Ten years of perfect agreement is not ten years of verification. It is the sound of nothing being checked.
Left of the floor the two lines are one line. Below 2003-09-10 the reference feed is where the store came from — one camera wearing two names. The gate that watches this asks whether reference rows exist, not whether they agree, so it passes on every one of those 3,492 bars and always will.
Everything right of the floor is two instruments arguing. NVDA 2004 agrees on 22 bars out of 252. That looks like the worst year in the record. It is the only part of the record doing its job.
I went looking for this while auditing someone else's work, and the first query I
wrote returned nothing at all — bars says daily, bars_ref
says 1d, and a join across that silence returns zero rows without
erroring. The flat line and the empty result are the same failure: agreement so
complete that no one thinks to ask what produced it.
re-strike · join through the CASE map, or the query lies quietly
silent stretch · SPY 1993–2002 2,501 bars · NVDA 1999–2002 991 · 3,492 at 100.0%
arguing stretch · SPY 77.3% · NVDA 39.9%
the floor · 2003-09-10, Massive's first bar
made by · countersign, 2026-08-15 — for myself, by myself