, series_csv)
-            session.commit()
-    except CatalogImportError as error:
-        typer.echo(str(error), err=True)
-        raise typer.Exit(code=1) from error
-
-    logger.info("Upserted %s authors and %s series", author_count, series_count)
-
-
-def upsert_authors_from_csv(session: Session, authors_csv: Path) -> int:
-    """Upsert authors from a CSV file."""
-    count = 0
-    for row_number, row in csv_rows(authors_csv):
-        name = required_csv_value(row, authors_csv, row_number, NAME_COLUMN)
-        upsert_author(session, name, csv_id(row, authors_csv, row_number))
-        count += 1
-    return count
-
-
-def upsert_series_from_csv(session: Session, series_csv: Path) -> int:
-    """Upsert series from a CSV file."""
-    count = 0
-    for row_number, row in csv_rows(series_csv):
-        series_name = required_csv_value(row, series_csv, row_number, NAME_COLUMN)
-        author_name = required_csv_value(row, series_csv, row_number, AUTHOR_NAME_COLUMN)
-        author = find_author_by_name(session, author_name)
-        if author is None:
-            msg = f"{series_csv}:{row_number}: author not found: {author_name}"
-            raise CatalogImportError(msg)
-        upsert_series(session, series_name, author, csv_id(row, series_csv, row_number))
-        count += 1
-    return count
-
-
-def upsert_author(session: Session, name: str, author_id: int | None) -> AudiobookAuthor:
-    """Upsert one author by id or exact name."""
-    if author_id is not None:
-        author = session.get(AudiobookAuthor, author_id)
-        if author is None:
-            author = AudiobookAuthor(id=author_id, name=name)
-            session.add(author)
-        else:
-            author.name = name
-        session.flush()
-        return author
-
-    author = find_author_by_name(session, name)
-    if author is None:
-        author = AudiobookAuthor(name=name)
-        session.add(author)
-        session.flush()
-    return author
-
-
-def upsert_series(
-    session: Session,
-    name: str,
-    author: AudiobookAuthor,
-    series_id: int | None,
-) -> AudiobookSeries:
-    """Upsert one series by id or exact author/name match."""
-    if series_id is not None:
-        series = session.get(AudiobookSeries, series_id)
-        if series is None:
-            series = AudiobookSeries(id=series_id, name=name, author=author)
-            session.add(series)
-        else:
-            series.name = name
-            series.author = author
-        session.flush()
-        return series
-
-    series = find_series_by_name_and_author(session, name, author.id)
-    if series is None:
-        series = AudiobookSeries(name=name, author=author)
-        session.add(series)
-        session.flush()
-    return series
-
-
-def find_author_by_name(session: Session, name: str) -> AudiobookAuthor | None:
-    """Find one author by exact name."""
-    return session.scalar(select(AudiobookAuthor).where(AudiobookAuthor.name == name))
-
-
-def find_series_by_name_and_author(
-    session: Session,
-    name: str,
-    author_id: int,
-) -> AudiobookSeries | None:
-    """Find one series by exact name and author."""
-    return session.scalar(
-        select(AudiobookSeries).where(
-            AudiobookSeries.name == name,
-            AudiobookSeries.author_id == author_id,
-        ),
-    )
-
-
-def csv_rows(csv_path: Path) -> list[tuple[int, dict[str, str | None]]]:
-    """Read a CSV file as numbered rows."""
-    with csv_path.open(newline="", encoding="utf-8") as file:
-        reader = csv.DictReader(file)
-        if reader.fieldnames is None:
-            msg = f"{csv_path}: missing CSV header"
-            raise CatalogImportError(msg)
-        return [(row_number, row) for row_number, row in enumerate(reader, start=2)]
-
-
-def required_csv_value(
-    row: dict[str, str | None],
-    csv_path: Path,
-    row_number: int,
-    column: str,
-) -> str:
-    """Read a required CSV value."""
-    value = row.get(column)
-    if value and value.strip():
-        return value.strip()
-    msg = f"{csv_path}:{row_number}: missing required column value: {column}"
-    raise CatalogImportError(msg)
-
-
-def csv_id(row: dict[str, str | None], csv_path: Path, row_number: int) -> int | None:
-    """Read an optional id field from a CSV row."""
-    value = row.get(ID_COLUMN)
-    if value is None or not value.strip():
-        return None
-    try:
-        return int(value)
-    except ValueError as error:
-        msg = f"{csv_path}:{row_number}: id must be an integer: {value}"
-        raise CatalogImportError(msg) from error
-    return None
-
-
-if __name__ == "__main__":
-    typer.run(main)

