Adopted RFC-0003 literal source alignment
Versioned draft snapshot
RFCs, reference catalogs, and evidence reflect Spec 3b2e3e7a114fb826b1445d39a48756b8fff629cd. Counts and pending/status notes belong to that snapshot. Read the dated project update for later progress.
Source repositories are private during staging. Links marked “repository access required” need access to the linked repository.
This is a trace and gap record from the pinned draft specification. It does not establish proof or implementation conformance.
This group implements two routine literal corrections at source base
0d6e70c541a531b5aae8f59104dfd059c1f99592. The authored RFC already fixes
both meanings; no new language, identity, ontology, or diagnostic contract is
selected. The owning protobuf source is kg/source-claims/RFC-0003.textproto;
tools/check_projections.py --write emits its claim Turtle.
Exact scope
Both claims retain their https://gnomon.sh/claim/RFC-0003/sec-5/ IRIs,
subjects, predicates, kinds, confidence, and emitted-section provenance.
| Claim | Literal correction | Existing source |
|---|---|---|
sensitive-field-warning | claimText adds by an active policy after marked Sensitive. | Authored §5a constraint 3, narrative line 273. Lean’s M03223 diagnostic, lines 81–83, retains the warning and runtime visibility requirement; it does not define a broader trigger. |
widget-stable-identity | claimValue changes stable identity id and typed prop references to stable identity id. | Authored §5 item 4, narrative line 235, lists identity IDs and typed prop references separately. The unchanged claim text retains both requirements. Lean b19 and its widget-canonical-identity-lookup invariant use identity for lookup; the Widget term retains typed props. |
The second correction does not define the format of a widget ID. The first does not claim that compiler or runtime enforcement is implemented.
Execution and retained failures
The scripts are records of this local execution. Do not rerun the mutating
align.py against an adopted tree: it requires the exact pre-correction base
and an absent actual directory.
The first controller preflight returned exit 1 in tool chunk d2d106 while
reading a nonexistent lean/proof-reviews.textproto path. No maintained source,
generated output, review, child gate, or actual evidence directory had changed.
The path was corrected to the owning lean/semantic-reviews.textproto ledger.
The original failed driver SHA-256 was
0e33be79b3ce7c7c83671a86451d86401c63ae15f0c7e706f1e9273ec75d9030;
this is a derivative account of the returned error, not an original process log.
The corrected controller ran these six child commands once, all with actual exit 0 and PID equal to PGID:
tools/check_projections.py --writetools/check_projections.pytools/check_projection_reviews.pytools/check_claim_domain_links.py --rfc RFC-0003tools/check_kg_links.py --turtle-dir generated/turtletools/check_claim_queries.py --turtle-dir generated/turtle
The fourth command is an inventory, not its strict release gate: it reports
42 unresolved references to 15 domain identities. One of them is the existing
sensitive-field-warning subject https://gnomon.sh/domain/field/View/show.
Its restored condition agrees with the authored clause, but its whole-node
verdict remains mismatched until canonical identity ownership is resolved.
The widget claim’s existing subject has an authored owner.
The controller then returned exit 1 in tool chunk 893810: raw whole-graph
set comparison treated independently parsed blank-node identifiers as changed
terms. The untouched partial text record (repository access required),
partial binary record (repository access required), executed controller (repository access required),
and six original logs preserve that execution. The partial record is not a
successful completion receipt.
A whole-graph canonicalization attempt (repository access required) was
interrupted during rdflib’s color refinement (tool chunk 6bf0fe, terminal
exit 130 in 370994). The linked file is a later byte-identical copy of that
attempt’s source, not its original execution path. It produced no completion
receipt and repeated no child gate or maintained-source edit.
The next ownership-audit attempt (repository access required) returned exit 1
in 3f04ee because it asserted that both corrected claims had authored subject
owners. Its check exposed the missing View/show owner. The source copy is
retained as a later copy, not its original execution path. The Sensitive-field
review was restored to MISMATCHES_SOURCES and its finding now records both
the corrected condition and unresolved owner. The original partial receipt
predates this review correction and is not the final ledger evidence.
The completion verifier (repository access required) compares every bound input against the immutable base or unchanged current bytes. It proves that all RDF documents except the one changed claim file are byte-identical and parses that common input once, sharing the unchanged blank-node mapping across both graphs. It proves the changed claim file has no blank nodes, then compares the exact triple sets, all review changes, and all 38 actual query result multisets. It calls the existing review integrity checker against the final ledger and preserves every unrelated review. It does not repeat projection source mutations or child gates. Its schema imports the unchanged original alignment schema (repository access required) through completion.proto (repository access required).
The recorded Python runtime is the existing Python 3.14.7 leaf and the private preview’s protobuf/rdflib module path. This execution invokes no Roma, Cargo, Lean compilation, Java rebuild, full package build, or publication.
Verified actual result
Status: implemented and verified locally. The completion process exited 0 in
tool chunk f1ed98 (session 18136). The authoritative final records are
completion.textproto (repository access required) and
completion.pb (repository access required). They bind the untouched partial record,
executed sources, original command logs, immutable preimages, and final outputs.
- Exactly two RDF triples are removed and two added, one
claimTextand oneclaimValue; no IRI or predicate changes. - All 38 checked-in SELECT query result multisets and columns are unchanged.
- All 6,959 ledger rows retain their order and content except the two related node hashes and findings. Only the widget claim’s verdict changes to match; the Sensitive-field claim remains mismatched on its missing domain owner.
- The ledger has 2,404 mismatches globally, 53 in RFC-0003, and zero unreviewed.
- All domain projections, Lean sources, reference/formal catalogs, and both proof review ledgers retain their bound bytes. The inherited strict identity and provenBy findings are not cleared or rerun by this literal correction.
- The RFC-0003 ownership inventory remains 42 references to 15 identities.
The independent parses used for failure diagnosis produce 17,866 removed and 17,866 added triples involving parse-local blank-node labels. Those are not reported as a semantic delta: the final proof uses the shared mapping for the byte-identical common documents and proves the changed document has no blank nodes. The two actual claim-literal replacements are then the exact full-graph set difference under that explicit bijection.
Invocation from the spec checkout, with the private module path and tool-bin PATH recorded in the partial receipt:
/opt/homebrew/Cellar/python@3.14/3.14.7/Frameworks/Python.framework/Versions/3.14/bin/python3.14 build/evidence/rfc0003-source-alignment-20261014/finish_alignment.py
This local receipt does not establish external exact-text consumer agreement, parser conformance, a new formal proof, a strict normative release, a final clean package, or publication. The site, dependency pins, framework duplicate sources, and deployment state were not changed.