100%

FOLD-IN + SUPERSEDE + GRIND-HERE (plan-wide standing directive — see overview Mission §Bug Fold-In + Supersession, ABSOLUTE). Every failing burden/RC-emission floor cell this section touches is worked HERE as part of this plan’s grind; any BUG-XX-NNN covering it, including BUG-04-118, BUG-04-142, BUG-04-156, the #150 same-allocation-identity class, and every sibling RC-emission bug, is FOLDED INTO this plan and SUPERSEDED — NEVER deferred to / routed away as separate bug tracking, a “separate architecture-level pass”, or “already tracked elsewhere”. Dead-end / deferred-compound / reclassified classifications (decisions/LEDGER.md §6) shape WHAT the fix is; they NEVER exit the cell from this plan. If the correct fix is an architecture-level pass, building it IS this plan’s work. Calculus-first per arc.md §CP-1; one terminal gate = ./test-all.sh zero failing cells.

RC-REMARKS CO-EVOLUTION (plan-wide standing directive — see overview Mission §8, ABSOLUTE). This grind is the rc-remarks RC-survivor observability system’s first real test: every floor cell / surviving-RC cluster this section works MUST be grounded via the burden-sole rc-remark stream (ori build --emit-rc-remarks + intel-query.sh rc-survivors --by-cause / rc-cause-rank) before hypothesizing a fix, AND when the stream/analyzer/query falls short for a real cell, improving rc-remarks (producer ori_arc/.../rc_remark.rs, analyzer tools/ori-rc-remarks, intel import_rc_remarks.py + query_graph.py rc-* handlers) IS this section’s work for that cell — never deferred as separate observability work. Burden-sole verdict discipline (arc.md §STOP) unchanged.

Section 06: Phase 7 Mechanical Lowering

Goal: Phase 7 has no semantic predicate stack. AIMS decides logical owner-credit/release events and mutation obligations before this boundary; each physical target mechanically selects an encoding that satisfies those frozen facts.

Vocabulary boundary: Burden* is the transitional Phase-5/6 carrier for logical ownership events. Rc*, CowMode, IsShared, copying, and runtime calls are current Phase-7 compiled-adapter choices, never AIMS calculus outputs. VM, compiled, native-WebAssembly, and JIT planners consume the same facts without re-analysis.

Context: Per proposal §Design — Phase 7 (Realization) — Mechanical Lowering. realize_rc_reuse becomes thin pass; realize_annotations (Phase 2 post-merge) stays where it is.

Reference implementations: Existing realize_rc_reuse for shape; existing realize_annotations for COW/reuse/drop-hint/arg-ownership.

DAG predecessor (frontmatter depends_on: ["05"]): §05 — the single immediately-prior section in the §05 → §06 → §07 → §09 chain, per the INV-19 linear-DAG single-predecessor invariant.

Consumes outputs from (transitively available through the §05 linear-chain predecessor — INPUTS §06 reads, NOT additional DAG predecessors; declaring them in depends_on: would form a diamond and violate INV-19):

  • §03 (logical additional-credit/release events emitted through transitional Burden* carriers at Phase 5 — the surface §06 mechanically lowers).
  • §04A (minimal lattice adaptation — Phase 6 logical-event elimination surface §06.2 compound obligation lowering reads from).
  • §05.2 (Phase 6 lattice rewrite produces the optimized burden baseline with DP-2/DP-3 annotations).
  • §02 (compute_drop_info wrapper consumed by element_fn_gen + instr_dispatch).

Predecessor gate — §04B Prototype Gate: Per codex Finding 1 + §04B.N decision table: any §04B criterion FAIL halts the cascade §05 → §06 → §07-§10. §06 is conditional on §04B.N decision = PASS. If §04B.N flips FAIL, §06 voids alongside §05 and the proposal-rescope branch (direct Perceus per 00-overview.md §Alternative 1) supersedes this section.

Intelligence Reconnaissance

Queries:

  • scripts/intel-query.sh --human file-symbols "compiler_repo/compiler/ori_arc/src/aims/realize" --repo ori
  • scripts/intel-query.sh --human callers realize_rc_reuse --repo ori
  • scripts/intel-query.sh --human callers class_alive_after --repo ori — confirm zero callers post-§09

Queried: 2026-05-08.

Results summary [ori] — concrete symbol inventory grounding §06 implementation:

  • realize_rc_reuse entry point: compiler_repo/compiler/ori_arc/src/aims/realize/mod.rs (per arc.md §Pipeline Step 5).
  • Predicate-stack query sites to be eliminated in §06.1 (per §00-overview Mission paragraph identifying class_alive_after, pin4_class_emits_dec_set, pin6_any_ancestor_will_cover, class_payload_of, ssa_alias_classes-driven emission): currently resident across compiler_repo/compiler/ori_arc/src/aims/realize/{mod.rs, walk.rs, walk_dec.rs, emit_unified.rs, burden_elim.rs} and compiler_repo/compiler/ori_arc/src/aims/emit_rc/dead_cleanup/emission_site.rs. §06.1 rewrites the lowering site realize/mod.rs and grep-enforces zero residual call sites THERE post-rewrite. The §09-retiring sibling files (walk.rs, walk_dec.rs, emit_unified.rs, emit_rc/dead_cleanup/emission_site.rs) retain their predicate definitions + call-sites in-tree until §09 wholesale-retires them per 00-overview.md absorbed_bugs[BUG-04-118]; the whole-realize/-dir zero-reference assertion is the §09 close-gate, NOT the §06 gate (a §06 grep over the whole dir would be unsatisfiable pre-§09).
  • BurdenInc/BurdenDec instruction surface §06 lowers: defined under compiler_repo/compiler/ori_arc/src/ir/instr.rs (per agy Finding 1 evidence) and emitted at §03 sites per §03 outcome HISTORY.
  • Compound-instruction lowering sites stay in realize_annotations Phase 2 post-merge per arc.md §Pipeline Step 10 + aims-rules.md §7 PL-4.

No parallel paths, no shadow trackers — extension within the unified model per missions.md §AIMS invariant 5.

06.1 realize_rc_reuse mechanical rewrite

  • DELIVERED (RL-4 live-out BurdenDec-placement matrix, burden_lower/tests.rs): conditional-transfer (owned param live-out, Branch), multi-block live-out (FRESH, Jump) net-0, dead-out-keeps-dec narrowness pin. Semantic (verify_burden_balance empty) + negative (per-block dec count) pairing — pins the faithful Phase-5 emission §06.1 shipped (the lowering SITE deliverable). This IS the §06.1 test deliverable; the real-RC FileCheck-on-RcDec pins moved to §09.2 with the real-RC activation.

§06.1 delivers the clean lowering SITE only. realize/mod.rs is established as the thin dispatcher verified zero of the 6 predicates; faithful Phase-5 emission (VF-1=0) + mirror removal + grep-zero are DONE. The real-RC ACTIVATION (BurdenInc→RcInc / BurdenDec→RcDec flipping from codegen no-ops to real RcInc/RcDec) is NOT a §06.1 deliverable: while the predicate stack (walk_dec.rs/emit_unified.rs) still emits its own RcInc/RcDec, flipping burden ops to real RC double-emits → double-frees (LEDGER §B.2). The activation is the sole-RC-emitter outcome of §09.2’s predicate-stack retirement — relocated to §09.2 per routing.md §4 (move the item to its owning section, NOT a <!-- blocked-by --> annotation). See §09.2 checkboxes “BurdenInc→RcInc / BurdenDec→RcDec mechanical mapping” + its real-RC FileCheck pins.

Subsection close-out (06.1) per protocol.

  • Remove/neutralize compiler_repo/compiler/ori_arc/src/aims/realize/burden_mirror.rs::reconcile_burden_ledger (the coexistence-only masking balancer) BEFORE whole-var BurdenInc→RcInc / BurdenDec→RcDec lowering lands — per §03A.3a forward-anchor. Its padding (inc-injection / terminal-dec / merge-equalize beyond the consensus dec-mirror) is safe TODAY only because whole-var burden ops are codegen no-ops (instr_dispatch.rs:509); once they lower to real RC the padding becomes SPURIOUS RcInc/RcDec → reintroduced double-frees (unless Phase-6 DP-2/DP-3 prunes the adjacent pairs first). Sequencing gate: mirror neutralized + VF-1 still clean via faithful Phase-5 emission BEFORE the mechanical mapping below. DONE (2026-06-02 a08): VF-1=0 corpus-wide via faithful Phase-5 emission (RL-4 live-out fix), THEN reconcile_burden_ledger + burden_mirror.rs + burden_mirror/tests.rs + BURDEN_MIRROR_DISABLED + ORI_DISABLE_BURDEN_MIRROR deleted; zero references corpus-wide.
  • Verify zero references to class_alive_after, pin4_class_emits_dec_set, pin6_any_ancestor_will_cover, class_payload_of, canonical_rep_for, var_emits_dec_in_block, syntactic-liveness fallback within compiler_repo/compiler/ori_arc/src/aims/realize/mod.rs (the rewritten lowering site); grep-enforced at section close. SCOPE NARROWED to realize/mod.rs (not the whole realize/ dir): the §09-retiring predicate definitions + call-sites intentionally REMAIN inside realize/ at walk_dec.rs + walk.rs + emit_unified.rs + emit_rc/dead_cleanup/emission_site.rs per 00-overview.md absorbed_bugs[BUG-04-118] until §09 wholesale-retires them — a grep over the whole realize/ dir CANNOT return zero pre-§09 and would make this gate unsatisfiable. realize/mod.rs is verified clean of all six predicates today; the §06.1 rewrite preserves that. (The whole-realize/-dir zero-reference assertion IS the §09 close-gate, not the §06 close-gate.) DONE (2026-06-02 a08): grep -nE 'class_alive_after|pin4_class_emits_dec_set|pin6_any_ancestor_will_cover|class_payload_of|canonical_rep_for|var_emits_dec_in_block' realize/mod.rs returns zero.

