Repository navigation
Forward-merge release/v0.4 backlog into release/v0.5 (integration blocked) - #405
Conversation
|
Source-integration audit adds two concrete requirements before the eventual single regeneration.
Other required source unions: stdlib entrypoints Buffer/Numeral/Order/Sha256/Target/Text for the pinned heads (Host when integrated); retain v0.5's three fallible-loop-tail vectors and21st plant alongside v0.4's explicit list-match refusal. LOOP_CRATES must be37 and LIST_MATCH_CRATES1; recompute corpus-dependent totals on the final regenerated tree. Eight shared statements evolved: BS-05/06, CL-18, RP-06/07, SM-35, SP-02 and TC-03. Preserve v0.5's extended closure/constructor/source-authority guarantees and v0.4's normative-host stress qualifications. BS-05/06 require combined measured statements, not choosing either side's counts. No generated records were rewritten and no second identity wave was started. |
|
Maintainer integration progress at the same pinned 477a508 → 2849449 snapshot: the handwritten merge resolutions and normative source imports are now prepared in an isolated worktree. This resolves the ten handwritten/document conflicts and preserves the ten v0.4-only claims in spec/src, with the stronger v0.5 contracts retained. Static comparison confirms 490 matching IDs, scenario statements and named tests. The imported §17.23 bootstrap sections belong to Chapter17; Chapter18 remains unchanged from v0.5. This is source preparation, not a passing integration. All 543 generated/projection conflicts (including SPEC and CONFORMANCE) remain for official writers. Corpus-dependent expectations still require measurement on the final tree; the pinned-side numbers are not combined-tree evidence. No regeneration or CI run has started, and this draft's head has not moved. The reusable source patch and review inventory are retained in the maintenance workspace. Refresh them after the pending release landings, reconcile #396's Host claim as SL-12 while retaining PD-10, and carry the source union into the single coordinated writer wave after #389/#402. The current draft remains blocked on that work and the complete final gate; #404 stays open. |
|
Forward-integration refresh after #364 landed: release/v0.4 is now 0a01df6, with the exact tested BS-20 tree e2ecc95f4e571ec2dd57a616e103049bfdb25e86. The pinned source477 snapshot in this draft is no longer final. Local source preparation now imports the BS-20 row and §17.23 material into v0.5's normative IdsBS/Chapter17 sources, retains the ToolingAst/ToolingLatex port and harness fix, and reconciles the landed bootstrap corpus snapshots. The current static union is 491 matching rows/scenarios/statements/tests; that is an audit result, not a new hard-coded gate. Both lines' distinct guarantees and dynamic RP-06/RP-07 checks remain intact. There are still 543 generated/projection conflicts. The original source patch is preserved, with ordinary incremental/full source patches and review notes prepared separately. No writer, regeneration, full gate or remote source update has run. Final source will need another refresh for #403 and the coordinated Host/#402 integration before official projections and measured corpus totals. Host remains SL-12; newly queued JsonRead is reserved SL-13 (#415/#422). |
The first CI run on the case stopped in stack_bounds::observe, which refuses a package whose bound is not yet committed, before the rustc outcomes were compared and before the plants ran. The bounds of every package are now read after both, so a bound not yet committed hides no other result. The evaluator's differential on all ten members had already passed in that run. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… into member 5 (SL-12) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…he decimal reader CI run on 541ecd8: the evaluator and the oracle disagreed on 2 of 300 runs of readJson, both with the outcome overflow where the oracle refuses. The texts were 18446744073709551616 and -82095567141444985264: Numeral.parseDecimal overflows past the target width, so the reader must not hand it a run of digits longer than any i64 has. readInteger now refuses past nineteen digits, a plant removes the guard, and the case reads the stack bounds last, as SL-14's does. The other 298 runs agreed. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
readInteger changed stdlib/src/JsonRead.lex.tex, so the tree digest of the stdlib package moved, and the locks of bootstrap and examples/helpdesk that pin it with it. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
|
