deleting data_science code

this code was moved to https://gitea.tmmworkshop.com/Nornsight/weave

.github/workflows/build_systems.yml

@@ -23,6 +23,6 @@ jobs:
     steps:
       - uses: actions/checkout@v4
       - name: Build default package
-        run: "nixos-rebuild build --flake ./#${{ matrix.system }}"
+        run: "nixos-rebuild build --accept-flake-config --flake ./#${{ 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

@@ -1,30 +0,0 @@
-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

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

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

.gitignore

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

.vscode/settings.json

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

AGENTS.md

@@ -1,12 +0,0 @@
-## 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

@@ -23,7 +23,10 @@
   boot = {
     tmp.useTmpfs = true;
     kernelPackages = lib.mkDefault pkgs.linuxPackages_6_12;
-    zfs.package = lib.mkDefault pkgs.zfs_2_4;
+    zfs = {
+      package = lib.mkDefault pkgs.zfs_2_4;
+      forceImportRoot = lib.mkDefault false;
+    };
   };
 
   hardware.enableRedistributableFirmware = true;
@@ -37,10 +40,17 @@
 
   nixpkgs = {
     overlays = builtins.attrValues outputs.overlays;
-    config.allowUnfree = true;
+    config = {
+      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

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

common/optional/monitoring-agent.nix

@@ -0,0 +1,256 @@
+{
+  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 = "TBRULKD-7DZPGGZ-F6LLB7J-MSO54AY-7KLPBIN-QOFK6PX-W2HBEWI-PHM2CQI"; # cspell:disable-line
+      phone.id = "JPVQKQW-CFXOJXT-Q5G5F3H-QIDHDRE-GKHPTQB-GXZUQSP-U7FR7F7-INP3AAH"; # 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 = "github:RichieCahill/dotfiles";
+    flake = "git+https://gitea.tmmworkshop.com/richie/dotfiles?ref=main";
     allowReboot = true;
     dates = "Sat *-*-* 06:00:00";
   };

