Add PPi-dependent non-oxidative pentose phosphate pathway

[realmarcin]

Add MetaCyc:PWY-8178, the PPi-dependent non-oxidative pentose phosphate pathway, with six reversible core reactions, 20 chemical-side observations, 51 mechanistic edges, 14 endpoint qualifiers, and 53 references. The generated pathway page includes its network graph and complete edge-local evidence; indexes and KGX downloads are regenerated. Assay, construct, strain and source limitations remain explicit.

The record cites a thesis using hdl:1993/32279. Add a bounded Handle grammar to reference validation in Python and LinkML, with coverage for malformed references, undeclared evidence, biological-node rejection, identifier collectors, rendered citations and exported metadata.

Adopt the actual native authority-build outputs and retain the independent scientific review, five corrected findings, and source provenance. The integration guard also binds source-cache file types, literal symlink text and resolved destinations.

Persist the complete native scientific review with explicit independent review inheritance and current integration evidence. Native validation, append-only save and review checks passed (108 valid review bundles).

Validation:

• All nine maintained checks passed: rendering, skill validation, full QC, tests, lint, KGX export, vendored sync, whitespace and page-current checks.
• Pytest: 2,462 passed, 3 skipped, 3 warnings.
• All 51 graph edges, 14 qualifiers, 120 evidence uses and 53 owning bibliography entries match the record.
• All 231 existing records/pages, 107 structured review bundles, previous history, 11,536 edge rows and 7,668 source contexts are preserved. Existing cache bodies and link identities remain exact.
• The final inventory guard detected one generated bytecode file. Its exact source identity was verified, its bytes retained in the evidence archive, and that single file removed. Strict guards passed afterward; the original failure and all original successful command receipts are retained.

Closes #455
Closes #456
Closes #457
Closes #458
Closes #459
Closes #463

[realmarcin]

Final adversarial review of PPi PR #470

Verdict: pass with stated review limits. No actionable findings.

PR: #470

Reviewed committed head: b9a07e28ed5c88c10cde5f6e84608feeae0d1c0f

Reviewed tree: d0712923d0f66d4c08b25b7fa0ed5adc9dbdc7fe

Parent/base: a33f81ebed11cc61880aa645e4d5b73bb8480435

The initial complete staged-tree capture and final committed tree are identical. All 785 changed blobs were compared to actual file bytes: 771 additions and 14 modifications, comprising 764 dossier files and 21 record, authority, schema/test, history, generated-output and review files. All are ordinary non-executable Git file modes. There are no deleted files or unrelated changed paths. Final HEAD, tree, parent, clean tracked worktree and committed whitespace checks passed. A direct read-only GitHub query independently confirmed PR #470 is open on this exact head, against main, with the expected branch, description and issue links. This is a review report; no GitHub approval, comment, issue mutation or merge was performed by this reviewer.

Runtime and namespace boundary

The complete maintained runtime/test diff is exactly four files: src/pathwaymech/schema.py, src/pathwaymech/schema/pathwaymech.yaml, tests/test_handle_references.py, and tests/test_linkml_schema.py. The Python Handle branch exists only in reference-ID validation. Its full-match grammar and LinkML Reference.id pattern permit the documented numeric dot-separated prefix and ASCII alphanumeric-leading suffix, with only dots, underscores and hyphens thereafter. Trailing whitespace/newlines, URL masquerading, fragments/query strings, extra path segments, case changes and unsupported namespaces remain outside that grammar. The global biological whitelist and all biological LinkML ID slots remain unchanged.

The 68 new focused tests cover valid/invalid reference syntax, every pre-existing reference prefix, undeclared evidence, record/taxon/participant/reaction rejection, edge endpoint resolution, closed authority namespaces/snapshots, unchanged source/ontology collector scope and page/KGX metadata. The former schema-prefix parity test is narrowed to biological classes while the new reference matrix separately proves reference parity. I read the entire four-file diff. I also read the separately independent Handle report: it records 147 focused tests, 349 syntax cases, 365 checks and frozen-record replay, with its historical 229-record scope and audit-only count correction retained. This final review does not relabel that historical scope as a fresh 232-record run.

The actual maintained nine-command gate run covered the adopted unchanged runtime and reports 2,462 passing tests, three skipped and three warnings. That broad suite was not rerun here. Fresh final-review execution did validate the exact adopted record, parse/validate the actual native review pair, and produce a pure in-memory target KGX export for exact comparison. No schema regression or namespace leakage was found.