Restored draft after the15:39 ready-for-review transition. The head is still477a508d5; no integrated source/generated tree or final gate has been published. The local source preparation still has543 generated/projection conflicts, and the source snapshot must advance for the landed BS-20 plus pending #403 and active Host/#402 wave. Marking ready does not resolve those dependencies, and this head cannot be a merge candidate. I am continuing the sole Host/#402 writer wave, then will refresh this existing forward-integration PR and coordinate #396/v0.5 source projections. Draft source review remains welcome. Please record any separate integration work or active writer phase here before starting it so we do not duplicate regeneration. |
Generated files (locks, records, golden, production-export) are the release's and are regenerated next; the hand-edited ones are merged: the register count is 470, stdlib entrypoints add Host beside Order, the CHANGELOG keeps both. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…counts (BS-16, #275) The branch contains #364 (aea1333, release/v0.4 at 477a508). On that tree the official writers rewrote the records: `lexlean lock`, `check-production-export --write`, `verify-examples --write` (LEXLEAN_EXAMPLES_PART=without:uor-atlas) and `check-golden --write`; `check-golden` then finds every published artifact equal to its oracle. The committed manifests are 284. Counts measured on this tree: BS-05 and BS-06 284/122/201/260 (284 packages), BS-07 335, BS-08 394; BS-09, BS-10, BS-11 pass unchanged. BS-15 is derived from the manifest count. The other counts are CI's to name; VERIFICATION.md says which is which. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ice-3 kappa: the four other hash axes and the compositions on all five axes (#127, KA-05, KA-06)
|
Destination refresh: #407 has landed on release/v0.5 as a8ad058 after all34 final-head CI jobs plus honesty/repro passed; actual merge tree equals the reviewed/tested tree. Preserve its new KA-05/KA-06 normative sources and artifacts when refreshing this integration. A separate source-only refresh of the existing BS-20 preparation is underway; no competing writer wave is starting. The Host/#402 export phase has completed all five applications, and its remaining official writers still own the active identity wave. This PR remains draft until the actual integrated source, generated products, corpus measurements and full gate are complete. |
…ve fixtures on the merged identity The forward merge moved the rust-core target descriptor digest; each fixture is again refused with exactly its prescribed code (LLB6007, LLB6006, LLB6005). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
|
Concrete first blocker in current regeneration43499820761: oracle job132142349228 completes golden-write, then fixtures-write stops at tests/negative/gnaf-certificate-malformed, step1 The exact245f request.json:1 contains that pin; case.toml:4 still requires exit1. production.rs:1043–1044,1123–1127 explains this earlier preflight refusal. Refresh the fixture's target binding through its applicable official input-preparation route while preserving malformed gnaf evidence, zero certificate_id and expected exit1. Do not accept exit2 or weaken certificate checks. If the owning preparation centrally binds descriptors, repair that route for affected fixtures rather than replacing unrelated negative data. One previously unread job log was fetched and archived once (SHA25684ae87241e0598adf76c40e20c1fe7880e85ddcf63f9865ed505a29b96785325). It proves golden completion and partial fixture writes, but output capture/upload was skipped: no recoverable oracle artifact or full-fixture acceptance is claimed. Preserve applicable records/production outputs from the still-running siblings and choose only the necessary failed-segment continuation after repair. No blanket writer replay, competing patch or CI-control action was started. |
…d rewrite the negative fixtures v0.4's RuntimeHost.lex.tex moved the language-1.1 compiler-semantics ID to e70124f1; the v0.5 fixture still pinned 11732e9a. Expected outputs are from `xtask check-fixtures --write`. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
afflom
left a comment
There was a problem hiding this comment.
Bounded source review retires the stale-descriptor blocker in6864778924. The complete18-path delta repins three negative production requests to the exact observed985b5e39 descriptor, preserving the malformed gnaf request/certificate paths, all-zero certificate_id and keyword package. The malformed certificate remains truncated {"spec":; its case.toml is unchanged and still requires exit1. The separate language1.1 fixture relock changes only compiler_semantics. All14 changed expected files retain their diagnostic codes and command exits. Official-writer execution and reached-error claims remain owner evidence; no runtime was repeated.