06.2 Compound instruction lowering (COW / reuse / drop-hint / arg-ownership)

Design intent — WHICH phase emits WHAT (resolves the Phase-6-fact vs Phase-7-lowering site ambiguity): Phase 7 mechanically maps logical owner-credit/release events to target operations; the current compiled adapter uses BurdenInc→RcInc / BurdenDec→RcDec (§06.1). It does not make new COW/reuse/drop-hint semantic decisions. Compound realization stays at the target-owned sites; those sites consume the CONVERGED Phase-6 facts (Uniqueness, Shape, borrow safety) and satisfy the frozen COW/reuse/drop-hint obligations without re-analysis. Arg ownership is a separate consumer at the Step-4b prelude.

Canonical COW input: DP-9 exposes cow_obligation, computed by cow_obligation_from_uniqueness and refined by borrow safety: SameIdentityPermitted, SharingObservationRequired, or AliasIsolationRequired. The current compiled compatibility adapter maps these obligations through decide_cow/CowMode to StaticUnique/Dynamic/StaticShared, with IsShared only for the sharing-observation projection.

SITE-vs-ACTIVATION + annotation-surface correction (2026-06-02, a09 — see §06 HISTORY): the original §06.2 design intent assumed Phase 6 lands distinct “COW/reuse/drop-hint MARKERS on surviving burden ops” and §06.2 wires a Step-10 consumer to read those markers. The actual proven AIMS architecture carries NO such surface and MUST NOT: the burden-op variants (BurdenInc/BurdenDec/BurdenDecPartial/BurdenDecField/BurdenDecVariant at ir/instr.rs) carry ONLY var/field/base — adding a compound-annotation field would be a parallel SoT / shadow uniqueness tracker, banned by aims-rules.md Invariant-5. COW/reuse/drop-hint are NOT marked on burden ops; they are read from the converged lattice DIMENSIONS at the realize sites (DP-4/DP-5/DP-6/DP-9; RL-7/RL-9/RL-11/RL-11a/RL-12). §05.2’s dimension-wiring audit established exactly this (“each consumer reads the converged lattice dimension… burden-transparent. Role survives the emission→elimination shift”). The §06.2 SITE deliverable is therefore the consumers reading converged dimensions at the correct sites — which are wired + active TODAY, not a new annotation-reader. NO §09.2 RELOCATION applies to §06.2: the compound instructions (COW Set/IsShared/copy via cow_annotations, Reset/Reuse, drop-hints) are the long-standing dimension-driven realizations that emit REAL LLVM instructions today (ori_llvm/.../arc_emitter/emitter_utils.rs consumes CowMode for real in-place-vs-copy emission). They were NEVER burden-op codegen no-ops, so flipping them carries NO double-emit hazard against the live predicate stack — the LEDGER §B.2 double-emit hazard that relocated the §06.1 whole-var BurdenInc→RcInc flip to §09.2 applies ONLY to the whole-var burden ops, NOT to these compound realizations.

Subsection close-out (06.2) per protocol.

  • COW SITE consumer reads converged Uniqueness plus borrow-disjointness at the existing realize_annotations Step-10 site and consumes DP-9 cow_obligation: SameIdentityPermitted, SharingObservationRequired, or AliasIsolationRequired. It is verified wired + active and does not read a transitional burden-op marker. The current compiled compatibility adapter then maps the obligation through decide_cow to CowMode; that adapter is projection evidence, not the semantic seam.

  • Reuse SITE consumer reads converged Shape+Uniqueness at the existing Step-5 realize site (realize/decide.rs:decide_reuse + emit_reuse/{detect,dynamic,planner}.rs; DP-6 + RL-11/RL-11a/RL-12) — verified wired + active today.

  • Drop-hint SITE consumer reads converged Uniqueness+collection-shape at realize_annotations Phase 2 post-merge (realize/decide.rs:decide_drop_hint, unchanged location per aims-rules.md §7 PL-4) — verified wired + active today.

  • arg-ownership placement per aims-rules.md §7 PL-5 + §3 TF-11 IC-3: ArcParam.ownership populated at Step 2 apply_aims_ownership (pipeline/aims_pipeline/batch.rs) from MemoryContract; call-site arg_ownership populated at Step-4b-prelude emit_arg_ownership (pipeline/aims_pipeline/mod.rs) — NOT colocated with COW/reuse/drop-hint at Step 10. Verified wired + active today. Plan-frontmatter goal-line phrasing collapsing “arg-ownership emission” into “stay in realize_annotations Phase 2 post-merge” is an inherited drift; §08 rule-file sync section closes the goal-line drift at plan close-out.

  • Tests: SITE-consumer pins per aims-rules.md §9 VF-1+VF-2+VF-3 + tests.md §Matrix Testing Rule — the compound-realization SITE is pinned by existing matrix coverage (no new annotation-reader to pin; pinning a non-existent marker-reader would be test-inflation per tests.md §Anti-patterns):

    • COW: canonical cow_obligation matrix (realize/tests.rs, DP-9 × Uniqueness × borrow-disjointness, semantic + negative pairing) pins all three obligations. Existing decide_cow/CowMode assertions and FileCheck codegen (ori_llvm/tests/codegen/cow/{unique_mutation_no_copy,shared_mutation_triggers_copy,mutation_triggers_store_ops}.ori + cow_is_shared_check.ori, RL-9 contraction) remain current compiled-projection evidence. tests/spec/collections/cow/ pins eval/LLVM behavior parity with leak checks.
    • Reuse: decide_reuse candidacy pins (realize/tests.rs, DP-6 × Shape × Uniqueness) + emit_reuse detect/dynamic/planner + AOT realize_rc_reuse.rs (RL-11 same-block / RL-11a dynamic). Cross-block RL-12 emission is exercised by the AOT corpus; an explicit cross-block FileCheck pin is the remaining additive coverage if a gap surfaces at §06 close-gate.
    • Drop-hint: decide_drop_hint matrix (realize/tests.rs, Uniqueness × collection-shape × borrowed-param × rc-incremented, semantic + negative pairing).
    • arg-ownership: arg_ownership/tests.rs (ArgOwnership::Owned/Borrowed at indirect calls, mono-name merge) + apply_aims_ownership exercised pipeline-wide. ArcParam.ownership population verified at apply_ownership Step 2.

06.3 Atomic vs non-atomic realization

Backend-neutral supersession: RL-19/20/21 now freeze ThreadReachability in the shared AIMS calculus and impose separate VM and compiled satisfaction obligations. AIMS does not select atomic instructions, runtime helpers, headers, or an RcAtomicity value. The neutral theorem split is green at 130/130 dual-discharge obligations, the section-08 suite is green at 40 proofs (18 shipped neutral-conformance proofs plus 20 target-only proofs), and the exact VM profile is green at 1/1. The remaining production work belongs to plans/repr-opt/section-10-thread-local-arc.md: generate and bind the neutral facts by stable identity, then validate each physical planner’s synchronization choice against them. The June carrier record below is retained as migration provenance only; RcAtomicity is not the production cross-backend seam.

Historical SITE-vs-ACTIVATION record (2026-06-02, a10 — see §06 HISTORY): §06.3 delivered the RcAtomicity CARRIER SITE only. Grounding the original §06.3 design against actual code established that the then-planned RL-19/20/21 dispatch that populated NonAtomic could not land there: the shipped runtime RC primitives (ori_rc_inc/ori_rc_dec at compiler/ori_rt/src/rc/mod.rs) were unconditionally atomic (AtomicI64::fetch_add/fetch_sub); there was no non-atomic runtime path, no emit_rc_inc/emit_rc_dec atomicity branch, and no thread-escape analysis in ori_arc. Under that superseded design, the thread-escape analysis (ThreadLocality in compiler/ori_repr/src/escape/thread.rs), non-atomic runtime (ori_rc_inc_nonatomic/ori_rc_dec_nonatomic), whole-program RL-21 optimization, and atomic-vs-non-atomic codegen selection were relocated together to plans/repr-opt/section-10-thread-local-arc.md (§10.1 + §10.2). The carrier was observably inert and default-Atomic, as confirmed by VF-1=0 and byte-identical AIMS snapshots.

