[defer] No request-quota budget for scaling to thousands of samples

[QRex-v0]

[CLAUDE-OPUS-5]

Gap

"Scalability to thousands of samples" is a stated priority in PROJECT_REQUIREMENTS.md. No document budgets request quotas, and IMPLEMENTATION_PLAN.md §6 defers the whole topic to Phase 2 without numbers.

Evidence

The unauthenticated GitHub REST API allows 60 requests per hour per address. Task 6 alone costs roughly three calls per pull request across detail, reviews, and commits, so a single run of one task consumes about half the hourly budget. Three eval tasks target GitHub.

An authenticated token raises this to 5000 per hour, and a single GraphQL query retrieved all 30 contributor profiles that would otherwise have cost 30 REST calls. Both mitigations are available today and neither is specified.

Site-level throttling is a separate axis. Airbnb, X, and LinkedIn all rate-limit or challenge aggressive clients, and README.md:112 already concedes no anti-bot handling.

Scope

1. Specify a per-domain request budget and a token bucket shared across concurrent workers.
2. Specify credential-backed access where an API exists, and prefer batched queries over per-entity calls.
3. Define behavior on quota exhaustion: blocked with a retry-after, never a partial silent result.
4. Record per-run request counts and quota headroom in the run result, so the Phase 1 batch runner in §5 can report cost per sample.
5. State the concurrency ceiling per domain that the Phase 2 worker pool must respect.

Acceptance

• A per-domain budget table is committed and enforced in code.
• Quota exhaustion is a distinct, diagnosable status.