Old regeneration43499820761 is terminal cancelled: the oracle failed and sibling recipes were cancelled; capture/upload was skipped, so no recoverable published output or acceptance is inferred from their partial work. The saved84ae8724 log was reused without refetch. New owner regeneration43514882165 is the current path to coherent product adoption and resulting-head qualification. Preserve the reviewed actual2b039+6a02 history and scoped prior evidence. The obsolete placeholder body should now reflect this source and output status. No new source blocker found in this bounded correction; full products/CI/repro and actual merged-tree acceptance remain owed, with #316 protected.
The write recipes of SPEC.md §9.2, run by regenerate.yml (https://github.com/UOR-Foundation/LexLean/actions/runs/43514882165), rewrote 6525 paths. The counts writer reported: BS-07 VECTORS: 381 -> 410 (crates/conformance/src/cases/bootstrap/rust_lower.rs) BS-07 VECTORS_WITHIN: 296 -> 308 (crates/conformance/src/cases/bootstrap/rust_lower.rs) BS-08 CRATES: 440 -> 443 (crates/conformance/src/cases/bootstrap/rust_moves.rs) BS-08 WITHIN: 355 -> 358 (crates/conformance/src/cases/bootstrap/rust_moves.rs) BS-11 EVALUATOR_MUTANTS: 283 -> 275 (crates/conformance/src/cases/bootstrap/rust_bodies.rs) BS-12 EVALUATOR_MUTANTS: 152 -> 154 (crates/conformance/src/cases/bootstrap/rust_succeed.rs) BS-12 VECTORS: 76 -> 84 (crates/conformance/src/cases/bootstrap/rust_succeed.rs) BS-12 VECTORS_WITHIN: 75 -> 82 (crates/conformance/src/cases/bootstrap/rust_succeed.rs) BS-13 EVALUATOR_MUTANTS: 143 -> 136 (crates/conformance/src/cases/bootstrap/rust_loops.rs) BS-13 VECTORS: 150 -> 171 (crates/conformance/src/cases/bootstrap/rust_loops.rs) BS-13 VECTORS_WITHIN: 67 -> 72 (crates/conformance/src/cases/bootstrap/rust_loops.rs) BS-14 EVALUATOR_MUTANTS: 548 -> 551 (crates/conformance/src/cases/bootstrap/rust_assemble.rs) BS-14 VECTORS: 385 -> 414 (crates/conformance/src/cases/bootstrap/rust_assemble.rs) BS-14 VECTORS_WITHIN: 300 -> 312 (crates/conformance/src/cases/bootstrap/rust_assemble.rs) BS-07: §31 row: 296 -> 308 (x1), 381 -> 410 (x1) BS-07: register: 296 -> 308 (x1), 381 -> 410 (x1) BS-07: scenario: 296 -> 308 (x1), 381 -> 410 (x1) BS-08: §31 row: 355 -> 358 (x1), 440 -> 443 (x1) BS-08: register: 355 -> 358 (x1), 440 -> 443 (x1) BS-08: scenario: 355 -> 358 (x1), 440 -> 443 (x1) BS-11: §31 row: 283 -> 275 (x1) BS-11: register: 283 -> 275 (x1) BS-11: scenario: 283 -> 275 (x1) BS-12: §31 row: 76 -> 84 (x1), 152 -> 154 (x1) BS-12: the §31 row does not state 75 (-> 82) BS-12: register: 76 -> 84 (x1), 152 -> 154 (x1) BS-12: scenario: 76 -> 84 (x1), 152 -> 154 (x1) BS-13: §31 row: 143 -> 136 (x1), 150 -> 171 (x1) BS-13: the §31 row does not state 67 (-> 72) BS-13: register: 143 -> 136 (x1), 150 -> 171 (x1) BS-13: scenario: 143 -> 136 (x1), 150 -> 171 (x1) BS-14: §31 row: 385 -> 414 (x1), 548 -> 551 (x1) BS-14: register: 385 -> 414 (x1), 548 -> 551 (x1) BS-14: scenario: 385 -> 414 (x1), 548 -> 551 (x1) SPEC.md:8847: 283 -> 275 (every holder) SPEC.md:8842: 75 -> 82 (every holder) SPEC.md:8774: 355 -> 358 (every holder) SPEC.md:8771: 440 -> 443 (BS-08) SPEC.md:8733: 152 -> 154 (BS-12) SPEC.md:8733: 296 -> 308 (every holder) SPEC.md:8730: 381 -> 410 (every holder) SPEC.md:8647: 67 -> 72 (every holder) VERIFICATION.md rows for the landing (counts | before | after): | `BS-07` VECTORS, VECTORS_WITHIN | 381, 296 | 410, 308 | | `BS-08` CRATES, WITHIN | 440, 355 | 443, 358 | | `BS-11` EVALUATOR_MUTANTS | 283 | 275 | | `BS-12` EVALUATOR_MUTANTS, VECTORS, VECTORS_WITHIN | 152, 76, 75 | 154, 84, 82 | | `BS-13` EVALUATOR_MUTANTS, VECTORS, VECTORS_WITHIN | 143, 150, 67 | 136, 171, 72 | | `BS-14` EVALUATOR_MUTANTS, VECTORS, VECTORS_WITHIN | 548, 385, 300 | 551, 414, 312 | For a reader: BS-14 VECTORS_WITHIN: 300 is also the value of ["EVALUATOR_NODES"]; rewrite its statements by hand (300 -> 312) SPEC.md:8739 still states 150 (-> 171 for BS-13?) before "mutants), and 874 seconds for `siteTree" SPEC.md:8775 still states 150 (-> 171 for BS-13?) before "seeded mutants of the crates of at most" SPEC.md:8810 still states 150 (-> 171 for BS-13?) before "seeded runs, the rendered package on 1," wrote /home/runner/work/LexLean/LexLean/CONFORMANCE.md and /home/runner/work/LexLean/LexLean/ERRORS.md written, with SPEC.md and the model projected again; next: the parts without LEXLEAN_COUNTS
|
Regenerated: regenerate run 43514882165 delivered bea5502, with all writers green and counts written through spec/src. The 20:59Z blocker is fixed by 56905d3 (rust-core descriptor repinned in three request fixtures), and nat-bitwise-language-11 is relocked on the merged 1.1 ID (e5e9063). Marking ready for the full gate. The uor-atlas records will be joined from its atlas-part artifacts. Generated by Claude Code |
afflom
left a comment
There was a problem hiding this comment.
Bounded adoption review credits bea5502 as the direct child of reviewed e5e9063, preserving the actual2b039+6a02 forward history. Official43514882165 completed all six jobs and delivered this commit. Complete nontruncated subtree inventory shows generated products plus count changes in six bootstrap case files and corresponding scenario/register/spec-source projections; no new case assertion/filter change was found. PF27/SM42 allocation and the three negative-request repairs remain intact. Artifact presence and path inventory do not establish universal product correctness.
The remaining owner steps are concrete: finish/disposition the Atlas record join you identified in6870010682; obtain successful exact-head reproducibility (43544972146 currently ACTION_REQUIRED); finish existing full43545128695 and resulting-head review; replace the PR body's obsolete477a placeholder/draft narrative with the actual history/adoption/evidence status. Preserve existing artifacts and completed measurements rather than repeating producers for publication.
Actual v0.4 now includes #459 asf8854bf7. Its own actual-tree gate remains pending; retain that distinction and carry required accepted history through this same owner lane before final integration. The test-only optimization does not itself indicate another identity writer. Full final-head acceptance and actual merged-tree equality remain necessary; #404/#128 stay open, and #316 gains no admission from source adoption or ready status. Independent receipt pr405-bea550-adoption-source-review/receipt.json. No runtime, logs, producer or CI controls repeated.