  • The historical atomic-vs-non-atomic activation and its tests were relocated to plans/repr-opt/section-10-thread-local-arc.md §10.2. Production §10 now replaces that carrier-population design with neutral ThreadReachability fact production plus independent VM and compiled physical-plan validation. The channel-send #skip companion remains tracked as BUG-02-037.

Subsection close-out (06.3) per protocol.

  • Define RcAtomicity enum per codex Finding 2 BEFORE atomic-vs-non-atomic dispatch: variants { Atomic, NonAtomic } in compiler_repo/compiler/ori_arc/src/ir/instr.rs (carrier on RcInc/RcDec ArcInstr variants). DONE (2026-06-02 a10): RcInc/RcDec carry atomicity: RcAtomicity; every construction site sets RcAtomicity::default_atomic() (= Atomic), reproducing the shipped unconditionally-atomic runtime RC primitives bit-for-bit. Exported from ir/mod.rs + crate root. The NonAtomic-SELECTING dispatch (RL-19/20/21 thread-locality derived from Locality + call-graph per RL-18a) is RELOCATED to plans/repr-opt/section-10-thread-local-arc.md §10.2 per routing.md §4 — that section owns the thread-escape analysis (§10.1) + non-atomic runtime (§10.2) the dispatch requires.
  • RcAtomicity representation (discriminant placement / layout for the carrier on lowered RcInc/RcDec, per RL-18/RL-19) in compiler_repo/compiler/ori_arc/src/ir/repr.rs alongside the existing RcStrategy repr — this is the in-body rationale for the ir/repr.rs entry in touches:. DONE (2026-06-02 a10): RcAtomicity enum + default_atomic() + is_atomic() + fmt_atomicity_suffix() (renders "" for Atomic — byte-identical AIMS-snapshot output — " [non-atomic]" for NonAtomic) in ir/repr.rs + ir/format/mod.rs.
  • Tests (SITE pins — carrier definition + default; the NonAtomic-SELECTION tests relocated to §10.2 with the dispatch): per tests.md §Matrix Testing Rule semantic + negative pairing — compiler/ori_arc/src/ir/repr/tests.rs: rc_atomicity_default_is_atomic (semantic pin — default_atomic() == Atomic; reverting the default to NonAtomic fails it, guarding against silent non-atomic emission before the §10.2 runtime exists), rc_atomicity_is_atomic_distinguishes_variants, rc_atomicity_atomic_renders_no_suffix (Atomic dumps byte-identically — AIMS snapshots depend on it), rc_atomicity_non_atomic_renders_suffix (negative-space pin — the dimension IS observable once §10.2 selects NonAtomic). The realize-path all-Atomic invariant is pinned corpus-wide by the AIMS snapshot suite (any realize site emitting NonAtomic would surface the [non-atomic] suffix in a baseline).
  • TPR checkpoint — /tpr-review AND /impl-hygiene-review covering 06.1-06.3. Map the /tpr-review exit reasons per the decisions/31 Option C exit-reason table (plans/completed/scripts-first-workflow-architecture/_archive/2026-05-15-pre-fold/skill-ecosystem-coherence/decisions/31-step-6-exit-reason-table-source.md): clean / resolved advance the subsection; cap_reached_with_substantive / cap_reached_clean route per that table rather than stalling the autopilot state-machine on an unclassified exit.

06.R Third Party Review Findings

3-round /tpr-review --skill review-plan --autonomous (exit_reason cap_reached_with_substantive); all findings CURED INLINE this session — zero open, zero deferred.

06.N Completion Checklist

  • /tpr-review passed (final); /impl-hygiene-review passed. /tpr-review: §06.3 checkpoint 3-round cap_reached_with_substantive, third_party_review.status=resolved, 4 findings cured. /impl-hygiene-review: ran 2026-06-03 (run impl-hygiene-ori_lang-qmsMg2rF), all 4 load-bearing invariants clean (Invariant-5/RcAtomicity/is_release_cleanup_instr/RL-4), 1 §06-attributable Major P3-F1 fixed (10 ephemeral test renames), residual BLOAT+cross-scope recorded additively to §06.R.H, zero open Critical/Major §06-attributable.
  • Phase 7 has zero predicate-stack queries. Mechanical lowering only. (2026-06-03 re-verify: grep-zero on realize/mod.rs for class_alive_after / pin4_class_emits_dec_set / pin6_any_ancestor_will_cover / class_payload_of / canonical_rep / syntactic-liveness — zero matches. Narrowed to realize/mod.rs per §06.1 opencode-F1; §09-retiring sibling files keep predicate sites until §09.)
  • eval + LLVM dual-execution parity preserved. (2026-06-03 re-verify: eval cargo st 4600 passed / 45 failed / 51 skipped AND LLVM AOT 2322 passed / 16 failed — BOTH at documented baseline, zero §06-attributable divergence. §06 touches the ori_arc→ori_llvm realization path only; both backends sit at baseline so no §06-introduced eval-vs-LLVM divergence. The 45 eval failures are the pre-existing type-error-blocked frontend baseline; the 16 AOT failures are the §09-coupled cohort — neither is §06 burden/RC.)
  • cargo t green; debug + release. (2026-06-03 re-verify: ori_arc lib 1539 passed / 0 failed; ori_llvm AOT debug 2322/16 AND release 2322/16 — byte-identical FastISel parity, zero NEW failures in either profile.)
  • timeout 150 ./test-all.sh — full backend-spec + LLVM gate, scoped to NO-NEW-failures-beyond-baseline per LEDGER §B.3 (the §06/§09-coupled baseline is AOT 2284/19 = 17 BUG-04-123 predicate-stack double-frees + 2 VF-1 loop residuals BUG-04-121, ALL coupled to §06 whole-var lowering + §09 retirement). §06 close requires zero NEW failures; the 19 baseline failures DISSOLVE at §09 (predicate-stack retirement) + §10/§12 verify, NOT at §06 close. A NEW failure introduced by §06 lowering blocks close per CLAUDE.md §Failing Tests Mid-Work. (2026-06-03 re-verify, run as components because the monolithic harness exceeds the 150s test cap and the AOT target needs LLVM-21 on PATH: full backend-spec cargo st 4600/45/51 baseline + LLVM AOT debug+release 2322/16 — zero NEW failures vs the §06-a08 2322/16 baseline. Relative to the original 2284/19 baseline, AOT passes rose +38 and failures dropped −3: §06 is strictly better, never regressive.)
  • Plan annotation cleanup. (2026-06-03: §06 body carries no stale <!-- blocked-by --> / TODO / FIXME / placeholder annotations; the only blocked-by mentions are HISTORY-prose documentation of the §06.1→§09.2 real-RC-activation relocation, a permitted-prose surface.)
  • Plan sync done; mission criterion-9 flip is §09.2’s deliverable, NOT §06’s. (2026-06-03: 00-overview criterion 9 (line 106) already routes SITE→§06 / ACTIVATION→§09.2 and states “Flips [x] at §09.2 close, NOT §06 close”; §06’s compiler-derived doc surfaces — arc.md ORI_DISABLE_BURDEN_MIRROR / ORI_DUMP_AFTER_BURDEN debug entries + the burden-balance diagnostic — landed with the §06 work in commit f2d078ff4. The full aims-rules.md / canon.md §7.1 / missions.md §AIMS sync is §08’s deliverable per 00-overview line 108. §06 does NOT flip criterion 9 — flipping it here would prematurely claim the real-RC ACTIVATION done before §09.2 retires the predicate stack.)

