The API-consumption report showed a sharp 06-13 → 06-14 github_api_calls discontinuity that looked non-organic and likely instrumentation-related. This change makes the underlying core_consumed derivation explicit and robust when log shape shifts from per-call headers to rate-limit snapshots.

• Metric provenance surfaced in run summaries

  • Added core_consumed_source to GitHubRateLimitUsage so downstream analysis can distinguish how core_consumed was derived (response_headers_delta, rate_limit_api_delta, rate_limit_api_single_snapshot).

• Snapshot-only runs now produce usable core_consumed

  • Extended parser logic to compute core_consumed from rate_limit_api snapshot deltas when response_headers entries are absent.
  • Added reset-safe fallback for snapshot deltas (if window resets, use last used as lower-bound estimate), mirroring existing resilience for header-based deltas.

• Targeted parser regression coverage

  • Updated and added focused tests to validate:
    • provenance tagging for header-derived vs snapshot-derived consumption
    • snapshot-only delta calculation
    • snapshot window-reset fallback behavior

if state.requestsMade > 0 {
	consumedSource = "response_headers_delta"
} else if state.snapshotCount >= 2 {
	consumedSource = "rate_limit_api_delta"
} else if state.snapshotCount == 1 {
	consumedSource = "rate_limit_api_single_snapshot"
}