python/tools/audiobook/llm_tool_calling.py

@@ -1,599 +0,0 @@
-"""LLM tool calling support for audiobook metadata resolution."""
-
-from __future__ import annotations
-
-import json
-import re
-import time
-from collections.abc import Callable
-from dataclasses import dataclass
-from typing import TYPE_CHECKING
-
-from sqlalchemy import or_, select
-
-from python.orm.richie import Audiobook, AudiobookAuthor, AudiobookSeries
-
-if TYPE_CHECKING:
-    from pathlib import Path
-
-    from sqlalchemy.orm import Session
-
-    from python.tools.audiobook.metadata_agent import AgentConfig
-
-CATALOG_SLUG_PATTERN = re.compile(r"^[a-z0-9]+(?:_[a-z0-9]+)*$")
-TITLE_SLUG_PATTERN = re.compile(r"^[a-z0-9]+(?:-[a-z0-9]+)*$")
-
-LogWriter = Callable[..., None]
-
-
-class MetadataResolutionError(ValueError):
-    """Metadata resolution failed validation."""
-
-
-@dataclass(frozen=True)
-class EnsuredBook:
-    """Book row plus whether it was created."""
-
-    book: Audiobook
-    action: str
-
-
-class CatalogToolRegistry:
-    """Controlled catalog tools exposed to the metadata model."""
-
-    def __init__(
-        self,
-        session: Session,
-        log_path: Path,
-        config: AgentConfig,
-        write_log: LogWriter,
-    ) -> None:
-        """Create a registry bound to one database session and audit log."""
-        self.session = session
-        self.log_path = log_path
-        self.config = config
-        self.write_log = write_log
-        self.seen_author_ids: set[int] = set()
-        self.seen_series_ids: set[int] = set()
-        self.seen_book_ids: set[int] = set()
-        self.created_author_ids: set[int] = set()
-        self.created_series_ids: set[int] = set()
-        self.created_book_ids: set[int] = set()
-
-    def tool_schemas(self) -> list[dict[str, object]]:
-        """Return Ollama tool schemas."""
-        schemas = [
-            {
-                "type": "function",
-                "function": {
-                    "name": "search_authors",
-                    "description": "Search canonical audiobook authors by slug or noisy source text.",
-                    "parameters": {
-                        "type": "object",
-                        "properties": {"query": {"type": "string"}},
-                        "required": ["query"],
-                    },
-                },
-            },
-            {
-                "type": "function",
-                "function": {
-                    "name": "search_series",
-                    "description": "Search canonical audiobook series by slug or noisy source text.",
-                    "parameters": {
-                        "type": "object",
-                        "properties": {
-                            "query": {"type": "string"},
-                            "author_id": {"type": ["integer", "null"]},
-                        },
-                        "required": ["query"],
-                    },
-                },
-            },
-            {
-                "type": "function",
-                "function": {
-                    "name": "search_books",
-                    "description": "Search canonical audiobook titles with optional author and series filters.",
-                    "parameters": {
-                        "type": "object",
-                        "properties": {
-                            "query": {"type": "string"},
-                            "author_id": {"type": ["integer", "null"]},
-                            "series_id": {"type": ["integer", "null"]},
-                        },
-                        "required": ["query"],
-                    },
-                },
-            },
-            {
-                "type": "function",
-                "function": {
-                    "name": "ensure_author",
-                    "description": "Normalize an author name to a catalog slug, then return or create that author.",
-                    "parameters": {
-                        "type": "object",
-                        "properties": {"name": {"type": "string"}},
-                        "required": ["name"],
-                    },
-                },
-            },
-            {
-                "type": "function",
-                "function": {
-                    "name": "ensure_series",
-                    "description": "Normalize a series name to a catalog slug, then return or create it for an author.",
-                    "parameters": {
-                        "type": "object",
-                        "properties": {
-                            "name": {"type": "string"},
-                            "author_id": {"type": "integer"},
-                        },
-                        "required": ["name", "author_id"],
-                    },
-                },
-            },
-            {
-                "type": "function",
-                "function": {
-                    "name": "ensure_book",
-                    "description": "Normalize a title to a book slug, then return or create it for an author/series.",
-                    "parameters": {
-                        "type": "object",
-                        "properties": {
-                            "title": {"type": "string"},
-                            "author_id": {"type": "integer"},
-                            "series_id": {"type": ["integer", "null"]},
-                            "series_index": {"type": "number", "multipleOf": 0.5},
-                        },
-                        "required": ["title", "author_id", "series_id", "series_index"],
-                    },
-                },
-            },
-        ]
-        enabled_tool_names = set(self.config.tool_names)
-        return [schema for schema in schemas if schema["function"]["name"] in enabled_tool_names]
-
-    def run(self, name: str, arguments: dict[str, object]) -> list[dict[str, object]]:
-        """Run one catalog tool and audit the call."""
-        handlers = {
-            "search_authors": self.run_search_authors,
-            "search_series": self.run_search_series,
-            "search_books": self.run_search_books,
-            "ensure_author": self.run_ensure_author,
-            "ensure_series": self.run_ensure_series,
-            "ensure_book": self.run_ensure_book,
-        }
-        handler = handlers.get(name)
-        if handler is None:
-            self.write_log(self.log_path, "tool_error", tool=name, arguments=arguments, error="unknown_tool")
-            msg = f"Unknown audiobook metadata tool: {name}"
-            raise MetadataResolutionError(msg)
-        if name not in self.config.tool_names:
-            self.write_log(self.log_path, "tool_error", tool=name, arguments=arguments, error="tool_not_enabled")
-            msg = f"Audiobook metadata tool is not enabled: {name}"
-            raise MetadataResolutionError(msg)
-
-        started = time.perf_counter()
-        self.write_log(self.log_path, "tool_call", tool=name, arguments=arguments)
-        result = handler(arguments)
-        duration_ms = round((time.perf_counter() - started) * 1000, 3)
-        self.write_log(
-            self.log_path,
-            "tool_result",
-            tool=name,
-            duration_ms=duration_ms,
-            result_count=len(result),
-            preview=result[:3],
-        )
-        return result
-
-    def get_author(self, author_id: int) -> AudiobookAuthor | None:
-        """Return an author by id."""
-        return self.session.get(AudiobookAuthor, author_id)
-
-    def get_book(self, book_id: int) -> Audiobook | None:
-        """Return a book by id."""
-        return self.session.get(Audiobook, book_id)
-
-    def get_series(self, series_id: int) -> AudiobookSeries | None:
-        """Return a series by id."""
-        return self.session.get(AudiobookSeries, series_id)
-
-    def prune_unused_created_rows(self, *, author_id: int, book_id: int | None, series_id: int | None) -> None:
-        """Remove catalog rows created during this run but not used by final metadata."""
-        used_book_ids = {book_id} if book_id is not None else set()
-        for created_book_id in self.created_book_ids - used_book_ids:
-            if book := self.get_book(created_book_id):
-                self.session.delete(book)
-
-        self.session.flush()
-        used_series_ids = {series_id} if series_id is not None else set()
-        for created_series_id in self.created_series_ids - used_series_ids:
-            series = self.get_series(created_series_id)
-            if series and not series.books:
-                self.session.delete(series)
-
-        self.session.flush()
-        for created_author_id in self.created_author_ids - {author_id}:
-            author = self.get_author(created_author_id)
-            if author and not author.books and not author.series:
-                self.session.delete(author)
-
-    def run_search_authors(self, arguments: dict[str, object]) -> list[dict[str, object]]:
-        """Search authors from tool arguments and remember returned ids."""
-        query = required_string(arguments, "query")
-        statement = select(AudiobookAuthor).order_by(AudiobookAuthor.name).limit(self.config.max_tool_results)
-        if terms := query_terms(query):
-            statement = statement.where(or_(*(AudiobookAuthor.name.ilike(f"%{term}%") for term in terms)))
-
-        authors = self.session.scalars(statement).all()
-        self.seen_author_ids.update(author.id for author in authors)
-        return [{"id": author.id, "name": author.name} for author in authors]
-
-    def run_search_series(self, arguments: dict[str, object]) -> list[dict[str, object]]:
-        """Search series from tool arguments and remember returned ids."""
-        query = required_string(arguments, "query")
-        author_id = optional_int(arguments.get("author_id"), "author_id")
-        statement = select(AudiobookSeries).order_by(AudiobookSeries.name).limit(self.config.max_tool_results)
-        if terms := query_terms(query):
-            statement = statement.where(or_(*(AudiobookSeries.name.ilike(f"%{term}%") for term in terms)))
-        if author_id is not None:
-            statement = statement.where(AudiobookSeries.author_id == author_id)
-
-        series_rows = self.session.scalars(statement).all()
-        self.seen_series_ids.update(series.id for series in series_rows)
-        self.seen_author_ids.update(series.author_id for series in series_rows)
-        return [
-            {
-                "id": series.id,
-                "name": series.name,
-                "author_id": series.author_id,
-                "author": series.author.name,
-            }
-            for series in series_rows
-        ]
-
-    def run_search_books(self, arguments: dict[str, object]) -> list[dict[str, object]]:
-        """Search books from tool arguments and remember returned ids."""
-        query = required_string(arguments, "query")
-        author_id = optional_int(arguments.get("author_id"), "author_id")
-        series_id = optional_int(arguments.get("series_id"), "series_id")
-        statement = select(Audiobook).order_by(Audiobook.title).limit(self.config.max_tool_results)
-        if terms := query_terms(query):
-            statement = statement.where(or_(*(Audiobook.title.ilike(f"%{term}%") for term in terms)))
-        if author_id is not None:
-            statement = statement.where(Audiobook.author_id == author_id)
-        if series_id is not None:
-            statement = statement.where(Audiobook.series_id == series_id)
-
-        books = self.session.scalars(statement).all()
-        self.seen_book_ids.update(book.id for book in books)
-        self.seen_author_ids.update(book.author_id for book in books)
-        self.seen_series_ids.update(book.series_id for book in books if book.series_id is not None)
-        return [
-            {
-                "id": book.id,
-                "title": book.title,
-                "author_id": book.author_id,
-                "author": book.author.name,
-                "series_id": book.series_id,
-                "series": book.series.name if book.series else self.config.standalone_series,
-                "series_index": book.series_index,
-            }
-            for book in books
-        ]
-
-    def run_ensure_author(self, arguments: dict[str, object]) -> list[dict[str, object]]:
-        """Ensure an author from tool arguments and return a tool result."""
