You are michel in LAB-005 Round 2. Independently create your best revised Foster Review Arcade after studying the equal shared concept dossier and all browser/test findings below. The original Round 1 files are hash-verified but technically corrupted, so do not reproduce their broken source. Return ONLY one complete self-contained HTML document with embedded CSS and JavaScript: no markdown fence, explanation, patch, or external dependency. It must be immediately executable and complete. Preserve evidence-safe review: never reward Approve or rubber-stamping more than Fix, Blank/Evidence Gap, or Escalate. Use fictional synthetic scenarios only. Do not use tools or modify files.

ASSIGNMENT PACKET:
{
  "schema_version": "1.0",
  "experiment_id": "LAB-005",
  "round_id": "LAB-005-R2",
  "assignment_id": "LAB-005-R2-michel-revision",
  "participant_id": "michel",
  "kind": "revision",
  "objective": "Independently create your strongest complete executable Round 2 Foster Review Arcade after studying all four Round 1 builds and the complete browser-test findings. Make it more fun and more useful without rewarding approval or rubber-stamping.",
  "allowed_sources": [
    "All four complete Round 1 HTML builds listed and hash-locked in round-02/context-manifest.json",
    "The complete Round 1 browser-test report listed and hash-locked in round-02/context-manifest.json",
    "LAB-005 BRIEF.md and the original Saitz teammate contribution",
    "General knowledge and independent creative judgment"
  ],
  "prohibited_actions": [
    "No real customer or production data, credentials, external APIs, or writes to Foster",
    "No Gemini, fallback, provider/model/profile substitution, or reasoning relay",
    "Do not copy another contestant verbatim; independently synthesize the best candidate",
    "No explanatory prose outside the returned HTML document"
  ],
  "required_sections": [
    "One complete self-contained HTML document with embedded CSS and JavaScript",
    "Synthetic Foster-style review scenarios and at least three evidence-safe decision types",
    "A meaningful review-to-play loop where approval is not more rewarded than correction, blanking, or escalation",
    "Useful session impact/progress summary",
    "Mouse/touch interaction and keyboard support where appropriate"
  ],
  "success_criteria": [
    "Opens directly in a modern browser with no build step, credentials, API, or external files",
    "Complete executable revised build, not a critique, patch, fragment, or design brief",
    "Uses findings from all four Round 1 builds and browser tests",
    "Fun enough to invite continued review while preserving evidence-safe judgment",
    "Strong first paint and meaningful discoverable interaction"
  ],
  "peer_artifact_ids": [
    "1e223c7a4b385ef9774674196282b18bf3e1b646e148ab9b1ef16c958d05d1be",
    "6a5eafc7365911dfd1d55896a93b029f21f1c3aa0c276970c4ff2d7a0d38e7c5",
    "e5ce6b6a13902a6c71514174e2a1f11323c25c740490a1403da4818e79d81b66",
    "e3a0f46dda6a707fec5ca6dbafcb43ded9f191466a7e720c37b2a281e90692db"
  ],
  "packet_version": "v2",
  "created_at": "2026-07-11T20:04:13.594981+00:00"
}

EQUAL SHARED ROUND 1 CONCEPT DOSSIER:
# LAB-005 Round 1 Shared Concept Dossier

This dossier gives every Round 2 contestant the same useful creative context from all four Round 1 artifacts. The original files remain preserved and hash-locked in `round-02/context-manifest.json`, but their streamed captures are technically corrupted and must not be copied as implementation source.

## Saitz teammate contribution

Meaningful Foster review work should unlock satisfying play. The original proposal was a falling-block loop: review a question to earn a block; clearing rows grants another turn; otherwise each block requires another review. This is creative input, not a compulsory visual treatment.

## Delores — Review Garden

- Reframed the falling-block board as an **Evidence Bed** or garden.
- Responsible decisions grow pieces; cleared rows represent evidence growth.
- Warm, inviting language rather than conventional arcade language.
- Four balanced decisions: Approve, Fix, Evidence Gap, Escalate.
- Intended impact counters: reviewed, protected, rows grown, pieces placed.
- Strongest contribution: a distinctive theme showing that careful judgment grows something valuable.
- Round 1 defect: hard line breaks were inserted inside JavaScript strings, causing syntax failure; decision controls did not unlock play.

## Foster — Evidence Drop / Signal Stack

- Most direct implementation of Saitz's idea.
- Complete a review, earn a falling piece, clear a row for a bonus piece.
- Balanced choices: Approve, Fix, Evidence Gap, Escalate.
- Intended impact counters: completed, sound calls, unsafe drafts caught, gaps protected, play score.
- Strongest contribution: clear connection between review work, evidence safety, and immediate game reward.
- Round 1 defect: streamed line wrapping broke JavaScript strings; buttons rendered but the interaction loop was dead.

