RP004 — Package interpretation and disposition review

Date: 2026-10-11. Explicit internal post-outcome review of the finite Cycle 01 commitment. Package RP-004; Test TST-RP004001; Result R-RP004001. Decision D-0006 records the disposition. This review uses preserved evidence, collects no new research data and is not external peer review.

Review question and acceptance basis

Has the declared branching-lineage study delivered G0–G4 constructions, construction-derived truth, R0–R3 encodings, exact inquiries, complete verification and the separation of state equality from lineage relations? The frozen baseline and research program set that goal. PC01 requires an explicit subject, transformation and continuation criterion; PC02 makes representation sufficiency relative to a declared inquiry family. The pre-outcome supplement operationalizes six synthetic queries without introducing a metaphysical identity predicate.

Accept the specified finite commitment. This closes a six-construction reproducibility audit, not every continuation or realized-history question. Unknown answers and information-loss counterexamples are retained outcomes. Wider domains are not retrospectively added to the acceptance gate or silently treated as solved.

Deliverable-by-deliverable review

Required work	Preserved evidence and assessment
G0 linear succession	Three fresh entities form a chain. Strict ancestry, roots, immediate parents, birth kind and consumption follow the declared construction. Every specified query is saved and independently replayed.
G1 copy with original surviving	Two copies share the original's state and ancestry; copying does not consume a. Both copied outputs and the original remain live. The complete query audit preserves this distinction.
G2 symmetric fork	One operation creates two equal-state outputs and consumes the input. Neither output is privileged. G1/G2 have the same untyped parent DAG but different birth kinds and original-input availability.
G3 independent same-state duplicates	Three roots have equal payloads without shared roots or ancestry between distinct entities. Payload equality never creates a parent edge.
G4 merge	Two distinct live inputs become immediate parents of c and are consumed; their root sets combine. Neither input becomes an ancestor of the other. Contributions are structural, without inferred content weights or physical fusion.
Auxiliary C4 control	One-parent continuation is distinguished from G4 in complete topology. R1 drops b from the merge and gives the same projection for both constructions, exposing six conflicting query answers. This is an extra control, not a changed baseline case.
R0–R3 contracts	The common opaque-handle/state universe is explicit. R1 retains one lexicographically selected parent, R2 all parents, R3 parents plus birth kinds. Case labels, chronology, live lists and expected answers are excluded from decoder inputs.
Exact inquiry family and scoring	State equality, strict ancestry, shared roots, complete parents, birth kind and availability have fixed arities and answer types. UNKNOWN remains separate from False/empty sets and stays in every denominator. Non-UNKNOWN answers are all sound in the audited corpus.
Complete coverage and witnesses	All 216 construction/query answers, 864 decoder rows and 60 pair/projection comparisons independently replay. Every same-projection/different-truth witness is saved; order and complete inventory are checked, not inferred from a producer count.
Registration and provenance	Implementation freeze db7f8ee475e2509cd46db665a27fc00c41ffb7a9 precedes corpus outcomes. All 15 frozen input hashes and six output hashes match. The seven-file run, Test, Result and audit-stage report remain unchanged.
Execution and failure handling	One deterministic attempt, no audit network, Python 3.12.14, 0.085009708 monotonic seconds and 28.94 MiB peak RSS satisfy 60 seconds/128 MiB. No failed run, automatic retry or outcome-dependent protocol change occurred.
Prior art and interpretation	The bounded primary-source collision review acknowledges familiar provenance/DAG machinery and states its inspection limits. Human intuition is outside truth construction. No novelty clearance, statistical confirmation or metaphysical identity theory follows.

The preserved-audit and preparation checkers passed again for this review, without creating a new run. Repository validation also compares complete run bytes with their first committed snapshot. Passing checks are the integrity basis; the interpretation and disposition judgment is documented separately here.

All decoder and collision outcomes retained

Representation	Correct / total	Unknown	Conflicting identical-projection pairs / 15	Witness queries
R0	90 / 216	126	15	129
R1	141 / 216	75	2	9
R2	201 / 216	15	1	3
R3	216 / 216	0	0	0

The total is 648 correct and 216 UNKNOWN answers. There are 141 case-pair/representation/query witness records; repeated witnesses across representations are not independent discoveries. R1 intentionally abstains on some recoverable negatives, so these are policy-specific answer counts, not optimal capacity rankings. Every audited state is A: same_state has only positive answers in this corpus. Separate engineering controls test state changes, but are not added to these denominators.

