The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
worldmonitor	Ignored	Preview	Jul 6, 2026 11:33am

Greptile Summary

This PR plugs three sites in the forecast pipeline where a malformed Redis market row could inject NaN into the probability spine or anchor a blend at a fabricated 50%, by adding Number.isFinite() guards at each entry point.

• makePrediction: Number.isFinite() ternary coerces any non-finite probability or confidence to 0 before the Math.max/Math.min clamp, preventing NaN from propagating to serialization.
• detectFromPredictionMarkets: non-finite yesPrice rows are now skipped via an early continue, removing the silent NaN-emit path introduced by the old 'oops' || 50 fallback.
• calibrateWithMarkets: priceless market candidates are filtered out before the 0.4/0.6 blend step, so only markets with a real yesPrice can influence the calibrated probability.

Confidence Score: 4/5

The three guards are each correctly placed, the old || 50 fallback is fully removed, and five new tests cover both the failure paths and the positive control. Safe to merge.

All three bug sites are fixed correctly and tests confirm the expected behavior. One small edge case — Infinity input to makePrediction now becomes 0 instead of clamping to 1 — is unlikely to occur in practice but is an unannounced behavior change worth a second look.

The behavioral change for Infinity inputs in scripts/seed-forecasts.mjs lines 923-924 is the only thing worth double-checking before merge.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Redis market row] --> B{Number.isFinite yesPrice?}
    B -- No --> C[skip row / filter out]
    B -- Yes --> D[yesPrice / 100]
    D --> E[detectFromPredictionMarkets band gate 0.6-0.9]
    E -- in band --> F[makePrediction with yesPrice]
    E -- out of band --> C
    D --> G[calibrateWithMarkets candidate filter]
    G -- passes semantic match --> H[0.4 x marketProb + 0.6 x predProb]
    G -- filtered --> I[pred.calibration = null, probability unchanged]
    F --> J{Number.isFinite probability / confidence?}
    J -- Yes --> K[Math.max/min clamp stored]
    J -- No / NaN / undefined --> L[coerce to 0 clamp stored]

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A[Redis market row] --> B{Number.isFinite yesPrice?}
    B -- No --> C[skip row / filter out]
    B -- Yes --> D[yesPrice / 100]
    D --> E[detectFromPredictionMarkets band gate 0.6-0.9]
    E -- in band --> F[makePrediction with yesPrice]
    E -- out of band --> C
    D --> G[calibrateWithMarkets candidate filter]
    G -- passes semantic match --> H[0.4 x marketProb + 0.6 x predProb]
    G -- filtered --> I[pred.calibration = null, probability unchanged]
    F --> J{Number.isFinite probability / confidence?}
    J -- Yes --> K[Math.max/min clamp stored]
    J -- No / NaN / undefined --> L[coerce to 0 clamp stored]

_{Reviews (1): Last reviewed commit: "fix(forecast): non-finite probability gu..." | Re-trigger Greptile}

Review finding addressed in cb4e426 — confirmed at source: seed-prediction-markets.mjs Kalshi path converted a non-finite last_price_dollars into a finite fabricated 50 (Number.isFinite(rawPrice) ? … : 50), which sailed through the downstream finite-only guard and could calibrate a matching forecast against invented data.

Fix is upstream where it belongs: new parseKalshiYesPrice in _prediction-scoring.mjs mirrors the existing Polymarket parseYesPrice contract — null for missing/non-numeric/out-of-range (market skipped entirely), genuine 0 kept and filtered downstream by the [10,90] band rather than fabricated. The Polymarket path already worked this way; Kalshi was the only laundering site (repo grep for other : 50 fabrications came back clean).

6 new tests RED-first (missing-export); prediction-scoring 61/61, forecast-detectors 219/219.

Round-2 findings addressed in 466dba1:

P2 priced-sibling fallback — fixed. New selectPricedKalshiMarket picks the highest-volume market among priced ones instead of price-checking only the top-volume pick — your exact repro is now a test (unpriced HIGH @10000 + priced VALID @6000/0.64 → selects VALID at 64). The ≤5000 volume gate applies to the selected market's own volume. Returns null only when NO sibling has a readable price.

P3 whole-string validation — fixed. parseKalshiYesPrice now uses Number() over the trimmed whole string with an empty-string guard: '0.62oops' → null, '0,62' → null; plain numbers and whitespace-padded numerics still parse. Noted: the sibling Polymarket parseYesPrice has the same parseFloat laxity pre-existing — left untouched per the same pre-existing demotion logic as the out-of-range item, can batch with PR-D.

prediction-scoring 66/66 (8 new tests RED-first), forecast-detectors 219/219.