gnomonlearn
Docs/RFC-0049 projection algebra claim review

RFC-0049 projection algebra claim review

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: 128 of 128 claim nodes reviewed on 2026-10-05. Owners: Compiler, Graph Runtime, RDF ontology maintainers. No normative facts changed. Results: 89 source matches and 39 whole-node mismatches.

The declaration, mode, authorization, precedence, and lowering text agrees with the authored narrative and diagnostic catalog. Several claims reference projection/auth, projection/materialize, or projection/body Field identities that have no definition in any authored domain source. The authoring slots exist, but the RDF identities need an explicit field definition and canonical declaration owner; object references do not constitute definitions.

The source-bound ledger distinguishes these identity mismatches from complete source matches. A matching declaration claim does not establish implemented algebra or authorization enforcement. View materialization also depends on the pending RFC-0053 security correction; a mode declaration cannot establish safe native materialized-view access.

Acceptance: inventory every projection-field reference, prepare canonical Field definitions and type/owner relations, obtain owner approval, regenerate RDF and validate query consumers. Keep the algebra endpoint-closure and elevated-authorization findings from the domain review as separate contract obligations.

Query and mutation lowering

The next twenty claims review typed query inputs, pure reads, mutation guards, idempotency, provenance, template chains, transactions, and lineage lookup. Additional query/mutation field and invariant aliases lack authored domain definitions; text agreement does not resolve these links.

The caller-only query authorization claim also conflicts with §10.4: service and system projections evaluate sources with elevated authority and then must reapply caller filtering before exposing results. §11.2’s description must distinguish evaluation authority from exposure authority.

The idempotency source match is bounded by §14.2’s carrier state, parameters, temporal qualifiers, and principal. It is not a guarantee of identical results across different permission contexts. Transaction and lineage source matches still require runtime failure and nested-template conformance tests.

Composition, compatibility, and diagnostics

Twenty more nodes cover §14 algebraic identities, purity, authorization, §15 transport lowering and legacy compatibility, and the Lean diagnostic catalog. Five whole-node mismatches remain in this group, including missing internal identity definitions.

The universal empty-stable result assertion conflicts with §8.1: count and sum return zero, while avg, min, and max return null. Resolve the result contract together with the existing aggregate type-model gap. Closure permits composition on each groupBy partition; the enclosing map is an exit value.

The declared M49000–M49099 diagnostic range excludes catalog entries through M49140. An informal overflow description does not allocate those codes. The catalog confirms M49120 is a warning and M49130 is an error.

The KP32 completeness statement matches the authored status assertion; it provides no implementation evidence. Legacy traversal compatibility and idempotency lowering likewise require actual runtime conformance checks.

Definitions and source boundaries

Twenty more claims cover definitions, endpoint closure, materialization, and the first source signatures. Four whole-node mismatches are recorded:

  • Aggregation is called the single exit boundary, although materialization is explicitly a distinct exit boundary.
  • The scalar/tuple aggregation signature excludes groupBy’s typed map result.
  • The carrier-instance definition is attached to Subgraph rather than Carrier; an arbitrary subgraph is a subset of that universe.
  • byId accepts a gnomon ID but promises zero edges, leaving a statement ID’s result incompatible with endpoint closure or undefined.

Acceptance: distinguish the carrier universe from its subgraphs in structured subjects, define aggregate exit shapes including partition maps, and resolve byId’s object/statement behavior under the pending endpoint-closure proposal. Source matches for self-containment describe the structural endpoint invariant; they do not verify canonical serialization, hashing, or storage implementation.

Source, refinement, and composition operators

Another twenty-four claims cover pattern matching, temporal defaults, source filtering, lifting, refinement, and composition. Six whole-node mismatches are recorded. Unconditional tenant and caller-policy assertions omit the elevated service/system evaluation and audited cross-tenant exceptions. Selection and set difference omit the incident-edge rule necessary for endpoint closure; these remain tied to the pending endpoint-closure decision. Missing internal identity definitions are separately recorded in the ledger.

The explicit edges operator retains endpoints, and lifted sequences auto-apply closure. These source matches do not authorize adding a hidden endpoint. Acceptance requires testing effective evaluation scope, caller refiltering, revocation, and endpoint pruning together, rather than testing each operator only on fully authorized vertex-only inputs.

Temporal qualifiers, aggregation, and materialization

The final twenty-four nodes have nineteen source matches and five whole-node mismatches. Aggregation’s scalar/tuple restriction omits groupBy’s partition map. Empty aggregate null results conflict with the universal empty-or-zero rule and the null-free type model. Authorization redaction must prune incident edges when an endpoint is removed. Undefined internal identities remain a separate repair requirement.

The §9.1 canonical wire example permits name: text|null, while §9.2 and RFC-0034 §5.2 prohibit outbound null. Define absence consistently before publishing a wire schema or compatibility claim. Latest-time union also needs an explicit ordering for the two temporal axes and a rule for reconciling historical payloads; source agreement is not evidence of those runtime rules.

Completion scope

All 128 claims and all 68 domain nodes now have source-bound reviews. Combined results are 139 source matches and 57 mismatches. This completes the semantic inventory for RFC-0049, not its normative repair or implementation acceptance. Pending endpoint-closure and provenance proposals, aggregate result types, authorization boundaries, temporal normalization, wire absence semantics, identity definitions, and diagnostic allocation remain release obligations.