docs/zfs_failed_root_import_recovry.md

@@ -0,0 +1,76 @@
+# 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": 1773979456,
-        "narHash": "sha256-9kBMJ5IvxqNlkkj/swmE8uK1Sc7TL/LIRUI958m7uBM=",
+        "lastModified": 1781150628,
+        "narHash": "sha256-b4mp8l3qWuSCyYYo9HSngDtcB3PpecYiOXjULrjwwlw=",
         "owner": "rycee",
         "repo": "nur-expressions",
-        "rev": "81e28f47ac18d9e89513929c77e711e657b64851",
+        "rev": "753319310f4673a2dabbfab87482187b40bf9bac",
         "type": "gitlab"
       },
       "original": {
@@ -29,11 +29,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1774007980,
-        "narHash": "sha256-FOnZjElEI8pqqCvB6K/1JRHTE8o4rer8driivTpq2uo=",
+        "lastModified": 1781189114,
+        "narHash": "sha256-5inaamLgUMWy+MOBE9ChF9QAF1o/74LFuHkI0W/9rqc=",
         "owner": "nix-community",
         "repo": "home-manager",
-        "rev": "9670de2921812bc4e0452f6e3efd8c859696c183",
+        "rev": "486595d2cf49cfcd649b58a284fa11ac0e34da22",
         "type": "github"
       },
       "original": {
@@ -43,12 +43,15 @@
       }
     },
     "nixos-hardware": {
+      "inputs": {
+        "nixpkgs": "nixpkgs"
+      },
       "locked": {
-        "lastModified": 1774018263,
-        "narHash": "sha256-HHYEwK1A22aSaxv2ibhMMkKvrDGKGlA/qObG4smrSqc=",
+        "lastModified": 1781168557,
+        "narHash": "sha256-LOnLQ2tpYF9gqIDDr3+j3DbpJJr/QCH6zPRT2GzEUOE=",
         "owner": "nixos",
         "repo": "nixos-hardware",
-        "rev": "2d4b4717b2534fad5c715968c1cece04a172b365",
+        "rev": "6358ff76821101c178e3ab4919a62799bfe3652e",
         "type": "github"
       },
       "original": {
@@ -60,27 +63,24 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1773821835,
-        "narHash": "sha256-TJ3lSQtW0E2JrznGVm8hOQGVpXjJyXY2guAxku2O9A4=",
-        "owner": "nixos",
-        "repo": "nixpkgs",
-        "rev": "b40629efe5d6ec48dd1efba650c797ddbd39ace0",
-        "type": "github"
+        "lastModified": 1767892417,
+        "narHash": "sha256-8bW3q88CEg2u4hSP66Vf4lpbLonHz7hqDNBMcCY7E9U=",
+        "rev": "3497aa5c9457a9d88d71fa93a4a8368816fbeeba",
+        "type": "tarball",
+        "url": "https://releases.nixos.org/nixos/unstable/nixos-26.05pre924538.3497aa5c9457/nixexprs.tar.xz"
       },
       "original": {
-        "owner": "nixos",
-        "ref": "nixos-unstable",
-        "repo": "nixpkgs",
-        "type": "github"
+        "type": "tarball",
+        "url": "https://channels.nixos.org/nixos-unstable/nixexprs.tar.xz"
       }
     },
     "nixpkgs-master": {
       "locked": {
-        "lastModified": 1774051532,
-        "narHash": "sha256-d3CGMweyYIcPuTj5BKq+1Lx4zwlgL31nVtN647tOZKo=",
+        "lastModified": 1781229721,
+        "narHash": "sha256-ORvqDbb/LYxiJljGIejapjkc/kJbVote2N1WSb9W45I=",
         "owner": "nixos",
         "repo": "nixpkgs",
-        "rev": "8620c0b5cc8fbe76502442181be1d0514bc3a1b7",
+        "rev": "173d0ad7a974f8543a9ab01d2271b2e290341b33",
         "type": "github"
       },
       "original": {
@@ -106,12 +106,28 @@
         "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",
+        "nixpkgs": "nixpkgs_2",
         "nixpkgs-master": "nixpkgs-master",
         "nixpkgs-stable": "nixpkgs-stable",
         "sops-nix": "sops-nix",
@@ -125,11 +141,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1773889674,
-        "narHash": "sha256-+ycaiVAk3MEshJTg35cBTUa0MizGiS+bgpYw/f8ohkg=",
+        "lastModified": 1780547341,
+        "narHash": "sha256-Gq8KNx5A7hBB3uGJaj6eQfLDIz5YdLu92gqBcvHvoUo=",
         "owner": "Mic92",
         "repo": "sops-nix",
-        "rev": "29b6519f3e0780452bca0ac0be4584f04ac16cc5",
+        "rev": "9ed65852b6257fbeae4355bc24ecfea307ca759a",
         "type": "github"
       },
       "original": {

overlays/default.nix

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

pyproject.toml

@@ -12,7 +12,6 @@ dependencies = [
     "alembic",
     "apprise",
     "apscheduler",
-    "huggingface-hub",
     "httpx",
     "python-multipart",
     "polars",
@@ -27,11 +26,7 @@ dependencies = [
 [project.scripts]
 database = "python.database_cli:app"
 van-inventory = "python.van_inventory.main:serve"
-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"
+whisper-transcribe = "python.tools.whisper.transcribe:main"
 
 [dependency-groups]
 dev = [
@@ -56,6 +51,7 @@ 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]
@@ -84,20 +80,10 @@ 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

@@ -1,50 +0,0 @@
-"""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

@@ -1,72 +0,0 @@
-"""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

@@ -1,153 +0,0 @@
-"""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

@@ -1,58 +0,0 @@
-"""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/signal_bot/versions/2026_03_17-seprating_signal_bot_database_6eaf696e07a5.py

