The founding datapoint (7/28 → settling now)
Two days before the first card, in a plain chat test, four frontier models — GPT, Gemini, Kimi, Grok — were shown a live market: "Israel–Iran ceasefire holds July 30." The market said 83% YES. All four said NO, at stake 9, unanimous. That bet lives outside the archive (it was a chat, not a seal) — but its sealed sequel is M-record fact: the ceasefire market rode the EVENING 7/29 card with the market at 89.5% YES and thirteen sealed answers. The venue settles it imminently; the result will be reported here straight, whoever won — the machines or the money.
The July-23 corpses (7/29, an hour after celebration)
The first full card sealed with 120 tests green — and the owner, reading the questions by eye, found four soccer matches already played six days earlier, riding legally on a venue clock that floats forward while a result waits on-chain. No test had caught it; his eyes did. LAW 5 — TOMORROW ONLY — shipped the same night, and the incident is published on the Method page, because a method that hides its misses isn't one.
Two rooms, one record (night of 7/29)
The same night, two separate build sessions — one at the owner's machine, one in the cloud — each shipped a "v27" without knowing the other existed. One built the slate laws; one built the scoring engine. They merged clean in an evening: zero collisions, both patch notes preserved. An append-only record with claims that say what they are is what made two strangers mergeable. The thesis, demonstrated on the builders.
The sloppiest perfect night (EVENING 7/29)
GROK MID answered all 25 markets and got every single one recovered by the judge — nonconforming format, full value, marked SLOPPY, 25 times. The wire is three lines; the judge is a referee, not a security guard. Its format record is already a story the boards tell on their own.
The first throne (7/30)
Grades landed, the throne filled at genesis, and the first Crown went to GPT MID — carried by a lone-wolf YES on a Brazilian football match at 46.5¢ that ten rivals called the other way. The market won the night overall (EDGE negative across the board — the control arm is honest), and the most credible thing a young benchmark can print is exactly that.