-        name = normalize_catalog_slug(required_string(arguments, "name"))
-        validate_catalog_slug(name, "author")
-        author = self.session.scalar(select(AudiobookAuthor).where(AudiobookAuthor.name == name))
-        action = "existing"
-        if author is None:
-            author = AudiobookAuthor(name=name)
-            self.session.add(author)
-            self.session.flush()
-            self.created_author_ids.add(author.id)
-            action = "created"
-
-        self.seen_author_ids.add(author.id)
-        return [{"id": author.id, "name": author.name, "action": action}]
-
-    def run_ensure_series(self, arguments: dict[str, object]) -> list[dict[str, object]]:
-        """Ensure a series from tool arguments and return a tool result."""
-        name = normalize_catalog_slug(required_string(arguments, "name"))
-        author_id = required_int(arguments, "author_id")
-        validate_catalog_slug(name, "series")
-        author = self.required_author(author_id)
-        series = self.find_series_by_catalog_slug(name, author.id)
-        action = "existing"
-        if series is None:
-            series = AudiobookSeries(name=name, author=author)
-            self.session.add(series)
-            self.session.flush()
-            self.created_series_ids.add(series.id)
-            action = "created"
-
-        self.seen_author_ids.add(author.id)
-        self.seen_series_ids.add(series.id)
-        return [self.series_result(series, action)]
-
-    def run_ensure_book(self, arguments: dict[str, object]) -> list[dict[str, object]]:
-        """Ensure a book from tool arguments and return a tool result."""
-        title = required_string(arguments, "title")
-        author_id = required_int(arguments, "author_id")
-        series_id = optional_int(arguments.get("series_id"), "series_id")
-        series_index = required_series_index(arguments, "series_index")
-        ensured = self.ensure_book(title, author_id, series_id, series_index)
-        return [self.book_result(ensured.book, ensured.action)]
-
-    def ensure_book(
-        self,
-        title: str,
-        author_id: int,
-        series_id: int | None,
-        series_index: float,
-    ) -> EnsuredBook:
-        """Return an existing book row, or create it after validating ownership."""
-        title = normalize_title_slug(title)
-        validate_title_slug(title)
-        author = self.required_author(author_id)
-        series = None
-        if series_id is None:
-            if series_index != 0:
-                msg = "standalone books must use series_index 0"
-                raise MetadataResolutionError(msg)
-        else:
-            series = self.required_series(series_id)
-            if series.author_id != author.id:
-                msg = f"series_id {series_id} does not belong to author_id {author_id}"
-                raise MetadataResolutionError(msg)
-            if series_index <= 0:
-                msg = "series books must use a positive series_index"
-                raise MetadataResolutionError(msg)
-
-        statement = select(Audiobook).where(
-            Audiobook.title == title,
-            Audiobook.author_id == author.id,
-        )
-        if series is None:
-            statement = statement.where(Audiobook.series_id.is_(None))
-        else:
-            statement = statement.where(Audiobook.series_id == series.id)
-        book = self.session.scalar(statement)
-        if book is None:
-            book = Audiobook(title=title, author=author, series=series, series_index=series_index)
-            self.session.add(book)
-            self.session.flush()
-            self.created_book_ids.add(book.id)
-            action = "created"
-        else:
-            action = "existing"
-
-        self.seen_book_ids.add(book.id)
-        self.seen_author_ids.add(author.id)
-        if book.series_id is not None:
-            self.seen_series_ids.add(book.series_id)
-        return EnsuredBook(book=book, action=action)
-
-    def required_author(self, author_id: int) -> AudiobookAuthor:
-        """Return an author or fail metadata resolution."""
-        author = self.get_author(author_id)
-        if author is None:
-            msg = f"author_id {author_id} does not exist"
-            raise MetadataResolutionError(msg)
-        return author
-
-    def required_series(self, series_id: int) -> AudiobookSeries:
-        """Return a series or fail metadata resolution."""
-        series = self.get_series(series_id)
-        if series is None:
-            msg = f"series_id {series_id} does not exist"
-            raise MetadataResolutionError(msg)
-        return series
-
-    def find_series_by_catalog_slug(self, name: str, author_id: int) -> AudiobookSeries | None:
-        """Return a series by exact slug or underscore-insensitive slug."""
-        exact = self.session.scalar(
-            select(AudiobookSeries).where(
-                AudiobookSeries.name == name,
-                AudiobookSeries.author_id == author_id,
-            ),
-        )
-        if exact is not None:
-            return exact
-
-        compact_name = compact_catalog_slug(name)
-        series_rows = self.session.scalars(
-            select(AudiobookSeries).where(AudiobookSeries.author_id == author_id).order_by(AudiobookSeries.name),
-        ).all()
-        for series in series_rows:
-            if compact_catalog_slug(series.name) == compact_name:
-                return series
-        return None
-
-    def series_result(self, series: AudiobookSeries, action: str) -> dict[str, object]:
-        """Build a normalized series tool result."""
-        return {
-            "id": series.id,
-            "name": series.name,
-            "author_id": series.author_id,
-            "author": series.author.name,
-            "action": action,
-        }
-
-    def book_result(self, book: Audiobook, action: str) -> dict[str, object]:
-        """Build a normalized book tool result."""
-        return {
-            "id": book.id,
-            "title": book.title,
-            "author_id": book.author_id,
-            "author": book.author.name,
-            "series_id": book.series_id,
-            "series": book.series.name if book.series else self.config.standalone_series,
-            "series_index": book.series_index,
-            "action": action,
-        }
-
-
-def run_tool_calls(
-    messages: list[dict[str, object]],
-    message: dict[str, object],
-    tool_calls: list[tuple[str, dict[str, object]]],
-    registry: CatalogToolRegistry,
-    log_path: Path,
-    write_log: LogWriter,
-) -> str | None:
-    """Run tool calls, append tool messages, and return fatal error text when stopped."""
-    messages.append(message)
-    for tool_name, arguments in tool_calls:
-        try:
-            tool_result = registry.run(tool_name, arguments)
-        except MetadataResolutionError as error:
-            if is_fatal_tool_error(error):
-                return str(error)
-            write_log(log_path, "tool_error", tool=tool_name, arguments=arguments, error=str(error))
-            messages.append(
-                {
-                    "role": "tool",
-                    "tool_name": tool_name,
-                    "content": json.dumps({"error": str(error)}, sort_keys=True),
-                },
-            )
-            continue
-        messages.append(
-            {
-                "role": "tool",
-                "tool_name": tool_name,
-                "content": json.dumps(tool_result, sort_keys=True),
-            },
-        )
-    return None
-
-
-def parse_tool_calls(message: dict[str, object]) -> list[tuple[str, dict[str, object]]]:
-    """Parse Ollama tool calls from a response message."""
-    raw_tool_calls = message.get("tool_calls") or []
-    if not isinstance(raw_tool_calls, list):
-        msg = "tool_calls must be a list"
-        raise MetadataResolutionError(msg)
-
-    tool_calls = []
-    for raw_call in raw_tool_calls:
-        if not isinstance(raw_call, dict):
-            msg = "tool call must be an object"
-            raise MetadataResolutionError(msg)
-        function = raw_call.get("function")
-        if not isinstance(function, dict):
-            msg = "tool call is missing function"
-            raise MetadataResolutionError(msg)
-        name = function.get("name")
-        if not isinstance(name, str) or not name:
-            msg = "tool call is missing function name"
-            raise MetadataResolutionError(msg)
-        arguments = parse_tool_arguments(function.get("arguments", {}))
-        tool_calls.append((name, arguments))
-    return tool_calls
-
-
-def parse_tool_arguments(raw_arguments: object) -> dict[str, object]:
-    """Parse tool call arguments returned by Ollama."""
-    if isinstance(raw_arguments, dict):
-        return {str(key): value for key, value in raw_arguments.items()}
-    if isinstance(raw_arguments, str):
-        parsed = json.loads(raw_arguments) if raw_arguments else {}
-        if isinstance(parsed, dict):
-            return {str(key): value for key, value in parsed.items()}
-    msg = "tool arguments must be an object"
-    raise MetadataResolutionError(msg)
-
-
-def validate_title_slug(title: str) -> None:
-    """Validate a canonical book title slug."""
-    if not TITLE_SLUG_PATTERN.fullmatch(title):
-        msg = f"title slug is invalid: {title}"
-        raise MetadataResolutionError(msg)
-
-
-def validate_catalog_slug(value: str, label: str) -> None:
-    """Validate a canonical catalog slug."""
-    if not CATALOG_SLUG_PATTERN.fullmatch(value):
-        msg = f"{label} slug is invalid: {value}"
-        raise MetadataResolutionError(msg)
-
-
-def normalize_catalog_slug(value: str) -> str:
-    """Normalize noisy catalog names into lower snake-case slugs."""
-    return re.sub(r"[^a-z0-9]+", "_", value.strip().casefold()).strip("_")
-
-
-def compact_catalog_slug(value: str) -> str:
-    """Return a catalog slug comparison key that ignores underscores."""
-    return normalize_catalog_slug(value).replace("_", "")
-
-
-def normalize_title_slug(value: str) -> str:
-    """Normalize noisy book titles into lower kebab-case slugs."""
-    return re.sub(r"[^a-z0-9]+", "-", value.strip().casefold()).strip("-")
-
-
-def is_fatal_tool_error(error: MetadataResolutionError) -> bool:
-    """Return whether a tool error should stop the agent immediately."""
-    message = str(error)
-    return message.startswith(
-        (
-            "Unknown audiobook metadata tool",
-            "Audiobook metadata tool is not enabled",
-        ),
-    )
-
-
-def query_terms(query: str) -> tuple[str, ...]:
-    """Return text variants useful for matching noisy audiobook metadata."""
-    normalized = query.strip().casefold()
-    underscore_slug = normalize_catalog_slug(normalized)
-    compact_slug = compact_catalog_slug(normalized)
-    hyphen_slug = normalize_title_slug(normalized)
-    return tuple(dict.fromkeys(term for term in (normalized, underscore_slug, compact_slug, hyphen_slug) if term))
-
-
-def required_string(data: dict[str, object], key: str) -> str:
-    """Read a required string field."""
-    value = data.get(key)
-    if not isinstance(value, str) or not value.strip():
-        msg = f"{key} must be a non-empty string"
-        raise MetadataResolutionError(msg)
-    return value.strip()
-
-
-def required_int(data: dict[str, object], key: str) -> int:
-    """Read a required integer field."""
-    value = data.get(key)
-    if isinstance(value, bool) or not isinstance(value, int):
-        msg = f"{key} must be an integer"
-        raise MetadataResolutionError(msg)
-    return value
-
-
-def required_series_index(data: dict[str, object], key: str) -> float:
-    """Read a required whole-number or half-number series index."""
-    value = data.get(key)
-    if isinstance(value, bool) or not isinstance(value, int | float):
-        msg = f"{key} must be a number"
-        raise MetadataResolutionError(msg)
-    series_index = float(value)
-    if not (series_index * 2).is_integer():
-        msg = f"{key} must be a whole number or .5 increment"
-        raise MetadataResolutionError(msg)
-    return series_index
-
-
-def optional_int(value: object, key: str) -> int | None:
-    """Read an optional integer field."""
-    if value is None:
-        return None
-    if isinstance(value, bool) or not isinstance(value, int):
-        msg = f"{key} must be an integer or null"
-        raise MetadataResolutionError(msg)
-    return value