@@ -1,100 +0,0 @@
-"""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

@@ -1,72 +0,0 @@
-"""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

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

python/data_science/ingest_congress.py

@@ -1,613 +0,0 @@
-"""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

@@ -1,247 +0,0 @@
-"""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

@@ -83,20 +83,6 @@ 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

@@ -0,0 +1,347 @@
+"""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

@@ -0,0 +1,148 @@
+"""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

@@ -1,13 +1,9 @@
 """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

@@ -1,11 +0,0 @@
-"""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

@@ -1,52 +0,0 @@
-"""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

@@ -1,14 +0,0 @@
-"""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

@@ -1,66 +0,0 @@
-"""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

@@ -1,66 +0,0 @@
-"""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

@@ -1,79 +0,0 @@
-"""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

@@ -1,16 +0,0 @@
-"""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

@@ -1,11 +0,0 @@
-"""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

@@ -1,33 +0,0 @@
-"""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

@@ -1,17 +0,0 @@
-"""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

@@ -1,71 +0,0 @@
-"""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

@@ -1,13 +0,0 @@
-"""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/signal_bot/__init__.py

@@ -1,16 +0,0 @@
-"""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

@@ -1,52 +0,0 @@
-"""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

@@ -1,62 +0,0 @@
-"""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

@@ -1,25 +0,0 @@
-# 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

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

python/prompt_bench/batch_bill_summarizer.py

@@ -1,233 +0,0 @@
-"""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

@@ -1,162 +0,0 @@
-"""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

@@ -1,236 +0,0 @@
-"""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

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

python/prompt_bench/containers/finetune.py

@@ -1,165 +0,0 @@
-"""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

@@ -1,23 +0,0 @@
-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

@@ -1,70 +0,0 @@
-"""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

@@ -1,75 +0,0 @@
-"""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

@@ -1,214 +0,0 @@
-"""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

@@ -1,215 +0,0 @@
-"""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

@@ -1,30 +0,0 @@
-"""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

@@ -1,34 +0,0 @@
-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

@@ -1,114 +0,0 @@
-"""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

@@ -1,97 +0,0 @@
-"""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

@@ -1,68 +0,0 @@
-"""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

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

python/signal_bot/commands/__init__.py

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

python/signal_bot/commands/inventory.py

@@ -1,137 +0,0 @@
-"""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

@@ -1,64 +0,0 @@
-"""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

@@ -1,284 +0,0 @@
-"""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

@@ -1,80 +0,0 @@
-"""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

@@ -1,239 +0,0 @@
-"""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

@@ -1,97 +0,0 @@
-"""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

@@ -1,141 +0,0 @@
-"""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/whisper/Dockerfile

@@ -0,0 +1,17 @@
+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

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

python/tools/whisper/__init__.py

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

python/tools/whisper/inference.py

@@ -0,0 +1,136 @@
+"""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

@@ -0,0 +1,167 @@
+"""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

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

systems/bob/hardware.nix

@@ -28,9 +28,13 @@
         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

@@ -42,11 +42,14 @@
       "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/models";
+    models = "/zfs/storage/models";
     openFirewall = true;
   };
 }

systems/bob/scripts/zfs.sh

@@ -1,11 +0,0 @@
-#!/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 = 2
+hourly = 24
 daily = 0
 monthly = 0

systems/bob/syncthing.nix

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

systems/brain/hardware.nix

@@ -26,7 +26,6 @@
         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

@@ -4,17 +4,21 @@ let
 in
 {
   imports = [
-    "${inputs.self}/users/richie"
-    "${inputs.self}/users/math"
     "${inputs.self}/users/dov"
+    "${inputs.self}/users/math"
+    "${inputs.self}/users/richie"
+    "${inputs.self}/users/steve"
     "${inputs.self}/common/global"
     "${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
@@ -35,5 +39,10 @@ in
     zerotierone.joinNetworks = [ "a09acf02330d37b9" ];
   };
 
+  users.groups = {
+    nornsight = { };
+    nornsight-admin = { };
+  };
+
   system.stateVersion = "24.05";
 }

systems/jeeves/hardware.nix

@@ -9,7 +9,6 @@ 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

@@ -0,0 +1,426 @@
+{
+  "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": ""
+}

