Intelligence Reconnaissance

Queries run 2026-04-17:

• scripts/intel-query.sh --human file-symbols "ori_llvm/src/codegen/arc_emitter" --repo ori — inventory ArcIrEmitter symbols (variable-definition ordering, emitter_utils) before investigating the BUG-04-085 variable-not-yet-defined crash.
• scripts/intel-query.sh --human callers "compute_flags" --repo ori — blast radius of TypeFlags computation (including PROPAGATE_MASK) to trace §02.0’s effect on downstream consumers.
• scripts/intel-query.sh --human callers "body_type_map" --repo ori — identify all consumers of body_type_map that may have relied on pre-§02.0 Tag::Scheme HAS_VAR=false flag semantics.
• scripts/intel-query.sh --human similar "variable not defined arc emitter ordering" --repo swift,lean4 --limit 3 — prior art for SSA variable-definition ordering bugs in ARC IR emitters.

Results summary (≤500 chars) [ori]: ArcIrEmitter lives in ori_llvm/src/codegen/arc_emitter/; emitter_utils.rs handles variable definition tracking. body_type_map consumed by prepare_mono_cached in nounwind/prepare.rs. compute_flags in ori_types/src/pool/ — PROPAGATE_MASK sets HAS_VAR on Tag::Scheme items since §02.0. [swift]: SIL variable-ordering bugs manifest as “use before definition” in ownership verification — mirrors the var=N not yet defined crash signature.

07.PRE — Predecessor-Status Gate

Origin. Per routing.md §5 ordering primitives + state-discipline.md §4 predecessor-integrity contract: every section in depends_on: ["01", "02", "03", "04", "05", "06", "08", "09", "10", "11"] MUST satisfy status: complete AND reviewed: true before §07.0 may begin. Without an explicit precheck step, §07.0/§07.1 silently advance against not-started/in-progress/in-review predecessors (STRUCTURE:work-order-violation per impl-hygiene.md §Finding Categories).

Gate. §07.0 SHALL NOT begin (and §07.1–§07.4 SHALL NOT begin transitively per subsection_depends_on:) while the precheck reports failed.

Transitive deps note. §07’s direct depends_on: list includes §08; §04 carries its own depends_on: chain that transitively reaches §07 via §08. The precheck consults the full depends_on: set per routing.md §5 ordering primitives — predecessor verification is whole-graph, not single-hop.

• Run python -m scripts.plan_corpus check plans/typeck-inference-completeness/ — zero STRUCTURE:work-order-violation findings. Any predecessor with status != complete or reviewed != true fails the precheck.
• Inspect each predecessor’s frontmatter directly: §01, §02, §03, §04, §05, §06, §08, §09, §10, §11 ALL carry status: complete + reviewed: true. Predecessors flagged blocked (e.g., §04 cross-scope blocked on aims-burden §06 per 00-overview.md RESUME POINTER) require resolution BEFORE §07 advances; pulling the blocker into scope is the cure per CLAUDE.md §Failing Tests Mid-Work → Pull Blockers Into Scope.
• On precheck failure, §07 SHALL remain in-review/in-progress and emit next_action: blocked referencing the specific predecessor IDs that failed; resolution path documented in 00-overview.md HISTORY block.
• On precheck success, advance to §07.0. Record the verifying SHA in review_pipeline: frontmatter.

07.0 Investigate BUG-04-085 — LLVM spec runner crash (absorbed 2026-04-17; HARD BLOCKER on §07.1)

Origin: Absorbed from bug-tracker BUG-04-085 per CLAUDE.md §Ownership & Deferral “Plan-blocker bugs belong IN the plan”. The bug was filed 2026-04-15 with explicit notes that §02.0’s Tag::Scheme PROPAGATE_MASK change may be the causal root. Because this plan introduced §02.0 and the crash may be its downstream symptom, the plan owns investigating and resolving it as part of close-out.

Block-gate semantics (per subsection_depends_on: { "07.1": ["07.PRE", "07.0"] }). §07.0 hard-blocks §07.1. Filing a new bug + advancing to §07.1 with the crash unresolved is BANNED — CLAUDE.md §The One Rule (correctness above all) + §ZERO DEFERRAL apply: the crash IS this plan’s deliverable to fix, not to defer. The two legitimate exit paths are:

• Path A — §02.0 is causal: §02.0’s Tag::Scheme PROPAGATE_MASK fix is confirmed root cause; the downstream consumer (ArcIrEmitter variable-definition ordering, body_type_map, prepare_mono_cached) is fixed in THIS plan to work correctly with the new flag semantics. §07.0 closes; §07.1 advances.
• Path B — §02.0 is not causal: §02.0 is confirmed not-causal AND the true root cause (drop-fn pointer type mismatch / ArcIrEmitter ordering regression independent of §02.0 / other) is identified AND fixed IN THIS PLAN as a new sibling subsection §07.0.A under the same umbrella per routing.md §3 case (a) blocker discipline. §07.0 closes; §07.1 advances.

Path B-with-new-bug routing. A new bug filed during §07.0 investigation is routed per routing.md §3 case (a): it becomes sibling subsection §07.0.A under THIS umbrella, NOT a separate bug-tracker/plans/BUG-XX-XXX/ entry (sibling bug-tracker entry would be PLAN_ROUTING_DRIFT per impl-hygiene.md §Finding Categories). §07.1 STILL blocks on §07.0.A close. The previous text permitting “continue with §07.1” when §02.0 is not causal is BANNED per the §07.0 block-gate above.

