calibration: pick a data-backed global shipped default for quality_gate_score, then register the second ceiling knob

[JSONbored]

Problem

quality_gate_score stayed out of the registry (#8224's recorded finding): qualityGateMinScore is per-repo nullable with no global shipped default, and every registry entry anchors on the shipped constant. Rather than inventing a number, derive it: the readiness-score capture (#8223) plus realized outcomes can backtest candidate defaults the same way every other threshold earns its value.

Requirements

⚠️ Required pattern. The default is chosen by evidence, recorded on this issue BEFORE the constant ships: replay candidate defaults over the quality_gate_score corpus (once it clears sane floors), pick the value whose backtest dominates, document the comparison. Then the registry entry mirrors #8224's ceiling shape verbatim (report_only, orientation ceiling, written bounds rationale, unique seed, >=50/12 floors).

Deliverables

• Evidence-backed default recorded here (comparison tables)
• DEFAULT_QUALITY_GATE_SCORE shipped + consumed as the nullable field's documented default
• Registry entry + invariant tests, mirroring the slop knob

Links & Resources

#8224 (the ceiling precedent + recorded finding), #8223 (capture), src/services/loosening-knobs.ts

Boundaries

Report-only registration; the default must not change live gating behavior for repos with an explicit setting.

maintainer-only — gate-adjacent policy.

[linear-code]

JSO-5727

[JSONbored]

Data unblock: the deployment's own repos now run the readiness/quality gate in advisory mode with a configured minScore — no gate behavior change, but quality_gate_score capture (the #8223 writer requires a non-null configured minimum) starts flowing from every reviewed PR. At current review volume the corpus should clear sane replay floors within days-to-weeks, at which point the candidate-default backtest this issue specifies becomes runnable.