gnomonlearn
Docs/RFC-0054 DDL claim review

RFC-0054 DDL 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: all 85 claim nodes have source-bound reviews as of 2026-10-05; 77 match and eight have unresolved source or contract mismatches. Owner: Compiler and Storage. No RDF facts changed.

Eleven claims agree with authored RFC-0054 and its Lean catalog: purpose and derived migration artifacts, migration-window diff policy, existing-deployment incremental behavior, historical reverse-mapper derivability, CI round-trip requirements, diagnostic range, and M54030/M54031/M54040/M54041 conditions. This review establishes agreement with the specification. It does not verify the historical 92.5% measurement, deployed compatibility, or a working DDL pipeline. No Gnomon deployment currently exists.

M54050 severity mismatch

The claim RFC-0054/sec-12/m54050-fingerprint-mismatch adds escalation to error after retirement. Lean’s diagnostic entry defines M54050 as Error without a retirement condition. Authored §11.3 allows a fingerprint diff through the migration window, then makes the CI assertion strict. It does not assign different diagnostic severities to those phases.

Recommend preserving the unconditional Error diagnostic and describing the migration-window exception as CI assertion policy. The owner must resolve this public diagnostic contract before changing the claim. Acceptance: align the claim and emitted Turtle, keep the IRI stable, bind its review to the corrected source, and test both window and post-retirement CI behavior when the pipeline implementation exists.

Second source-review group

Twelve further claims were reviewed: M54060/M54070/M54071/M54090, canonicalization, composite/domain emission, create/alter/drop diff rules, risk classification, and destructive flagging. Ten agree with the sources. Two need contract reconciliation:

  • bcrisk-classifies-entries describes every entry’s risk, but §3.2 only assigns explicit classes to CREATE and DROP; ALTER classification is absent.
  • drop-flagged-for-gate mirrors §3.2 and Lean’s blanket DROP flagging, while §5.2 says DROP INDEX and DROP VIEW are never destructive. A potential-operation report could include them, but mandatory acknowledgment and destructive denial must have a coherent scope before this can be considered a matching public guarantee.

No facts changed. Acceptance requires a complete operation-to-risk and destructive-gate mapping, agreed by the owner, propagated through prose, Lean and RDF, with tests covering safe drops, unsafe alters, and genuinely destructive drops in the planner.

SQL emission and stage review

Twenty further claims were reviewed. Nineteen match the authored category mapping, planning stages, dependency order, canonicalization, and fingerprint contract. These verdicts establish source agreement, not working codegen or runtime reproduction.

The per-module-transaction claim remains mismatched because §3.3 and Lean require module-wide transaction atomicity while §7.1 and Lean permit atomic per-file or per-statement transactions. A failed later file could leave earlier file transactions committed. A nested savepoint implementation could reconcile those units within a module transaction, but the current source does not state that model. Acceptance requires an explicit transaction boundary, consistent rollback semantics across these sections and Lean, and failure tests spanning multiple files before claiming module-wide atomicity.

Completed remaining review

The final 41 claims cover canonical SQL, destructive-operation categories, deprecation, bootstrap inputs and lockfile semantics, audit, rollback, and cross-module ordering. Thirty-seven agree with the sources; four remain mismatched:

  • destructive-default-deny: authored §5.1 permits two-phase deprecation as an independent bypass, but Lean evidence omits it and §10.3 requires explicit acknowledgment.
  • unacknowledged-destructive-emits-m54010: the blanket rejection claim does not reconcile that deprecation bypass.
  • destructive-on-user-facing-via-deprecation: §5.3 states the lifecycle as a requirement while §10.10 describes it as recommended.
  • apply-default-per-file: repeats the transaction-boundary conflict already recorded for the per-module claim.

The owner must decide whether deprecation eligibility replaces explicit acknowledgment and whether user-facing deprecation is mandatory. Transaction scope, complete risk classification, destructive DROP filtering, and M54050 severity also remain open. Source agreement does not establish operational cast safety, data recovery after destructive changes, historical derivability measurements, or a working DDL emitter.

The review gate now reports 4,911 of 6,954 corpus nodes reviewed, with 1,165 mismatches and 2,043 unreviewed. No RDF fact changed during this review.