P-M2-1: the pairing run reproduces the oi band

id:        P-M2-1
filed:     2026-07-17, from spec 2105 doc 10 section 6.4, before the M2 pairing run
suite:     core 14
tier:      free
engine:    tomo-oi against tomo-cx, both at the M2-landed pins
band:      floor: oi passes 13 of 14.
           ceiling: oi passes 14 of 14, at under half of cx's total tokens (journal
           aggregate: about 100K against about 280K).
mechanism: the code-as-action surface sends structurally less per round than the
           tool-call surface at equal grade, and the guard family holds the last-mile
           failure shapes closed; the gate runs at n >= 10 where the journal number was
           fewer repeats, so the claim is about the ratio holding, not the absolute
           totals.
settled by: the M2 pairing run at n >= 10.
           go run ./cmd/lab run tomo-oi && go run ./cmd/lab run tomo-cx && go run ./cmd/lab report
result:    (open)
verdict:   (open)

If the band does not reproduce, that is a major finding about either the landed engine or the journal's original measurement, and M2 does not close until the discrepancy has a written explanation.