RP006 — actual pre-execution methods, graph and source review

Review date: 2026-10-12. Reviewed execution profile: one recorded attempt, two processes, no network, Python 3.11 or newer, 60 seconds and 128 MiB conservative process-tree RSS. The exact source commit, file hashes, Test and authorization are bound by EXECUTION_FREEZE.json and D-0016. This review concerns the frozen finite audit only.

Methods and interpretation

The frozen question is finite field-ablation sufficiency for FULL and TOPOLOGY_STATE over the six already known RP004 event logs G0, G1, G2, G3, G4 and C4. The three indivisible fields are complete state, complete parent sets, and birth operation (S/P/K). All eight uniform masks are used. The protocol fixes 216 FULL truth rows, 180 topology/state truth rows, 48 projections, 120 case-pair/mask comparisons and 16 family/mask verdicts. Equality uses canonical values with exact types; feasibility checks aligned complete answers only when projections are equal. Minimality is inclusion-minimality among these eight masks.

The six cases and their behavior have already appeared in RP004 and RP006 engineering preflight. The known classifications in RP006_IMPLEMENTATION_REVIEW.md are disclosed. The run is reproducible completion and replay of this finite audit, not blind confirmation, fresh evidence, a sampled population, a new theorem, or a novelty test. Inference stops at the declared corpus and mask class. It does not resolve Q-3 or Q-4, establish general representation necessity, or show sufficiency for unseen constructions.

The producer reconstructs truth from event logs. The verifier independently reconstructs truth, projections, pair comparisons, conflict witnesses, certificates and minimal masks. The prior-art review identifies functional-dependency and provenance-model overlap and limits the contribution to an explicit finite inquiry-relative certificate; it is not a systematic survey or external review. The exact RP004 fixture source is included in the execution hash inventory alongside DOMAIN.json, so the domain's source pointer is checked against the data actually reviewed.

Forward and backward graph trace

Forward: RP-006 remains PLANNED in Cycle 02 and tests only the scopes in Q-3 (which distinctions preserve specified inquiries) and Q-4 (which realized-history distinctions remain recoverable). Q-3 and Q-4 remain ACTIVE. The RP006 protocol, domain, entry plan, prior-art review, implementation review, supervisor review, producer, verifier, worker, recorded-attempt runner, resource helper, engineering preflight receipts, implementation/supervisor controls and this review are the complete source/method chain. SRC-FD2026 records the inspected functional-dependency handout; the prior-art review also cites W3C PROV-DM and PROV-O. D-0013 through D-0015 accept entry, software preparation and supervisor preparation only. D-0016 is the actual checkpoint for this exact Test and source freeze.

Backward: Q-3 and Q-4 trace to the selected baseline's foundational questions and remain open in the graph. RP-006 has no incoming scientific support relation and no prior Result. The six event logs trace to RP004 fixtures, with preserved provenance in RP004_FIXTURES.json, RP004 lineage/replay code, and the frozen RP006 domain. Earlier RP004 outcomes and the RP006 preflight are known disclosures, not independent observations. No Cycle 01 result, D-0015 preparation decision, or source publication is treated as execution evidence. The immutable cycle-02-open anchor, rp006-preparation-001 tag, and Cycle 01 close anchor are retained.

Host, resources and stop behavior

The current execution host is macOS 26.3 arm64 with Python 3.12.13 at /Users/omerasimozturk/.local/bin/python3.12. The default python3 resolves to Python 3.9 and is outside protocol; the recorded attempt must be invoked with the explicit 3.12.13 path. The earlier engineering preflight used Python 3.12.14 and reported 0.02634 seconds and 27,426,816 bytes for a one-process computation. That measurement does not establish the two-process recorded run's resource use. The current runner enforces mach_continuous_time, monotonic and civil-clock agreement, a 60-second whole-attempt limit and a conservative sum of parent/worker lifetime RSS peaks under 128 MiB. The worker blocks at each guard, imports no network or child-process facility, and uses an audit hook to deny sockets and descendants. A threshold breach, clock discrepancy, unsupported host/runtime, input or provenance mismatch, missing output, verifier disagreement or serialization failure leaves an INCOMPLETE record and failure evidence; the runner cannot overwrite or automatically retry an attempt.

The two-process profile is the pre-outcome scope revision accepted by D-0015. The source-bound Test fixes the same finite methods and output inventory; it does not add a second process as a scientific variable. D-0016 accepts the exact frozen source and one attempt only. It does not accept a Result or resolve any Cycle 02 obligation. A failure remains an incomplete attempt; any corrected code requires a new review, freeze, Test/checkpoint decision and unique authorization.

Checkpoint conclusion

The method and interpretation boundary, known-source overlap, full graph trace, execution host/runtime, resource measurements and limits, and fail-closed stop contract have been reviewed against the exact inventory in EXECUTION_FREEZE.json. D-0016 accepts this review, the fixed-input Test TST-RP006001, and authorization for one recorded attempt on the frozen source. No output is inspected or generated by this review.