The AI Private-Valuation Marks-vs-Reality Ledger (v1): a pre-registered, reproducible scoreboard for how private AI valuations survive a real public price

A reproducible, primary-sourced ledger that scores how private AI-company valuations survive contact with a real public price. It is not a forecast, a price target, or a bubble call; it is a referee layer. A private mark is recorded, then checked against the next real repricing event, and the gap is reported. Four properties: reproducible (full data plus a standard-library script), primary-sourced (S-1s, IPO prints, post-IPO placements, fund N-PORT marks), motive-neutral (no book, fee, or subscription to sell), and pre-registered (predictions recorded before the event, then graded). Headline (SpaceX, June 2026): two pre-registered calls resolved. (1) The honest private-to-public step-up was about 1.42x, not the widely-cited 5.1x (a stale-baseline artefact that used a December-2024 mark and ignored four later,higher ones); against the correct last private mark, the roughly $1.25T xAI-merger valuation, the $1.77T IPO price is 1.42x. (2) A day-one pop, not a drop: SPCX opened $150, peaked $176.52, and closed at $160.95 (up 19.2%, about $2.11T); the open was the day's low and never broke the issue price. The open prediction, publicly time-stamped here, is that any real correction lands at lock-up expiry, not on debut (Morningstar fair value about $780B, roughly 56% below the $1.77T price). Changes in v1.1 (standards refresh, not the lock-up version). Corrects one isolated stray figure (day-one intraday high $165 to the web-confirmed $176.52; the rest of the document already carried $176.52); standardises the Morningstar gap (about 56% below the $1.77T price, about 55% below the $1.75T target); tidies the Anthropic Series H note ($65B round, $965B post-money). Repackaged to portfolio standard: Markdown source plus a house-style PDF,replacing the v1 .docx. The substantive v2 follows at SpaceX lock-up expiry. Author: N. Milton (ORCID 0009-0003-4213-7769), independent. The author holds no position in, and receives no compensation from, any entity named. Produced with AI assistance (Claude Opus 4.8, Anthropic), which is disclosed for transparency as the model's maker is within the sector analysed; the analysis relies on cited primary and public sources. Open-science documentation only, not investment advice. CC BY 4.0.