RFC-0001 RDF claim-model proposal
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.
Status: complete batch owner-approved, implemented, and verified locally.
Owner: spec projection and ontology maintainers.
The backend-only wording, conditional suggested fix, module cardinality, and two further authored-clause omissions have been corrected through their maintained sources. Those routine alignments select no new contract. The subsequent owner-approved batch resolves nine claim-subject scopes, one permission kind, and three module-type value identities. The current ledger has 6,959 reviewed nodes, 2,405 mismatches, no unreviewed nodes, and no remaining RFC-0001 projection mismatches. Earlier sections preserve the historical source audit and measured alternatives; their row counts are historical observations.
Approved batch and local adoption (2026-10-14)
The owner approved the complete batch in reply call359cf486. The
exact reviewed seven-file patch (repository access required)
combines the existing claim-model and module type value
proposals with the consumer-query adoption they require. It was applied with
SHA-256 e54815060cba87c1de66ebfd96cc2e90226ac13d399307b305c756e46f7c20ba
at source base c0abe3ed640b1fa0258b933d0f27848d671bb276, preserving every claim
IRI and the authored language/diagnostic contracts.
Paths in the following table are relative to https://gnomon.sh/domain/.
All nine changes replace only claimSubject:
| Claim suffix | Former subject | Approved subject |
|---|---|---|
sec-4/agent-model-must-resolve | field/module/agents | Existing declaration/agent |
sec-4/ai-actor-requires-agent | field/module/agents | Existing declaration/action |
sec-4/empty-list-warning | field/module/objects | New authoring/module-explicit-file-lists |
sec-5/path-must-resolve | field/module/objects | New authoring/module-explicit-file-lists |
sec-4/explicit-listing-naming-recommendation | field/module/objects | New authoring/explicit-listing-files |
sec-4/forbidden-generic-filenames | field/module/objects | New authoring/explicit-listing-files |
sec-4/rollback-hints-q2-vs-phase1 | field/operations-entry/rollbackCheckpoint | New authoring/rollback-hints |
sec-4/unrecognized-file-warning | declaration/file/module.gnm | New authoring/concept-folder-files |
sec-9/diagnostic-allocation | declaration/module | New diagnostic-namespace/RFC-0001 |
The five new handles are typed only as the existing DomainEntity: they
describe authoring or allocation scopes, not new DSL declarations or fictitious
DSL FQNs. They retain RFC/section provenance and source-backed scope summaries;
the file-list and rollback groups reference their existing constituent fields.
The diagnostic handle attributes the existing allocation without deciding the
separate exact-code registry and legacy-exception conflict.
imports-may-be-empty changes from DefaultValue/defaults-to/empty or omitted
to Permission/may-be-empty-or-omitted/when no upstream dependencies exist,
adding the approved scoped-MAY Permission claim kind. Its subject and text
remain stable.
The three declaration/module-manifest/{backend-only,hybrid,ui-only} nodes move
to value/module.type/{backend-only,hybrid,ui-only}, become FieldChoice, and
link through choiceForField to the existing field/module/type. Four claim
references move with them. Existing choice FQNs, summaries, and provenance
remain stable. The ontology and JSON-LD context additions are in the same patch.
Concrete consumer consequences
The initial five-file identity counterfactual showed a consumer assumption:
orphan-claims.rq calls any subject without a DSL FQN an orphan, even when it
has an emitted domain identity. The approved batch therefore also adopts:
- Classify orphan subjects by the absence of a declared type in the
DomainEntityhierarchy, rather than absence offqn. - Change
exclusions.rq’s public output column fromsubjectFqntosubject, with its FQN when present and exact subject IRI otherwise. Consumers of that column name must adopt the new result contract.