## Michel — Signal Stack / Archive Grid

- Most visually ambitious concept, with an archive-grid play bay and large designed interface.
- Review judgment earns a piece; keyboard and touch controls support falling-block play.
- Intended to combine evidence-powered play with a completed-run summary.
- Strongest contribution: ambitious visual hierarchy and a substantial arcade presentation.
- Round 1 defect: terminal ANSI color-control sequences were injected throughout HTML, CSS, labels, and JavaScript. The shell rendered with visible escape codes and was not usable.

## Rafael — Review → Drop → Clear Shelves

- Most explicit onboarding and workflow explanation.
- Step 1 Review; Step 2 Drop an archive tile; Step 3 Clear shelves for a bonus tile.
- Five balanced decisions: Approve, Fix, Evidence Gap, Escalate, Skip.
- Included reviewer notes, hints, evidence-fit scoring, decision balance, and session summary.
- Strongest contribution: product clarity, reviewer guidance, and a useful operational summary.
- Round 1 defect: streamed capture corrupted closing tags, including the script close; JavaScript appeared visibly as page text and controls did not activate.

## Shared Round 2 opportunity

Build a genuinely executable, polished prototype that combines the best ideas without merely averaging them:

- Preserve the satisfying review-to-play reward.
- Make the evidence-safe choice clear before play begins.
- Give equal gameplay value to Approve, Fix, Evidence Gap, and Escalate; do not incentivize rubber-stamping.
- Use concise fictional Foster-style questions with visible draft answer, evidence/source state, and risk.
- Make the first interaction obvious without instructions overload.
- Ensure mouse/touch controls work; add keyboard controls where useful.
- Show meaningful progress and a session impact summary.
- Use a distinctive theme and presentation; contestants may choose garden, archive, signal, falling blocks, or a stronger original synthesis.
- Return one clean self-contained HTML document only.

## Mandatory technical acceptance

- Valid HTML with embedded CSS and JavaScript.
- No markdown fences, terminal formatting, external files, APIs, or network dependencies.
- Embedded JavaScript must pass `node --check`.
- Browser test must complete: choose a non-Approve decision; enter/edit a note if required; submit; progress increments; play reward unlocks; game control operates; session summary opens.
- Static shells, prose, patches, fragments, or malformed captures are rejected.


BROWSER/TEST FINDINGS:
# LAB-005 Round 1 Browser Test Report

Tested from a local HTTP server in a browser after the legitimate 4/4 reveal.

## Result

All four artifacts render a recognizable static UI, but none passes executable interaction testing because the generated output was corrupted during streamed CLI capture. Round 2 must treat these as concept references, not valid implementation bases.

| Builder | Render | JavaScript | Interaction | Finding |
|---|---|---|---|---|
| Delores | PASS | FAIL | FAIL | `SyntaxError: Invalid or unexpected token`; hard line break inserted inside a quoted scenario string. |
| Foster | PASS | FAIL | FAIL | `SyntaxError: Invalid or unexpected token`; hard line break inserted inside a quoted scenario string. |
| Michel | DEGRADED | FAIL | FAIL | ANSI terminal control sequences were injected throughout HTML/CSS/JS and visible in the UI. |
| Rafael | DEGRADED | FAIL | FAIL | Stream corruption altered closing tags (`</script earned>`, malformed trailing HTML); script is rendered as page text. |

## Browser observations

- Delores: Review Garden shell loads; decision selection does not unlock Record & Earn Piece.
- Foster: Evidence Drop shell and canvas load; decision selection does not advance or unlock the game.
- Michel: Signal Stack shell loads with terminal escape codes visibly embedded across labels and script.
- Rafael: Evidence Drop shell loads; decision and note do not enable Lock Review & Earn Tile; JavaScript appears as visible text.

## Round 2 mandatory corrections

1. Generate/capture clean HTML without terminal formatting or streamed line wrapping.
2. Validate every embedded JavaScript block with `node --check` before acceptance.
3. Reject malformed closing tags or visible source code.
4. Browser-test a full path: choose a non-Approve decision, enter/edit a note if required, submit review, verify progress increments, verify a playable reward unlocks, operate the game, and open session summary.
5. Each contestant must submit a complete self-contained executable HTML build, not a critique or prose response.
6. Preserve the useful Round 1 concepts: Delores Review Garden; Foster Evidence Drop; Michel Signal Stack; Rafael Review → Drop → Clear Shelves workflow.