Same-family-with-§08 resolution status (per 00-overview.md Known Bugs table, 2026-04-20). BUG-04-085 is classified Same-family with §08 (cross-module pool-merge / monomorphization corruption surfacing in ArcIrEmitter). §08 (BUG-04-042) shipped the JIT-only fix; the LLVM AOT path tracked via BUG-04-119. §07.0 investigation MUST either (a) confirm the §08.3 remap-aware re-intern fix closes BUG-04-085 in the LLVM path as well (no separate cure needed; flip §07.0 complete with §08.3 SHA as resolved_in), OR (b) identify the residual LLVM-specific cure surface AND fix it in this plan as §07.0.A. Either path REQUIRES the crash to be gone before §07.1.

Failure mode (from bug entry): timeout 150 ./test-all.sh shows Ori spec (LLVM backend) CRASHED with cluster of errors on assert_eq$m$int (monomorphized assert_eq<int>):

• LLVM IR verification failed after codegen (emit_prepared_functions) with Call parameter type does not match function signature! on ori_rc_dec({ i64, i64, ptr }, ptr @"_ori_drop$N")
• Three ArcIrEmitter: variable not yet defined errors (var=11, var=2, var=7) — mirrors resolved BUG-04-024 signature
• error[E4003]: ARC internal error: index assignment reached ARC lowering before desugaring (BUG-04-070 class)
• thread 'ori-main' has overflowed its stack → SIGABRT
• Interpreter passes all 4444 spec tests; only --backend=llvm crashes

Causal hypothesis: compiler/ori_types/src/pool/mod.rs:655-661 — §02.0’s Tag::Scheme PROPAGATE_MASK fix sets HAS_VAR on scheme Idx values where previously unset. If the mono pipeline’s body_type_map or ArcIrEmitter’s variable-definition ordering relied on the old flag semantics (specifically: “scheme types have !HAS_VAR so skip the substitute step”), then setting HAS_VAR correctly now surfaces as “variable not yet defined” in the downstream consumer.

• Reproduce the crash cleanly: timeout 150 ./test-all.sh on a clean post-§08/§09/§10/§11 tree — confirm the specific crash line, the 3 var=N errors, the assert_eq$m$int function name, and the SIGABRT stack overflow.
• Confirm §08.3 status against the crash: re-run the repro with HEAD at §08.3’s 131c771ea (or later) and §11.1 complete. If the crash IS GONE post-§08.3, the §08 same-family fix closed it; record the SHA in §07.0 HISTORY + the bug-tracker resolved_in: field + advance to §07.0 close. If the crash PERSISTS, continue to the isolation step below.
• Isolate §02.0’s causal role (only if §08.3 did not close it): revert compiler/ori_types/src/pool/mod.rs:655-661’s Tag::Scheme contribution to PROPAGATE_MASK on a scratch branch; re-run test-all.sh. If the crash disappears, §02.0 is causal (Path A). If the crash persists, §02.0 is not causal (Path B); identify the true root cause via ORI_DUMP_AFTER_LLVM=1 + ORI_VERIFY_ARC=1 + compiler_repo/diagnostics/dual-exec-debug.sh per tooling-first.md §4 (use existing tooling first; extend if gap exists).
• Path A (§02.0 causal): identify the downstream consumer that depended on the old flag semantics. Candidate sites: ArcIrEmitter variable-definition ordering (emitter_utils.rs), the mono pipeline’s body_type_map construction, prepare_mono_cached’s scheme handling. Fix the consumer to work correctly with the new (correct) flag semantics. Land as commit fix(typeck): downstream consumer respects Tag::Scheme HAS_VAR semantics (BUG-04-085). Do NOT revert §02.0.
• Path B (§02.0 not causal): fix the identified true root cause IN THIS PLAN as §07.0.A sibling subsection per routing.md §3 case (a). Do NOT route to a separate bug-tracker/plans/BUG-XX-XXX/ entry (would be PLAN_ROUTING_DRIFT). §07.1 stays blocked on §07.0.A close.
• TDD: add a spec test or Rust unit test that exercises the failing path — assert_eq<int> in a file that also uses polymorphic lambdas (overlap with §08 surface). Verify the test fails on the current code and passes after the fix.
• Matrix: poly-lambda × imported generic × assert_eq type — confirm no crash for int, str, bool, float variants.
• Verify gate: timeout 150 ./test-all.sh no longer shows Ori spec (LLVM backend) CRASHED. §07.1 unblocks.
• Update 00-overview.md: amend Known Bugs table BUG-04-085 row Status column from “Same-family with §08 per 2026-04-20 resolution” to “Complete (§07.0 confirms §08.3 closes the LLVM path | Fixed via §07.0.A | other)” and add a HISTORY entry recording the verifying SHA. Resolves PLAN_COHERENCE_DRIFT between §07.0 unchecked boxes and 00-overview prose claim.

Section 07: Close-out + Supersession

Status: In Progress (per frontmatter status: in-progress; flips to complete post-§07.4 + umbrella 00-overview.md close-out per state-discipline.md §4) Goal: Formally close BUG-04-074, BUG-04-084, BUG-04-042, BUG-04-085 and this plan. Verify all code changes are clean, all plan annotations are removed, the LLVM spec-runner crash is fixed, and the bug-tracker is updated.

Depends on: All previous sections complete (§01–§06, §08, §09, §10, §11) per depends_on: frontmatter; §07.PRE verifies.

07.1 Final Test Verification

timeout 150 ./test-all.sh          # debug build — must be green; Ori spec (LLVM backend) row no longer CRASHED
timeout 150 cargo test --release -p ori_types   # release build
timeout 150 cargo test --release -p ori_llvm    # LLVM + AOT release

Dual-execution parity verification covers the FULL work arc (§05 empty-list corpus + §08–§11 representative paths). Run on each path; diagnostics/dual-exec-verify.sh MUST report zero parity divergences across all listed paths.