The reviewed counterfactual (repository access required) records every approved triple and query row change. The actual adoption record (repository access required) matches that exact delta and all 38 reviewed query comparisons. It replaces 28 triples and adds 71, growing the graph from 89,301 to 89,344 triples. Six of 38 query result multisets change; 32 remain identical:
| Query | Measured change |
|---|---|
claims-by-kind | Default Value decreases 167→166; Permission gains one. |
cross-rfc-claims | Agent and action each gain the correctly scoped claim; module loses the diagnostic-allocation claim. |
exclusions | Thirty rows change through FQN-or-IRI display, including existing formerly unbound subjects; the output column also changes explicitly. |
orphan-claims | Thirty-nine rows change after correcting existing false-orphan classification and the 50-row page. Displaced rows are not deleted RDF claims. |
predicate-vocabulary | defaults-to decreases 90→89; the new permission predicate lies below the top-40 page. |
entity-by-type | NamedDeclaration decreases 235→232; FieldChoice gains three; direct DomainEntity count increases 3→8. |
These are local catalog measurements. They do not establish acceptance by dynamic or future consumers. Nothing is deployed to migrate. The framework’s duplicate spec and released artifact consumers still require cutover. The compact review patch was verified to produce exactly the same seven file payloads as the measured candidate; only diff context was shortened. The original candidate and measurements remain retained in the evidence record.
Recorded decision and observed acceptance
The owner accepted the complete precise-scope/value/permission model and both explicit query changes together. The approved source patch was applied and projections regenerated. Seventeen existing node hashes were refreshed, thirteen existing mismatch verdicts corrected, three value review IRIs moved, and five new source-bound matching reviews added. Every unrelated finding and verdict, including the four already-matching choice-reference claims, was preserved.
Projection drift/review, RFC-0001 domain ownership, emitted RDF links, claim-query column contracts, domain-link inventory, proof-review integrity, and provenBy-review integrity passed. The runtime and baseline checks (repository access required) also ran the existing live-corpus and malformed-document Jena SHACL fixtures successfully with a bound 24-jar cached Jena 5.0.0 classpath and JDK 21.
The strict whole-corpus domain-identity gate remains red on 18 FQN and 21 kind-class conflicts. Its entire finding set is identical to the source base; this batch introduces none of those conflicts. The separate 715 provenBy mismatches and formal-model/release gaps remain. Zero RFC-0001 projection mismatches is a local source-review result, not full RFC implementation or cross-RFC taxonomy conformance.
Downstream released-artifact adoption, duplicate removal, final clean package generation, publication, and dependency-pin changes remain separate steps. The approval authorizes this implemented RDF/ontology/query batch only.
Earlier source contract and proposed triples
RFC-0001 §4 says an empty explicit objects, operations, or agents list
emits M01065. The current empty-list-warning claim has the correct text but
uses only field/module/objects as claimSubject. Change that one object IRI
to the emitted declaration/module node, which owns all three lists. Keep the
claim IRI, text, kind, predicate, value, and section link. Every current claim
has one subject; assigning three subjects here would introduce a new
cardinality pattern and duplicate some join results.
RFC-0001 §4 permits imports to be empty or omitted when there are no upstream
dependencies. It does not define a default value. The current
imports-may-be-empty claim is DefaultValue with predicate defaults-to
and value empty or omitted. Introduce ontology claim kind Permission for
normative MAY statements, then use:
| Property | Current | Proposed |
|---|---|---|
claimKind | DefaultValue | Permission |
claimPredicate | defaults-to | may-be-empty-or-omitted |
claimValue | empty or omitted | when no upstream dependencies exist |
The claim IRI, claimSubject (field/module/imports), exact source text, and
section link stay stable. Define Permission in the ontology as a ClaimKind
with a label and a scope-aware description. Using Exception would imply a
general obligation to provide imports that RFC-0001 does not impose; using
Behavior would turn permission into observed runtime behavior.
Three further §4 reviews found declaration-level subject and value errors. Keep each claim IRI stable and propose these source-aligned replacements:
| Claim | Property | Current | Proposed |
|---|---|---|---|
agent-model-must-resolve | claimSubject | domain/field/module/agents | Emitted domain/declaration/agent. The rule concerns each agent’s model reference, not the module file list. |
ai-actor-requires-agent | claimSubject | domain/field/module/agents | Emitted domain/declaration/action. M01054 is triggered by an action naming an AI actor without a resolvable agent. |
backend-only-no-views | claimText | Forbids view and route | Forbids composition and view, matching authored §4 and Lean M01002. |
backend-only-no-views | claimValue | view or route declarations | composition or view declarations. |
The view/route pair is the separate ui-only M01003 presence rule. These
changes correct four consumer-visible triples and do not alter section links,
claim IRIs, or node counts.