06.R.H — Hygiene Findings

Already resolved this review run

  • RESOLVED: [BLOAT:decorative-banner — 22 instances] compiler/ori_arc/src/aims/verify/burden_balance/tests.rs — hygiene-lint --fix --apply stripped 22 decorative comment banners. Auto-fixed in Phase 0.
  • RESOLVED: [STRUCTURE:co-evolution — postprocess.rs:75] compiler/ori_arc/src/pipeline/aims_pipeline/postprocess.rs:75 — mangled doc-comment per \`` + /// Surface` Layer 1` dangling-backtick fragment (C-B13 non-spec-pointer leak). Rewrote to self-contained present-tense description of verify/debug_assertions gating. Fixed in Phase 0.5.
  • RESOLVED: [STRUCTURE:removal-trail — project_escape.rs:70] compiler/ori_arc/src/aims/realize/project_escape.rs:70 — C-B3 change-trail // Previously this used a single parent->project_dst map … Now each arg gets its own RcInc. Rewrote to present-tense WHAT-only. Fixed in Phase 0.5.
  • RESOLVED: [LEAK:swallowed-error — emit_unified.rs:590] compiler/ori_arc/src/aims/realize/emit_unified.rs:590 — .unwrap_or_default() mechanical match. Documented-invariant cure: added // INVARIANT: comment (legitimate contract-absence per IC-5/TF-6 CONSERVATIVE default; NOT lossy fallback). Fixed in Phase 0.5.
  • RESOLVED: [LEAK:swallowed-error — emit_unified.rs:611] compiler/ori_arc/src/aims/realize/emit_unified.rs:611 — .unwrap_or_default() mechanical match. Same documented-invariant cure as :590. Fixed in Phase 0.5.
  • RESOLVED: [NAMING:provenance-leak — burden_lower/tests.rs — 10 sites] compiler/ori_arc/src/lower/burden_lower/tests.rs lines 510, 602, 1921, 2547, 3669, 433, 1942, 1991, 2040, 2074 — 10 test function names carrying ephemeral plan-section/pair-cycle IDs (_per_03_4_cycle_50b, _cycle_42, _cycle_46, _per_03_3_rule_5, etc.) banned per impl-hygiene.md §Test Function Naming. Renamed in-place; provenance moved to /// doc comments. test-naming-lint clean, 57 tests green. Fixed post-Phase-3 (commit 6ba573306). (P3-F1)
  • RESOLVED: [NAMING:litter-pickup — closure_drop.rs + ir/tests.rs] compiler/ori_llvm/src/codegen/function_compiler/closure_drop.rs doc-ref + compiler/ori_arc/src/ir/tests.rs sibling — two litter-pickup renames alongside P3-F1 cure. Fixed post-Phase-3. (P3-F1 litter-pickup)

Open — §06-attributable BLOAT (additive, opportunistic split when next touched)

  • DEFERRED (anchored): [BLOAT:file-length — walk_dec.rs] compiler/ori_arc/src/aims/realize/walk_dec.rs — 960 lines (+460 in f2d078ff4). PIN-4/PIN-5/PIN-6 SSA-alias-class dec-suppression flow + §06 Burden* pass-through arm. Anchor: this file is owned by §09 retirement; split anchor = §09 retire-file close-gate (extract PIN-4/PIN-5 class-suppression filter into sibling submodule at §09 retirement). Opportunistic until §09. (P3-F3)
  • DEFERRED (anchored): [BLOAT:file-length — edge_cleanup.rs] compiler/ori_arc/src/aims/emit_rc/edge_cleanup.rs — 804 lines (+304 in f2d078ff4). Added release_with_burden / release_with_burden_edge / release_with_burden_into_block family (coexistence handshake). Anchor: opportunistic-split-when-next-touched per impl-hygiene.md §File Organization; extract release_burden.rs sibling for the three helpers. (P3-F11)
  • DEFERRED (anchored): [BLOAT:file-length — burden_lower/mod.rs] compiler/ori_arc/src/lower/burden_lower/mod.rs — 788 lines (+288 in f2d078ff4). Suppression-set helpers + RL-4 liveness pair. Anchor: opportunistic-split-when-next-touched; extract suppression helpers → suppression.rs, RL-4 liveness pair → live_out.rs. (P3-F2)
  • DEFERRED (anchored): [BLOAT:file-length — emit_unified.rs] compiler/ori_arc/src/aims/realize/emit_unified.rs — 778 lines (+278 in f2d078ff4). §09-retiring file. Anchor: §09 retirement; file retired wholesale at §09, split not required pre-§09. Opportunistic until §09. (P3-F4)
  • DEFERRED (anchored): [BLOAT:file-length — dead_cleanup/mod.rs] compiler/ori_arc/src/aims/emit_rc/dead_cleanup/mod.rs — 629 lines (+129 in f2d078ff4). §06 touched the dead-cleanup path for burden-paired release. Anchor: opportunistic-split-when-next-touched. (P3-F10)
  • DEFERRED (anchored): [BLOAT:file-length — walk.rs] compiler/ori_arc/src/aims/realize/walk.rs — 733 lines (+233 in f2d078ff4). §09-retiring file. Anchor: §09 retirement. Opportunistic until §09. (P3-F5)
  • DEFERRED (anchored): [BLOAT:file-length — burden_lower/emit.rs] compiler/ori_arc/src/lower/burden_lower/emit.rs — 559 lines (+59 in f2d078ff4). Per-instr + per-terminator burden emission. Anchor: opportunistic-split-when-next-touched; extract emit_terminator.rs sibling. (P3-F6)
  • DEFERRED (anchored): [BLOAT:file-length — ir/instr.rs] compiler/ori_arc/src/ir/instr.rs — 543 lines (+43 in f2d078ff4). §06 added Burden* variants + helpers. Anchor: opportunistic-split-when-next-touched; methods could move to instr/methods.rs if growth continues. (P3-F7)
  • DEFERRED (anchored): [BLOAT:file-length — ir/mod.rs] compiler/ori_arc/src/ir/mod.rs — 513 lines (+13 in f2d078ff4). Marginally over cap. Anchor: opportunistic-split-when-next-touched. (P3-F8)
  • DEFERRED (anchored): [BLOAT:file-length — realize/decide.rs] compiler/ori_arc/src/aims/realize/decide.rs — 512 lines (+12 in f2d078ff4). Marginally over cap. Anchor: opportunistic-split-when-next-touched. (P3-F9)

Open — cross-scope record-only (owned by concurrent parallel-session AIMS work; NOT §06 close-blockers)

  • CROSS-SCOPE (parallel-session-owned): [INVERTED-TDD:ghost-test — Critical — intraprocedural/tests.rs:311] compiler/ori_arc/src/aims/intraprocedural/tests.rs:311 — analyze_function_terminates_on_back_edge_loop runs but assertion never checked. file NOT touched by f2d078ff4; owned by the concurrent parallel-session AIMS work, NOT §06’s deliverable. Owning session must add missing assertion or remove ghost test. NOT a §06 close-blocker.
  • CROSS-SCOPE (parallel-session-owned): [INVERTED-TDD:ghost-test — Critical — intraprocedural/tests.rs:3460] compiler/ori_arc/src/aims/intraprocedural/tests.rs:3460 — compute_project_alias_sources_cross_block_let ghost test. same cross-scope owner as above. NOT a §06 close-blocker.
  • CROSS-SCOPE (parallel-session-owned): [INVERTED-TDD:ghost-test — Critical — intraprocedural/tests.rs:5283] compiler/ori_arc/src/aims/intraprocedural/tests.rs:5283 — negative_pin_no_synthetic_context_hole_typeid_minting_pathway_exists ghost test. same cross-scope owner. NOT a §06 close-blocker.
  • CROSS-SCOPE (parallel-session-owned): [INVERTED-TDD:ghost-test — Critical — ir/validate/tests.rs:49] compiler/ori_arc/src/ir/validate/tests.rs:49 — unresolved_type_var_is_copy ghost test. file NOT touched by f2d078ff4; cross-scope parallel-session owner. NOT a §06 close-blocker.
  • CROSS-SCOPE (parallel-session-owned): [LEAK:catch-all-arms — Major — operators/mod.rs:159,339] compiler/ori_llvm/src/codegen/arc_emitter/operators/mod.rs lines 159 + 339 — two _ => unreachable!() catch-all arms (float unary op; Lt/Gt/LtEq/GtEq comparison). file NOT in §06 commit; cross-scope parallel-session owner. NOT a §06 close-blocker.
  • CROSS-SCOPE (parallel-session-owned): [LEAK:swallowed-error — Major — 5 cross-scope sites] compiler/ori_arc/src/aims/intraprocedural/block.rs:142, intraprocedural/post_convergence.rs:70, compiler/ori_llvm/src/codegen/arc_emitter/construction.rs:120, arc_emitter/drop_enum.rs:397, arc_emitter/rc_helpers.rs:455 — five .unwrap_or_default() sites NOT in f2d078ff4; unverified (may be legitimate-absence or genuine lossy-fallback). cross-scope parallel-session owner must classify each. NOT a §06 close-blocker.
  • CROSS-SCOPE (parallel-session-owned): [PLAN_ROUTING_DRIFT:stale-plan-annotation — Minor — generics.rs] compiler/ori_llvm/tests/aot/generics.rs — 6 stale-resolved TPR annotations (TPR-04-002/003/007) referencing repr-opt §04. file NOT in §06 commit; cross-scope (repr-opt §04 owner). NOT a §06 close-blocker.
  • CROSS-SCOPE (parallel-session-owned): [PLAN_DELIVERY_DRIFT:MISSING_PROPAGATION — Minor — bug-tracker/plans/completed/] 16 closed bug plans missing delivered_to_roadmap field (Phase 0 delivery-drift). pre-existing cross-scope issue; no §06 bug plans closed in f2d078ff4; tracker-maintenance owner. NOT a §06 close-blocker.

