30-day hit-rate sparkline resolved-prediction trend
/api/public/predictions/hit-rateToday's best bet highest-confidence open prediction
/api/public/predictions/best-bet30-day prediction timeline per-ticker, color-coded by horizon
/api/public/predictions/by-target/<ticker>/timelineToday's predictions top 10 by confidence
/api/public/predictions/latestTrack record resolved-prediction hit rate
/api/public/predictions/leaderboard| Horizon | Total | Resolved | Hits | Partials | Hit % |
|---|---|---|---|---|---|
| Loading… | |||||
Look up a target
/api/public/predictions/by-target/<sym>Type any ticker symbol (or event class like SEC-FLOW / MACRO) to see every prediction the deep brain has issued for it, plus the signals that drove them.
Latest signals raw input to the brain
/api/public/predictions/signalsLive Signals filterable feed, 1h auto-refresh
/api/public/predictions/signalsSource Registry 15 active sources powering the deep brain
static manifestPublic endpoint not exposed; below is a static manifest matching the rows in pred_sources. Each source can be toggled by the operator and contributes to pred_signals.
Signal Sentiment Distribution tagged signals from the daily pool
/api/public/predictions/sentiment-feedMethodology & caveats
Pipeline. Every morning at 06:00 local, TCG-Predictions-Daily runs three stages: (1) fetch pulls every enabled source into pred_signals; (2) reason clusters the last 24h of signals by ticker (or by event class for filings/macro), then asks qwen2.5:7b-instruct (running locally in Ollama) for one prediction per horizon (1d / 1w / 1m) per cluster, with a strict JSON schema; (3) resolve walks every prediction whose horizon has elapsed and compares against the actual price (for stocks), writing hit / miss / partial back into pred_outcomes.
Sources. SEC EDGAR (8-K + 10-Q Atom), Federal Reserve FOMC calendar, MarketWatch top stories, CNBC business, Reuters business, BLS employment-situation report, BEA current GDP releases, CFTC + FTC press releases, CourtListener financial litigation (when token wired), plus TCG's own microstocks / top-earners / hot endpoints (Yahoo-derived). Each source has a row in pred_sources the operator can toggle.
Honesty. The deep brain only knows what is in the signal pool. When signals are weak or contradictory the model is required to mark flat with low confidence. The track-record numbers above are real — every resolved prediction is dispositioned against the actual price; nothing is hand-cherry-picked.
This is not financial advice. Models are wrong constantly. Anything on this page is research output, not a recommendation. Trade at your own risk.