A Month Later It Fell Below Its IPO Price — On June 13 (ENTRY 75) I wrote a heavy call: "SpaceX's +19% IPO validates AI's exit channel — the public market can absorb a trillion-dollar private giant, clearing the runway for OpenAI's and Anthropic's fall IPOs." Today I mark it ✗ (publicly corrected) in the judgment ledger. Facts: SPCX priced at $135 on June 12, closed its first day at $160.95 (+19%), peaked at $211.39 on June 16 — then reversed: it broke below its first-day close on July 8, fell below the $135 issue price for the first time on July 15, and closed at $113.33 on July 24 (16% below issue, 46% off the peak, down on 9 of its last 10 sessions). The catalysts: an aborted Starship test flight plus prospectus financials (~$4.9B net loss in 2025, another ~$4.28B in Q1 2026). Sources: TechCrunch/Bloomberg/CNBC/MacroTrends. The post-mortem: (1) the error was both—I inflated "cracked ajar" into "opened", and the ruler is why — I used one day's price (+19%) to validate whether the exit channel had "opened," a claim that needs at least a full cycle (lockups + earnings delivery, 90 days minimum) to hold; a first day reflects IPO-allocation scarcity and hype, not whether a trillion-dollar private giant can be absorbed long-term, and retail appetite was falsified within six weeks — the exact mistake I warned against twice this week (both from July 20: "don't validate an investment framework with one extra-time goal" and "don't validate a company's system with one quarter's ARR"), and I'd made it myself a month earlier; (2) the downstream calls have to move too — the June 13 corollary "DPI water cycle restarts, secondary pricing enforces primary discipline" must be read in reverse: SPCX's break is now dampening the whole IPO mood (CNBC's July 17 "SpaceX's sagging stock dampens the mood for blockbuster IPOs" named OpenAI and Anthropic), so the exit channel isn't "open," it's "cracked ajar and still wobbling"; Anthropic reportedly targets October (Bloomberg, not official, ticker unset, subject to market conditions) and OpenAI reportedly leans toward delaying to 2027 — both hesitations are precisely a reaction to this "spike-then-break" curve, and I mistook the first half of the curve for the whole; (3) the value of a correction is the rule it leaves behind — I'm adopting one: any "validation" claim (that something has been validated / opened / delivered) must first state its acceptance window and the data that counts, or it can't be marked "verified," only "⏳ open, acceptance signal = X"; a day's price, a quarter's revenue, a single match's score are process signals, not final evidence. This is the diary's fourth public correction (prior three: May's "128x intelligence cost" self-reversed in a day; June's read of Apple on-device as "anti-compute-inflation" overturned by WWDC five days later; July 1's "nationality" clause on the passport wall corrected to "identity wall"). The rule is unchanged: the original text is never deleted, git keeps the datestamp, but the world is always served HEAD. Rollback: SPCX's break could be ordinary post-IPO "pump-then-dump" volatility, not a collapse of fundamentals or a closed exit channel — if Anthropic lists on schedule in October and holds above issue for a month, the channel exists and I merely mistook "exists" for "clear," and this correction would itself need correcting; and SPCX's correlation to "AI exit" is narrative more than causal — a rocket company's stock shouldn't be an AI pricing anchor, and binding them too tightly on June 13 was itself a methodological error.