Informational (non-blocking; no action required)

  • NOTE [emit_unified.rs:590,611] Two .unwrap_or_default() hygiene-lint hits are mechanical false positives: contracts.get(&func.name) returns Option<&MemoryContract>; None is legitimate absence (TF-5/TF-6 CONSERVATIVE default for absent contract per IC-5). The // INVARIANT: comments (Phase 0.5 cure) are architecturally correct; residual lint hit is a comment-unaware syntactic detector. No action. (P3-F12)
  • NOTE [walk.rs:82,426] BurdenInc/BurdenDec pass-through in Phase-7 realize walk is correct §06 SITE-establishment (markers preserved verbatim; codegen no-op at instr_dispatch.rs:519). Real-RC activation relocated to §09.2 per LEDGER §B.2 double-emit hazard. NOT a GAP. (P3-F13)
  • NOTE [instr_dispatch.rs:556,606] debug_assert!(false) arms guarding BurdenDecPartial-on-non-struct and BurdenDecVariant-on-non-enum are sound cross-crate defense-in-depth (not impossible-state WASTE): compute_drop_info return shape is not statically guaranteed to match the burden-op variant at the codegen site. (P3-F14)
  • NOTE [postprocess.rs:75,88-89] Phase 0.5 co-evolution cure (mangled doc-comment) verified clean post-fix. No action. (P3-F15)
  • NOTE [project_escape.rs:70] Phase 0.5 removal-trail cure verified clean post-fix (present-tense WHAT-only). No action. (P3-F16)
  • NOTE [burden_elim/tests.rs:1186] Semantic-pin doc comment STRICTLY SHRINKS … This pin FAILS on revert is testing-discipline vocabulary (tests.md §Matrix Clamping), not a C-B3 change-trail. Mechanical detector false-fired; left unchanged correctly. No action. (P3-F17)
  • NOTE [tooling-gap] Hygiene-lint swallowed-error check lacks comment-awareness: documented-invariant .unwrap_or_default() sites re-fire perpetually despite adjacent // INVARIANT: comments. Candidate for /improve-tooling comment-aware refinement (add a suppression-comment parser to the swallowed-error detector). Non-blocking; informational.

