Public API

Search by free text, optionally scoped to one entity type.

Query params: q (required) · type: all | figure | topic | statement (default all)

List figures.

Query params: topicId · page · pageSize

Get one figure by slug, with stats and statements.

Get one figure by id.

Overall and per-topic FigureStats rows for a figure.

Overall accuracy, per-topic accuracy breakdown, resolved/total counts.

Accuracy and resolved-prediction count per year (bucketed by when each outcome was resolved, not made).

Create a figure.

Update a figure.

Delete a figure.

Validate a batch of figures (CSV-style rows) without creating anything — returns per-row errors/resulting slugs.

Create the figures from a previously-previewed batch.

List categories with their sub-topics nested.

List top-level categories only.

Every topic with domain difficulty, statement count, and figure count.

List one category's sub-topics.

Get one topic, with its figures and statements.

Domain difficulty (mean accuracy across every figure active in this topic) and a per-figure ranking.

Create a topic.

Update a topic.

Delete a topic.

List/filter statements.

Query params: figureId · topicId · status · predictionType · q · dateMadeFrom · dateMadeTo · page · pageSize

Get one statement, with its outcome if resolved.

Groups of near-duplicate statements (shared topic, close dates, overlapping keywords — no AI).

List outcomes.

Get one outcome.

Edit history for one outcome.

Create a statement.

Update a statement.

Delete a statement.

Create (resolve) an outcome.

Update an outcome.

Delete an outcome.

AI-generated outcome suggestion (market data or news evidence) for one unresolved statement. Suggest-only — never writes to the database on its own. Spends real money on the Anthropic API per call.

The same suggestion, for up to 20 statements in one request (processed sequentially). Each call still costs money per statement.

Query params: body: { statementIds: string[] }

Current price for a ticker symbol.

Query params: symbol (required)

Historical daily closes for a symbol and date range.

Query params: symbol (required) · start (required, ISO date) · end (required, ISO date)

All recent articles from every configured feed, flattened into one list.

Fetch a URL and extract clean article text via Readability. Guarded against SSRF (rejects localhost, private/link-local IPs, and the cloud metadata address, including through redirects).

Query params: body: { url: string }

Normalize raw transcript text — strips timestamps and bracketed stage directions like [applause], but keeps speaker labels (needed for figure attribution in the next step).

Query params: body: { text: string }

LLM call: extract candidate predictions (figure, topic, statement text, type, market symbol, target date, confidence) from clean text. Returns at most 20 candidates; nothing is written to the database.

Query params: body: { text: string } (50–20,000 characters)

Create real statements from up to 20 reviewed/edited candidates. Each item needs a real figureId (and optional topicId) — no auto-creation of new figures or topics. Per-item success/failure, not all-or-nothing.

Top figures by accuracy. Adding startDate/endDate computes live from outcomes resolved in that window instead of the all-time cache.

Query params: topicId · sortBy: averageAccuracyScore | correctPredictions | resolvedPredictions · limit (max 100) · minPredictions · startDate (ISO date) · endDate (ISO date)

One figure's rank within the full (unlimited) sorted leaderboard.

Query params: topicId · sortBy · minPredictions

Accuracy, rank, per-topic breakdown, and recent activity for each figure.

Query params: slugs (required, comma-separated, 2–5 figures)

Trending figures, recent predictions/resolutions, and top topics by activity.

This week’s featured figure — deterministically picked from the trending pool by the current week number, not stored or randomly generated, so it’s stable all week and rotates on its own.

Basic liveness check.

Admin login. Single hardcoded admin credential (email/password from environment variables) — there is no public user-account system yet.

Query params: body: { email: string, password: string }