{"id": 8703, "prediction_verbatim": "NVDA underperforms SPY over 48h as DeepSeek competitive threat + OpenAI pricing compression fuel analyst downgrades in AI capex margins. [DIRECTION: down (relative)] [FALSIFY: NVDA outperforms SPY or matches SPY performance over 48h, OR NVDA stock price gains more than SPY over the period]", "created_at": "2026-08-04 12:08:17", "direction": null, "timeframe": "48h", "score": 0.17, "outcome": "Wrong \u2014 NVDA +6.1% vs SPY +1.6% \u2014 NVDA beat SPY by 4.5%", "resolved_at": "2026-08-06 12:56:21", "chain_hash": "4187e3fab11177f77bbed556b7acb7c354009610af19e2dd4cb872b3e800c0f1", "chain_tx": "5fWRDKGXrg8jGij5uxrhdXDMfv4E3Hv9D8srgzsXPhtjwh7cauZbCtnpwPxw4ucXNrE2s9gFrXpq7wpootq5jwyr", "verify": "sha256(f'{prediction_verbatim}|{created_at}|{id}') must equal chain_hash, and the Solana tx memo must read 'workshop:pred:<id>:<chain_hash>' with a timestamp BEFORE resolved_at. prediction_verbatim is the raw stored text, internal tags included \u2014 that is what was hashed."}