python/tools/audiobook/metadata_agent.py

@@ -1,575 +0,0 @@
-"""Resolve audiobook metadata with a controlled Ollama tool loop."""
-
-from __future__ import annotations
-
-import json
-import re
-from dataclasses import asdict, dataclass, is_dataclass, replace
-from os import PathLike
-from typing import TYPE_CHECKING
-
-import httpx
-from sqlalchemy.orm import Session
-
-from python.common import utcnow
-from python.tools.audiobook.llm_tool_calling import (
-    CatalogToolRegistry,
-    MetadataResolutionError,
-    normalize_title_slug,
-    optional_int,
-    parse_tool_calls,
-    required_int,
-    required_series_index,
-    required_string,
-    run_tool_calls,
-    validate_catalog_slug,
-    validate_title_slug,
-)
-
-if TYPE_CHECKING:
-    from pathlib import Path
-
-    from sqlalchemy.engine import Engine
-
-    from python.orm.richie import AudiobookAuthor
-
-FENCED_JSON_PATTERN = re.compile(r"^```(?:json)?\s*(?P<json>.*?)\s*```$", re.IGNORECASE | re.DOTALL)
-
-
-@dataclass(frozen=True)
-class AgentConfig:
-    """Runtime settings for the audiobook metadata agent."""
-
-    model: str = "deepseek-v4-flash:cloud"
-    ollama_chat_url: str = "https://ollama.com/api/chat"
-    http_timeout_seconds: int = 300
-    max_agent_turns: int = 8
-    max_tool_results: int = 10
-    min_confidence: float = 0.85
-    invalid_final_retries: int = 1
-    standalone_series: str = "standalone"
-    tool_names: tuple[str, ...] = (
-        "search_authors",
-        "search_series",
-        "search_books",
-        "ensure_author",
-        "ensure_series",
-        "ensure_book",
-    )
-
-
-@dataclass(frozen=True)
-class StandardBookMetadata:
-    """Canonical metadata for the final audiobook path."""
-
-    author_id: int
-    author: str
-    book_id: int | None
-    title: str
-    series_id: int | None
-    series: str
-    series_index: float
-    confidence: float
-    needs_review: bool
-    evidence: list[str]
-
-
-@dataclass(frozen=True)
-class FinalMetadataFields:
-    """Raw model fields after schema validation."""
-
-    author_id: int
-    book_id: int | None
-    title: str
-    series_id: int | None
-    series_index: float
-    confidence: float
-    evidence: list[str]
-
-
-@dataclass(frozen=True)
-class ResolvedBookFields:
-    """Book fields after optional catalog book resolution."""
-
-    book_id: int | None
-    title: str
-    series_id: int | None
-    series_index: float
-
-
-@dataclass(frozen=True)
-class AgentStepResult:
-    """Outcome from one model response."""
-
-    metadata: StandardBookMetadata | None
-    invalid_final_count: int
-    should_continue: bool
-
-
-def standard_book_metadata(
-    aax_file_name: str,
-    aax_metadata_from_ffprobe: dict[str, str],
-    engine: Engine,
-    log_path: Path,
-    ollama_api_key: str,
-    config: AgentConfig,
-) -> StandardBookMetadata:
-    """Resolve canonical audiobook metadata with the configured Ollama Cloud model."""
-    with Session(engine) as session:
-        registry = CatalogToolRegistry(session, log_path, config, write_agent_log)
-        agent = AudiobookMetadataAgent(
-            registry=registry, log_path=log_path, ollama_api_key=ollama_api_key, config=config
-        )
-        metadata = agent.run(aax_file_name, aax_metadata_from_ffprobe)
-        if metadata.needs_review:
-            session.rollback()
-        else:
-            registry.prune_unused_created_rows(
-                author_id=metadata.author_id,
-                book_id=metadata.book_id,
-                series_id=metadata.series_id,
-            )
-            session.commit()
-        return metadata
-
-
-class AudiobookMetadataAgent:
-    """Ollama-backed metadata resolver with a fixed local tool registry."""
-
-    def __init__(
-        self,
-        *,
-        registry: CatalogToolRegistry,
-        log_path: Path,
-        ollama_api_key: str,
-        config: AgentConfig,
-    ) -> None:
-        """Create an Ollama metadata agent."""
-        self._registry = registry
-        self._log_path = log_path
-        self._ollama_api_key = ollama_api_key
-        self._config = config
-
-    def run(self, aax_file_name: str, aax_metadata_from_ffprobe: dict[str, str]) -> StandardBookMetadata:
-        """Resolve metadata for one AAX file."""
-        messages = [
-            {"role": "system", "content": system_prompt()},
-            {"role": "user", "content": user_prompt(aax_file_name, aax_metadata_from_ffprobe)},
-        ]
-        invalid_final_count = 0
-        result: StandardBookMetadata | None = None
-
-        for turn in range(1, self._config.max_agent_turns + 1):
-            step = self.run_step(messages, turn, invalid_final_count)
-            invalid_final_count = step.invalid_final_count
-            if step.should_continue:
-                continue
-            result = step.metadata
-            break
-
-        if result is None:
-            return self.force_final_response(messages)
-        return result
-
-    def run_step(
-        self,
-        messages: list[dict[str, object]],
-        turn: int,
-        invalid_final_count: int,
-    ) -> AgentStepResult:
-        """Run one model turn and return the next agent-loop action."""
-        data = self.chat(messages, turn)
-        message = data.get("message")
-        if not isinstance(message, dict):
-            return AgentStepResult(
-                metadata=review_metadata("Ollama response did not include a message", self._config),
-                invalid_final_count=invalid_final_count,
-                should_continue=False,
-            )
-
-        try:
-            tool_calls = parse_tool_calls(message)
-        except (json.JSONDecodeError, MetadataResolutionError) as error:
-            return AgentStepResult(
-                metadata=review_metadata(str(error), self._config),
-                invalid_final_count=invalid_final_count,
-                should_continue=False,
-            )
-        if tool_calls:
-            fatal_error = run_tool_calls(messages, message, tool_calls, self._registry, self._log_path, write_agent_log)
-            if fatal_error is not None:
-                return AgentStepResult(
-                    metadata=review_metadata(fatal_error, self._config),
-                    invalid_final_count=invalid_final_count,
-                    should_continue=False,
-                )
-            return AgentStepResult(metadata=None, invalid_final_count=invalid_final_count, should_continue=True)
-        return self.handle_final_message(messages, message, invalid_final_count)
-
-    def handle_final_message(
-        self,
-        messages: list[dict[str, object]],
-        message: dict[str, object],
-        invalid_final_count: int,
-    ) -> AgentStepResult:
-        """Validate a final model message or request one retry."""
-        content = message.get("content")
-        if not isinstance(content, str):
-            return AgentStepResult(
-                metadata=review_metadata("Ollama final response did not include string content", self._config),
-                invalid_final_count=invalid_final_count,
-                should_continue=False,
-            )
-
-        try:
-            resolved = self.validate_final(parse_final_json_content(content))
-        except (json.JSONDecodeError, MetadataResolutionError) as error:
-            return self.handle_invalid_final(messages, error, invalid_final_count)
-
-        write_agent_log(self._log_path, "final_metadata", metadata=resolved)
-        return AgentStepResult(metadata=resolved, invalid_final_count=invalid_final_count, should_continue=False)
-
-    def handle_invalid_final(
-        self,
-        messages: list[dict[str, object]],
-        error: json.JSONDecodeError | MetadataResolutionError,
-        invalid_final_count: int,
-    ) -> AgentStepResult:
-        """Log invalid final JSON and either retry or return review metadata."""
-        invalid_final_count += 1
-        write_agent_log(
-            self._log_path,
-            "final_validation_error",
-            error=str(error),
-            invalid_final_count=invalid_final_count,
-        )
-        if invalid_final_count > self._config.invalid_final_retries:
-            return AgentStepResult(
-                metadata=review_metadata(str(error), self._config),
-                invalid_final_count=invalid_final_count,
-                should_continue=False,
-            )
-        messages.append(
-            {
-                "role": "user",
-                "content": (
-                    "Your previous final answer was invalid. Return only valid JSON matching the required "
-                    f"schema. Validation error: {error}"
-                ),
-            },
-        )
-        return AgentStepResult(metadata=None, invalid_final_count=invalid_final_count, should_continue=True)
-
-    def force_final_response(self, messages: list[dict[str, object]]) -> StandardBookMetadata:
-        """Request a no-tool final answer after the normal turn limit."""
-        messages.append({"role": "user", "content": forced_final_prompt()})
-        write_agent_log(self._log_path, "forced_final_request", reason="max_turns")
-        data = self.chat(messages, self._config.max_agent_turns + 1, tools_enabled=False)
-        message = data.get("message")
-        if not isinstance(message, dict):
-            return review_metadata("Ollama forced final response did not include a message", self._config)
-        content = message.get("content")
-        if not isinstance(content, str):
-            return review_metadata("Ollama forced final response did not include string content", self._config)
-        try:
-            resolved = self.validate_final(parse_final_json_content(content))
-        except (json.JSONDecodeError, MetadataResolutionError) as error:
-            return review_metadata(f"Ollama forced final response was invalid: {error}", self._config)
-        write_agent_log(self._log_path, "final_metadata", metadata=resolved)
-        return resolved
-
-    def chat(self, messages: list[dict[str, object]], turn: int, *, tools_enabled: bool = True) -> dict[str, object]:
-        """Send one chat request to Ollama and log the request and response."""
-        payload = {
-            "model": self._config.model,
-            "messages": messages,
-            "stream": False,
-            "options": {"temperature": 0.1},
-        }
-        tool_names = []
-        if tools_enabled:
-            payload["tools"] = self._registry.tool_schemas()
-            tool_names = self._config.tool_names
-        write_agent_log(
-            self._log_path,
-            "model_request",
-            model=self._config.model,
-            turn=turn,
-            message_count=len(messages),
-            tool_names=tool_names,
-            tools_enabled=tools_enabled,
-        )
-        write_agent_log(
-            self._log_path,
-            "llm_messages_sent",
-            model=self._config.model,
-            turn=turn,
-            messages=messages,
-            tools_enabled=tools_enabled,
-        )
-        response = httpx.post(
-            self._config.ollama_chat_url,
-            headers={"Authorization": f"Bearer {self._ollama_api_key}"},
-            json=payload,
-            timeout=self._config.http_timeout_seconds,
-        )
-        response.raise_for_status()
-        raw_data = response.json()
-        if not isinstance(raw_data, dict):
-            return {}
-        data = {str(key): value for key, value in raw_data.items()}
-        message = data.get("message", {})