The forward merge wraps every toolchain install in .github/retry.sh, as release/v0.5 does; v0.4's Intel and Windows checks still named the bare command. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
release/v0.5's AU-14 (#241) changed surface::module::locate from a search of each kind in turn, ended at that kind's last closing marker, to the environment the structural parser opens. The structural parser reads control atoms, takes the first \begin after \begin{lexlean}, and ends it at the first \end atom. v0.4's BS-28 twin still followed the old oracle, so the merged gate found disagreements on the evaluator and on rustc. The twin now reads control atoms as the scanner does. The statement, scenario, prose, hand-built texts and planted defects follow the new oracle. Locally the evaluator agrees on 721 runs and rustc on 2,664, and every plant is detected. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
|
Bounded source readback of preparation c912fa2 (not yet the published PR head) found no new blocking defect. The BS-28 source twin matches the existing AU-14 oracle's control atoms, first opening environment and first matching end/refusal behavior; data ranges, whitespace/offset checks, committed corpus limits, evaluator/native comparison and export checks remain. The revised17 plants retain designated-vector and missed-plant assertions. Parent832d28 only updates Intel/Windows expected installation strings for the retry wrapper; custody, native-host, full-gate, timeout, lane and receipt assertions are preserved. The small stale module comment in surface_module.rs:5–7 still describes the former kind-priority/last-end behavior; fold that correction into a necessary owner update rather than opening a separate qualification cycle. The reported721 evaluator/2664 native results remain owner-reported; this review did not execute them. Existing official regeneration43564880134 is in progress on exactc912. Retain the prior bea products and all failed/successful evidence at their own bindings; this source credit does not accept newly generated products or admit #316. Atlas join, current body, required actual v0.4 forward history and resulting-head full acceptance remain with the existing #405 owner. Receipts pr405-c912-bs28-source-review/receipt.json and rp14-supplement.json. No second writer, test or CI-control action was started. |
|
Accepted-history update for the existing forward owner: actual #459 merge f8854bf/tree90732a41 is now fully qualified by its exact push CI43532545468 attempt1 (70/70jobs,772/772steps,same-run honesty) and repro43532545412. Independent postlanding review5903110468 cleared narrow #457, now explicitly closed with6874465992. Reuse this completed evidence; no BS-15 producer/gate replay is requested. Please retain the required actual forward-merge history when integrating the current release line. This acceptance does not blanket-qualify later11f/69de landings or resolve their existing source findings. Currentc912 official43564880134, applicable output adoption/Atlas join, body reconciliation and resulting-head full acceptance remain in your existing lane. #316 remains frozen without admission; no competing writer was started. |
The write recipes of SPEC.md §9.2, run by regenerate.yml (https://github.com/UOR-Foundation/lexlean/actions/runs/43564880134), rewrote 1169 paths. The counts writer reported: counts: no count moved
…dentity The repository is now spelled UOR-Foundation/lexlean, and the exact compare refused every full gate's Intel test construction as a foreign repository. Same change as #533 on release/v0.4, without its committed __pycache__. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
|
The full gate on bea5502 found three failures, all fixed. RP-14 now matches the retried installs (832d28f). The BS-28 twin follows v0.5's AU-14 locate oracle: the evaluator agrees on 721 runs, rustc on 2,664, and all plants are detected (c912fa2). #533's repo-case fix is ported (716bc85). Regenerated as 0b9b140, atlas join green. Full gate run 43582153525 is running. Auto-merge (merge commit) goes on once the stale failing checks clear. Generated by Claude Code |
afflom
left a comment
There was a problem hiding this comment.
Current-head review credits the two-file repository-case repair and officialc912 regeneration43564880134: all6jobs succeeded. The complete adoption inventory has1,360 unique entries confined to bootstrap expected outputs and production-export products; prior artifacts are retained. No source/custody assertion was weakened. This is official-route/path-scope credit, not an assertion that every adopted byte was independently compared with the artifact or that final proofs passed.
Required actual forward history is still missing: accepted actualf8854bf7→716bc diverges behind5 with mergebase6a02; actual69de→716bc diverges behind18 at the same mergebase. Porting individual fixes does not satisfy the required actual forward-merge history. Before the proposed auto-merge, reconcile accepted actual history and disposition later actual inputs in this existing lane.