systems/jeeves/monitoring/dashboards/process-history-grouped.json

@@ -0,0 +1,216 @@
+{
+  "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": "percentunit"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "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": "topk(10, rate(namedprocess_namegroup_cpu_seconds_total[5m]))",
+          "legendFormat": "{{instance}} {{groupname}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Grouped CPU",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "bytes"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 12,
+        "y": 0
+      },
+      "id": 2,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(10, namedprocess_namegroup_memory_bytes{memtype=\"resident\"})",
+          "legendFormat": "{{instance}} {{groupname}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Grouped Resident Memory",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "Bps"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 0,
+        "y": 10
+      },
+      "id": 3,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(10, rate(namedprocess_namegroup_read_bytes_total[5m]))",
+          "legendFormat": "{{instance}} {{groupname}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Grouped Read I/O",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "Bps"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 12,
+        "y": 10
+      },
+      "id": 4,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(10, rate(namedprocess_namegroup_write_bytes_total[5m]))",
+          "legendFormat": "{{instance}} {{groupname}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Grouped Write I/O",
+      "type": "timeseries"
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "style": "dark",
+  "tags": [
+    "monitoring",
+    "process"
+  ],
+  "templating": {
+    "list": []
+  },
+  "time": {
+    "from": "now-7d",
+    "to": "now"
+  },
+  "timepicker": {},
+  "timezone": "",
+  "title": "Process History Grouped",
+  "uid": "monitor-process-history",
+  "version": 1,
+  "weekStart": ""
+}

systems/jeeves/monitoring/dashboards/process-live-pid.json

@@ -0,0 +1,224 @@
+{
+  "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-pid-short"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "percentunit"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 0,
+        "y": 0
+      },
+      "id": 1,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-pid-short"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, rate(namedprocess_namegroup_cpu_seconds_total[2m]))",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{groupname}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top PID CPU",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-pid-short"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "bytes"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 12,
+        "y": 0
+      },
+      "id": 2,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-pid-short"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, namedprocess_namegroup_memory_bytes{memtype=\"resident\"})",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{groupname}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top PID RSS",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-pid-short"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "Bps"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 0,
+        "y": 10
+      },
+      "id": 3,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-pid-short"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, rate(namedprocess_namegroup_read_bytes_total[2m]))",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{groupname}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top PID Read I/O",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-pid-short"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "Bps"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 10,
+        "w": 12,
+        "x": 12,
+        "y": 10
+      },
+      "id": 4,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-pid-short"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, rate(namedprocess_namegroup_write_bytes_total[2m]))",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{groupname}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top PID Write I/O",
+      "type": "table"
+    }
+  ],
+  "refresh": "15s",
+  "schemaVersion": 39,
+  "style": "dark",
+  "tags": [
+    "monitoring",
+    "process"
+  ],
+  "templating": {
+    "list": []
+  },
+  "time": {
+    "from": "now-10m",
+    "to": "now"
+  },
+  "timepicker": {},
+  "timezone": "",
+  "title": "Process Live PID",
+  "uid": "monitor-process-pid",
+  "version": 1,
+  "weekStart": ""
+}

systems/jeeves/monitoring/dashboards/storage-zfs.json

@@ -0,0 +1,351 @@
+{
+  "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": 8,
+        "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 * (zfs_pool_allocated_bytes / zfs_pool_size_bytes)",
+          "legendFormat": "{{instance}} {{pool}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Pool Usage",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "bytes"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 8,
+        "x": 8,
+        "y": 0
+      },
+      "id": 2,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "zfs_pool_free_bytes",
+          "legendFormat": "{{instance}} {{pool}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Pool Free Bytes",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "bytes"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 8,
+        "x": 16,
+        "y": 0
+      },
+      "id": 3,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, zfs_dataset_used_bytes{type=\"filesystem\"})",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{name}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Top Filesystems by Used Bytes",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "ns"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 0,
+        "y": 8
+      },
+      "id": 4,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, zpool_iostat_total_wait_read_ns{vdev!=\"_pool\"})",
+          "legendFormat": "{{host}} {{pool}} {{vdev}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "ZFS Read Wait",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "ns"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 12,
+        "y": 8
+      },
+      "id": 5,
+      "options": {
+        "legend": {
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "topk(20, zpool_iostat_total_wait_write_ns{vdev!=\"_pool\"})",
+          "legendFormat": "{{host}} {{pool}} {{vdev}}",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "ZFS Write Wait",
+      "type": "timeseries"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "celsius"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 0,
+        "y": 17
+      },
+      "id": 6,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": true,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "smartctl_device_temperature{temperature_type=\"current\"}",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{device}}",
+          "refId": "A"
+        }
+      ],
+      "title": "Disk Temperature",
+      "type": "table"
+    },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prom-main"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "short"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 12,
+        "y": 17
+      },
+      "id": 7,
+      "options": {
+        "cellHeight": "sm",
+        "showHeader": true,
+        "sortBy": [
+          {
+            "desc": false,
+            "displayName": "Value"
+          }
+        ]
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prom-main"
+          },
+          "editorMode": "code",
+          "expr": "smartctl_device_smart_status",
+          "format": "table",
+          "instant": true,
+          "legendFormat": "{{instance}} {{device}}",
+          "refId": "A"
+        }
+      ],
+      "title": "SMART Health",
+      "type": "table"
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "style": "dark",
+  "tags": [
+    "monitoring",
+    "zfs"
+  ],
+  "templating": {
+    "list": []
+  },
+  "time": {
+    "from": "now-24h",
+    "to": "now"
+  },
+  "timepicker": {},
+  "timezone": "",
+  "title": "Storage and ZFS",
+  "uid": "monitor-storage",
+  "version": 1,
+  "weekStart": ""
+}

