業績,點樣驗? A track record, verified how?

AI 選股工具個個都話自己有業績。與其問「邊個最準」,不如問「邊個啲數驗得到」。以下五條問題,任何工具答唔出,答案已經喺度。 Every AI stock picker claims a track record. Don't ask "who's most accurate" — ask "whose numbers can I check?" Five questions. Any tool that can't answer them has already answered.

1. 預測,有冇即場封存?Are predictions frozen the moment they're made?

   先講後驗。每週名單一出即凍結,改唔到、刪唔到。冇封存嘅「歷史戰績」,隨時係尋日先寫嘅。Call first, verify later. Every weekly list is frozen on publication — can't be edited or deleted. An "historical record" that isn't frozen may have been written yesterday.

2. 淨值,有冇密碼學簽名?Is the NAV cryptographically signed?

   Merkle + Ed25519 簽咗嘅日淨值,任何人攞 API 數據落嚟都可以離線複算。改一個位,簽名即刻穿崩。Daily NAV signed with Merkle + Ed25519; pull the API and recompute offline. Change one bit and the signature breaks.

3. 歷史,可唔可以回放?Can history be replayed?

   每個凍結週有自己嘅永久網址——?w=2026-10-16 睇到嗰日講咗乜。RSS 檔案都係 backdated,唔係得最新一期。Every frozen week has a permanent URL — ?w=2026-10-16 shows exactly what was said that day. Even the RSS archive is backdated, not just the latest issue.

4. 有冇第三方對照?Is there an independent cross-check?

   今週 Top-10 裏 9 票同獲 Danelfin 獨立 AI 評分 ≥7(2026-10-24 快照);MRNA 僅 4 分——異見都照記。第三方唔係證明,但方向一致值得知。9 of this week's Top-10 also score ≥7 on Danelfin's independent AI (2026-10-24 snapshot); MRNA sits at 4 — the dissent stays on record. A third party proves nothing, but agreement is worth knowing.

5. 佢肯唔肯講邊度會錯?Does it admit where it can be wrong?

   回測永遠靚仔。模擬倉喺 IBKR 撮合引擎真金實盤規則成交,先至接近現實。仲肯話你知「紙上模擬唔等於實盤」嘅,先係老實。Backtests always look pretty. Paper fills on IBKR's real matching engine come closer to reality. A tool that tells you "simulated ≠ live" is being honest.

模擬倉業績唔等於實盤;過往表現唔保證將來。驗證得嚟嘅嘢,係「佢冇講大話」,唔係「佢次次啱」。 Simulated performance isn't live trading; the past guarantees nothing. What verification buys you is "it isn't lying" — not "it's always right".