Red-main repos: 'typical PR waits Xs' should carry a success-only-basis caveat with the observed failure rate

Migrated from the pre-public development archive (issue #130 there); archive PR/issue numbers in the body refer to that archive.

Pre-flip audit (live run on yasm/yasm, 8/10 default-branch runs failing), LOW, not gating.

Timing is correctly computed over successful runs only (failures don't pollute percentiles — verified live), but a red-main repo gets a clean-looking report: "A typical PR waits 1m 50s" with no hint most runs fail. OPT48 is volume-gated (>=100 runs/mo) AND advisory-only, so low-volume red repos disclose nothing.

Defensible as-designed (speed scope, not reliability), but one caveat line — "measured over successful runs; N% of default-branch runs failed in the window" — would close it. The audit that surfaced this ran the full pipeline against yasm/yasm on 2026-07-21; re-running it there reproduces the shape.