P-M3-A: the spend-ceiling slice is invisible at defaults
The spend-ceiling slice changes nothing on any suite at defaults; a pure rail is invisible until a runaway makes it visible.
id: P-M3-A
filed: 2026-07-17, from spec 2105 doc 10 section 7.4, before the M3 gate runs
suite: core 14 and swebench-live
tier: all tiers the M3 gate runs
engine: any engine under the spend ceiling at default values
band: floor: no pass count, token median, or cost column moves on any suite when
the ceiling slice lands at defaults.
ceiling: the ceiling's first visible act, whenever it comes, is a capped
entry on a runaway, not a changed graded row.
mechanism: a pure-rail slice adds a bound that no healthy run approaches, so at
defaults it should be observationally absent until a runaway crosses it.
settled by: the M3 gate sweep, compared column by column against the pre-slice report.
go run ./cmd/lab report && go run ./cmd/lab report --suite swebench-live
result: (open)
verdict: (open)
This entry complements P-C3. P-C3 predicts where the ceiling fires first; this one predicts that landing the ceiling changes nothing before that first fire. A miss here means the defaults were set inside the envelope healthy runs actually use, which reopens the knob before any spend claim can lean on it.