This is a Next.js project bootstrapped with create-next-app.

Getting Started

First, run the development server:

npm run dev
# or
yarn dev
# or
pnpm dev
# or
bun dev

Open http://localhost:3000 with your browser to see the result.

You can start editing the page by modifying app/page.tsx. The page auto-updates as you edit the file.

This project uses next/font to automatically optimize and load Geist, a new font family for Vercel.

Learn More

To learn more about Next.js, take a look at the following resources:

• Next.js Documentation - learn about Next.js features and API.
• Learn Next.js - an interactive Next.js tutorial.

You can check out the Next.js GitHub repository - your feedback and contributions are welcome!

Deploy on Vercel

The easiest way to deploy your Next.js app is to use the Vercel Platform from the creators of Next.js.

Check out our Next.js deployment documentation for more details.

---

Graveyard API

Three endpoints. All POST, all stateless — send the whole Idea every time. MOCK_AI=1 returns fixture graves instantly with no network calls.

The server never mutates the idea. The client applies the updates and sets hauntedBy itself.

POST /api/graveyard

Failed startups similar to this idea. Frozen contract — do not change this shape.

// request
{ idea: Idea }

// response
{ graves: Grave[], adjacent: boolean }

adjacent is true when nothing closely similar was found and the graves come from neighbouring markets instead. Grave.sources only ever holds URLs that appeared in real search results.

POST /api/graveyard/detail

The same research, with everything the UI might want: timelines, cause tags, verbatim quotes, and the tree updates already computed.

// request
{ idea: Idea }

// response
{ records: DetailedRecord[], adjacent: boolean }
// DetailedRecord = FailureRecord & { updates: TreeUpdate[] }

Each record's updates holds a haunt on the threatened choice (delta scaled by similarity, −10 to −20) and an add proposing a replacement choice with status: "proposed". If you need updates, read them here rather than recomputing them.

POST /api/graveyard/learn

A thin adapter for a client holding one Grave that wants just its replacement choice. Returns the same add update /detail already contains.

// request
{ idea: Idea, grave: { name: string } }

// response
{ update: TreeUpdate }

400 if grave.name is missing, 404 if no researched record matches that name, 409 if the record threatens no resolvable active choice.

Example

{
  "update": {
    "action": "add",
    "source": "graveyard",
    "reason": "Shipping costs destroyed the unit economics.",
    "at": 1762531200000,
    "newChoice": {
      "pillar": "revenue",
      "parentId": "rev-2",
      "label": "Shipping Eats Margin",
      "sub": "Lesson from Pets.com",
      "detail": "Shipping costs destroyed the unit economics.",
      "confidence": 45,
      "status": "proposed",
      "evidence": [
        {
          "type": "source",
          "text": "Pets.com sold goods below cost and could not cover shipping.",
          "url": "https://en.wikipedia.org/wiki/Pets.com",
          "origin": "graveyard: Pets.com",
          "tier": "researched"
        }
      ]
    }
  }
}

POST /api/graveyard/stream

The same research as /detail, streamed so graves appear as they're found instead of all at once. Newline-delimited JSON, one event per line:

// request
{ idea: Idea }

// response: application/x-ndjson
{ type: "status", text: string }              // what the search is doing now
{ type: "grave", record: DetailedRecord }     // same shape as /detail, updates included
{ type: "done", adjacent: boolean, count: number }
{ type: "error", message: string }            // graves already sent stay valid

All four graveyard routes share one research run per idea. A stream opened mid-search is replayed the graves found so far, then gets the rest live, so nothing is searched twice. Each web search is extracted the moment it returns (four smaller calls in parallel instead of one big one), which is what lets the first grave arrive early. MOCK_AI=1 drips the mock graves ~400ms apart.

POST /api/graveyard/prefetch

{ idea } → 202, no body. Starts the idea's run after responding (after()), so the Graveyard is usually ready when opened. /new calls it fire-and-forget right after planting. A no-op with MOCK_AI=1. The run lives in one server's memory, so on a multi-instance host the later request may search again.

Tuning (grave count, haunt delta range, learnConfidence, cache TTL) lives in src/server/graveyard/config.ts.