# Issue #302 Evidence — canonical bluff → guess → reveal → scoreboard regression

## Runnable command

```bash
python manage.py migrate --noinput
python manage.py smoke_staging --artifact docs/artifacts/issue-302-canonical-loop-smoke.json
```

`migrate` is the normal local bootstrap precondition when the database has not been initialized yet; the regression evidence itself is produced by `smoke_staging`.

## What the regression proves

`smoke_staging` now exercises one full canonical round and fails fast with step-specific diagnostics if any of these break:

1. `start_round` lands the session in `lie` and returns a concrete `round_question_id`.
2. Final `submit_lie` auto-advances the session to `guess` and exposes mixed answers containing both the correct answer and player bluffs.
3. Final `submit_guess` auto-advances the session to `reveal` and returns the canonical reveal payload.
4. The reveal payload includes:
   - correct answer
   - all lies
   - all guesses
   - fooled-player references for bluff hits
5. The first canonical state read after reveal promotes the session to `scoreboard`.
6. Scoreboard promotion preserves the same reveal payload and exposes a leaderboard with `scoreboard_ready=true`.

## Artifact shape

When `--artifact` is provided, the JSON file records:

- the exact smoke command
- session code and round question id
- deterministic guess plan used to produce both bluff hits and one correct guess
- per-step evidence for:
  - `create_session`
  - `join_players`
  - `start_round`
  - `auto_guess_transition`
  - `submit_guesses`
  - `auto_reveal_transition`
  - `auto_scoreboard_transition`
  - `finish_game`
- reveal summary (`correct_answer`, lie/guess counts, fooled-player ids, correct guess player ids)
- promoted scoreboard leaderboard payload

## Targeted test coverage

Backend regression coverage lives in `lobby/tests.py`:

- `test_smoke_staging_command_runs_full_flow`
- `test_smoke_staging_writes_phase_evidence_artifact_when_requested`

Together they ensure the command stays runnable in normal workflow and that the evidence artifact contains phase-by-phase proof instead of only a generic pass/fail.

Refs #287 #302