# §05 — empty-list corpus (Section 05 carryover)
diagnostics/dual-exec-verify.sh tests/spec/types/collections/empty_list/empty_list_push_unified_dual_exec.ori
diagnostics/dual-exec-verify.sh tests/spec/types/collections/empty_list/empty_list_element_struct.ori
diagnostics/dual-exec-verify.sh tests/spec/types/collections/empty_list/empty_list_iter_chain_dual_exec.ori

# §09.3/§09.5 — Result<T, user-Error> LHS propagation + method-call return BD-2
diagnostics/dual-exec-verify.sh tests/spec/traits/into/str_to_error.ori

# §09.4 — lambda-parameter propagation from receiver element type
diagnostics/dual-exec-verify.sh tests/spec/traits/iterator/methods.ori

# §10.1 — rigid-receiver bound-chain dispatch
diagnostics/dual-exec-verify.sh tests/spec/declarations/traits.ori

# §10.2 — capability-method dispatch
diagnostics/dual-exec-verify.sh tests/spec/capabilities/propagation.ori

# §11.1 — polymorphic-constructor defaulting regression guard
diagnostics/dual-exec-verify.sh tests/spec/types/empty_literals/

Coverage map (§07.1 dual-exec rows × surface owner):

07.2 Plan Annotation Cleanup Sweep + Post-§09.3 Helper Removal

Per CLAUDE.md §Compiler Coding Guidelines: “Plan annotations are temporary scaffolding. They MUST be removed when the plan completes.”

07.2.a Annotation marker sweep

Run the sweep over all three trees (compiler source, spec tests, and stdlib), using only this plan’s specific markers. The 2026-04-23 plan rename from empty-container-typeck-phase-contract to typeck-inference-completeness means BOTH pre-rename and post-rename markers must appear in the sweep regex until every annotation site has been migrated:

# Plan-specific markers in compiler source (Rust files) — must return zero hits
rg 'BUG-04-074|BUG-04-084|BUG-04-042|BUG-04-085|empty-container-typeck|typeck-inference-completeness' compiler/ --glob '*.rs'

# Plan-specific markers in AOT fixture files (Ori files in compiler tree) — must return zero hits
# These are created in §05.3 at compiler/ori_llvm/tests/aot/fixtures/empty_list/*.ori
rg 'BUG-04-074|BUG-04-084|BUG-04-042|BUG-04-085|empty-container-typeck|typeck-inference-completeness' compiler/ --glob '*.ori'

# Plan-specific markers in spec tests (regression comments added in §06.2)
rg 'BUG-04-074|BUG-04-084|BUG-04-042|BUG-04-085|empty-container-typeck|typeck-inference-completeness' tests/ --glob '*.ori'

# Plan-specific markers in stdlib (edits made in §06.3)
rg 'BUG-04-074|BUG-04-084|BUG-04-042|BUG-04-085|empty-container-typeck|typeck-inference-completeness' library/ --glob '*.ori'

Note: Generic §NN tokens (e.g. §01, §03) are intentionally excluded from the sweep pattern because they exist in unrelated plans (e.g., compiler/ori_repr/ contains §03 and §04 references from a different plan). Searching for bare §NN produces false positives and sends the implementer chasing unrelated code. Use only the plan-specific identifiers above.

Every hit must be removed or converted to a permanent spec comment (Spec: Clause N.M format). No plan-referencing comments in shipped code.

07.2.b Post-§09.3 helper removal (traceable/ workarounds)

Scope: the @mk_ok / @mk_err helper-function workarounds applied during §06.2C remediation to tests/spec/traits/traceable/definition.ori + tests/spec/traits/traceable/result_delegation.ori. These are CLEANUP DEBT — they stay in main-branch source only until §09.3 lands. Post-§09.3, the direct-constructor form compiles clean without the helpers, and keeping them is a hygiene violation (dead-ish workarounds mascuerading as legitimate helpers).

Predecessors (MUST all be complete before this subsection starts):

• §09.3 (Result<T, user-Error> LHS propagation) is status: complete in its frontmatter.

• §06.4.4’s “Per-file INVERTED-TDD reconciliation for 2 traceable/ helper-workaround files” step is marked [x] and reports the scratch-revert as GREEN (proving the direct-constructor form compiles under the shipped §09.3 fix).

• Confirm §09.3 frontmatter status: complete.

• Confirm §06.4.4’s traceable/ scratch-revert step is [x] with GREEN result. If the scratch-revert failed, §07.2 helper removal is BLOCKED pending §09.3 completeness investigation — file a bug naming the specific residual failure.

• In main-branch source, strip the @mk_ok / @mk_err helper calls AND their helper definitions from tests/spec/traits/traceable/definition.ori + tests/spec/traits/traceable/result_delegation.ori. Restore each file to the direct-constructor form (Ok(x: ...) / Err(x: ...) in try { ... } with outer let r: Result<T, E> = ... LHS annotation — same shape as the §06.4.4 scratch-revert fixture).

• Run timeout 150 cargo stf tests/spec/traits/traceable/ — both files green. Run timeout 150 diagnostics/dual-exec-verify.sh on both files — zero parity divergences.

• Run timeout 150 ./test-all.sh — confirm no regression outside the traceable/ directory (if a regression appears, the removal is unmasking a different compiler gap; pull the unmasked compiler gap into §07.2 scope (or sibling subsection) and fix per CLAUDE.md §Failing Tests Mid-Work; never back out + defer).