-        content = message.get("content") if isinstance(message, dict) else ""
-        write_agent_log(
-            self._log_path,
-            "llm_message_received",
-            model=self._config.model,
-            turn=turn,
-            message=message,
-        )
-        write_agent_log(
-            self._log_path,
-            "model_response",
-            model=self._config.model,
-            turn=turn,
-            has_tool_calls=bool(isinstance(message, dict) and message.get("tool_calls")),
-            content_chars=len(content) if isinstance(content, str) else 0,
-        )
-        return data
-
-    def validate_final(self, raw_metadata: object) -> StandardBookMetadata:
-        """Validate final model metadata against catalog rows."""
-        fields = parse_final_metadata_fields(raw_metadata)
-        fields = replace(fields, title=normalize_title_slug(fields.title))
-        author = self.validate_author(fields.author_id)
-        validate_title_slug(fields.title)
-        book_fields = self.resolve_book_fields(fields)
-        series = self.validate_series(fields.author_id, book_fields.series_id, book_fields.series_index)
-
-        return StandardBookMetadata(
-            author_id=fields.author_id,
-            author=author.name,
-            book_id=book_fields.book_id,
-            title=book_fields.title,
-            series_id=book_fields.series_id,
-            series=series,
-            series_index=book_fields.series_index,
-            confidence=fields.confidence,
-            needs_review=fields.confidence < self._config.min_confidence,
-            evidence=fields.evidence,
-        )
-
-    def validate_author(self, author_id: int) -> AudiobookAuthor:
-        """Validate that an author id was seen and exists."""
-        if author_id not in self._registry.seen_author_ids:
-            msg = f"author_id {author_id} was not returned by search_authors"
-            raise MetadataResolutionError(msg)
-        author = self._registry.get_author(author_id)
-        if author is None:
-            msg = f"author_id {author_id} does not exist"
-            raise MetadataResolutionError(msg)
-        validate_catalog_slug(author.name, "author")
-        return author
-
-    def resolve_book_fields(self, fields: FinalMetadataFields) -> ResolvedBookFields:
-        """Resolve final book fields from a seen book id or created book."""
-        if fields.book_id is None:
-            ensured = self._registry.ensure_book(
-                fields.title,
-                fields.author_id,
-                fields.series_id,
-                fields.series_index,
-            )
-            return ResolvedBookFields(
-                book_id=ensured.book.id,
-                title=ensured.book.title,
-                series_id=ensured.book.series_id,
-                series_index=ensured.book.series_index,
-            )
-
-        if fields.book_id not in self._registry.seen_book_ids:
-            msg = f"book_id {fields.book_id} was not returned by search_books"
-            raise MetadataResolutionError(msg)
-        book = self._registry.get_book(fields.book_id)
-        if book is None:
-            msg = f"book_id {fields.book_id} does not exist"
-            raise MetadataResolutionError(msg)
-        if book.author_id != fields.author_id:
-            msg = f"book_id {fields.book_id} does not belong to author_id {fields.author_id}"
-            raise MetadataResolutionError(msg)
-        return ResolvedBookFields(
-            book_id=fields.book_id,
-            title=book.title,
-            series_id=book.series_id,
-            series_index=book.series_index,
-        )
-
-    def validate_series(self, author_id: int, series_id: int | None, series_index: float) -> str:
-        """Validate final series fields and return the canonical series slug."""
-        if series_id is None:
-            if series_index != 0:
-                msg = "standalone books must use series_index 0"
-                raise MetadataResolutionError(msg)
-            return self._config.standalone_series
-
-        if series_id not in self._registry.seen_series_ids:
-            msg = f"series_id {series_id} was not returned by search_series"
-            raise MetadataResolutionError(msg)
-        series = self._registry.get_series(series_id)
-        if series is None:
-            msg = f"series_id {series_id} does not exist"
-            raise MetadataResolutionError(msg)
-        if series.author_id != author_id:
-            msg = f"series_id {series_id} does not belong to author_id {author_id}"
-            raise MetadataResolutionError(msg)
-        if series_index <= 0:
-            msg = "series books must use a positive series_index"
-            raise MetadataResolutionError(msg)
-        validate_catalog_slug(series.name, "series")
-        return series.name
-
-
-def write_agent_log(log_path: Path, event: str, **fields: object) -> None:
-    """Append one JSONL audit event."""
-    log_path.parent.mkdir(parents=True, exist_ok=True)
-    record = {
-        "created": utcnow().isoformat(),
-        "event": event,
-        **{key: json_log_value(value) for key, value in fields.items()},
-    }
-    with log_path.open("a", encoding="utf-8") as file:
-        file.write(json.dumps(record, sort_keys=True))
-        file.write("\n")
-
-
-def json_log_value(value: object) -> object:
-    """Return a JSON-serializable value for audit logs."""
-    if is_dataclass(value) and not isinstance(value, type):
-        return json_log_value(asdict(value))
-    if isinstance(value, dict):
-        return {str(key): json_log_value(item) for key, item in value.items()}
-    if isinstance(value, list | tuple):
-        return [json_log_value(item) for item in value]
-    if isinstance(value, set):
-        return [json_log_value(item) for item in sorted(value, key=str)]
-    if isinstance(value, PathLike):
-        return str(value)
-    return value
-
-
-def system_prompt() -> str:
-    """Return the stable system prompt."""
-    return """You standardize Audible audiobook metadata against a private catalog.
-
-Rules:
-- You must use the provided tools before returning final metadata.
-- Only use author_id, series_id, or book_id values returned by tools.
-- Return final metadata as JSON only. Do not wrap it in Markdown.
-- The final JSON object must contain author_id, book_id, title, series_id, series_index, confidence, and evidence.
-- title must be a canonical title slug using lower-case words separated by hyphens.
-- Use series_id null and series_index 0 for standalone books.
-- If you use a series_id, series_index must be a whole number or .5 value greater than 0.
-- Treat series slugs that differ only by underscores as the same series. Prefer the existing catalog row instead of
-  creating a new series.
-- Detect omnibus or box-set editions that contain multiple numbered novels, books, or novellas.
-- For an omnibus, make a best-effort range from the filename, tags, and catalog rows. Keep series_index as the
-  first covered book number and include the range in the title when the source title includes it, for example
-  books-1-3.
-- Be careful with omnibuses of novels or novellas later published as one book: keep the omnibus as the audiobook's
-  book record unless catalog rows clearly identify a better match.
-- Do not create publisher collections or author collections as series unless the book metadata clearly gives a
-  numbered series.
-- Series belong to authors. Use a series_id only when it belongs to the selected author_id.
-- Always search for the author before creating one. If no exact author slug exists, call ensure_author.
-- Always search for a series with author_id before creating one. If no exact series slug exists, call ensure_series.
-- Always search for a book before creating one. If no exact title slug exists, call ensure_book.
-- If a tool returns an error, correct your tool arguments or final metadata before continuing.
-- confidence must be a number from 0 to 1.
-- evidence must be a short list of strings explaining which filename, tags, and catalog rows support the answer."""
-
-
-def forced_final_prompt() -> str:
-    """Return the no-tools finalization prompt."""
-    return (
-        "Stop calling tools. Return final metadata as JSON only using the tool results already provided. "
-        "If search_books returned no matching rows but author and series are known, use book_id null and resolve "
-        "the title slug from the AAX filename and ffprobe tags. The validator will create the missing book. "
-        "Use only author_id and series_id values returned by earlier tool results."
-    )
-
-
-def user_prompt(aax_file_name: str, metadata: dict[str, str]) -> str:
-    """Build the user prompt from source metadata."""
-    return (
-        "Resolve this Audible audiobook.\n\n"
-        f"AAX file name: {aax_file_name}\n\n"
-        "ffprobe format tags:\n"
-        f"{json.dumps(metadata, indent=2, sort_keys=True)}"
-    )
-
-
-def parse_final_json_content(content: str) -> object:
-    """Parse final model content, accepting bare or fenced JSON."""
-    stripped = content.strip()
-    if match := FENCED_JSON_PATTERN.fullmatch(stripped):
-        stripped = match.group("json").strip()
-    return json.loads(stripped)
-
-
-def parse_final_metadata_fields(raw_metadata: object) -> FinalMetadataFields:
-    """Parse the model's final JSON object into typed fields."""
-    if not isinstance(raw_metadata, dict):
-        msg = "Final metadata must be a JSON object"
-        raise MetadataResolutionError(msg)
-    data = {str(key): value for key, value in raw_metadata.items()}
-    return FinalMetadataFields(
-        author_id=required_int(data, "author_id"),
-        book_id=optional_int(data.get("book_id"), "book_id"),
-        title=required_string(data, "title"),
-        series_id=optional_int(data.get("series_id"), "series_id"),
-        series_index=required_series_index(data, "series_index"),
-        confidence=required_float(data, "confidence"),
-        evidence=required_string_list(data, "evidence"),
-    )
-
-
-def review_metadata(reason: str, config: AgentConfig) -> StandardBookMetadata:
-    """Return a metadata result that must be reviewed manually."""
-    return StandardBookMetadata(
-        author_id=0,
-        author="unknown_author",
-        book_id=None,
-        title="unknown-title",
-        series_id=None,
-        series=config.standalone_series,
-        series_index=0,
-        confidence=0,
-        needs_review=True,
-        evidence=[reason],
-    )
-
-
-def required_float(data: dict[str, object], key: str) -> float:
-    """Read a required float field."""
-    value = data.get(key)
-    if isinstance(value, bool) or not isinstance(value, int | float):
-        msg = f"{key} must be a number"
-        raise MetadataResolutionError(msg)
-    confidence = float(value)
-    if confidence < 0 or confidence > 1:
-        msg = f"{key} must be between 0 and 1"
-        raise MetadataResolutionError(msg)
-    return confidence
-
-
-def required_string_list(data: dict[str, object], key: str) -> list[str]:
-    """Read a required list of strings."""
-    value = data.get(key)
-    if not isinstance(value, list) or not value or not all(isinstance(item, str) for item in value):
-        msg = f"{key} must be a non-empty list of strings"
-        raise MetadataResolutionError(msg)
-    strings = [item.strip() for item in value if item.strip()]
-    if not strings:
-        msg = f"{key} must include at least one non-empty string"
-        raise MetadataResolutionError(msg)
-    return strings

