[YinkaMetrics]

Summary

• retain the regenerated 2026-08-05 Mem0 evidence as verified retrospectively by quota reconciliation
• record that provenance was not self-recorded at execution time — the reason invariant 11 now exists
• cite the founder's independent live Mem0 SEARCH reading of 599 on 2026-08-05, against approximately 885 before the live work
• require future live-provider runs to self-record executor, environment, SEARCH quota before/after, and the exact results filename
• make the diagnostic refuse missing executor/environment and persist provenance before mutations
• correct the customer-facing full 15×2 cost from the historical ~106 estimate to at least 182 SEARCH under invariant 10

Retrospective verification

The founder independently read Mem0's live SEARCH counter at 599 on 2026-08-05, against approximately 885 before the live work. Mem0 owns this counter outside the repository; repository code cannot write it.

The approximately 286-unit movement is consistent with:

Bucket	SEARCH units
Seven-arm diagnostic	41
Full 15×2 execution	at least 182
Unallocated remainder	at most 63
Total movement	approximately 286

Status: verified retrospectively by quota reconciliation; provenance not self-recorded at execution time — the reason the rule now exists.

The unallocated remainder may include extra convergence polls, standalone quota probes, or earlier development calls. The surviving artifacts do not support assigning it more precisely.

Publication status

The current 2026-08-05 report and run ID 1986edd5512147dca783bc513029b4f3 remain the published evidence. This PR does not roll them back. Existing Mem0 findings are unchanged: pack identifiers do not slug-collide, the run used seeds=2, there was no concurrency, and the answering layer was quoting.

Invariant 11 applies prospectively: a future run missing any required provenance field is reported, unverified and cannot support a gated publication claim.

Customer impact

The README now states that a default 15×2 Mem0 run costs at least 182 SEARCH, not ~106. The minimum comprises reset pre-reads, write confirmations, delete reads/confirmations, and scenario queries; additional convergence polls increase the total.

Verification

• pytest -q: 89 passed, 4 credential-gated tests skipped
• diagnostic dry-run: exits 0 with no provider access
• diagnostic live invocation without executor/environment: exits 2 before credential or provider access
• JSON report parses successfully
• git diff --check: clean
• no additional Mem0 calls or quota spend for this correction
• GitHub Actions memorycheck-ci run #70: PASS

Required label: needs-founder-review. Keep this PR draft and do not merge without explicit founder approval.