• Commit: refactor(tests): remove @mk_ok/@mk_err helpers from traceable/ — §09.3 closes the compiler gap directly (§07.2).

• Update §06.2C table: change the Classification for the 2 traceable/ rows from helper workaround — CLEANUP DEBT to REMOVED (direct form, §09.3 closed gap) and the Owning column to — (resolved via §07.2).

• Update §03.N Known Failing Tests table (section-03-bodies-pass-integration.md:~1243-1280): strike the two traceable/ rows as [REMEDIATED in §07.2 — helpers removed post-§09.3].

07.2.c R7-006 Validator Refinement — Discriminate Lambda return-vs-param in E2005 Site Classifier

Origin. Absorbed from TPR Round 7 finding R7-006 (codex medium) per §06.R line 701 anchor. The finding cites compiler_repo/compiler/ori_types/src/check/validators/mod.rs:204 where site_from_expr_kind is invoked at the SINGLE per-ExprIndex body-walk call site, classifying every Lambda-sited Tag::Var as LambdaParam and producing the closure-parameter diagnostic wording even when the unresolved var is on the lambda’s RETURN type. The §06.1 spec bullet at :151 defines LambdaParam as specifically “ExprKind::Lambda { params, .. } where a parameter has unresolved Tag::Var” — return-type vars violate this classification.

Sequencing — moved from §07.5 to §07.2.c (per /tp-help gemini High SEQ_GAP). Original §07.5 placement after §07.3 meant validator changes could ship to production without going through §07.3’s final TPR + hygiene gate. §07.2.c routes validator refinement BEFORE §07.3, so §07.3 covers it. subsection_depends_on: { "07.3": ["07.2.c"] } enforces.

Why post-§09/§10/§11. Running the validator refinement against a body-inference surface that still has §09 BD-2 gaps or §11.1 defaulting holes produces false diagnostic wording regressions and thrashes the test matrix. §09/§10/§11 must land FIRST so the only remaining E2005 sites are genuine residuals the validator refinement must classify correctly.

Call-site precision (per /tp-help opencode Low DOC_PRECISION). The validator has ONE Lambda-aware call site at compiler_repo/compiler/ori_types/src/check/validators/mod.rs:204 inside the body-expr walk loop (the site = site_from_expr_kind((ctx.kind)(expr_idx)); line). The signature-position walk at :192-195 always uses AmbiguousTypeSite::Expression (no ExprKind available). The LambdaParam narrow-span helper at :213-217 (narrow_to_lambda_param_span) is the secondary surface that consumes the classifier output. Fix scope is narrow: refine site_from_expr_kind + thread VarPosition + extend narrow_to_lambda_param_span (or sibling helper) for the return-type case.

Fix shape.

• Read compiler_repo/compiler/ori_types/src/check/validators/mod.rs:95-100 (site_from_expr_kind definition) + :200-220 (body-walk call site + narrow_to_lambda_param_span) + surrounding context to understand the current signature (what does it receive? just the ExprKind, or does it have access to the specific Tag::Var position that triggered the E2005?).
• If the signature doesn’t currently carry position info, extend it: add a VarPosition enum ({ LambdaParam(param_index), LambdaReturn, FunctionParam(param_index), FunctionReturn, LetBinding, MatchArm, ... }) and thread it through the validator’s Tag::Var-reporting call site so each E2005 emission knows which slot introduced the unresolved var.
• Split LambdaParam site class into LambdaParam { index } + LambdaReturn — emit distinct E2005 diagnostic messages per site class (parameter wording stays verbatim per §06.1 spec; return wording is new — propose “cannot infer the return type of this lambda; add a return annotation or constrain the body expression”).
• Extend narrow_to_lambda_param_span (or add sibling narrow_to_lambda_return_span) so return-position vars narrow to the body span (or return-annotation span when present), NOT the whole-lambda span.
• Write positive + negative test pair: test_e2005_lambda_return_uses_return_wording (positive — synthesize an ExprKind::Lambda with unresolved return-position Tag::Var, verify the emitted E2005 message uses the return wording) + test_e2005_lambda_param_uses_param_wording_unchanged (regression — verify param-position classification still produces the pre-refinement wording).
• Dual-exec parity: the validator refinement is diagnostic-only (no AST change, no canonicalization change, no codegen change) so diagnostics/dual-exec-verify.sh should show zero parity divergences — verify explicitly on one of the §07.1 dual-exec paths.
• Update §06.1’s E2005 wording success criterion at section-06-diagnostics-audit.md:21 if the return-position wording is added (currently only cites the empty-list example).

Gate. §07.2.c MUST be status: complete before §07.3 final TPR + hygiene gate runs. §07.3 scope COVERS §07.2.c per the expanded scope line below.

Commit shape. One commit scoped to the validator refinement alone (no bundling with §07.0 BUG-04-085 investigation or §07.2 helper removal). Commit message: fix(typeck): validator E2005 site classifier discriminates Lambda return vs param (R7-006).

07.3 Final TPR + Hygiene Gate

Run this BEFORE updating any external bug trackers. All validation gates must pass before declaring any bug resolved — marking bugs complete before gates pass leaves the tracker out of sync if a gate fails.

First, validate the entire plan directory against the plan corpus schema:

python -m scripts.plan_corpus check plans/typeck-inference-completeness/

This command must pass with zero schema errors before the plan can be marked complete. If it fails on sections 01–04 or index.md (sections written before the plan_corpus schema was current), update their frontmatter to satisfy the schema before proceeding.