python/tools/whisper/Dockerfile

@@ -1,17 +0,0 @@
-FROM nvidia/cuda:12.4.1-cudnn-runtime-ubuntu22.04
-
-ENV DEBIAN_FRONTEND=noninteractive \
-    PYTHONDONTWRITEBYTECODE=1 \
-    PYTHONUNBUFFERED=1
-
-RUN apt-get update \
-    && apt-get install -y --no-install-recommends python3 python3-pip ffmpeg \
-    && rm -rf /var/lib/apt/lists/*
-
-RUN pip3 install --no-cache-dir --upgrade pip \
-    && pip3 install --no-cache-dir faster-whisper requests
-
-WORKDIR /app
-COPY python/tools/whisper/inference.py /app/inference.py
-
-ENTRYPOINT ["python3", "/app/inference.py"]

python/tools/whisper/Dockerfile.dockerignore

@@ -1,2 +0,0 @@
-*
-!python/tools/whisper/inference.py

python/tools/whisper/__init__.py

@@ -1 +0,0 @@
-"""Whisper transcription tools (host orchestrator and container entrypoint)."""

python/tools/whisper/inference.py

@@ -1,136 +0,0 @@
-"""Container entrypoint that transcribes a directory of audio files with faster-whisper.
-
-Run inside the whisper-transcribe docker image; segment timestamps are grouped
-into one-minute buckets so the output reads as ``[HH:MM:00] text``.
-"""
-
-from __future__ import annotations
-
-import argparse
-import logging
-from pathlib import Path
-
-from faster_whisper import WhisperModel
-
-logger = logging.getLogger(__name__)
-
-AUDIO_EXTENSIONS = {".mp3", ".wav", ".m4a", ".flac", ".ogg", ".opus", ".mp4", ".mkv", ".webm", ".aac"}
-BUCKET_SECONDS = 60
-BEAM_SIZE = 5
-SECONDS_PER_HOUR = 3600
-SECONDS_PER_MINUTE = 60
-
-
-def format_timestamp(total_seconds: float) -> str:
-    """Render a whole-minute timestamp as ``HH:MM:00``.
-
-    Args:
-        total_seconds: Offset in seconds from the start of the audio.
-
-    Returns:
-        A zero-padded ``HH:MM:00`` string.
-    """
-    hours = int(total_seconds // SECONDS_PER_HOUR)
-    minutes = int((total_seconds % SECONDS_PER_HOUR) // SECONDS_PER_MINUTE)
-    return f"{hours:02d}:{minutes:02d}:00"
-
-
-def transcribe_file(model: WhisperModel, audio_path: Path, output_path: Path) -> None:
-    """Transcribe one audio file and write the bucketed transcript to disk.
-
-    Args:
-        model: Loaded faster-whisper model.
-        audio_path: Source audio file.
-        output_path: Destination ``.txt`` path.
-    """
-    logger.info("Transcribing %s", audio_path)
-    segments, info = model.transcribe(
-        str(audio_path),
-        language="en",
-        beam_size=BEAM_SIZE,
-        vad_filter=True,
-    )
-    logger.info("Duration %.1fs", info.duration)
-
-    buckets: dict[int, list[str]] = {}
-    for segment in segments:
-        bucket = int(segment.start // BUCKET_SECONDS)
-        buckets.setdefault(bucket, []).append(segment.text.strip())
-
-    lines = [f"[{format_timestamp(bucket * BUCKET_SECONDS)}] {' '.join(buckets[bucket])}" for bucket in sorted(buckets)]
-    output_path.write_text("\n\n".join(lines) + "\n", encoding="utf-8")
-    logger.info("Wrote %s", output_path)
-
-
-def find_audio_files(input_directory: Path) -> list[Path]:
-    """Collect every audio file under ``input_directory``.
-
-    Args:
-        input_directory: Directory to walk recursively.
-
-    Returns:
-        Sorted list of audio file paths.
-    """
-    return sorted(
-        path for path in input_directory.rglob("*") if path.is_file() and path.suffix.lower() in AUDIO_EXTENSIONS
-    )
-
-
-def configure_container_logger() -> None:
-    """Configure logging for the container (stdout, INFO)."""
-    logging.basicConfig(
-        level=logging.INFO,
-        format="%(asctime)s %(levelname)s %(message)s",
-    )
-
-
-def parse_arguments() -> argparse.Namespace:
-    """Parse CLI arguments for the container entrypoint.
-
-    Returns:
-        Parsed argparse namespace.
-    """
-    parser = argparse.ArgumentParser(description=__doc__)
-    parser.add_argument("--input", type=Path, default=Path("/audio"))
-    parser.add_argument("--output", type=Path, default=Path("/output"))
-    parser.add_argument("--model", default="large-v3")
-    parser.add_argument(
-        "--download-only",
-        action="store_true",
-        help="Download the model into the cache volume and exit without transcribing.",
-    )
-    return parser.parse_args()
-
-
-def main() -> None:
-    """Load the model, then either exit (download-only) or transcribe the directory."""
-    configure_container_logger()
-    arguments = parse_arguments()
-
-    logger.info("Loading model %s on CUDA", arguments.model)
-    model = WhisperModel(arguments.model, device="cuda", compute_type="float16")
-
-    if arguments.download_only:
-        logger.info("Model ready; exiting (download-only mode)")
-        return
-
-    arguments.output.mkdir(parents=True, exist_ok=True)
-
-    audio_files = find_audio_files(arguments.input)
-    if not audio_files:
-        logger.warning("No audio files found in %s", arguments.input)
-        return
-
-    logger.info("Found %d audio file(s)", len(audio_files))
-    for audio_path in audio_files:
-        relative = audio_path.relative_to(arguments.input)
-        output_path = arguments.output / relative.with_suffix(".txt")
-        output_path.parent.mkdir(parents=True, exist_ok=True)
-        if output_path.exists():
-            logger.info("Skip %s (already transcribed)", relative)
-            continue
-        transcribe_file(model, audio_path, output_path)
-
-
-if __name__ == "__main__":
-    main()