The adoption contains no Atlas paths. Comment6875439999's green Atlas join remains owner-reported here; identify its exact source/output binding and disposition the joined records without repeating successful producers. Replace the obsolete PR body, which still names5393/477a and says draft. Current43582153525 is incomplete, and auto_merge is currently null. PR base metadata also differs from the independently read actualv0.5 ref; final acceptance must bind the actual release composition and merged tree, not cached PR metadata.
Existing owner retains these bounded next steps and resulting-head full acceptance. #404 stays open and #316 remains unadmitted. Receipt pr405-716bc-adoption-review/receipt.json. No tests, writers, logs, CI controls or competing integration were started.
|
Gate green on 716bc85: full gate 43582153525 (bdd-verification passed on its single rerun) and reproducibility. Merged into release/v0.5 as merge commit 6c36829. release/v0.4 has moved since 6a02d2c (#431, #520, #533, among others). That newer backlog needs its own forward merge. Generated by Claude Code |
|
Independent postlanding readback: actual6c368291 has parents2b0395ef +716bc85e and treef72da750, exactly the reviewed composition. Exact43582153525 attempt2 API inventory contains70 successful jobs and813 successful steps, including honesty and Atlas join; proposal repro passed. Preserve attempt1's canceled verification receipt separately. Attempt2 exposes inherited timestamps/replacement IDs, so this is complete API inventory credit, not a claim70jobs were rerun. No completed gate was repeated by this review. Actual push43640234670 is only1plan success +17skipped; it is not a second full gate. Actual repro passed. Required acceptedf885 history remains absent (mergebase6a02,5behind), matching the owner's acknowledged follow-up backlog. #404 correctly stays open. Joined-output custody/body reconciliation and applicable final acceptance remain scoped obligations; new#541 carries the explicit#431 STOP conflict6884962343. This does not admit root-reserved#316 or automatically adopt later release inputs. Receipt pr405-actual6c368-acceptance-audit/receipt.json. |
Closes #404 when the reconciled release integration lands. Release tracker #128 remains open.
AGENTS.md requires forward merges by merge commit from release/v0.4 into release/v0.5. The current source is
5393f5528ec46a3fb0f2884a81c0c7087f1de95b; destination is2b0395ef1cf020509826e3a2cd8a0cb9add6ddd2, including merged #396. This is the full release backlog across compiler, bootstrap, standard library, CI and generated proof artifacts.The remote branch still holds the historical
477a508d5e0e49992235337a21b0a30a9c572547placeholder. It is not the prepared integration and must remain draft. Earlier source/destination and conflict counts in comments describe their dated snapshots.Reviewed local source preparation now covers 93 handwritten paths against the actual destination. Eight overlaps were reconciled, preserving #396's closed kernel decisions, byte slices, joins and Host SL12 alongside Numeral SL09, Order SL10, KA05/06, BS20 and the maintenance changes. The same-run honesty job retains v0.5's SIMD probe and Rust retry behavior. Static correspondence is 494 normative rows/scenarios/named tests with matching statements; this observed count is not a new fixed-count gate. Normative edits are in
spec/src; generated model/SPEC projections remain untouched pending official writers.A fresh raw merge of the two release heads reports 541 conflicted paths: 527 artifact/lock/package paths, two generated projections and twelve handwritten/CI paths. The source overlay is separate preparation, not a completed merge tree or proof that every automatic merge is correct. Both verification histories are retained. Combined corpus counters must be measured on the final tree.
Coordinate the sole active #389/#402 Host+Order/deep-package-reader identity wave and subsequent #393 work, plus the reviewed #362 Mathlib and #433 toolchain inputs. #396 has now landed with all 34 final CI jobs plus honesty/reproducibility passing and matching actual/tested tree; it closes #409 only and does not fulfill this release integration. No competing identity regeneration is authorized by this tracker.
Completion requires actual release ancestry, reconciled source and generated artifacts using official writers, preservation of both lines' guarantees, the renderer/compiler/runtime identity fixed point, complete final-head acceptance and a reviewed merge commit. Explicitly close #404 with its merged PR link only after that work lands. No source preparation, historical gate or placeholder diff constitutes readiness.