HISTORY

  • 2026-06-02 a10 — §06.3 RcAtomicity carrier SITE landed; RL-19/20/21 NonAtomic-selecting dispatch relocated to repr-opt/section-10 per routing.md §4. Grounding §06.3 against actual code established the SITE-vs-ACTIVATION split for the atomicity carrier: (1) the shipped runtime RC primitives ori_rc_inc/ori_rc_dec (compiler/ori_rt/src/rc/mod.rs) are UNCONDITIONALLY atomic (AtomicI64::fetch_add/fetch_sub); there is NO non-atomic runtime, NO emit_rc_inc/emit_rc_dec atomicity branch in ori_llvm, and NO thread-escape analysis in ori_arc. RL-19/20/21 are entirely target-only. SITE delivered NOW (carrier + default-Atomic, observably a no-op): RcAtomicity { Atomic, NonAtomic } enum on RcInc/RcDec ArcInstr variants (ir/instr.rs); repr + default_atomic() + is_atomic() + fmt_atomicity_suffix() (ir/repr.rs + ir/format/mod.rs); exported from ir/mod.rs + crate root; every construction site across aims/emit_rc/{mod,forward_walk,coalesce}, aims/emit_reuse/dynamic, aims/realize/{walk,walk_dec,emit_unified,project_escape,mod} sets RcAtomicity::default_atomic() (= Atomic), reproducing today’s all-atomic emission bit-for-bit. atomic_rc_dec helper extracted in walk_dec.rs (DRY + restores the 100-line emit_defined_dead cap). The atomicity formatter suffix is empty for Atomic, so AIMS-snapshot output is byte-identical for the carrier (the 2 realize_rc_reuse snapshots re-blessed in this session changed ONLY for the pre-existing uncommitted §06.1 a08 faithful-emission RL-4 delta — +RcDec %2, -burden_inc/dec — NOT the atomicity carrier; verified zero [non-atomic] lines in the bless diff). ACTIVATION relocated to plans/repr-opt/section-10-thread-local-arc.md §10.2 per routing.md §4: the RL-19/20/21 dispatch that POPULATES NonAtomic requires §10.1’s ThreadLocality thread-escape analysis (compiler/ori_repr/src/escape/thread.rs) + §10.2’s non-atomic runtime (ori_rc_inc_nonatomic/ori_rc_dec_nonatomic) + §10.2’s atomic-vs-non-atomic codegen selection (emit_atomic_rc/emit_nonatomic_rc) — all already owned by §10 (status: not-started). Populating NonAtomic in §06.3 without §10’s backend would be an inert no-op deliverable (carrier ignored by unconditionally-atomic codegen). NOT a §09.2 relocation — §09.2 CONSUMES this carrier (its line references “atomicity carrier per §06.3 RcAtomicity”); §10.2 POPULATES the NonAtomic value. SITE tests landed (ir/repr/tests.rs: 4 RcAtomicity pins, semantic + negative). The 3 NonAtomic-SELECTION tests (cross-thread-escape FFI-export → Atomic; intra-thread Sendable → NonAtomic; whole-program RL-21 → all NonAtomic) + the channel-send #skip companion (blocked by E2040 → BUG-02-037, filed this session) relocated to §10.2 with the dispatch — they require §10’s selecting backend to exercise. §06.3 checkboxes 1+2 flipped [x] (carrier + repr DONE); the SITE tests checkbox flipped [x]. GATES: LLVM_SYS_211_PREFIX=/usr/lib/llvm-21 cargo build --all-targets GREEN ori_arc+ori_llvm debug+release; cargo test -p ori_arc --lib 1527/0 debug + 1523/0 release; clippy clean ori_arc+ori_llvm; VF-1=0 distinct corpus-wide (diagnostics/burden-balance.sh over all 459 spec files, before AND after — carrier is emission-neutral); AIMS snapshot suite green (2 re-blessed for the pre-existing faithful-emission delta). §10 frontmatter gains the relocated dispatch + tests. Working tree carries the carrier + plan edits + BUG-02-037 uncommitted (no git commit per charter).
  • 2026-06-02 a09 — §06.2 compound-instruction SITE: annotation-surface premise corrected; consumers wired+active today; ZERO §09.2 relocation. Grounding the §06.2 design intent against actual code surfaced that the premise — “Phase 6 lands COW/reuse/drop-hint MARKERS on surviving burden ops; §06.2 wires a Step-10 consumer to READ those markers” — does not match the proven AIMS architecture and MUST NOT. The burden-op variants (BurdenInc/BurdenDec/BurdenDecPartial/BurdenDecField/BurdenDecVariant, compiler/ori_arc/src/ir/instr.rs) carry ONLY var/field/base — no annotation field. Phase 6 (realize/burden_elim.rs:eliminate_burden_ops) is pure DP-2/DP-3 removal; it does not (and a removal-only census guard asserts it MUST not) construct or annotate burden ops. Adding a compound-annotation field to a burden op would be a parallel SoT / shadow uniqueness tracker, banned by aims-rules.md Invariant-5. The architecturally-correct (and shipped) mechanism: COW/reuse/drop-hint are read from the CONVERGED lattice DIMENSIONS (Uniqueness, Shape) at the realize sites — realize/decide.rs:decide_cow (DP-9, RL-7/RL-8), decide_reuse + emit_reuse/{detect,dynamic,planner}.rs (DP-6, RL-11/RL-11a/RL-12), decide_drop_hint (PL-4 post-merge) — exactly as §05.2’s dimension-wiring audit established (“each consumer reads the converged lattice dimension… burden-transparent. Role survives the emission→elimination shift”). arg-ownership: ArcParam.ownership at Step 2 apply_aims_ownership (pipeline/aims_pipeline/batch.rs) + call-site arg_ownership at Step-4b-prelude emit_arg_ownership (pipeline/aims_pipeline/mod.rs), per PL-5 + TF-11 IC-3. ALL FOUR SITE consumers are wired + active TODAY. No §05.2 dependency gap: §05.2 (status: complete) delivered the converged-dimension reads, NOT markers; the §06.2 dependency is satisfied. ZERO §09.2 relocation for §06.2: the compound instructions emit REAL LLVM instructions today (ori_llvm/.../arc_emitter/emitter_utils.rs consumes CowMode for real in-place-vs-copy; Reset/Reuse real; drop-hints real) — they were NEVER burden-op codegen no-ops, so flipping carries NO double-emit hazard against the live predicate stack. The LEDGER §B.2 double-emit hazard that relocated the §06.1 whole-var BurdenInc→RcInc flip to §09.2 applies ONLY to whole-var burden ops, NOT to these dimension-driven compound realizations. SITE-consumer test coverage is the existing matrix (77-cell realize/tests.rs decision matrix for COW/reuse/drop-hint + COW FileCheck codegen + cow spec corpus + arg_ownership/tests.rs); fabricating a marker-reader test would be test-inflation per tests.md §Anti-patterns (no marker-reader exists to pin). GATES: dev binary rebuilt; VF-1=0 distinct corpus-wide (diagnostics/burden-balance.sh) BEFORE and AFTER (no code change — plan-doc-only edit); cargo test -p ori_arc --lib 1527/0 debug; clippy clean ori_arc. §06.2 body design-intent + checkboxes reworded SITE-vs-activation; NO §09 plan edit (no activation relocated).
  • 2026-06-02 — Linear-execution-blocked auto-cure: real-RC activation relocated §06.1 → §09.2 (routing.md §4): §06.1’s two §09-coupled [ ] items (the BurdenInc→RcInc / BurdenDec→RcDec REAL-RC mechanical-mapping checkbox + its real-RC FileCheck-on-RcDec test pins) carried banned <!-- blocked-by: §09 --> annotations (routing.md §4 prohibits the annotation as a long-term cure; INV-19 forbids §06.1 carrying a forward-dependency on downstream §09). Per routing.md §4 MOVE-the-item discipline, the real-RC ACTIVATION was relocated to §09.2 (predicate-stack retirement) where the burden ops become the sole RC emitter and the flip is safe (LEDGER §B.2 double-emit hazard dissolves WITH the predicate-stack removal). §06.1’s residue is the SITE deliverable only (faithful Phase-5 emission VF-1=0 + mirror removal + grep-zero + RL-4 live-out matrix — all DONE), plus a non-checkbox citation pointing to §09.2; the §09-coupled items are no longer actionable-[ ] in §06.1. success_criterion 3 reworded SITE-vs-activation; §09 frontmatter gains the activation success-criterion + realize/mod.rs touch. §06.1 now closes on its actual SITE deliverable.
  • 2026-05-22 — Autopilot auto-cure: phantom_ids=[§05]; per decisions/10-target-self-drift-cure.md 2026-05-11 autopilot carve-out
  • 2026-05-22 — /review-plan Step 5 editor cycle: applied surgical edits per /tp-help reviewer findings (claude-ds Findings 1-3 + codex Findings 1-3 + agy Findings 1-2). Concretely: depends_on extended ["05", "02"] → ["03", "04A", "05", "02"] (claude-ds 3); success_criterion 6 rewritten from auto-suggest placeholder to concrete §05.2/§03/§02 dep statement (claude-ds 2); Intel recon block replaced boilerplate with concrete symbol inventory (claude-ds 1); §06.1/§06.2/§06.3 Tests bullets tightened to matrix-test + semantic-pin + parity + leak-check rigor (codex 3); §06.3 RcAtomicity defined as carrier-on-ArcInstr before atomic-vs-non-atomic dispatch (codex 2); body Depends on: section adds §04B Prototype Gate predecessor + rescope-on-FAIL branch (codex 1); §06.2 arg-ownership resolved to Step 4b-prelude per aims-rules.md PL-5 with plan-frontmatter goal-line drift flagged for §08 sync (agy 1); success_criterion 4 extends element_fn_gen to include instr_dispatch as second cross-crate compute_drop_info consumer (agy 2).
  • 2026-06-02 — Stale review_pipeline: marker cleared by /continue-roadmap orchestrator: marker carried stage: ?, next_step: ?, updated: ?. Per /review-plan SKILL.md §Step 1a stale-marker rule (reviewed: false + marker present → STALE by definition), marker invalid; prior diagnosis preserved here for traceability. Cure rooted in scripts/plan_orchestrator/markers.py:clear_stale_marker_if_unreviewed.
  • 2026-06-02 — /review-plan Step 5 editor cycle (pre-implementation plan-shape review): applied surgical edits per agy + opencode reviewer findings (codex returned skeleton-only; survivor-mode 2-of-3). Concretely: promoted burden_mirror reconcile_burden_ledger neutralization sequencing to a section-level ORDERING-GATE success_criterion (was §06.1 checkbox only) — mirror neutralized AND VF-1 clean BEFORE any whole-var BurdenInc→RcInc / BurdenDec→RcDec lowering, per LEDGER §B.2 spurious-RC-after-lowering risk (opencode-F1, Major); added ir/instr.rs (RcAtomicity carrier) + ir/repr.rs (atomic-vs-non-atomic representation per RL-18/19) to touches: for orchestrator conflict-detection + locked-design guards (opencode-F2); added §06.2 design-intent block stating Phase 7 realize_rc_reuse does only whole-var mapping (no compound code), compound COW/reuse/drop-hint stays at Step 10 realize_annotations, arg-ownership at Step 4b-prelude (opencode-F3); narrowed the §06.1 verify-zero gate from whole realize/ dir to realize/mod.rs (the rewritten site) — the §09-retiring sibling files keep predicate sites until §09, so a whole-dir grep is unsatisfiable pre-§09; reconciled the Intel-recon block to match (audit gap-1); mapped the §06.3 TPR checkpoint exit reasons to decisions/31 Option C to avoid autopilot state-machine stalls (agy-F1); re-pathed the §06.3 cross-thread atomic-RC test to an FFI-pointer-export exercise with a #skip channel-send companion anchored to E2040 resolution (agy-F2); added timeout 150 ./test-all.sh (no-NEW-failures-beyond-baseline per LEDGER §B.3, baseline AOT 2284/19) to the §06.N close-gate (agy-F3); added §02-under-enumeration provenance note to success_criterion 4 so the two-consumer superset reads as deliberate (cohesion gap-2).
  • 2026-06-02 — §06.1 Step A (faithful Phase-5 burden emission) landed; ORDERING-GATE Steps B/C pending AOT-path verification: release_with_burden(func, var, strategy) + carries_burden(func, var) added to compiler/ori_arc/src/aims/emit_rc/mod.rs, wired into predicate-stack release sites (emit_rc/trampoline.rs, emit_rc/edge_cleanup.rs, emit_rc/dead_cleanup/mod.rs) so every release RcDec co-emits a faithful paired BurdenDec gated on carries_burden — “one SSOT, two emitters” (Invariant-5-clean, no parallel pass; grounded RL-2/RL-4/RL-5). ORI_DISABLE_BURDEN_MIRROR=1 bisection flag added at pipeline/aims_pipeline/mod.rs + documented in arc.md/compiler-ops.md. Build green (ori_arc + ori_llvm --all-targets, debug + release). cargo st mirror-ON ≡ mirror-OFF (4600 passed / 45 failed / 51 skipped, byte-identical) — Step A introduced ZERO spec regressions; the 45 failures are the pre-existing type-error-blocked frontend baseline (fmt declarations / http caller), NOT §06 burden/RC. ORDERING-GATE remainder pending: conclusive VF-1 verify_burden_balance 656→0 measurement via the AOT/LLVM path (the eval spec harness does not surface the debug-assert warn) → gated Step B reconcile_burden_ledger removal (ONLY after VF-1=0, per LEDGER §B.2 spurious-RC-after-lowering hazard) → Step C mechanical BurdenInc→RcInc/BurdenDec→RcDec verification in realize/mod.rs → matrix tests. Verification + B/C continuation is code-authoring chartered to an Opus agent, transiently server-rate-limited across three consecutive dispatches; next autopilot firing retries when the rate-limit clears.
