gnomonlearn
Docs/RFC-0029 conditional batch scope and reference hold

RFC-0029 conditional batch scope and reference hold

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.

The owner approved the prepared conditional-batch patch on 2026-10-16. It was applied once at clean Spec baseline d97c9084f6a48fede98fc7bd2417d7c0310e64cb.

The maintained-source change is exactly one literal in each of the RFC-0029 claim source, domain source, and Lean invariant catalog. Each now begins “If batch support is implemented”. Authored §8.6 already makes support SHOULD and conditions the ten-procedure limit and automatic splitting on implemented batch support. No identity, diagnostic, numeric limit, transport profile, or batch-support obligation changed.

Verified available work

CheckActual result
Claim/domain owning emissionTwo literal replacements: two removed and two added RDF triples
Combined RDF graph89,351 triples before and after
Projection reviewsAll 105 RFC-0029 Lean fingerprints revalidated; only the batch claim changes verdict
Complete review ledger6,962 reviewed; 2,307 mismatches; zero unreviewed
Cache holdsAll three remain mismatches, with their prior findings and node bytes preserved
Diagnostic inventoryOnly RFC-0029’s whole-source hash changes; all 742 definitions remain unchanged
Diagnostic coverageUnchanged: 654 authored tokens, 742 definitions, 545 shared codes
Displayed query comparisons37 unchanged; governance/invariants-by-rfc.rq replaces exactly the approved summary in its 637-row result
Complete relations without outer LIMITAll fourteen unchanged

verify_adoption.py (repository access required) proves the exact three-string source delta against the baseline and unchanged authored prose. It preserves every unrelated review message and textual block. The only other review fields changed are the two affected node hashes/findings, the 105 Lean hashes, and the approved batch claim’s verdict. This revalidation uses the unchanged semantic inputs outside the exact approved delta; it is not a new proof or implementation conformance audit.

The typed query effects (repository access required) use the existing query-effects.proto (repository access required) schema. They bind the actual graph, query, source, review, and comparison inputs, compare all 38 SELECT result multisets and columns, and restore the baseline graph in memory. The comparison is bounded to 300 seconds of wall time, 240 seconds of CPU, 250,000 rows per relation, and an 8 MiB effect file.

The normal owning Python tools returned zero:

These commands use the retained Python 3.14.7 runtime and existing protoc from spec-site-approved-projections-preview-links-v1b-20261015. No tool installation or acquisition ran. The final comparison returned zero in 89222a (launch b6c583, session 77467). Generation returned zero in ef96e0; the existing checks returned zero in bc93c7.

Explicit generated-reference hold

tools/check_lean_corpus.py still requires external framework commit 3725bd9f6d2711aa9bb179deea84f51e4424382a. That pin has the documented consistency defects in the formal-model proposal. Its corrected upstream release and Gnomon pin adoption are pending. The all-55 native Lean/catalog pipeline was not launched for this change.

Consequently, the checked-in reference/rfc-0029.md:58 still contains the old unconditional invariant summary. Its invariant ordering is owned by the Lean Markdown emitter and must also be regenerated through that tool. The retained generated/turtle corpus and generation evidence are not a fresh emission of this edited Lean source. They were used only as the existing formal RDF context for the link and query checks above. No generated Lean output was edited by hand, and no new preview package or consumer pin was published. Full generated-reference coherence is an open gate.

The three cache holds concern pre-extraction invalidation, a canonical hash preimage, and a complete cached result/lock contract. They remain owned by RFC-0024/RFC-0029 and the existing cache/lock proposals. Runtime batching fixtures, Lean proof, Java consumers, Site rebuild, packaging, published dependency integration, and normative release acceptance are not claimed.