python/tools/whisper/transcribe.py

@@ -1,167 +0,0 @@
-"""Build and run the whisper transcription docker container on demand.
-
-The container is started fresh for each invocation and removed on exit
-(``docker run --rm``). The model is cached in a named docker volume so
-only the first run pays the download cost.
-"""
-
-from __future__ import annotations
-
-import logging
-import subprocess
-from pathlib import Path
-from typing import Annotated
-
-import typer
-
-from python.common import configure_logger
-
-logger = logging.getLogger(__name__)
-
-
-class Config:
-    """Paths and names for the whisper-transcribe Docker workflow."""
-
-    image_tag = "whisper-transcribe:latest"
-    model_volume = "whisper-models"
-    repo_root = Path(__file__).resolve().parents[3]
-    dockerfile = Path(__file__).resolve().parent / "Dockerfile"
-    huggingface_cache = "/root/.cache/huggingface"
-
-
-def run_docker(arguments: list[str]) -> None:
-    """Run a docker subcommand, streaming output and raising on failure.
-
-    Args:
-        arguments: Arguments to pass to the ``docker`` binary.
-
-    Raises:
-        subprocess.CalledProcessError: If docker exits non-zero.
-    """
-    logger.info("docker %s", " ".join(arguments))
-    subprocess.run(["docker", *arguments], check=True)
-
-
-def build_image() -> None:
-    """Build the whisper-transcribe image using the repo root as build context."""
-    logger.info("Building image %s", Config.image_tag)
-    run_docker(
-        [
-            "build",
-            "--tag",
-            Config.image_tag,
-            "--file",
-            str(Config.dockerfile),
-            str(Config.repo_root),
-        ],
-    )
-
-
-def model_cache_present(model: str) -> bool:
-    """Check whether the given model is already downloaded in the cache volume.
-
-    Args:
-        model: faster-whisper model name (e.g. ``large-v3``).
-
-    Returns:
-        True if the HuggingFace cache directory for the model exists in the volume.
-    """
-    cache_directory = f"hub/models--Systran--faster-whisper-{model}"
-    completed = subprocess.run(
-        [
-            "docker",
-            "run",
-            "--rm",
-            "--volume",
-            f"{Config.model_volume}:/cache",
-            "alpine",
-            "test",
-            "-d",
-            f"/cache/{cache_directory}",
-        ],
-        check=False,
-    )
-    return completed.returncode == 0
-
-
-def download_model(model: str) -> None:
-    """Download the model into the cache volume and exit.
-
-    Args:
-        model: faster-whisper model name.
-    """
-    logger.info("Downloading model %s into volume %s", model, Config.model_volume)
-    run_docker(
-        [
-            "run",
-            "--rm",
-            "--device=nvidia.com/gpu=all",
-            "--ipc=host",
-            "--volume",
-            f"{Config.model_volume}:{Config.huggingface_cache}",
-            Config.image_tag,
-            "--model",
-            model,
-            "--download-only",
-        ],
-    )
-
-
-def transcribe(input_directory: Path, output_directory: Path, model: str) -> None:
-    """Run transcription on every audio file under ``input_directory``.
-
-    Args:
-        input_directory: Host path containing audio files (mounted read-only).
-        output_directory: Host path for ``.txt`` transcripts.
-        model: faster-whisper model name.
-    """
-    logger.info("Transcribing %s -> %s (model=%s)", input_directory, output_directory, model)
-    run_docker(
-        [
-            "run",
-            "--rm",
-            "--device=nvidia.com/gpu=all",
-            "--ipc=host",
-            "--volume",
-            f"{input_directory}:/audio:ro",
-            "--volume",
-            f"{output_directory}:/output",
-            "--volume",
-            f"{Config.model_volume}:{Config.huggingface_cache}",
-            Config.image_tag,
-            "--model",
-            model,
-        ],
-    )
-
-
-def main(
-    input_directory: Annotated[Path, typer.Argument(help="Directory of audio files to transcribe.")],
-    output_directory: Annotated[Path, typer.Argument(help="Directory to write .txt transcripts to.")],
-    model: Annotated[str, typer.Option(help="faster-whisper model name.")] = "large-v3",
-    *,
-    force_download: Annotated[
-        bool,
-        typer.Option("--force-download", help="Re-download the model even if already cached."),
-    ] = False,
-) -> None:
-    """Build the image, ensure the model is cached, then transcribe and stop."""
-    configure_logger()
-
-    resolved_input = input_directory.resolve(strict=True)
-    output_directory.mkdir(parents=True, exist_ok=True)
-    resolved_output = output_directory.resolve()
-
-    build_image()
-
-    if force_download or not model_cache_present(model):
-        download_model(model)
-    else:
-        logger.info("Model %s already cached in volume %s", model, Config.model_volume)
-
-    transcribe(resolved_input, resolved_output, model)
-    logger.info("Done. Container stopped.")
-
-
-if __name__ == "__main__":
-    typer.run(main)

systems/bob/default.nix

@@ -1,13 +1,11 @@
 { inputs, pkgs, ... }:
 {
   imports = [
-    "${inputs.self}/users/math"
     "${inputs.self}/users/richie"
-    "${inputs.self}/users/steve"
+    "${inputs.self}/users/math"
     "${inputs.self}/common/global"
     "${inputs.self}/common/optional/docker.nix"
     "${inputs.self}/common/optional/scanner.nix"
-    "${inputs.self}/common/optional/monitoring-agent.nix"
     "${inputs.self}/common/optional/steam.nix"
     "${inputs.self}/common/optional/syncthing_base.nix"
     "${inputs.self}/common/optional/systemd-boot.nix"
@@ -28,12 +26,7 @@
   networking = {
     hostName = "bob";
     hostId = "7c678a41";
-    firewall = {
+    firewall.enable = true;
-      enable = true;
-      allowedTCPPorts = [
-        8000
-      ];
-    };
     networkmanager.enable = true;
   };
 

systems/bob/hardware.nix

@@ -28,13 +28,9 @@
         allowDiscards = true;
         keyFileSize = 4096;
         keyFile = "/dev/disk/by-id/usb-Samsung_Flash_Drive_FIT_0374620080067131-0:0";
+        fallbackToPassword = true;
       };
     };
-
-    zfs.extraPools = [
-      "storage"
-    ];
-
     kernelModules = [ "kvm-amd" ];
     extraModulePackages = [ ];
   };

systems/bob/llms.nix

@@ -42,14 +42,11 @@
       "qwen3:8b"
       "qwen3.5:27b"
       "qwen3.5:35b"
-      "qwen3.6:27b"
-      "qwen3.6:35b"
-      "rinex20/translategemma3:12b"
       "translategemma:12b"
       "translategemma:27b"
       "translategemma:4b"
     ];
-    models = "/zfs/storage/models";
+    models = "/zfs/models";
     openFirewall = true;
   };
 }

systems/bob/scripts/zfs.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+
+# zpools
+
+# storage
+sudo zpool create -f -o ashift=12 -O acltype=posixacl -O atime=off -O dnodesize=auto -O xattr=sa -O compression=zstd -m /zfs/storage storage mirror
+sudo zpool create -o ashift=12 -O acltype=posixacl -O atime=off -O dnodesize=auto -O xattr=sa -O compression=zstd -m /zfs/storage storage
+
+
+# storage datasets
+sudo zfs create storage/models -o recordsize=1M

systems/bob/snapshot_config.toml

@@ -24,6 +24,6 @@ monthly = 0
 
 ["root_pool/models"]
 15_min = 4
-hourly = 24
+hourly = 2
 daily = 0
 monthly = 0

systems/bob/syncthing.nix

@@ -31,15 +31,5 @@
       ];
       fsWatcherEnabled = true;
     };
-    "recordings" = {
-      path = "/home/richie/recordings";
-      devices = [
-        "jeeves"
-        "phone"
-        "rhapsody-in-green"
-      ];
-      fsWatcherEnabled = true;
-    };
-
   };
 }

systems/brain/hardware.nix

@@ -26,6 +26,7 @@
         allowDiscards = true;
         keyFileSize = 4096;
         keyFile = "/dev/disk/by-id/usb-USB_SanDisk_3.2Gen1_03021630090925173333-0:0";
+        fallbackToPassword = true;
       };
     };
     kernelModules = [ "kvm-intel" ];

systems/jeeves/default.nix

@@ -4,21 +4,17 @@ let
 in
 {
   imports = [
-    "${inputs.self}/users/dov"
-    "${inputs.self}/users/math"
     "${inputs.self}/users/richie"
-    "${inputs.self}/users/steve"
+    "${inputs.self}/users/math"
+    "${inputs.self}/users/dov"
     "${inputs.self}/common/global"
     "${inputs.self}/common/optional/docker.nix"
-    "${inputs.self}/common/optional/monitoring-agent.nix"
     "${inputs.self}/common/optional/ssh_decrypt.nix"
     "${inputs.self}/common/optional/syncthing_base.nix"
     "${inputs.self}/common/optional/update.nix"
     "${inputs.self}/common/optional/zerotier.nix"
-    ./monitoring
     ./docker
     ./services
-    ./web_services
     ./hardware.nix
     ./networking.nix
     ./programs.nix
@@ -39,10 +35,5 @@ in
     zerotierone.joinNetworks = [ "a09acf02330d37b9" ];
   };
 