  • 2026-06-02 follow-up — §06.1 faithful-emission redesign: VF-1 350→25 via 3 RL-2-grounded emission fixes; ORDERING GATE still OPEN (25≠0), Steps B/C NOT attempted. MEASUREMENT CORRECTION: the staged working tree already carries the prior session’s eliminate_whole_function (function-wide Phase-6 elim) in burden_elim.rs, so the true AOT-path VF-1 baseline is 350 / 93 files (all net=-1/-2 orphan decs), NOT the LEDGER §B.2 “656”. DOMINANT GAP re-localized (per ORI_DUMP_AFTER_BURDEN=1 + ORI_LOG=ori_arc::aims::verify::burden_balance=debug per-block deltas): a FRESH heap value passed at a BORROWED Invoke/InvokeIndirect arg position (the assert_* message-string family dominates; 319/324 files import std.testing) gets a terminator-last-use BurdenDec from the burden walk AND a scope-exit release BurdenDec from the predicate-stack edge cleanup (release_with_burden) → double-count → net -1 per terminal path. THREE EMISSION FIXES LANDED in compiler/ori_arc/src/lower/burden_lower/ (all RL-2-grounded, no parallel path, Invariant-5-clean): (i) compute_borrowed_alias_vars (mod.rs) excludes Let-Var aliases of borrowed params from owned_vars_needing_rc (a borrow’s alias is a borrow per TF-2/TF-4 → no burden ops; cures §B.2 gap (b)); (ii) compute_transfer_via_move_alias (mod.rs) backward-propagates terminator/owned-position transfer through Let { Var } MOVE-alias chains (src used once) to a fixpoint, suppressing the move source’s last-use dec + symmetric inc (cures id<T>(x)=x owned-param-returned-via-alias orphan dec); (iii) invoke_terminator_borrowed_args (emit.rs) — THE dominant fix (345→25) — suppresses the terminator-last-use BurdenDec for a value at a borrowed Invoke/InvokeIndirect arg position (value survives the borrowed call; released at the successor by the predicate-stack edge cleanup whose release_with_burden co-emits the paired scope-exit dec; emitting the terminator dec too double-counts). GATE STATUS: LLVM_SYS_211_PREFIX=/usr/lib/llvm-21 cargo build --all-targets GREEN ori_arc+ori_llvm debug+release; cargo test -p ori_arc --lib 1530/0 green (debug + release 1526/0); AOT suite 2322 passed / 16 failed — the 16 failures are IDENTICAL with-vs-without my changes (verified by restoring the staged baseline and re-running) = PRE-EXISTING dirty-tree state (incl. the aims_burden_alias::test_burden_alias_inner_survives_result_destructure pin, which fails at the staged baseline too), ZERO new failures from my edits; VF-1 = 25 (was 350 staged baseline). 3 NEW dead-ends + 1 untried-risky candidate recorded in LEDGER §B.2 (Step-4b unwind-edge pass INERT/972, block-local-pair elim guard 2433, dec-only-retention guard 3055, Apply-body borrowed-call deferral 475). RESIDUAL 25 / 15 files (the §06.1 checkbox is NOT satisfiable until 0): diverse Step-A coexistence-pairing long-tail — dominant = owned-param-to-Invoke-OWNED-arg transfer where release_with_burden over-pairs a burden_dec on the transferred-out var (the callee consumes it; the predicate-stack unwind-edge release double-counts against the burden ledger). NEXT SURFACE to reach 0 (the inverse of fix (iii)): gate release_with_burden’s carries_burden so the predicate-stack edge/scope-exit release does NOT pair a burden_dec for a var transferred OUT at an owned Invoke/InvokeIndirect arg position; secondary: a project-escape-aware Project-dst borrow exclusion (a blanket Project-dst exclusion is UNSAFE — project-escape RL-15a adds incs); plus niche-Option<str> payload, Never-propagation, TRMC residuals. Steps B (mirror removal) + C (mechanical lowering + grep-zero) + matrix tests REMAIN GATED on VF-1=0. Working tree left with the 3 emission fixes uncommitted (per charter: no git commit).
  • 2026-06-02 a07 — §06.1 faithful-emission: VF-1 19→10 via 3 RL-grounded fixes + permanent burden-op diagnostics; ORDERING GATE still OPEN (10≠0), Steps B/C NOT attempted. MEASUREMENT CORRECTION: prior “25/15” is a raw-grep over-count; the DISTINCT (var, exit_block) baseline via the new diagnostics/burden-balance.sh is 19 / 15 files. Built permanent burden-op-aware diagnostics (the dump-correlation gap that made this hard): diagnostics/burden-balance.sh (distinct-imbalance corpus measurement, stale-binary-aware), burden_sites= in the verify_burden_balance imbalance trace, binc/bdec in trace_phase_snapshot, per-block burden_sites= in the AIMS pipeline-checkpoint INFO log, function-name in the imbalance WARN. DOMINANT net=-1 localized: the predicate-stack edge/dead-at-entry cleanup co-emitted a BurdenDec on the unwind/dead edge for a var transferred OUT at an OWNED Invoke/InvokeIndirect arg — already balanced at the transfer point, so the edge dec double-counts. 3 FIXES LANDED (RL-2/RL-4/RL-5-grounded, Invariant-5-clean, no parallel path): (i) release_with_burden_edge (emit_rc/mod.rs; wired at edge_cleanup.rs apply_edge_decs + trampoline.rs) suppresses the paired BurdenDec (keeps RcDec) for an owned-transfer arg of the PRED terminator — inverse of a06 fix (iii); (ii) release_with_burden_into_block (same SSOT; wired at the 3 dead_cleanup emit_dead_at_entry_decs sites) — successor-keyed companion suppressing the block-entry BurdenDec for an owned-transfer Invoke arg’s successor block; (iii) compute_borrowed_terminator_invoke_args (burden_lower/mod.rs → inc_suppressed_vars) suppresses the FRESH-site BurdenInc for a FRESH value passed at a BORROWED terminator Invoke arg (callee stores it into its result; a06 fix (iii) already suppresses its terminator dec, so the var ends carrying NO burden ops → fully predicate-stack-managed). GATE STATUS: cargo build --all-targets GREEN ori_arc+ori_llvm debug+release; cargo test -p ori_arc --lib 1530/0 debug + 1526/0 release; clippy clean (fixed one pre-existing detached-doc warning in burden_elim.rs litter-pickup); AOT suite 2322 passed / 16 failed = IDENTICAL to the charter baseline (the 16 = BUG-04-123 predicate-stack double-frees + BUG-04-121 VF-1 loop residuals: aims_burden_alias/borrow_independence/for_yield_option/generics/higher_order/match_alias/tagless_enum), ZERO new AOT failures from my edits. test-all.sh width_tests_spec + width_tests_comprehensive (ori_fmt) fail on the untracked tests/spec/traits/index_set/updated_method.ori line 88 (>60 chars) — a dirty-tree artifact from a parallel session staged BEFORE this session (I touched no ori_fmt / .ori files); NOT a burden regression. 1 NEW dead-end recorded in LEDGER §B.2 (FRESH-borrowed terminator-dec retention → 12→190). RESIDUAL 10 / 7 files: all net=-1/-2 ORPHAN-DEC, all in the JIT/eval conservative-contract realization path (NOT the AOT codegen path, which ORI_DUMP_AFTER_ARC confirms balanced for the same functions) — comparable find_max/find_min, cow/{substring,string_cow,matrix_string}, option_str, index/definition, debug/escape (a single-inc value receiving a BurdenDec on >1 CFG branch). Next surface: contract-precision analysis of why the conservative-contract JIT realization emits per-branch decs for a single-inc value. Steps B (mirror removal) + C (mechanical lowering + grep-zero) + matrix tests REMAIN GATED on VF-1=0. Working tree carries the 3 emission fixes + permanent diagnostics uncommitted (no git commit per charter).
  • 2026-06-02 a08 — §06.1 VF-1 10→0 (ORDERING GATE MET) + Step B mirror removal; Step C real-RC flip is §09-coupled (blocker, NOT defer). VF-1 10→0 in ONE RL-4-grounded fix (the entire residual was one bug class). ROOT (per a07 localization, confirmed via ORI_DUMP_AFTER_BURDEN + per-block verify_burden_balance deltas): detect_last_uses in lower/burden_lower/mod.rs is purely PER-BLOCK — it registers a “last use” (and thus a BurdenDec) in EVERY block where a var appears, with no cross-block liveness. For a var live-OUT of a block (used in a reachable successor — comparable::find_max %0 aliased in bb0 AND bb1; cow::shared_substring/debug::escape FRESH str aliased across bb0→bb1), the in-block last-use dec is SPURIOUS: the value lives on, so emitting a dec there releases a reference still needed downstream → VF-1 net=-1/-2 on the live-out path. Per Spec: Annex E §AIMS RL-4 (edge-specific decs: a var live at block exit but dead at successor entry releases on the dying CFG EDGE, not unconditionally in the block), the genuine releases are (a) the dead-OUT block’s own last-use dec (kept) and (b) the predicate-stack edge / dead-at-entry cleanup (release_with_burden_*) on the dying edges (already emitted by realize). FIX (RL-4-grounded, Invariant-5-clean, no parallel path): compute_live_out_owned (lower/burden_lower/mod.rs) — standard backward liveness restricted to owned_vars_needing_rc (gen/kill mirrors liveness::compute_liveness, keyed on the burden walk’s own owned-RC set — no parallel ownership tracker); threaded into BurdenAnalysisCtx.live_out_per_block and consumed at BOTH emit_last_use_decs (instruction-position) and emit_terminator_burden_decs (terminator-position) in emit.rs to SUPPRESS the in-block last-use BurdenDec for any var live-out of the block. STEP B (mirror removal) DONE: reconcile_burden_ledger call + BURDEN_MIRROR_DISABLED static + ORI_DISABLE_BURDEN_MIRROR flag removed from pipeline/aims_pipeline/mod.rs; aims/realize/burden_mirror.rs + burden_mirror/tests.rs deleted; mod burden_mirror + pub(crate) use reconcile_burden_ledger removed from realize/mod.rs; ORI_DISABLE_BURDEN_MIRROR debug_flags.rs entry + arc.md/compiler-ops.md/diagnostic.md doc lines + burden_delta.rs comment scrubbed; burden-balance.sh updated (mirror flag dropped — faithful emission is the sole RC path, script KEPT). Zero reconcile_burden_ledger/burden_mirror/BURDEN_MIRROR_DISABLED/ORI_DISABLE_BURDEN_MIRROR references corpus-wide; check-debug-flags.sh clean (0 stale/orphan/undocumented). VF-1=0 holds mirror-REMOVED corpus-wide. STEP C grep-zero DONE (site established): realize/mod.rs verified zero of the 6 predicates (class_alive_after/pin4_class_emits_dec_set/pin6_any_ancestor_will_cover/class_payload_of/canonical_rep_for/var_emits_dec_in_block) — it is a thin dispatcher; the predicates live in the §09-retiring siblings (walk_dec.rs/walk.rs/emit_unified.rs) per line 144 scope. STEP C real-RC flip (checkbox 2 + success-criterion 3) BLOCKED on §09, NOT deferred: burden ops are codegen no-ops (instr_dispatch.rs:509); the predicate stack STILL emits the real RcInc/RcDec (in the §09-retiring walk_dec.rs/emit_unified.rs). Flipping BurdenInc→RcInc to REAL RC NOW would DOUBLE-emit RC alongside the live predicate stack → double-frees — exactly the LEDGER §B.2 / success-criterion-1 ORDERING-GATE hazard (line 20-22). The architecturally-correct path per The One Rule is to NOT reintroduce double-frees; the no-op→real-RC activation is structurally coupled to §09 removing the parallel predicate emission (the whole-realize/-dir zero-reference assertion IS the §09 close-gate per line 144). realize/mod.rs is established as the clean lowering site; §09 flips the activation. MATRIX TESTS DELIVERED (lower/burden_lower/tests.rs, 3 RL-4 semantic+negative pins): owned_param_live_out_of_block_gets_no_in_block_last_use_dec (conditional-transfer Branch), fresh_value_live_across_blocks_nets_burden_balance_zero (multi-block Jump; verify_burden_balance empty semantic pin), dead_out_value_keeps_its_single_block_last_use_dec (narrowness negative-space — suppression fires ONLY for live-out, not dead-out). Reverting RL-4 → bb0-dec reappears → VF-1 imbalance (negative pin via per-block dec count). GATE: cargo build --all-targets GREEN ori_arc+ori_llvm debug+release; clippy clean (split compute_live_out_owned + extracted compute_use_counts_and_dup_aliases for the 100-line cap; fixed a litter-pickup decorative-banner in tests.rs); cargo test -p ori_arc --lib 1527/0 (was 1524 post-mirror-test-removal + 3 RL-4 = 1527); AOT suite 2322 passed / 16 failed = IDENTICAL charter baseline (16 = BUG-04-123/121, enumerated above), ZERO new AOT failures; aims-proof check 172/0 regressions (locality divergent_pending non-fatal); prose-lint clean (docs + touched .rs). test-all.sh residual NON-mine failures: 2 ori_fmt width_tests (parallel-session index_set/updated_method.ori dirty artifact), 45 interpreter (no method 'updated' ×18 parallel-session index_set feature + blocked by type errors ×26 + eq on Point ×1 frontend baseline) — none RC/burden. LEAK NOTE: tests/valgrind/cow/cow_list_set.ori reports “1 RC allocation not freed” — PROVEN pre-existing predicate-stack (not mine): persists IDENTICALLY with ORI_DISABLE_BURDEN_OPS=1 (all burden emission off → pre-§04A baseline); burden ops + mirror are codegen no-ops so CANNOT alter real RC codegen. §09-retirement / BUG-04-123 class, outside §06.1 burden-emission scope. ORDERING GATE (mirror-neutralized + VF-1-clean) MET; checkbox-2 real-RC flip §09-coupled. Working tree carries the RL-4 fix + Step B/C + matrix tests + diagnostics uncommitted (no git commit per charter).