Exact scientific target and corrected findings

The record is byte-identical to the independently accepted scientific successor: 20541f677f9228b670aac99a94e4d19ef89b5c20e0f6fdbb288d18ec7a9d98f5. It contains six explicitly reversible core reactions, 20 complete chemical-side observations, 51 mechanistic edges, 14 endpoint qualifiers, 53 references and 120 total edge/qualifier evidence observations.

The full description, reference metadata, graph structure and accepted successor dispositions were reviewed for integration hazards. The route remains PPi-dependent and distinct from ATP-dependent S7P phosphorylation, conventional transaldolase routing and Calvin condensation. Generic native compound classes and incompatible protonation/cross-reference identities remain explicit, without invented exact mappings. Activity-level Entamoeba observations are not assigned to modern sequences; DSM 1313, ATCC 27405 and thermosuccinogenes preparations remain distinct. Rabbit coupling does not establish native clostridial SBP cleavage. Rex remains unsigned sequence-specific DNA binding, and paired transketolase genetics do not claim independent complete subunit catalysis or a measured assembly.

All five accepted corrections remain in the exact current target:

• PPi pathway: Correct the CDQ83_07225 coupling-enzyme preparation and its evidence #455: lyophilized rabbit auxiliary enzymes and separate PFK ultrafiltration; ammonium evidence retains the ATP/F6P comparator and source formula uncertainty.
• PPi pathway: Separate the Pfp mixed-substrate activity assay from the MS identity experiment #456: the mixed-substrate FBP activity assay is distinguished from S7P-only MS, with bounded competition interpretation.
• PPi pathway: Use the actual titles of the 1982 and 2022 primary articles #457: corrected 1982 and 2022 reference titles and unchanged reference identities/source metadata.
• PPi pathway: Carry the six known reversible directions into reaction metadata #458: all six reaction directions remain reversible and are exposed in native output metadata; layout-oriented direction is not substituted for reversibility.
• PPi pathway: Bind construct features to the exact tested enzyme and nearest primary passage #459: direct construct/preparation evidence, explicit unresolved Pfp linker/codon optimization and separate F6P versus S7P assay scope.

I inherit the accepted complete original scientific review and its independently accepted 31-change successor. Exact hashes and complete assessment transport are checked here. This is not a fresh full literature review, a new source-exhaustion claim, or independent authorship of earlier judgments.

Authority provenance and retained originals

All four adopted authority artifacts are byte-exact native build outputs: source authority, ontology authority, source manifest and Rhea updates. This independently confirms that maintained files use the actual native serialization rather than replacement proposal serialization. Full native execution and four full SQL-original hashes are inherited from the accepted canonical-build/audit records; no full builder or SQL rehash ran in this review.

Every prior authority object, nested object order and relative old-key order was compared against actual merged-base Git blobs. Source authority grows from 2,474 to 2,507 terms and from 595 to 617 metadata objects; ontology grows from 1,963 to 1,969 terms while retaining its five metadata objects. Every prior object remains exact. The 758-entry source manifest remains the exact ordered prefix of the current 782 entries; 24 whole-original definitions are appended. Rhea updates remain byte-identical to the merged base. Native first-source precedence and prior PPI/proton objects remain retained; unused later candidate metadata is not substituted for prior authority identity. Unnamed source reaction IDs remain truthful source-context labels.

A fresh read-only invocation of the previously accepted source-cache guard verified all 780 selected original bindings, the complete hidden/ignored cache inventory, all 765 prior cache bodies and path identities including 90 symlinks, and 24 exact new regular copies. This confirms the operative #463 correction: file type, literal symlink text and resolved destination remain guarded alongside bytes. That prior guard/audit code is explicitly inherited, not presented as a new independent review of my own earlier audit work. No source cache was written.

Archive closure and genuine failures

All 758 rows across the three actual archive maps were independently verified against their bound source bytes. The original adoption map has 476 mappings (437 exact files, 38 lossless UTF-8 wrappers and one prepared provenance README). Final-gate archival has 229 mappings (227 exact, one lossless JSON text wrapper and one deterministic gzip retained-bytecode artifact). Native finalization has 53 exact mappings. The maps plus three explicit actual receipt/README files account for every one of the 764 changed dossier files. There are no silent dossier additions, source substitutions, executable Python files or symlink entries in those mapped artifacts.