Run final TPR over the FULL work arc:

/tpr-review

The scope covers the full work arc: all compiler changes in Sections 01–04, new test files from Section 05, annotation changes from Section 06, §07.0 BUG-04-085 LLVM-crash resolution (and §07.0.A if Path B fired), §07.2.a annotation sweep + §07.2.b traceable/ helper removal, §07.2.c R7-006 validator refinement, §08 poly-lambda mono fix, all §09 body-inference BD-2 extensions (§09.1 try-block, §09.2 def-impl Self, §09.3 Result<T,E> LHS, §09.4 lambda-param, §09.5 method-call return BD-2, §09.6 Map lambda-param, §09.7 file splits), all §10 dispatch-on-rigid-receiver changes (§10.0 file splits, §10.1 bound-chain, §10.2 capability), and §11.1 Never-fallback predicate extension + body_inference_complete debug_assert + Some-exclusion. The review must return zero actionable findings before the plan can be marked complete.

Run hygiene review:

/impl-hygiene-review

Auto-scope mode covers the FULL work arc (sections §01–§06, §07.0, §07.0.A if any, §07.2.a/§07.2.b/§07.2.c, §08–§11). Must pass.

Run tooling improvement sweep:

/improve-tooling

Sweep covers the FULL work arc — review test harness, diagnostic script, or developer tooling gaps surfaced during the implementation of §01–§06 AND §07.0 LLVM-crash investigation AND §07.2.a/§07.2.b/§07.2.c cleanup work AND §08–§11 compiler extensions. Surface gaps in dual-exec-verify.sh, test-all.sh, ARC diagnostic flags (ORI_DUMP_AFTER_*, ORI_VERIFY_ARC, ORI_TRACE_RC), typeck diagnostics (E2005 site classifier infrastructure from §07.2.c), and validator tooling (the body-walk machinery exercised by §11.1’s body_inference_complete debug_assert).

Run CLAUDE.md sync sweep:

/sync-claude

Sync covers the FULL work arc — verify no new APIs, commands, or invocation patterns introduced by §01–§06 AND §07.0/§07.0.A AND §07.2.a/§07.2.b/§07.2.c AND §08–§11 are missing from CLAUDE.md / .claude/rules/*.md. Particular surfaces to verify:

• typeck.md §BD-2 — §09.1/§09.3/§09.4/§09.5 expected-type-propagation gates extending the check_expr SSOT pattern (BD-2 propagation table).
• typeck.md §EX-4 — §10.1 bound-chain dispatch on Tag::RigidVar receivers extending TraitRegistry::lookup_method_checked; §10.2 capability handler dispatch via same registry policy.
• canon.md §4.2 / typeck.md §PC-2 — §11.1 defaulting predicate extension to polymorphic-constructor roots (None/Ok/Err) AND body_inference_complete debug_assert in body_finalize/mod.rs.
• typeck.md §DI-1 — §07.2.c E2005 site classifier discriminating Lambda return-vs-param.
• compiler.md §Commands / repr.md — any new ORI_* env vars or diagnostic flags introduced during §07.0 investigation tooling work.
• bug-tracker/00-overview.md — reduced open-bug count after §07.4 closes four bugs.

07.4 Bug-Tracker Updates — Close ALL FOUR Bugs Atomically

Only update bug trackers AFTER 07.3 validation gates pass. This sequencing ensures no bug is prematurely marked complete.

Scope (per frontmatter supersedes: + 00-overview.md Scope Absorption + 2026-04-17/2026-04-23 absorption record). This plan formally supersedes / closes FOUR bugs in a single §07.4 transaction:

Atomic-flip discipline (per state-discipline.md §4). All four bug-tracker entries flip from open-bugs.json to closed-bugs.json via scripts/plan_corpus/bug_tracker/api.py:flip_bug_status under fcntl.flock cross-file atomic transaction. Partial closure (closing 1–3 of 4 bugs) is BANNED — §07.4 incomplete until all four flips succeed.

Per-bug update steps (apply for each of the four bugs):

1. Update bug-tracker/plans/BUG-04-NNN/ (active → completed):

   • Set status: complete in 00-overview.md frontmatter
   • Add pointer: "Superseded by plans/typeck-inference-completeness/section-NN-*.md" (the owning section per the table above)
   • Set resolved_in: to the commit SHA of the relevant landing (§03 landing for BUG-04-074/BUG-04-084; §08.3 SHA 131c771ea for BUG-04-042; §07.0 closing SHA for BUG-04-085)
   • Move plan directory from bug-tracker/plans/BUG-04-NNN/ to bug-tracker/plans/completed/BUG-04-NNN/

2. Atomic tracker flip via scripts/plan_corpus/bug_tracker/api.py:flip_bug_status:

   • Removes entry from bug-tracker/open-bugs.json
   • Appends entry to bug-tracker/closed-bugs.json
   • Writes write-ahead journal at bug-tracker/_inflight/BUG-04-NNN.json BEFORE source mutation per state-discipline.md §4 cross-file flip discipline
   • Lock-order: acquire closed-bugs.json.lock BEFORE open-bugs.json.lock (alphabetic)

3. Update bug-tracker/00-overview.md:

   • Decrement open bug count by 4
   • Increment closed bug count by 4
   • Add HISTORY entry: YYYY-MM-DD — plans/typeck-inference-completeness/ §07.4 closed BUG-04-074, BUG-04-084, BUG-04-042, BUG-04-085 atomically via api.flip_bug_status; resolution summaries above.