systems/jeeves/monitoring/default.nix

@@ -0,0 +1,186 @@
+{
+  lib,
+  pkgs,
+  ...
+}:
+let
+  vars = import ../vars.nix;
+
+  prometheusDataRoot = "${vars.database}/prometheus";
+  mainPrometheusDataDir = "${prometheusDataRoot}/main";
+  pidPrometheusDataDir = "${prometheusDataRoot}/pid-short";
+
+  prometheusYaml = pkgs.formats.yaml { };
+
+  mkPrometheusConfig =
+    name: cfg:
+    let
+      configFile = prometheusYaml.generate "${name}.yaml" cfg;
+    in
+    pkgs.runCommand "${name}-checked.yaml"
+      {
+        nativeBuildInputs = [ pkgs.prometheus.cli ];
+      }
+      ''
+        promtool check config ${configFile}
+        cp ${configFile} $out
+      '';
+
+  mkTarget = host: address: {
+    targets = [ address ];
+    labels.instance = host;
+  };
+
+  mainPrometheusConfig = mkPrometheusConfig "prometheus-main" {
+    global = {
+      scrape_interval = "30s";
+      scrape_timeout = "10s";
+      evaluation_interval = "30s";
+    };
+    scrape_configs = [
+      {
+        job_name = "node";
+        static_configs = [
+          (mkTarget "jeeves" "192.168.90.40:9100")
+          (mkTarget "bob" "192.168.90.25:9100")
+        ];
+      }
+      {
+        job_name = "process_grouped";
+        static_configs = [
+          (mkTarget "jeeves" "192.168.90.40:9256")
+          (mkTarget "bob" "192.168.90.25:9256")
+        ];
+      }
+      {
+        job_name = "smartctl";
+        static_configs = [
+          (mkTarget "jeeves" "192.168.90.40:9633")
+          (mkTarget "bob" "192.168.90.25:9633")
+        ];
+      }
+      {
+        job_name = "zfs";
+        static_configs = [
+          (mkTarget "jeeves" "192.168.90.40:9134")
+          (mkTarget "bob" "192.168.90.25:9134")
+        ];
+      }
+    ];
+  };
+
+  pidPrometheusConfig = mkPrometheusConfig "prometheus-pid-short" {
+    global = {
+      scrape_interval = "15s";
+      scrape_timeout = "10s";
+      evaluation_interval = "15s";
+    };
+    scrape_configs = [
+      {
+        job_name = "process_pid";
+        static_configs = [
+          (mkTarget "jeeves" "192.168.90.40:9257")
+          (mkTarget "bob" "192.168.90.25:9257")
+        ];
+      }
+    ];
+  };
+
+  mkPrometheusService =
+    {
+      dataDir,
+      configFile,
+      port,
+      retention,
+    }:
+    {
+      after = [
+        "zfs-media-database-prometheus.mount"
+        "network.target"
+      ];
+      requires = [ "zfs-media-database-prometheus.mount" ];
+      wantedBy = [ "multi-user.target" ];
+      unitConfig.RequiresMountsFor = [ dataDir ];
+      serviceConfig = {
+        ExecStart = "${lib.getExe pkgs.prometheus} ${
+          lib.escapeShellArgs [
+            "--config.file=${configFile}"
+            "--storage.tsdb.path=${dataDir}"
+            "--storage.tsdb.retention.time=${retention}"
+            "--web.listen-address=127.0.0.1:${toString port}"
+          ]
+        }";
+        User = "prometheus";
+        Group = "prometheus";
+        Restart = "always";
+        RestartSec = "5s";
+        WorkingDirectory = dataDir;
+        ReadWritePaths = [ dataDir ];
+        CapabilityBoundingSet = [ "" ];
+        DeviceAllow = [ "/dev/null rw" ];
+        DevicePolicy = "strict";
+        LockPersonality = true;
+        MemoryDenyWriteExecute = true;
+        NoNewPrivileges = true;
+        PrivateDevices = true;
+        PrivateTmp = true;
+        ProtectClock = true;
+        ProtectControlGroups = true;
+        ProtectHome = true;
+        ProtectHostname = true;
+        ProtectKernelLogs = true;
+        ProtectKernelModules = true;
+        ProtectKernelTunables = true;
+        ProtectProc = "invisible";
+        ProtectSystem = "strict";
+        RemoveIPC = true;
+        RestrictAddressFamilies = [
+          "AF_INET"
+          "AF_INET6"
+          "AF_UNIX"
+        ];
+        RestrictNamespaces = true;
+        RestrictRealtime = true;
+        RestrictSUIDSGID = true;
+        SystemCallArchitectures = "native";
+        SystemCallFilter = [
+          "@system-service"
+          "~@privileged"
+        ];
+      };
+    };
+in
+{
+  users = {
+    groups.prometheus = { };
+    users.prometheus = {
+      isSystemUser = true;
+      group = "prometheus";
+      description = "Prometheus daemon user";
+    };
+  };
+
+  systemd = {
+    services = {
+      prometheus-main = mkPrometheusService {
+        configFile = mainPrometheusConfig;
+        dataDir = mainPrometheusDataDir;
+        port = 9090;
+        retention = "90d";
+      };
+
+      prometheus-pid-short = mkPrometheusService {
+        configFile = pidPrometheusConfig;
+        dataDir = pidPrometheusDataDir;
+        port = 9092;
+        retention = "10m";
+      };
+    };
+
+    tmpfiles.rules = [
+      "d ${prometheusDataRoot} 0755 root root - -"
+      "d ${mainPrometheusDataDir} 0750 prometheus prometheus - -"
+      "d ${pidPrometheusDataDir} 0750 prometheus prometheus - -"
+    ];
+  };
+}

