gupi-osint-board/docs/scene-7-todo.md

# Scene 7: prove Barricelli was an inventor

Status: **implemented and verified** on `scene-7-evidence-goal`
Demo: **GUPI Demo 1 — The Barricelli Files**

## Outcome

Scene 7 teaches one idea: a screenshot found during an OSINT search can become
source evidence.

The player opens a minimal OSINT board, reads **“Demonstrate OSINT skill: prove
Nils Aall Barricelli was an inventor,”** finds the relevant Google Patents result,
and pastes or uploads one screenshot. The board creates a document, extracts its
text, recognizes the source, and clears the level. A URL and written report are
not required.

```text
paste screenshot -> document appears -> text is read -> source is verified
                 -> Scene 7 complete -> continue to Scene 8
```

## Product decisions

- One suitable Google Patents screenshot is sufficient.
- Clipboard paste and file upload use the same server path.
- The image and OCR remain a real source document on the player's board.
- A known-source fuzzy match is the fast, deterministic victory path.
- A small semantic judge is a fallback for other credible sources and for
  distinguishing Nils from his father. It cannot overrule the trusted path.
- Evidence about Barricelli's father may award an optional discovery but does not
  clear the assignment unless it also supports the claim about Nils.
- The boarding-house fire belongs to the later age/rescue assignment, not Scene 7.
- Scene 7 records `scene7.nils_inventor_proved`. Scene 8 owns the ceremony and
  awards `barricelli_luggage`.
- Inconclusive evaluation never deletes or penalizes uploaded evidence.

## Shared identifiers

| Purpose | Key |
|---|---|
| Goal | `barricelli.inventor-proof` |
| Scene 7 completion | `scene7.nils_inventor_proved` |
| Optional father discovery | `scene7.father_inventor_discovered` |
| Scene 8 reward | `barricelli_luggage` |

All identifiers and content are template data. There must be no Barricelli
conditional in React or server business logic.

## Existing foundation

- `025_level_document_flags.sql`: clonable document gates and level flags.
- `026_achievements.sql`: playthrough achievements.
- `027_evidence_text_matching.sql`: asset OCR, board match rules/anchors,
  level-owned evaluations, and flag provenance.
- `028_level_goals.sql`: board-owned goals and normalized flag requirements.
- `029_semantic_evidence_judging.sql`: clonable semantic rules, level-owned
  evaluations, and semantic flag provenance.
- `030_evidence_match_source_metadata.sql`: author-only canonical source metadata.
- `server/ocr.ts`: plain-text extraction and Tesseract.
- `server/evidenceMatching.ts`: OCR-tolerant fuzzy passage matching.
- `POST /api/levels/:id/documents`: persistent upload plus OCR and matching.
- The story runtime already tracks `current_node_id` and `current_level_id`.

## Architecture contract

### Recognition and completion are separate

Recognition answers what a document supports. A goal answers whether the level's
authored requirements have been satisfied. `level_goals` and
`level_goal_flag_requirements` clone with a template. Goal completion is derived
from level flags; there is no second mutable completion boolean.

Play mode receives a goal's key, title, instructions, completion copy, status,
and completion time. IDs, enabled state, required flags, target text, and judging
prompts remain author-only.

### Known-source fast path

The Scene 7 template owns an `evidence_match_rule` with distinctive text visible
in the real Google Patents result: a combination of patent number/title, inventor
name, and invention language. A name alone is too generic. When enough anchors
match, the existing matcher awards `scene7.nils_inventor_proved` in the upload
transaction and the goal becomes complete immediately.

Reference OCR, thresholds, source metadata, and copy live in the manifest/database,
not TypeScript constants. The expected text is never returned to play mode.

### Semantic fallback

A provider-neutral `EvidenceJudge` receives only allowlisted goal data and OCR
text. It returns strictly validated structured data:

```ts
type EvidenceVerdict = {
  subject: 'target' | 'related' | 'ambiguous' | 'neither'
  supportsClaim: boolean
  evidenceExcerpt: string
  confidence: number
}
```

OCR is untrusted quoted material. The judge prompt explicitly ignores instructions
inside it. Provider/model, timeout, input limit, and confidence threshold are
environment configuration. Semantic configuration clones with the board;
evaluation history belongs to the level/document/extraction and records evaluator
version, provider/model, verdict, excerpt, confidence, timestamps, and sanitized
failure state.

| Verdict | Mutation |
|---|---|
| Target + assertion supported at threshold | authored success flag (`scene7.nils_inventor_proved`) |
| Related subject only + assertion supported | authored related flag (`scene7.father_inventor_discovered`) |
| Ambiguous, unsupported, or below threshold | none |
| Provider failure | none; retryable |

The known-source pass avoids an LLM call. After a deterministic miss, the client
automatically calls an idempotent semantic endpoint. This keeps document upload
durable even if an external provider times out, without requiring a demo job queue.

### Story progression

Advancing from a level is server-authoritative. The server verifies the JWT user,
active playthrough/current level, and enabled goal state. It promotes the Scene 7
completion fact idempotently and only then follows the level terminal. The browser
cannot grant its own achievement.

The player must see the verification result before navigation. Success exposes a
single Continue action to Scene 8. Scene 6 only wires into the Scene 7 level node;
Scene 8 owns the luggage reward.