4. Update bug-tracker/section-04-codegen-llvm.md (or wherever the four bugs are tracked in section-scoped trackers):

   • Mark each entry as [x] resolved
   • Add plan pointer per the table above

Verification. After §07.4 atomic flip:

• bug-tracker/open-bugs.json no longer contains BUG-04-074, BUG-04-084, BUG-04-042, BUG-04-085 entries.
• bug-tracker/closed-bugs.json contains all four entries with status: complete + resolved_in: populated + plan pointer populated.
• bug-tracker/_inflight/ is empty (no orphaned journal entries — successful flip cleans up).
• python -m scripts.plan_corpus check bug-tracker/ reports zero BUG_BLOCKER_DEAD_REF / BUG_BLOCKER_CYCLE findings (closed bugs remain referenceable via the supersession pointer chain).

07.R Third Party Review Findings

Round 2 — Dual-source TPR on sections 05, 06, 07 (Codex + Gemini). Findings addressed in this revision.

[[TPR-07-001-codex]] [MEDIUM] Scope the annotation cleanup sweep to this plan’s markers and all affected trees

Location: plans/empty-container-typeck-phase-contract/section-07-closeout.md:77 Reviewer: Codex | Status: Fixed

Evidence: Section 07 originally grepped compiler/ for BUG-04-074|empty-container-typeck|§01|§02|§03|§04. That misses temporary plan comments in spec tests (section-05-test-matrix.md instructs implementers to add Regression: BUG-04-074 comments under tests/spec/...). It also produces unrelated false-positive hits because generic §03/§04 markers already exist in compiler/ori_repr/src/narrowing/tests.rs:1 and compiler/ori_parse/src/dispatch.rs:134 (from a different plan).

Fix: Split the sweep into two targeted commands that use only plan-specific identifiers (BUG-04-074|empty-container-typeck) and cover both compiler/ and tests/. Dropped generic §NN tokens from the pattern to eliminate false positives from unrelated plans. Updated the success_criteria in the frontmatter to match.

[[TPR-07-002-gemini]] [LOW] Expand plan annotation cleanup regex to cover all sections

Location: plans/empty-container-typeck-phase-contract/section-07-closeout.md:77 Reviewer: Gemini | Status: Fixed (subsumed by TPR-07-001-codex)

Evidence: Section 07.2 used a regex that explicitly omitted §05, §06, and §07. If developers left code annotations referencing those sections, the final cleanup sweep would miss them.

Fix: Same as TPR-07-001-codex — the updated sweep uses plan-specific identifiers only (BUG-04-074|empty-container-typeck) which covers all sections by design: any annotation referencing this plan’s work will include the bug ID or plan name.

[[TPR-07-003-codex]] [MEDIUM] Add a whole-plan corpus validation gate before close-out

Location: plans/empty-container-typeck-phase-contract/section-07-closeout.md:104 Reviewer: Codex | Status: Fixed

Evidence: Section 07’s final gate originally ran only /tpr-review and /impl-hygiene-review. The command python -m scripts.plan_corpus check plans/empty-container-typeck-phase-contract/ currently fails on index.md, section-01-value-restriction.md, section-02-validator-module.md, and section-04-codegen-assertions.md. Without an explicit gate, Section 07 could declare the plan complete while the owning plan directory is still schema-invalid to the plan tooling.

Fix: Added python -m scripts.plan_corpus check plans/empty-container-typeck-phase-contract/ as the first step in 07.4. The gate must pass before TPR or hygiene review proceeds. Also added to the 07.N checklist as an explicit - [ ] item.

Round 3 — Dual-source TPR on sections 05, 06, 07 (Codex + Gemini). Findings addressed in this revision.

[[TPR-07-004-codex]] [MEDIUM] Sweep library/ directory for plan markers during close-out

Location: plans/empty-container-typeck-phase-contract/section-07-closeout.md:§07.2 Reviewer: Codex | Status: Fixed

Evidence: Section 06.3 directs implementers to annotate library/std/ code. Section 07.2 only swept compiler/ (with *.rs filter, missing .ori AOT fixtures) and tests/. The library/ tree was entirely missing from the cleanup sweep.

Fix: Added two new sweep commands: one for compiler/ --glob '*.ori' (AOT fixtures) and one for library/ --glob '*.ori' (stdlib). Updated success_criteria to reference all four trees. Updated 07.N checklist to explicitly call out all four trees.

[[TPR-07-005-codex]] [MEDIUM] Add release ori_llvm test and section-04 tracker update to 07.N

Location: plans/empty-container-typeck-phase-contract/section-07-closeout.md:§07.N Reviewer: Codex | Status: Fixed

Evidence: Section 07.1 specifies cargo test --release -p ori_llvm but the 07.N checklist item for “07.1 complete” only said “./test-all.sh green debug + release”. Section 07.3 (now 07.4) step 3 requires updating bug-tracker/section-04-codegen-llvm.md but the 07.N checklist for “07.3 complete” (now “07.4 complete”) omitted this.

Fix: Updated 07.N “07.1 complete” item to include cargo test --release -p ori_llvm. Updated “07.4 complete” item to explicitly include bug-tracker/section-04-codegen-llvm.md resolution.

[[TPR-07-006-gemini]] [HIGH] Reorder 07.3/07.4 so validation gates run before bug-tracker closure

Location: plans/empty-container-typeck-phase-contract/section-07-closeout.md:§07.3/07.4 Reviewer: Gemini | Status: Fixed

Evidence: Section 07.3 was “Bug-Tracker Updates” (marks BUG-04-074 complete) and 07.4 was “Final TPR + Hygiene Gate”. This ordering allows the bug to be declared resolved in external trackers before the final /tpr-review and /impl-hygiene-review pass. If a gate fails after the tracker is updated, the tracker is out of sync.

