Status: Completed on merge — PR #425 hosted-green, squash-merged at
0fad272b, and exact-main verified
Parent: Corrective prerequisite for CK-09
Recommended owner: test_engineer evidence-scale; Luna
Accounting: TASK_PACKETS.md
Central plan: REMAINING_EXECUTION_PLAN.md
Roadmap: AGENT_FIRST_CLEAN_CUTOVER.md
Goal: Prove bounded first/deep pages at both frozen scales.
Dependencies: CK-08R3A accepted, merged, and exact-main verified. CK-08R3A itself depends on accepted CK-08R0.
Owned files/interfaces: Evidence tests/workload/artifact; never production.
Produces: Evidence-scale v1: SQL, EXPLAIN, rows/bytes/RSS, five p95 samples.
Independent truth source: Typed selector/seven-part-order synthetic oracle.
Consumer seam: Actual EvidenceService, one query-only snapshot; consume CK-08R3A's exact blocker identity, which ran no scale/admission.
Parallelism: Read-only after CK-08R3A; disjoint from Wave 2.
Non-goals: Production fix, projection/backbone/API, index, or new budget.
Invariants: Typed provenance, stable late/replacement events, no gaps/duplicates, <=100 rows, <=16,384 bytes, exact count off.
Required tests/checks: Every view/scope/direction, ties/late/replacement/
byte truncation, 100,000 and 1,316,864-call fixtures, first/deep plans and
budgets at both scales, just v/vc.
Acceptance: All preceding invariants/tests pass at both scales.
Accepted evidence: The committed
docs/decisions/evidence/ck08r3/evidence-scale-qualification.json records
first_failure=null for the synthetic 100,000-call and 1,316,864-call
profiles. Each profile covers 14 selector scopes, seven views, two directions,
first/deep pages, typed seven-part truth, one late event, byte truncation,
cursor tamper/replacement rejection, query-only execution, and bounded
physical plans. PR #425 passed hosted Console and Python 3.10/3.14 before
squash merge and exact-main verification.
Failure/rollback: Retain first failure; stop scale/admission and route
physical defects separately. Never weaken gates, create R4, or invent
evidence_timeline_current.
Handoff: Evidence digest and classification input.
Cleanup/docs: CK-08R4 links the result.
Suggested commit: test: qualify evidence service scale