systems/jeeves/networking.nix

@@ -1,4 +1,13 @@
 {
+  # Docker loads br_netfilter on jeeves. Disable bridge netfilter so
+  # br-nix-builder behaves like a pure L2 bridge and bridged traffic
+  # does not hit the host firewall/rpfilter path.
+  boot.kernel.sysctl = {
+    "net.bridge.bridge-nf-call-arptables" = 0;
+    "net.bridge.bridge-nf-call-ip6tables" = 0;
+    "net.bridge.bridge-nf-call-iptables" = 0;
+  };
+
   networking = {
     hostName = "jeeves";
     hostId = "0e15ce35";
@@ -34,11 +43,18 @@
       };
     };
     networks = {
-      "10-1GB_Primary" = {
-        matchConfig.Name = "enp97s0f1";
+      "10-Primary" = {
+        matchConfig.Name = "enp97s0";
         address = [ "192.168.99.14/24" ];
+        dns = [
+          "192.168.99.1"
+          "2600:4040:abfb:d700::1"
+        ];
         routes = [ { Gateway = "192.168.99.1"; } ];
         vlan = [ "internet-vlan" ];
+        dhcpV4Config.UseDNS = false;
+        dhcpV6Config.UseDNS = false;
+        ipv6AcceptRAConfig.UseDNS = false;
         linkConfig.RequiredForOnline = "routable";
       };
       "50-internet-vlan" = {
@@ -49,23 +65,10 @@
       "60-br-nix-builder" = {
         matchConfig.Name = "br-nix-builder";
         bridgeConfig = { };
-        address = [ "192.168.3.10/24" ];
-        routingPolicyRules = [
-          {
-            From = "192.168.3.0/24";
-            Table = 100;
-            Priority = 100;
-          }
-        ];
-        routes = [
-          {
-            Gateway = "192.168.3.1";
-            Table = 100;
-            GatewayOnLink = false;
-            Metric = 2048;
-            PreferredSource = "192.168.3.10";
-          }
-        ];
+        networkConfig = {
+          IPv6AcceptRA = false;
+          LinkLocalAddressing = "no";
+        };
         linkConfig.RequiredForOnline = "no";
       };
     };