## Work packages

### S7-A — Goal model and cloning

- [x] Add normalized board goals and flag requirements in migration `028`.
- [x] Clone goals during freeze, instantiation, and reset.
- [x] Derive pending/complete state from level flags.
- [x] Hide goal requirements from play payloads.
- [x] Return newly completed goals from document upload.
- [x] Add admin CRUD repository/API contracts.
- [x] Test migration, cloning, deterministic completion, and isolation.
- [x] Clear level flags/reveal state when resetting from a template.

### S7-B — Barricelli content and trusted recognition

- [x] Add goals and evidence-match rules to the normal mystery manifest importer.
- [x] Create the Scene 7 template with its assignment and no solution-bearing
  starting document.
- [x] Add a reproducible screenshot-paste acceptance path that runs through local OCR.
- [x] Author three distinctive patent anchors.
- [x] Tune against the target and unrelated/father-only negative fixtures.
- [x] Award `scene7.nils_inventor_proved` and require it for the level goal.
- [x] Store canonical source metadata for administrators; keep player URL optional.

### S7-C — Semantic judge

- [x] Add provider-neutral interface and strict verdict validation.
- [x] Add clonable semantic rule configuration and level-owned evaluations.
- [x] Add semantic evaluation provenance to level flags.
- [x] Add provider/model/timeout/input/confidence environment configuration.
- [x] Send OCR text rather than raw image bytes.
- [x] Add ownership-checked, idempotent judge endpoint.
- [x] Route target and father verdicts to their authored flags.
- [x] Make disabled provider, timeout, quota, and malformed output safe/retryable.
- [x] Do not log evidence text or secrets.

### S7-D — Story bridge

- [x] Require completed enabled goals before a playthrough can leave a level.
- [x] Verify the current playthrough belongs to the JWT user and owns the level.
- [x] Promote required completion facts exactly once.
- [x] Return a useful pending-goal error rather than advancing early.
- [x] Return to the generic story runtime after success; the Scene 6/8 graph owner
  wires the actual neighboring nodes.
- [x] Keep the arbitrary achievement-grant route development-only.

### S7-E — Board experience

- [x] Present the active goal prominently on Scene 7.
- [x] Preserve clipboard paste, drag/drop, and file picker equivalence.
- [x] Show source import/OCR and semantic checking stages.
- [x] Highlight the accepted document and show **SOURCE VERIFIED — NILS AALL
  BARRICELLI: INVENTOR**.
- [x] Show Continue only after server-confirmed completion.
- [x] Acknowledge father-only discovery while asking for evidence about Nils.
- [x] Keep inconclusive evidence and offer neutral guidance.
- [x] Respect reduced-motion and mobile layouts.

### S7-F — Verification

- [x] Add a legally safe derived OCR fixture.
- [x] Test line breaks, punctuation, cropping, name hyphenation, unrelated patent,
  father-only text, empty OCR, and prompt-injection-like text.
- [x] Contract-test with a fake semantic provider; CI never calls a paid model.
- [x] Integration-test target upload -> document -> flag -> completed goal.
- [x] Test duplicate evaluation, provider failure/retry, two-user isolation, and
  story progression before/after completion.
- [x] Browser-test clipboard image paste through OCR and verification. Continue
  is covered by the generic story-gate integration until the Scene 8 node lands.
- [x] Run migrations on an empty database and one currently at `027`.
- [x] Run unit, integration, build, and Docker smoke tests.

## API shape

```ts
type LevelGoal = {
  key: string
  title: string
  instructions: string
  completionMessage: string
  status: 'pending' | 'complete'
  completedAt?: string
  newlyCompleted: boolean
}

type DocumentAnalysis = {
  extractionStatus: 'succeeded' | 'unsupported' | 'failed'
  matchedFlags: string[]
  awardedFlags: string[]
  goals: LevelGoal[]
}
```

`newlyCompleted` is response-local: a reload returns complete with
`newlyCompleted: false`.

## Security and cost limits

- Require identity and level ownership on player mutations.
- Limit upload bytes, OCR/model characters, output tokens, duration, and retries.
- Never expose reference anchors or judge instructions to play mode.
- Treat filenames, MIME declarations, OCR, and model output as untrusted.
- Validate model output and confidence before mutating flags.
- Persist enough provenance to explain completion without storing unnecessary raw
  provider payloads.

## Out of scope

- Scenes 1–6, Scene 8's ceremony/3D model, and Scene 9's fire mystery.
- Knowledge graph, Case Report, claims, red-thread reasoning, or multi-document
  synthesis.
- Web crawling, URL fetching, or requiring a URL for victory.
- Custom model training or the general story graph `llm_gate`.
- Deleting irrelevant evidence.

## Merge guidance

Prefer new modules and narrow glue commits. High-conflict files are
`server/index.ts`, `server/narrativeRepository.ts`, `src/App.tsx`, `src/main.tsx`,
and `src/play.tsx`; one integrator should own their final changes.

Suggested order: S7-A schema -> S7-B deterministic content -> S7-C judge -> S7-D
story bridge -> S7-E UI -> S7-F hardening. Reserve migration numbers before
parallel schema work and never renumber an already-shared migration silently.

## Definition of done

From a fresh playthrough, the player reaches Scene 7 and pastes one accepted
Google Patents screenshot. One source document appears; MinIO asset, OCR, match
evaluation, and `scene7.nils_inventor_proved` are persisted; the goal completes;
the UI explains what was proved and offers Continue; and the story enters Scene 8.
Reload and duplicate evaluation are idempotent, another player's level is
unaffected, no URL was required, and `barricelli_luggage` is not awarded until
Scene 8.