Keep a permanent, hash-chained ledger of public forecasts and see who has actually earned weight. Add calls before they resolve, add outcomes later, and get hit rate with confidence intervals, calibration, and naive versus track-record-weighted consensus on open questions. Edited or removed entries are detected. Offline deterministic script; calls recorded after the answer was known earn no weight.
Paste your Playwright, Puppeteer or Selenium symptom; get ranked causes, probe code and a read-back check.
Turn vague Expo bugs into safe, reproducible fix briefs.
You get: a seven-slide decision outline with an evidence ledger, decision brief, option matrix, and action register. Verdict: rollout, follow-up, or hold is shown only against your supplied criteria, with [UNVERIFIED] gaps labelled separately — so you do not invent significance, causality, or outcomes.
用 AI 做事最容易蒸發的不是成果，是決策理由。檔案還在，但「為什麼這樣設計、為什麼放棄另一個方案」隨著對話關閉一起消失，三週後你會完全想不起來。這支把一場對話收斂成四類寫進每日紀錄，封存前再逐檔回查驗證真的寫進磁碟——AI 說「已寫入」跟磁碟上真的有那個檔是兩件事，更陰險的是索引寫了但檔案本體從未落盤。
For a short you already posted. Send the video plus a screenshot of its retention graph (or your average % viewed) and get the curve redrawn with the drop-offs marked, the frame on screen at each drop, every leak ranked, a cut list with the new target length, and a verdict printed next to its threshold.
Paste an RFP and your company evidence, and get a requirement-by-requirement matrix — each row labelled supported, partially supported, a gap, a question, or out of scope — with cautious draft wording for supported rows, a red-flag list for unsupported commitments, and a prioritised verification queue. Verdict: ready to draft only where evidence supports it; gaps and approval-dependent claims stay in the queue, not in the response.
Paste rough notes, a transcript, or chat log. Get confirmed decisions, owner-and-deadline actions, blockers, open questions, a follow-up email, and next agenda without invented commitments.