systems/jeeves/programs.nix

@@ -3,5 +3,6 @@
   environment.systemPackages = with pkgs; [
     filebot
     docker-compose
+    ffmpeg
   ];
 }

systems/jeeves/runners/default.nix

@@ -1,20 +1,7 @@
-{ pkgs, ... }:
+{ ... }:
 {
   imports = [ ./nix_builder.nix ];
 
-  users = {
-    users.github-runners = {
-      shell = pkgs.bash;
-      isSystemUser = true;
-      group = "github-runners";
-      uid = 601;
-      openssh.authorizedKeys.keys = [
-        "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIA/S8i+BNX/12JNKg+5EKGX7Aqimt5KM+ve3wt/SyWuO github-runners" # cspell:disable-line
-      ];
-    };
-    groups.github-runners.gid = 601;
-  };
-
   services.nix_builder.containers = {
     nix-builder-00.enable = true;
     nix-builder-01.enable = true;

systems/jeeves/runners/nix_builder.nix

@@ -2,6 +2,7 @@
   config,
   lib,
   outputs,
+  utils,
   ...
 }:
 
@@ -9,6 +10,8 @@ with lib;
 let
   vars = import ../vars.nix;
   cfg = config.services.nix_builder;
+  runnerUsername = "gitea-runner";
+  runnerUserid = 601;
 in
 {
   options.services.nix_builder = {
@@ -23,37 +26,40 @@ in
         types.submodule (
           { name, ... }:
           {
-            options.enable = mkEnableOption "GitHub runner container";
+            options.enable = mkEnableOption "Gitea runner container";
           }
         )
       );
       default = { };
-      description = "GitHub runner container configurations";
+      description = "Gitea runner container configurations";
     };
   };
 