Fix: Swapped 07.3 and 07.4: 07.3 is now “Final TPR + Hygiene Gate” and 07.4 is “Bug-Tracker Updates (runs AFTER 07.3 validation passes)”. Added explicit note in 07.4 header that bug trackers must only be updated after 07.3 passes. Updated frontmatter sections list and success_criteria to reflect the new ordering.

[[TPR-07-007-gemini]] [LOW] section-04-codegen-llvm.md missing from 07.N 07.3 checklist item

Location: plans/empty-container-typeck-phase-contract/section-07-closeout.md:§07.N Reviewer: Gemini | Status: Fixed (subsumed by TPR-07-005-codex)

Evidence: The 07.N completion checklist for the bug-tracker step only listed bug-tracker/plans/BUG-04-074/ and 00-overview.md, omitting the section-04-codegen-llvm.md update that step 3 of the bug-tracker section explicitly requires.

Fix: Same as TPR-07-005-codex — the updated 07.N checklist for “07.4 complete” now explicitly includes bug-tracker/section-04-codegen-llvm.md entry resolved.

Round 4 — Dual-source TPR on sections 05, 06, 07 (Codex + Gemini). Findings addressed in this revision.

[[TPR-07-R4-001-codex]] [MEDIUM] /improve-tooling and /sync-claude dropped from 07.3 and 07.N

Location: plans/empty-container-typeck-phase-contract/section-07-closeout.md:§07.3 Reviewer: Codex | Status: Fixed

Evidence: 00-overview.md:165 and index.md:119 both specify Phase 6 / Section 07 close-out as /tpr-review → /impl-hygiene-review → /improve-tooling sweep → /sync-claude. After the 07.3/07.4 reorder in Round 3, the implementation of 07.3 and the 07.N checklist only included /tpr-review and /impl-hygiene-review. Both /improve-tooling and /sync-claude were silently dropped, leaving the section cross-inconsistent with the approved overview and creating two unanchored mandatory close-out tasks.

Fix: Added /improve-tooling and /sync-claude blocks to 07.3 body with instructions for each. Updated the success_criteria to name all five gates. Updated the 07.N “07.3 complete” checklist item to explicitly list /improve-tooling and /sync-claude.

07.R — Third Party Review Findings (Round 5 cap-exit — 2026-05-15)

Round 5 — /tpr-review cap-exit at cap_reached_max_rounds (5 rounds). Dual-source reviewer set (codex + gemini); opencode in transport failure → survivor mode 2-of-3. Three actionable findings filed below for resolution before §07 close-out.

• [TPR-07-R5-001-codex][HIGH] DRIFT — §07.4 atomic-flip transaction description stale: all four bugs already in closed-bugs.json

  • Location: section-07-closeout.md:352 (§07.4 body — “Atomic-flip discipline” paragraph)
  • Evidence: bug-tracker/closed-bugs.json already contains BUG-04-042 (line 68, status: resolved, resolved_date: 2026-05-14), BUG-04-074 (line 99), BUG-04-084 (line 119), BUG-04-085 (line 621) — each with plan_dir: bug-tracker/plans/completed/BUG-04-NNN/. bug-tracker/open-bugs.json contains no "id": lines for the four bugs. bug-tracker/section-04-codegen-llvm.md marks all four [x]. §07.4 body still describes the flip as a future transaction (“All four bug-tracker entries flip from open-bugs.json to closed-bugs.json”) even though §07.4 has status: not-started and the flips are already complete elsewhere.
  • Rule violated: .claude/rules/impl-hygiene.md — DRIFT; .claude/rules/state-discipline.md §4 (atomic-flip ownership, cross-file flip discipline)
  • Recommended fix: Rewrite §07.4 body to “verify already-complete flip state” (no-op transaction); list each bug + its closed-bugs.json line; add a verification checklist that confirms (a) all four in closed-bugs.json with status: resolved, (b) zero in open-bugs.json, (c) plan_dirs at bug-tracker/plans/completed/, (d) bug-tracker/_inflight/ empty. Update §07.4 title from “Bug-Tracker Updates — Close ALL FOUR Bugs Atomically” to “Bug-Tracker Updates — Verify ALL FOUR Bugs Closed Atomically”. Recompute counts from the JSON. Reconcile resolved-by-whom: file a HISTORY block at the bottom of 00-overview.md documenting which prior workflow / session performed the flip outside §07.4 scope (audit trail). NO content-level fix to closed-bugs.json itself required — the flip already executed correctly.

• [TPR-07-R5-002-codex][MEDIUM] PLAN_COHERENCE_DRIFT — 00-overview.md Quick Reference table shows §07 as Not Started but section frontmatter is status: in-review

  • Location: plans/typeck-inference-completeness/00-overview.md:390
  • Evidence: 00-overview.md:390 row reads | 07 | Close-out + Supersession (absorbs BUG-04-085 investigation) | section-07-closeout.md | Not Started |. section-07-closeout.md:5 carries status: in-review. index.md:128 carries Status: In Review. The umbrella overview is the only surface still showing Not Started.
  • Rule violated: .claude/rules/impl-hygiene.md — PLAN_COHERENCE_DRIFT
  • Recommended fix: Update 00-overview.md:390 Quick Reference row to | 07 | Close-out + Supersession (absorbs BUG-04-085 investigation) | section-07-closeout.md | In Review |. No body changes needed.

