Dune/docs/VERIFICATION_WORKFLOW.md

Repeatable verification

Use the development guide to choose checks appropriate to the change. These wrappers retain the existing test runner and simulation harness; they do not replace rules, browser or release acceptance.

Compact checks and source evidence

node --import tsx tools/verify.ts --out /tmp/dune-new-focused focused --focus guild-payment
node --import tsx tools/verify.ts --out /tmp/dune-new-check check
# Add build for app/build/dependency changes, integration for HTTP/auth/persistence.
node --import tsx tools/verify.ts --out /tmp/dune-new-release check build integration
# Only for relevant AI/gameplay work; the existing study covers Basic base factions.
node --import tsx tools/verify.ts --out /tmp/dune-new-study base-games --games 20 --seed 20260913

Each output must be a new directory outside the checkout, with an existing parent. It is created with private permissions. Full command output stays in step logs; report.json records commands, exit codes, timings, HEAD and a content fingerprint of tracked and nonignored untracked files before and after the run. Ignored local credentials, saved games and build caches are excluded. Do not supply secrets in command arguments. A failed child stops subsequent steps. Changed source or HEAD invalidates a successful run; finish all source/doc edits before broad checks. This checks the two observed snapshots, not changes made and reverted between them, nor the contents of dependencies, ignored files or an external HTTP server. An interrupted or incomplete report is never evidence of success.

Saved games and online backups

Reusable base and faction sample games

node --import tsx tools/faction-games.ts --out /tmp/dune-new-faction-games
# Ten genuine base samples: Basic and Advanced at each player count.
node --import tsx tools/faction-games.ts --out /tmp/dune-new-base-games --profile base
# Basic Moritani with the full Leader Skills deck, two through six seats.
node --import tsx tools/faction-games.ts --out /tmp/dune-new-moritani-skills --profile moritani-skills
node --import tsx tools/faction-games.ts --out /tmp/dune-new-base-three --profile base --players 3 --rules advanced
node --import tsx tools/faction-games.ts --out /tmp/dune-new-combined --profile combined --rules advanced --seed 20260926
node --import tsx tools/faction-games.ts --out /tmp/dune-new-resume --resume /private/failed-combined-advanced.json

Default and --profile all retain the original six fixed CHOAM/Richese, Ecaz/Moritani and combined Basic/Advanced samples. --profile base selects ten additional samples: Basic and Advanced with two through six players. --players 2..6 narrows only that base profile; omission or all retains all five counts. The fixed base roster adds Atreides, Harkonnen, Fremen, Emperor, Guild and Bene Gesserit in that order. These are sample rosters, not an assertion that these are the only rules-permitted player-count configurations.

--profile moritani-skills adds five Basic samples with Moritani, then Emperor, Guild, Harkonnen, Fremen and Bene Gesserit. --players also narrows this profile; Advanced requests fail before running. Its seed offsets are 16–20, leaving existing samples unchanged. The full fourteen-card skill census and ordinary Moritani/base Traitor census are checked throughout. --resume preserves this explicit profile; other optional modules still fail its admission checks.

All samples use genuine setup and saved AI profiles, cycling Easy, Medium, Hard and Brutal by seat. Smaller games necessarily contain fewer profiles; the matrix collectively exercises all four. Filters retain each scenario's seed offset: the original six use 0–5; base uses 6 + 2 × (players − 2) + Advanced, where Advanced is 0 or 1. --seed defaults to 20260926; --max-actions defaults to 3500 accepted actions per game. Every action checks physical card custody (including unsold auction cards), nonnegative integer forces/spice, twenty-force totals and the distinct Fremen/Emperor/Ixian elite totals and subsets. Base samples also check the exact Traitor inventory after dealing has finished; expansion Traitor/Face Dancer zones are not certified by this check. Every rejected candidate must leave its input unchanged. Every 37 actions checks all private views after JSON restoration and absence of rival hand, spice, Traitor and Face Dancer fields. Rejected candidates remain visible in private traces even when a later candidate succeeds. The ordinary AI arrival adapter filters known supported-scope reaction conflicts before attempting an action; remaining rejections are still reported, never hidden by the runner. Samples are development evidence, not calibration or complete rules certification.

The output is a new private directory outside the checkout. It contains private traces, incomplete-game snapshots, results.json and a source-bound report.json. An incomplete game or changed source exits nonzero. Resume requires a matching fixed base or expansion scenario snapshot with saved AI profiles and records its SHA-256; it cannot be combined with profile/rules/player-count filters. Its action budget starts at the snapshot, and its random stream restarts from the supplied seed plus scenario offset. Check setup provenance against the original report: accepting a supplied snapshot does not prove it came from genuine setup. No live room or database is accessed.

Preservation commands

node --import tsx tools/saved-games.ts snapshot --db /absolute/games.sqlite --out /private/new-baseline --backup
node --import tsx tools/saved-games.ts compare --db /absolute/games.sqlite --baseline /private/new-baseline/snapshot.json

The source database is opened read-only. Snapshots store room versions and state hashes, not private game state. New rooms are allowed; every changed or missing original room fails comparison. Legitimate human play can change a room: investigate and document it, never overwrite progress to make a comparison pass. Snapshot the current set before each maintenance operation so newly created rooms are protected. Online backup includes WAL contents and requires Node 22.16 or newer; snapshot and comparison retain the project's Node 22.13 minimum. Backup files contain private state and session records, remain outside Git, and never overwrite an earlier run. This tool does not delete, reset or restore the live database.

Authorized reset, 13 September 2026

The user requested clearing all old local games and starting a new checkpoint. With no development server running, a verified online backup preceded a single transaction deleting 3,644 rooms and cascading their seat/recovery/entry records. Schema, migrations and Wrangler state files were preserved. All five application tables were empty afterward; foreign-key validation passed.

Private archive: /home/quintine/.local/share/dune/checkpoints/20260913-before-reset/. Fresh empty database backup and snapshot: /home/quintine/.local/share/dune/checkpoints/20260913-fresh-baseline/. These machine-local paths are operational records, not portable repo artifacts. Old room IDs in historical browser evidence no longer identify active local games. Future games must be preserved. This was a one-time user-authorized reset, not a maintenance strategy or permission to clear future games.

Private-seat restoration

Create a private JSON file (mode 0600) outside the checkout with this structure; substitute credentials only in that private file, never in a command or tracked doc:

{"baseUrl":"http://127.0.0.1:3000","isolatedQa":true,"seats":[{"code":"ABCDEFGH","playerId":"fixture-seat-id","version":4,"token":"fixture_session_token"}]}

node --import tsx tools/restore-seats.ts --config /private/qa-seats.json

Use deliberately isolated, stable human QA rooms. The server's GET route can resume bots or automatic work, so this is not a generally side-effect-free read and must not probe arbitrary human games. Requests use loopback origins only, reject redirects and time out. Checks require the expected room, seat and version, the owning seat's private fields, and absence of selected rival private fields. No session tokens, response bodies or hands are printed. This is HTTP seat restoration, not recovery-key rotation, exact hand-content proof, a complete privacy audit or browser rendering. Pair it with saved-game comparison and feature-specific browser acceptance.