Two additional claims need a subject that covers files across explicit-listing
sections. explicit-listing-naming-recommendation names both Schema and
Operations files, and forbidden-generic-filenames applies M08703 to every
explicit-listing file. Both currently point at field/module/objects, which
narrows them to the objects list. The existing declaration/module node is a
possible broad subject. A dedicated explicit-listing file domain identity
would state the scope more precisely, but would add a public RDF node and
ontology relation. The owner should choose the identity after reviewing known
consumers; keep the claim IRIs, exact source text, and section links stable.
Measured local impact
An in-memory RDF counterfactual against the current generated catalog removes
four triples and adds seven: one subject replacement, three claim-property
replacements, and three ontology facts for Permission. It keeps the claim
IRIs and all typed claim counts unchanged. Distinct result rows from all 16
checked-in claim queries were compared. Thirteen result sets were unchanged:
| Query | Current row | Counterfactual row |
|---|---|---|
claims-by-kind.rq | Default Value: 167 | Default Value: 166; new Permission: 1 |
cross-rfc-claims.rq | module: 4, 36 | module: 4, 37 |
predicate-vocabulary.rq | defaults-to: 90 | defaults-to: 89 |
The new predicate falls outside the query’s top-40 limit, so its absence from that result is not evidence that it is absent from the graph. The comparison does not cover external queries, stored RDF results, or downstream ontology consumers. The class addition and subject change are public RDF contract changes, so external consumer review is required before publication.
The additional four-triple counterfactual was run against the same 16
checked-in claim queries. Fourteen result sets were unchanged. In
cross-rfc-claims.rq, the distinct agent row changed from (6, 29) to
(7, 30) and action from (5, 11) to (6, 12); in exclusions.rq, the
backend-only row’s object changed from view or route declarations to
composition or view declarations. The query comparison does not establish
that external RDF clients or persisted results can accept these changes.
For the two explicit-listing claims, an in-memory replacement of their
subjects with declaration/module changed three of 16 query result sets.
cross-rfc-claims.rq changed the module row’s claim count from 36 to 38;
exclusions.rq moved one row from module.objects to module;
implicit-conflicts.rq added a heuristic row pairing M08703 with an unrelated
RFC-0008 source-format claim. The broad subject causes that apparent conflict,
which is a reason to evaluate a dedicated identity before choosing the final
triples. External query behavior remains unmeasured.
The §5 path-must-resolve claim has the same subject-scope issue in a different
rule: M01064 covers paths in objects, operations, and agents, but its
subject is field/module/objects. Consider a domain identity for the module’s
explicit file-list path entries, shared by all three lists, and link that node
to each list field. Reusing the broad module declaration would avoid a new
identity but may pair this path rule with unrelated module rules in heuristic
queries. Preserve the claim IRI and section link; measure query behavior and
review external consumers before changing the subject triple.
The §9 diagnostic-allocation claim also needs a subject correction. Its text
and value assign M01000–M01999 to RFC-0001, while its subject is the module
declaration. A stable RFC or diagnostic-namespace identity should own this
claim. Coordinate the choice with DIAGNOSTIC_NAMESPACE_PROPOSAL.md; preserve
the claim IRI and section link, then measure local query and external consumer
effects before changing the RDF source.
Two final §4 subject corrections are also proposed. The Q2-to-Phase-1
rollback-hints-q2-vs-phase1 claim covers both checkpoint and dependency hints
but points only to rollbackCheckpoint; a shared rollback-hint domain identity
would carry the rule. unrecognized-file-warning points to module.gnm, while
M07902 applies to unrecognized files in a concept folder; a concept-folder or
concept-file identity is needed. Preserve both claim IRIs and section links,
then measure query effects and external consumers before changing either
subject triple.
Acceptance
After the owner accepts each mapping, edit the affected protobuf claim nodes and ontology source where needed, regenerate the RFC-0001 Turtle projection, and update their source-bound reviews. Run projection byte-drift, link, SHACL, and all 16 claim query gates. Check the site against a pinned spec revision and compare any known external or persisted RDF consumers before release. Keep the current nodes marked mismatched until the source and generated RDF agree.