[humuhumu33]

Lands ten self-hosting slices of the compiler (#269, #274) together, with one regeneration and one CI gate, instead of ten.

What is in it

ID	Component	Source PR
BS-25	the LSE reader and printer (LexiconLse)	#445
BS-26	the LRE reader and the facts of a render (LexiconLre)	#446
BS-27	the notation of a tooling module (SurfaceTooling)	#452
BS-28	where the module of a source is (SurfaceModule)	#453
BS-29	the canonical diagnostic (Diagnostics)	#455
BS-30	the terminal output of diagnostics (CliOutput)	#456
BS-31	the canonical text of the lock file (LockText)	#458
BS-32	the hash recipes of content identity (ContentId)	#460
BS-33	the canonical project manifest and its digest (ProjectText)	#463
BS-34	the build manifest (ManifestText)	#464

It also carries the count writer of #442, which the regeneration needs to write the moved counts.

Not included: RB-24 (#419). It changes language/semantics-1.2.toml, which moves every example's lock, UOR Atlas included, so it gets its own regeneration.

How it was put together

• Base: release/v0.4 at 7eb816284, then Bootstrap count cases: report every moved count in one pass, and write them #442 and BS-27 (BS-27: the notation of a tooling module as language-1.2 source (#269) #452, already regenerated) merged in.
• Each other slice: its own diff (its merge-base to its head) applied in dependency order: 25, 26, 28, 29, 30, 31, 32, 33, 34.
• Shared files, resolved by kind:
  • Generated files (CONFORMANCE.md, ERRORS.md, the bootstrap lock) were taken from the base and rewritten afterwards.
  • JSON plans (vv-rows.json, intel-host-lanes.json, roots.json) were merged by content. bdd-parts.json was merged as text, because it holds duplicate keys on release that a parse would drop.
  • Register, tests and scenarios (model/ids.toml, tests/conformance.rs, bootstrap.feature) were rebuilt with each slice's block whole, in ID order.
  • SPEC prose and VERIFICATION records got each slice's additions inserted whole. §31 rows are in register order (RP-07).
• Register count: 493.
  • SM-38 (SM-38: declaration IDs and statement IDs #469) landed on release during the batch. It was merged in at b8f8cb3f4. Each batch slice's CHANGELOG entry gives the register size in landing order after SM-38's 483: BS-27 484, then BS-25 to BS-34 485 to 493.
  • SM-38 moved only build oracles. check-golden --write added the declarations files of the ten new modules (0e26b6914). No export, record or lock moved.
• Checked by script: every run of lines each slice added appears unbroken in the result.
• Regeneration: regenerate.yml (ci: regenerate.yml, the write recipes on runners #467), run 43112441927. That covers the export, records and oracles, then the count writer, the model documents, and the stack bound of each new recursive package.

Verified before this PR

• Locally on the merged tree: cargo fmt --check, cargo clippy --workspace --all-targets --all-features -D warnings, validate-model (493 IDs), validate-spec-links, and all 17 conformance_rp_* cases pass.
• The regeneration run itself: 151 min from push to patch, all five jobs passed.
  • Writers: export 72 min, records 82, oracles 15.
  • Counts: the nine count parts all exited 0 under LEXLEAN_COUNTS=fast. The writer moved every count for the 22 new packages (308 → 330). It flagged four SPEC numbers for a reader, and each stays as written (see VERIFICATION.md).
  • Stack bounds: it found the two new recursive packages, print_lse and build_manifest, and measured them by running BS-25 and BS-34 whole. Both passed, which also checked the new counts.
• After applying the patch: the same local gates pass again, and lexlean lock --check is current.

The full gate on this PR is the first time the ten cases run whole together.

🤖 Generated with Claude Code

[afflom]

Bounded integration review at707ed confirms actual5dcc ancestry and the493-row register with the ten BS-25–34 statement/scenario/test mirrors. The one-producer batch avoids ten competing regenerations. It still carries concrete findings already assigned to the existing owner:

1. counts.rs remains blob a194f608: unchanged sum explanations are excluded before ambiguity resolution, and prose stops at the first matching anchor. Both #442 findings remain; the already reviewed #462 repair d8ee566 is absent. Integrate that existing coordinated repair rather than create a second implementation. A successful count run does not establish that the planner handles ambiguous inputs.
2. BS-15 still promises3,190 in its authority/mirrors. Current rust_package.rs retains (349 manifests +182 rules +2000 mutants) +(330 accepted +137 render rules +330 render mutants)=3,328 compiled runs. Correct the authority and regenerate its mirrors through the official model writer; do not drop corpus cases to match stale prose. The BS-07 paragraph likewise says1,392 while specifying381 cases at each of four roots (=1,524).
3. The inherited BS-27 VERIFICATION record still calls the third package render_tooling_latex instead of declaration_spans and attributes all three to the bounded subsets. The newer batch table does not correct that earlier causal claim. Make the record precise while preserving what the retained evidence actually shows.

The four separately retained SPEC numbers have plausible independent meanings (fixed windows/draws and historical timing); they should not be blindly rewritten. I did not retrieve their original counts.txt payload, so this is not an exact flag-to-line audit. The formerly missing BS-27 packages are present.

Producer43112441927 genuinely passed5/5 at583e917, but its output adoption and later SM-38/golden reconciliation are distinct from final707 qualification. Retain the existing producer/adoption receipts; no blanket regeneration, old-gate replay or log/artifact download was performed here. The full current gate remains nonterminal. This review covers selected integration findings, not all1,468 changed files or all ten feature implementations; #269/#274 and original components remain open until their actual delivered scope is established. Receipt dc3e3de8 records these limits.

[afflom]

Exact-head CI43135541484 is now terminal SUCCESS:64/64 materialized jobs,720/720 reported steps, including same-run honesty130910427049; repro43135541535 also passed. All74 listed artifacts are unexpired. This readback fetched no job logs or artifact payloads and repeated no tests or producers.

The head is unchanged, so the concrete source/count/provenance findings in review5867962064 remain unresolved. Passing execution does not establish that the count writer handles the ambiguity cases or correct the current source totals. Please batch the existing #462 repair and the already identified authority/mirror/history corrections before resulting-head qualification; no competing implementation or repeated old-head gate is requested. Existing owner retains the batch and landing; component issues and broad parents remain open.

[afflom]

Actual landing readback: this PR was externally merged at09:47:01 as aaf8455, ordered parents[5dccf9e,707ed976]. Tree e72ad3a57ca0423db1254f852d3094ab4fddef90 equals the reviewed head and recorded pre-merge synthetic b000f7e. Existing exact-head CI43135541484 (64/64 jobs,720 steps, same-run honesty) and repro43135541535 were reused without rerunning them or retrieving logs/artifact payloads.

This confirms landing identity, not resolution of reviews5867962064/5869622192. The #462 count-writer repair is still absent; BS-15 remains3,328 versus3,190, BS-07 remains1,524 versus1,392, and the inherited BS-27 provenance wording remains inaccurate. No final source-clearance review preceded this external merge. Please retain these as concrete owner follow-ups rather than losing them when the component carriers close.

Full-body reassessment keeps #269 and #274 open: BS-25–34 deliver slices, not the complete certified front end and every CLI command on all committed projects. The nine component PRs closed with explicit links to this merged PR remain superseded/unmerged; GitHub separately reports #442/#452 merged through reachable batch history. No further issue closure is justified by this bounded audit.