   config = {
+    users = {
+      users.${runnerUsername} = {
+        isSystemUser = true;
+        group = runnerUsername;
+        uid = runnerUserid;
+      };
+      groups.${runnerUsername}.gid = runnerUserid;
+    };
+
     containers = mapAttrs (
       name: containerCfg:
       mkIf containerCfg.enable {
         autoStart = true;
         privateNetwork = true;
         hostBridge = cfg.bridgeName;
-        ephemeral = true;
         bindMounts = {
-          storage = {
-            hostPath = "/zfs/media/github-runners/${name}";
-            mountPoint = "/zfs/media/github-runners/${name}";
-            isReadOnly = false;
-          };
           host-nix = {
             mountPoint = "/host-nix/var/nix/daemon-socket";
             hostPath = "/nix/var/nix/daemon-socket";
             isReadOnly = false;
           };
-          pat = {
-            hostPath = "${vars.secrets}/services/github-runners/runner_pat";
-            mountPoint = "${vars.secrets}/services/github-runners/runner_pat";
+          token = {
+            hostPath = "${vars.secrets}/services/gitea-runners";
+            mountPoint = "/run/secrets/gitea-runners";
             isReadOnly = true;
           };
         };
@@ -92,46 +98,69 @@ in
                 "nix-command"
               ];
               sandbox = true;
-              allowed-users = [ "github-runners" ];
+              allowed-users = [ "gitea-runner" ];
               trusted-users = [
                 "root"
-                "github-runners"
+                "gitea-runner"
               ];
             };
             nixpkgs = {
               overlays = builtins.attrValues outputs.overlays;
               config.allowUnfree = true;
             };
-            services.github-runners.${name} = {
+            users = {
+              users.${runnerUsername} = {
+                isSystemUser = true;
+                group = runnerUsername;
+                uid = runnerUserid;
+              };
+              groups.${runnerUsername}.gid = runnerUserid;
+            };
+            services.gitea-actions-runner.instances.${name} = {
               enable = true;
-              replace = true;
-              workDir = "/zfs/media/github-runners/${name}";
-              url = "https://github.com/RichieCahill/dotfiles";
-              extraLabels = [ "nixos" ];
-              tokenFile = "${vars.secrets}/services/github-runners/runner_pat";
-              user = "github-runners";
-              group = "github-runners";
-              extraPackages = with pkgs; [
+              name = "jeeves-${name}";
+              url = "http://192.168.99.14:6443/";
+              labels = [
+                "self-hosted:host"
+                "nixos:host"
+              ];
+              tokenFile = "/run/secrets/gitea-runners/registration-token";
+              hostPackages = with pkgs; [
+                bash
+                coreutils
+                curl
+                gawk
                 gitMinimal
-                gh
+                gnused
+                my_python
+                nix
                 nixfmt
                 nixos-rebuild
+                nodejs
                 treefmt
-                my_python
+                wget
               ];
             };
-            users = {
-              users.github-runners = {
-                shell = pkgs.bash;
-                isSystemUser = true;
-                group = "github-runners";
-                uid = 601;
+            systemd.services."gitea-runner-${utils.escapeSystemdPath name}" = {
+              serviceConfig = {
+                DynamicUser = mkForce false;
+                User = mkForce runnerUsername;
+                Group = mkForce runnerUsername;
               };
-              groups.github-runners.gid = 601;
             };
             system.stateVersion = "24.05";
           };
       }
     ) cfg.containers;
+
+    systemd.services = builtins.listToAttrs (
+      map (name: {
+        name = "container@${name}";
+        value = {
+          requires = [ "gitea.service" ];
+          after = [ "gitea.service" ];
+        };
+      }) (builtins.attrNames (filterAttrs (_: c: c.enable) cfg.containers))
+    );
   };
 }

systems/jeeves/scripts/zfs.sh

@@ -23,6 +23,7 @@ sudo zfs create media/secure/home_assistant -o compression=zstd-19
 sudo zfs create media/secure/notes -o copies=2
 sudo zfs create media/secure/postgres -o mountpoint=/zfs/media/database/postgres -o recordsize=16k -o primarycache=metadata
 sudo zfs create media/secure/postgres-wal -o mountpoint=/zfs/media/database/postgres-wal -o recordsize=32k -o primarycache=metadata -o special_small_blocks=32K -o compression=lz4 -o secondarycache=none -o logbias=latency
+sudo zfs create media/secure/prometheus -o mountpoint=/zfs/media/database/prometheus -o compression=lz4
 sudo zfs create media/secure/services -o compression=zstd-9
 sudo zfs create media/secure/share -o mountpoint=/zfs/media/share -o exec=off
 
@@ -41,3 +42,4 @@ sudo zfs create storage/secure/plex -o recordsize=1M -o compression=zstd-19
 sudo zfs create storage/secure/secrets -o compression=zstd-19 -o copies=3
 sudo zfs create storage/secure/syncthing -o compression=zstd-19
 sudo zfs create storage/secure/transmission -o recordsize=1M -o compression=zstd-9 -o exec=off -o sync=disabled
+sudo zfs create storage/secure/important -o compression=zstd-19 -o copies=2 -o mountpoint=/zfs/storage/important

systems/jeeves/services/audiobookshelf.nix

@@ -3,7 +3,10 @@ let
   vars = import ../vars.nix;
 in
 {
-  services.audiobookshelf.enable = true;
+  services.audiobookshelf = {
+    enable = true;
+    port = 8000;
+  };
   systemd.services.audiobookshelf.serviceConfig.WorkingDirectory =
     lib.mkForce "${vars.docker_configs}/audiobookshelf";
   users.users.audiobookshelf.home = lib.mkForce "${vars.docker_configs}/audiobookshelf";