-  users.groups = {
-    nornsight = { };
-    nornsight-admin = { };
-  };
-
   system.stateVersion = "24.05";
 }

systems/jeeves/hardware.nix

@@ -9,6 +9,7 @@ let
     inherit device;
     keyFileSize = 4096;
     keyFile = "/dev/disk/by-id/usb-XIAO_USB_Drive_24587CE29074-0:0";
+    fallbackToPassword = true;
   };
   makeLuksSSD =
     device:

systems/jeeves/monitoring/dashboards/overview.json

@@ -1,426 +0,0 @@
-{
-  "annotations": {
-    "list": [
-      {
-        "builtIn": 1,
-        "datasource": {
-          "type": "grafana",
-          "uid": "-- Grafana --"
-        },
-        "enable": true,
-        "hide": true,
-        "iconColor": "rgba(0, 211, 255, 1)",
-        "name": "Annotations & Alerts",
-        "type": "dashboard"
-      }
-    ]
-  },
-  "editable": true,
-  "fiscalYearStartMonth": 0,
-  "graphTooltip": 0,
-  "links": [],
-  "panels": [
-    {
-      "datasource": {
-        "type": "prometheus",
-        "uid": "prom-main"
-      },
-      "fieldConfig": {
-        "defaults": {
-          "unit": "percent"
-        },
-        "overrides": []
-      },
-      "gridPos": {
-        "h": 8,
-        "w": 6,
-        "x": 0,
-        "y": 0
-      },
-      "id": 1,
-      "options": {
-        "legend": {
-          "displayMode": "list",
-          "placement": "bottom"
-        },
-        "tooltip": {
-          "mode": "multi"
-        }
-      },
-      "targets": [
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "prom-main"
-          },
-          "editorMode": "code",
-          "expr": "100 * (1 - avg by (instance) (rate(node_cpu_seconds_total{mode=\"idle\"}[5m])))",
-          "legendFormat": "{{instance}}",
-          "range": true,
-          "refId": "A"
-        }
-      ],
-      "title": "CPU Used",
-      "type": "timeseries"
-    },
-    {
-      "datasource": {
-        "type": "prometheus",
-        "uid": "prom-main"
-      },
-      "fieldConfig": {
-        "defaults": {
-          "unit": "percent"
-        },
-        "overrides": []
-      },
-      "gridPos": {
-        "h": 8,
-        "w": 6,
-        "x": 6,
-        "y": 0
-      },
-      "id": 2,
-      "options": {
-        "legend": {
-          "displayMode": "list",
-          "placement": "bottom"
-        },
-        "tooltip": {
-          "mode": "multi"
-        }
-      },
-      "targets": [
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "prom-main"
-          },
-          "editorMode": "code",
-          "expr": "100 * (1 - (node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes))",
-          "legendFormat": "{{instance}}",
-          "range": true,
-          "refId": "A"
-        }
-      ],
-      "title": "RAM Used",
-      "type": "timeseries"
-    },
-    {
-      "datasource": {
-        "type": "prometheus",
-        "uid": "prom-main"
-      },
-      "fieldConfig": {
-        "defaults": {
-          "unit": "percent"
-        },
-        "overrides": []
-      },
-      "gridPos": {
-        "h": 8,
-        "w": 6,
-        "x": 12,
-        "y": 0
-      },
-      "id": 3,
-      "options": {
-        "legend": {
-          "displayMode": "list",
-          "placement": "bottom"
-        },
-        "tooltip": {
-          "mode": "multi"
-        }
-      },
-      "targets": [
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "prom-main"
-          },
-          "editorMode": "code",
-          "expr": "100 * (1 - (node_memory_SwapFree_bytes / node_memory_SwapTotal_bytes))",
-          "legendFormat": "{{instance}}",
-          "range": true,
-          "refId": "A"
-        }
-      ],
-      "title": "Swap Used",
-      "type": "timeseries"
-    },
-    {
-      "datasource": {
-        "type": "prometheus",
-        "uid": "prom-main"
-      },
-      "fieldConfig": {
-        "defaults": {
-          "unit": "short"
-        },
-        "overrides": []
-      },
-      "gridPos": {
-        "h": 8,
-        "w": 6,
-        "x": 18,
-        "y": 0
-      },
-      "id": 4,
-      "options": {
-        "legend": {
-          "displayMode": "list",
-          "placement": "bottom"
-        },
-        "tooltip": {
-          "mode": "multi"
-        }
-      },
-      "targets": [
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "prom-main"
-          },
-          "editorMode": "code",
-          "expr": "node_load1",
-          "legendFormat": "{{instance}} load1",
-          "range": true,
-          "refId": "A"
-        },
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "prom-main"
-          },
-          "editorMode": "code",
-          "expr": "node_load5",
-          "legendFormat": "{{instance}} load5",
-          "range": true,
-          "refId": "B"
-        },
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "prom-main"
-          },
-          "editorMode": "code",
-          "expr": "node_load15",
-          "legendFormat": "{{instance}} load15",
-          "range": true,
-          "refId": "C"
-        }
-      ],
-      "title": "Load",
-      "type": "timeseries"
-    },
-    {
-      "datasource": {
-        "type": "prometheus",
-        "uid": "prom-main"
-      },
-      "fieldConfig": {
-        "defaults": {
-          "unit": "Bps"
-        },
-        "overrides": []
-      },
-      "gridPos": {
-        "h": 9,
-        "w": 12,
-        "x": 0,
-        "y": 8
-      },
-      "id": 5,
-      "options": {
-        "legend": {
-          "displayMode": "list",
-          "placement": "bottom"
-        },
-        "tooltip": {
-          "mode": "multi"
-        }
-      },
-      "targets": [
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "prom-main"
-          },
-          "editorMode": "code",
-          "expr": "sum by (instance) (rate(node_disk_read_bytes_total[5m]))",
-          "legendFormat": "{{instance}} read",
-          "range": true,
-          "refId": "A"
-        },
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "prom-main"
-          },
-          "editorMode": "code",
-          "expr": "sum by (instance) (rate(node_disk_written_bytes_total[5m]))",
-          "legendFormat": "{{instance}} write",
-          "range": true,
-          "refId": "B"
-        }
-      ],
-      "title": "Disk Throughput",
-      "type": "timeseries"
-    },
-    {
-      "datasource": {
-        "type": "prometheus",
-        "uid": "prom-main"
-      },
-      "fieldConfig": {
-        "defaults": {
-          "unit": "percent"
-        },
-        "overrides": []
-      },
-      "gridPos": {
-        "h": 9,
-        "w": 12,
-        "x": 12,
-        "y": 8
-      },
-      "id": 6,
-      "options": {
-        "cellHeight": "sm",
-        "showHeader": true,
-        "sortBy": [
-          {
-            "desc": true,
-            "displayName": "Value"
-          }
-        ]
-      },
-      "targets": [
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "prom-main"
-          },
-          "editorMode": "code",
-          "expr": "100 * (1 - (node_filesystem_avail_bytes{mountpoint=~\"(/|/home|/var|/zfs.*)\",fstype!=\"\"} / node_filesystem_size_bytes{mountpoint=~\"(/|/home|/var|/zfs.*)\",fstype!=\"\"}))",
-          "format": "table",
-          "instant": true,
-          "legendFormat": "{{instance}} {{mountpoint}}",
-          "refId": "A"
-        }
-      ],
-      "title": "Filesystem Usage",
-      "type": "table"
-    },
-    {
-      "datasource": {
-        "type": "prometheus",
-        "uid": "prom-main"
-      },
-      "fieldConfig": {
-        "defaults": {
-          "unit": "percentunit"
-        },
-        "overrides": []
-      },
-      "gridPos": {
-        "h": 10,
-        "w": 12,
-        "x": 0,
-        "y": 17
-      },
-      "id": 7,
-      "options": {
-        "cellHeight": "sm",
-        "showHeader": true,
-        "sortBy": [
-          {
-            "desc": true,
-            "displayName": "Value"
-          }
-        ]
-      },
-      "targets": [
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "prom-main"
-          },
-          "editorMode": "code",
-          "expr": "topk(10, rate(namedprocess_namegroup_cpu_seconds_total[5m]))",
-          "format": "table",
-          "instant": true,
-          "legendFormat": "{{instance}} {{groupname}}",
-          "refId": "A"
-        }
-      ],
-      "title": "Top Grouped CPU",
-      "type": "table"
-    },
-    {
-      "datasource": {
-        "type": "prometheus",
-        "uid": "prom-main"
-      },
-      "fieldConfig": {
-        "defaults": {
-          "unit": "bytes"
-        },
-        "overrides": []
-      },
-      "gridPos": {
-        "h": 10,
-        "w": 12,
-        "x": 12,
-        "y": 17
-      },
-      "id": 8,
-      "options": {
-        "cellHeight": "sm",
-        "showHeader": true,
-        "sortBy": [
-          {
-            "desc": true,
-            "displayName": "Value"
-          }
-        ]
-      },
-      "targets": [
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "prom-main"
-          },
-          "editorMode": "code",
-          "expr": "topk(10, namedprocess_namegroup_memory_bytes{memtype=\"resident\"})",
-          "format": "table",
-          "instant": true,
-          "legendFormat": "{{instance}} {{groupname}}",
-          "refId": "A"
-        }
-      ],
-      "title": "Top Grouped Memory",
-      "type": "table"
-    }
-  ],
-  "refresh": "30s",
-  "schemaVersion": 39,
-  "style": "dark",
-  "tags": [
-    "monitoring"
-  ],
-  "templating": {
-    "list": []
-  },
-  "time": {
-    "from": "now-24h",
-    "to": "now"
-  },
-  "timepicker": {},
-  "timezone": "",
-  "title": "Overview",
-  "uid": "monitor-overview",
-  "version": 1,
-  "weekStart": ""
-}