R0 loses all non-self lineage distinctions represented by this corpus. R1's two collision pairs are G1/G2 (three operation/availability witnesses) and G4/C4 (six witnesses including the missing merge parent). R2's sole collision pair is G1/G2, with birth_kind(b), birth_kind(c) and alive(a) differing. R3 answers the declared family on all six constructions. Neither absence of a corpus collision nor a conservative decoder's abstention establishes a universal representation theorem.

Query-relative information requirements

Inquiry	Scoped assessment
State equality	Current states suffice for this exact comparison. Equality is not lineage evidence.
Strict ancestry	A complete parent DAG suffices; R0 G0/G3 and R1 G4/C4 show particular distinctions lost after projection. Equivalent reachability information could also answer the query.
Shared roots	Complete parent sets and root reachability suffice. A retained single-parent forest may lose a contributing root at merge.
Complete immediate parents	R2/R3 preserve them; G4/C4 show that a single pointer cannot recover the second parent without other information.
Birth operation	Untyped G1/G2 topology cannot distinguish copy from fork. Type labels or equivalent operation information are needed for those inquiries over a domain containing both cases.
Availability	Consumption matters: a survives copying but is consumed by fork; b survives C4 but is consumed by merge G4. Operation information suffices under this toy policy; an explicit availability record could serve this query instead.

This is not a proof that a full R3 graph is the weakest sufficient representation. Query-specific summaries may retain fewer distinctions. Event chronology, fork grouping, ownership and contribution weights are absent from R3 and outside the audited inquiry family. Closure accepts explicit requirements and counterexamples, not a universal need for full history.

Scientific limits and remaining questions

The corpus has six three-entity constructions, a common known universe, aligned opaque handles and equal states. It does not enumerate every admissible DAG. Parent selection in R1 is label-sensitive. Copy/fork consumption and two-input merges are declared synthetic rules. Calibration expectations were disclosed in advance; counts are deterministic checks, not empirical frequencies or confidence estimates. Prior-art inspection is targeted and external scientific review has not occurred.

Q-0 and Q-4 remain ACTIVE. Arbitrary graph sizes, changing states, multiway or repeated merges, richer lifecycle rules, chronology-sensitive inquiries, fork grouping, content contribution, uncertain provenance, minimal encodings and real-world continuation judgments remain open. Broader scope needs a new reviewed protocol. No broad Claim is promoted, no baseline definition is changed and no invented receiving Cycle is assigned unfinished work.

Package and Cycle disposition

Accept RP-004's declared Cycle 01 commitment as completed. Set RP-004 to CLOSED and resolve only the rp004 package-disposition obligation as completed under D-0006, with evidence links to this review, the audited report, scoped Result and decision. Retain the already resolved rp002a and rp003 entries and all 13 equal weights. The other 10 obligations stay unresolved; progress becomes 3/13 (23.1%) and Cycle 01 remains ACTIVE.

This review completes no checkpoint, graph review, whole-Cycle synthesis, Foundation Review, transition or close bundle. The audit-stage report retains its historical ACTIVE/pending-disposition and 2/13 statement; this review and D-0006 record the subsequent change. Test TST-RP004001, Result R-RP004001 and all run bytes remain unchanged.

Next work is RP005A protocol preparation: explicit graph/target construction, V0–V2 controls, planner horizon and tie-breaking rules, constraint-relative viability and coverage measures, resource and verification contracts. This review starts no RP005A experiment and asserts no restriction benefit. Its structural reachability inclusion and conditional interpretation lock remain intact.

Reopen RP-004 through a new recorded Decision if invalidation, an artifact or replay defect, unsound decoder answer, incorrect collision/coverage claim or material methodological interpretation error changes this assessment. Revisit its Cycle obligation at that time. Never overwrite the original run to repair the research record.

Evidence references

Source record

Repository path: research/RP004/RP004_PACKAGE_REVIEW.md

SHA-256: 1331e24156d1e58d4872c8980b12c4f25189855bcfa4b5cf901993bb837a4268

This reading view is generated from the source. The downloaded Markdown preserves the original bytes.