Interrogation Sim MVP

Local MVP for a shared-world, multi-character interrogation sim.

Run

1. Install dependencies:
   • npm install
2. Create a local .env (copy from .env.example) and set OPENAI_API_KEY.
3. Start the server:
   • npm run dev
4. Open http://localhost:3000.

Notes

• State is stored in-memory per session.
• Character responses use the OpenAI API when OPENAI_API_KEY is set.
• Agent prompt templates live in agents/.
• Language toggle (English/Greek) is available in the UI.
• Model mode toggle (auto/routine/critical) is available in the UI.

Story Logic Judge + Feedback Loop

• LLM judge endpoint: POST /api/story-judge
  • Accepts novel (or story_text) plus optional cast_list, timeline_notes, clue_list, solution_reveal.
  • Returns structured scores + a full markdown critique report.
• Feedback loop endpoint: POST /api/story-loop
  • Runs judge -> pass/fail gate -> minimal patch -> re-judge (up to max_rounds).
  • If gate passes, runs constructor next and (by default) writes a backlog entry to server/story_backlog/.
  • Returns each round’s judge report, patch summary, final verdict, constructor status, and backlog metadata.
• Unified endpoint: POST /api/storylab
  • Same inputs as above.
  • Set judge_only: true for judge-only mode; default runs loop mode.
  • Use persist_backlog: false to skip backlog write.

Quality gate (pass criteria):

• total score >= 80/100
• internal consistency >= 18/25
• means/opportunity/motive integrity >= 14/20
• clue fairness >= 14/20
• critical contradictions <= 0
• major contradictions <= 2

CLI:

• Judge only (text report): npm run story:judge -- <caseId> --file <story.txt>
• Judge only (JSON): npm run story:judge -- <caseId> --file <story.txt> --json
• Feedback loop JSON: npm run story:loop -- <caseId> --file <story.txt>
  • Reads from stdin when no --file is provided, so copy/paste works directly.
  • Add --no-backlog to skip backlog persistence.
• Batch judge+fix all library stories: npm run story:all -- --max-rounds 3 --write-casepacks
• Legacy storylab review: npm run storylab:review -- <caseId> <novel-file-path>

Vercel (serverless)

• API routes are in api/ and are compatible with Vercel Serverless Functions.
• Deploy the repo and set environment variables in Vercel:
  • OPENAI_API_KEY
  • Optional: OPENAI_MODEL_ROUTINE, OPENAI_MODEL_CRITICAL, OPENAI_MODEL_NARRATOR, OPENAI_MODEL_CHECKER, OPENAI_MODEL_STORY_JUDGE, OPENAI_MODEL_STORY_FIXER, OPENAI_MODEL_STORYLAB, OPENAI_MAX_OUTPUT_TOKENS, OPENAI_TEMPERATURE
• Note: in-memory sessions are ephemeral on serverless; persistence needs a DB/KV for production.