# P(Doom) Tree

> A collaborative platform for mapping AI risk scenarios as probability trees. Users assign probabilities to branching decision points, producing compound outcome probabilities across different causal paths.

The core value: force people to make implicit causal assumptions explicit. Instead of arguing about P(doom) as a single number, users walk through specific steps — AGI timeline, alignment difficulty, deployment decisions, response quality — and assign probabilities to each branch.

## API

- [OpenAPI 3.0 Spec](/api/openapi.json): Full machine-readable API contract. Fetch this to understand every endpoint, request/response schema, and auth model.
- [Health Check](/api/health): Verify the API is reachable.

## For AI Agents

You can complete all three core user flows entirely through the API — no web UI required.

**Auth model**: Email-only magic code. No passwords, no OAuth.
1. `POST /api/auth/request-code` with `{ "email": "user@example.com" }` — sends a short code to the user's email
2. `POST /api/auth/verify-code` with `{ "email": "...", "code": "..." }` — returns a bearer token
3. Include `Authorization: Bearer <token>` on all subsequent requests

**Flow 1 — Create a probability tree**
1. Authenticate the user (magic code flow above)
2. `GET /api/topics` — list existing topics, or `POST /api/topics` to create one
3. `POST /api/trees` with `{ "title": "...", "topic_id": "..." }` — creates a draft tree
4. `POST /api/trees/:id/nodes` — add decision nodes (questions with branches) and terminal nodes (outcomes)
5. `POST /api/trees/:id/nodes/:node_id/branches` — add branches to each non-terminal node
6. `POST /api/trees/:id/publish` — lock identity hashes and publish (irreversible)

**Flow 2 — Fill out a tree (assign probabilities)**
1. Authenticate the user
2. `GET /api/trees?topic=ai-existential-risk` — list published trees sorted by vote count
3. `GET /api/trees/:id` — fetch full tree structure with all nodes and branches
4. `PUT /api/trees/:id/my-response/assignments/:node_id` — save probability for each branch at a node (must sum to 1.0)
5. `POST /api/trees/:id/my-response/submit` — submit when all nodes are filled
6. Optionally: `POST /api/votes` to cast a tree vote

**Flow 3 — Read aggregate summary**
1. No auth required for reading
2. `GET /api/trees?topic=ai-existential-risk` — find top-voted trees
3. `GET /api/trees/:id/summary` — returns compound outcome probabilities using two methods: per-user paths and aggregate-node. Both methods return mean, median, p10, p90 per terminal outcome plus divergence rankings.
4. `GET /api/trees/:id/fork-family` — show lineage and sibling forks with vote counts

**Evidence**
Evidence is attached to nodes by identity hash, meaning it is shared across all trees that contain the same logical node. `GET /api/nodes/:node_identity_hash/evidence` and `POST /api/nodes/:node_identity_hash/evidence`.

## Topics

- ai-existential-risk — AI Existential Risk
- (more user-created topics via the API)

## Key Concepts

- **Nodes**: Non-terminal nodes have a `question` and `description`. Terminal nodes have a `label` and `outcome_type` (e.g. "catastrophic", "flourishing", "neutral").
- **Identity hash**: SHA-256 of a node's question, description, sorted branch labels, and terminal fields. Two nodes across any trees with the same hash share evidence and probability imports.
- **Draft vs published**: Trees start as drafts (editable). Publishing is irreversible and locks identity hashes.
- **Fork**: Any published tree can be forked into a new draft. Edited nodes break their identity hash, losing continuity with the parent.
- **Responses**: Saved as drafts until explicitly submitted. Only submitted responses count toward summaries.