The original scientific findings, author/probe failures, initial integration identity defect, Handle audit count correction, preparation lint/schema failures and original final inventory failure remain immutable historical evidence. Their earlier pending statements are not rewritten as later success. The finalization README distinguishes later actual execution from historical preparation.

The original nine maintained commands all passed, but their first final inventory closure correctly rejected one newly generated history-helper bytecode file. The separately accepted bounded recovery retained its bytes, proved the code/header matched unchanged source, removed only that path and refreshed strict guards without rerunning commands or waiving inputs. The bytecode writer remains unknown. This final review explicitly inherits the prior sealed independent recovery/support audit, then verifies the actual retained archive and subsequent saved-review/final-preservation results. It does not claim a fresh independent audit of that earlier auditor's own code.

Native review and history provenance

Actual native validate, append-only save and check receipts all returned zero. The saved YAML and Markdown match their receipt hashes, and the unchanged native bundle parser validates their semantic/display pairing. All 146 prepared assessments survive exactly, followed by two actual root assessments, for 148 total. All 1,129 declared source-input hashes and every linked evidence snapshot were checked against current bytes. The exact root read receipt precedes the final reviewed timestamp and binds the full assembled content.

The review truthfully declares integrator self_review, independent scientific inheritance, pass_with_limitations and a working-tree attestation at the real merged ATP base. It does not pretend those working-tree bytes were already committed at that base. This final adversarial report supplies the later exact committed-tree binding. Native findings/actions are empty, while external dispositions for issues 455–459 and 463 remain evidence-linked assessments; no nonexistent native predecessor or already-closed issue is invented.

The historical failed final guard remains an optional failed check, followed by the distinct successful bounded recovery. The native review does not assert future PR/CI/merge success. Two actual CREATE/schema EDIT histories precede rendering; neither review persistence nor final archival adds late history. Actual postarchive/staging checks report current pages for 232 pathways and 108 valid native review bundles. Root's final preservation receipt independently passed with 288 exact new additions and no old-file waiver, unexpected path, changed old history/review/record/page or cache defect.

Graph, page and KGX integrity

I independently compared every old KGX edge cell against the merged base: all 11,536 rows remain exact. Multiplicity-aware counters preserve all 7,668 old source-context occurrences. There are exactly 51 new edges, 50 new contexts and 39 new nodes. Every target edge, including qualifiers, evidence and reference metadata, matches a fresh native in-memory single-record export. Public downloads equal maintained KGX bytes.

The only changed old node display cells are the exact nine separately approved category unions across six chemical IDs; there are no label, identity or direction changes hidden in the context additions. Every new context belongs to this target. The Handle stays a citation with its exact issuing URL/title/source metadata and produces no biological node. Six reaction nodes retain their reversible source contexts.

The actual page has unique IDs and resolving internal fragment links. Its bound owner-specific QA verifies all 51 edges, 14 qualifiers, 120 evidence observations and 53 bibliography entries in their owning rows. I visually inspected the bound static graph overview: the reaction/chemical columns and experimental-construct shapes are visible, label blocks stay inside their shapes, and no obvious clipping occurs. The large left edge bundle is dense and the whole overview is tall; the detailed tables retain the precise evidence. No interactive browser, keyboard, focus or hover test is claimed. Existing graph implementation is unchanged in this PR.

Execution evidence and limits

Fresh final-review integration audit: actual 9335cb/0, session 60070. It emitted only the existing native topClass deprecation warning. Fresh inherited cache guard: 2ee233/0, session 77188. Final exact commit binding: c3e865/0, session 15866. Successful independent direct GitHub capture: d6490a/0.

Two intervening read-only network capture attempts failed to connect to api.github.com; their original helper/stdout/stderr and explicit observation record are retained. A later direct query succeeded on the exact expected head. These are acquisition failures, not failed product tests or fabricated remote success.

No maintained file, shared cache, Git index/object/ref, or GitHub object was mutated by this review. All audit artifacts are in the new external review directory. Full broad tests, full source builds, fresh SQL hashing and fresh full scientific rereview were deliberately not repeated after accepted exact checks. CI and merge readiness remain root-owned external checks; this report does not claim their future outcome. Any head/tree change invalidates this exact-head verdict until reviewed.

Root has read and accepted this report and verified the closed 60-artifact audit packet. Handoff SHA-256: d08da1cc9baa818a9551efc498201941aca86569d5491fe7425930f0e7e90cb7. This comment records an independent agent review; it is not a GitHub approving review by a separate human account.