• [TPR-07-R5-003-gemini][MEDIUM] PLAN_COHERENCE_DRIFT:mixed-working-directory — §07.1 + §07.2.a bash blocks use wrapper-root-relative paths without compiler_repo/ prefix or cd compiler_repo

  • Location: section-07-closeout.md:146-150 (test-all.sh + cargo blocks); :154-174 (dual-exec-verify.sh blocks); :201-214 (rg sweep blocks)
  • Evidence: Per CLAUDE.md Repo Layout — “wrapper-root relative; compiler-owned carry compiler_repo/ prefix. Compiler harnesses (./test-all.sh etc.) run in compiler_repo/; wrapper cmds (python -m scripts.plan_corpus, …) from wrapper root.” The bash blocks at lines 146-150 invoke ./test-all.sh and cargo test, lines 154-174 invoke diagnostics/dual-exec-verify.sh tests/spec/..., and lines 201-214 invoke rg ... compiler/ ... / rg ... tests/ ... / rg ... library/ ... — all wrapper-root-relative. Adjacent §07.3 block at line 278 invokes python -m scripts.plan_corpus check plans/typeck-inference-completeness/ which DOES run from wrapper root. Implementer reading top-to-bottom hits an inconsistent working directory and either prefixes wrong or cds wrong.
  • Rule violated: .claude/rules/impl-hygiene.md §Finding Categories — PLAN_COHERENCE_DRIFT:mixed-working-directory
  • Recommended fix: Prefix each bash block with an explicit cd compiler_repo line (for blocks at 146-150, 154-174, 201-214) OR rewrite each path with compiler_repo/ prefix (compiler_repo/diagnostics/dual-exec-verify.sh compiler_repo/tests/spec/..., compiler_repo/compiler/, compiler_repo/tests/, compiler_repo/library/). Pick the cd compiler_repo form for consistency with ./test-all.sh invocation (which is naturally relative). Leave §07.3 python -m scripts.plan_corpus blocks unchanged (wrapper-root native). Add a one-line note above the first cd compiler_repo block: “Working directory: compiler_repo/ for compiler harnesses and tree-relative paths; wrapper root for python -m scripts.plan_corpus invocations.” for implementer clarity.

07.N Completion Checklist

• 07.PRE complete — every predecessor (§01, §02, §03, §04, §05, §06, §08, §09, §10, §11) carries status: complete + reviewed: true; python -m scripts.plan_corpus check plans/typeck-inference-completeness/ returns zero STRUCTURE:work-order-violation findings
• 07.0 complete — BUG-04-085 LLVM spec runner crash resolved (Path A §02.0 downstream consumer fixed, OR Path B §07.0.A residual cure shipped); timeout 150 ./test-all.sh no longer shows Ori spec (LLVM backend) CRASHED; 00-overview.md Known Bugs BUG-04-085 row + HISTORY updated
• 07.1 complete — ./test-all.sh green debug + release; cargo test --release -p ori_types green; cargo test --release -p ori_llvm green; dual-exec parity verified on the 8 representative paths covering §05 empty-list corpus + §09.3/§09.5 (str_to_error.ori) + §09.4 (iterator/methods.ori) + §10.1 (declarations/traits.ori) + §10.2 (capabilities/propagation.ori) + §11.1 (empty_literals/)
• 07.2.a complete — zero BUG-04-074|BUG-04-084|BUG-04-042|BUG-04-085|empty-container-typeck|typeck-inference-completeness hits in all four trees: compiler/*.rs, compiler/*.ori (AOT fixtures), tests/*.ori, library/*.ori
• 07.2.b complete — @mk_ok / @mk_err helpers removed from tests/spec/traits/traceable/definition.ori + tests/spec/traits/traceable/result_delegation.ori; both files green under direct-constructor form after §09.3 shipped; §06.2C table + §03.N ledger updated to reflect REMOVED (direct form, §09.3 closed gap)
• 07.2.c complete — R7-006 validator refinement shipped: site_from_expr_kind discriminates Lambda return-vs-param via VarPosition threading at validators/mod.rs:204 call site; positive + negative test pair landed; §06.1 wording success criterion updated
• 07.3 complete — plan_corpus check plans/typeck-inference-completeness/ passes; final /tpr-review clean (scope: §01–§06, §07.0, §07.0.A if any, §07.2.a/§07.2.b/§07.2.c, §08–§11); /impl-hygiene-review clean (same scope); /improve-tooling sweep clean (covers §07.0 tooling gaps + §07.2.c diagnostic infrastructure + §08–§11 surfaces); /sync-claude sweep clean (typeck.md §BD-2 + §EX-4 + §DI-1; canon.md §4.2 PC-2; bug-tracker/00-overview.md count)
• 07.4 complete — atomic flip via api.flip_bug_status for ALL FOUR bugs: bug-tracker/plans/completed/BUG-04-074/ + BUG-04-084/ + BUG-04-042/ + BUG-04-085/ each carry status: complete + plan pointer + resolved_in: SHA; bug-tracker/open-bugs.json no longer contains the four; bug-tracker/closed-bugs.json contains all four; bug-tracker/_inflight/ empty; bug-tracker/00-overview.md open-count decremented by 4; bug-tracker/section-04-codegen-llvm.md entries marked [x] with plan pointers
• This section’s own status flipped from in-review to complete in frontmatter via flip_from_in_review_clean() per state-discipline.md §4 (after umbrella 00-overview.md close-out completes — final flip in line below)
• plans/typeck-inference-completeness/index.md updated: all sections show status: complete; umbrella 00-overview.md status: flipped from in-progress to complete

Exit criteria: BUG-04-074, BUG-04-084, BUG-04-042, BUG-04-085 all formally closed. The typeck inference completeness plan is complete. Empty-container typeck phase-contract enforcement, cross-module pool-merge var_id collision, LLVM spec-runner crash on monomorphized assert_eq, body-inference BD-2 gaps (§09), dispatch-on-rigid-receiver gaps (§10), polymorphic-constructor defaulting fallback (§11) — all no longer block the Ori compiler.