Fresh intel (regenerated)

This is not a mechanical-lowering design problem. It is a plan-state/code-state reconciliation problem wearing a design problem’s clothes.

  1. The deliverable already shipped, unannounced, in the codebase. compiler/ori_arc/src/aims/realize/{walk.rs,walk_dec.rs} and compiler/ori_arc/src/aims/emit_rc/dead_cleanup/** — the predicate-stack files this section’s own Intelligence Reconnaissance block (queried 2026-05-08) named as the retiring siblings §06.1 rewrites around — do not exist on disk. They were deleted in commit 72126402d (“chore(compiler): sweep AIMS RC, call-site generics, SCIP, fmt escape”, 2026-06-16, Bypass: --bypass flag passed by user). emit_edge_cleanup(..., burden_only) is hardcoded true at its one call site; lower_burden_ops_to_rc (the literal BurdenInc->RcInc / BurdenDec->RcDec function this section’s own checklist item names) runs unconditionally inside emit_rc_unified, gated on no environment variable at all. mod.rs’s own doc comment now reads: “The burden path emits RC via Phase 2.5 elimination + Phase 7 lowering… No output owns an independent decision procedure.” Burden is the sole RC emitter, in the default build, today — not merely under the ORI_DISABLE_PREDICATE_STACK_RC=1 probe arc.md still documents as the gate. [JOIN — code + git]
  2. The plan’s own tracked state disagrees with its own code. python -m scripts.plan_corpus.read --completion shows §06 (s-745db2fc, THIS section) status: in-progress, §07A (s-b14a3bfd) status: in-progress, §08 (s-d7008993, rule-file sync) status: not-started, and §09 (s-745fdec3, predicate-stack retirement) status: not-started. §07A’s own 2026-06-04 HISTORY entry says its ~40-file working tree was left uncommitted, waiting on a user-typed /commit-push --bypass. Commit 72126402d twelve days later bundles that exact class of change (AIMS RC consolidation) with unrelated work (turbofish, SCIP, fmt-escape) under a generic chore: message that never names the plan, §06, §07A, §09, or a single BUG-04-1xx id. symbol-plans lower_burden_ops_to_rc even shows §09’s OWN section body already citing that function name — the plan text anticipated the code, but nothing flipped the section status when the code arrived. [JOIN — plan-state + git + graph]
  3. The rule-file quartet is stale in the same window. arc.md’s §STOP — PREDICATE STACK IS LEGACY section and its ~62 documented ORI_DISABLE_* toggle entries were last content-edited 2026-05-28 (per its own HISTORY block); the one 2026-07-02 touch to arc.md is a single-line BUG-04-240 addition, not a predicate-stack resync. §08 (s-d7008993, the section literally named “rule-file-sync-arc-md-aims-rules-md-canon-md-7-1-missions”) is not-started. The rule files still describe ORI_DISABLE_PREDICATE_STACK_RC=1 as gating a real predicate-stack-vs-burden choice; that choice no longer exists in source. [JOIN — rule-file HISTORY + git]
  4. The mission’s own falsifiable collapse metric reads the wrong direction. mission.md §4 criterion 1 demands ls ownership_scans/*.rs | wc -l and the ORI_DISABLE_* toggle count both DECREASE toward zero with no new scan added. Measured today: ownership_scans/ = 31 files (baseline cited “~28”); distinct ORI_DISABLE_* names in ori_arc/src = 78 (whole compiler/ = 81; arc.md documents 62 of them) against a cited “~50” baseline. Both proxies have GROWN since the generalize-the-inference mission (2026-06-19) was written — the opposite of the stated goal — even while the underlying predicate-stack removal (the enabling milestone mission.md §5.2 retains) has already happened. [JOIN — /calc-verified source counts vs mission.md §4 probe]
  5. A live, actively-churning bug sits directly on this section’s lowering surface right now. BUG-04-240 (critical, in-progress in the tracker as of today) is the moved-out-fields fixpoint non-termination guard feeding Phase-5 burden emission that Phase 7 (this section) lowers; its calculus fix landed in aims-rules.md TODAY (IA-MF1, 2026-07-02) but the bug itself has not closed. BUG-04-238, BUG-04-219, BUG-07-286 are three more in-progress burden-domain bugs the section’s own “Bug Fold-In + Supersession” standing directive requires be worked HERE, not routed away.

Decisions this package forces before touching code:

  • Do not “design” the Phase-7 activation — it is DONE. The remaining work is (a) verify the current unconditional-burden-emitter behavior against every criterion §06’s mission checklist item + the relocated §09.2 activation criterion actually demand (VF-1 corpus-wide, dual-exec parity, leak-clean, ./test-all.sh zero-new-failures), (b) flip the section-state and mission-checklist artifacts to match reality (or file the specific gap if verification finds one), and (c) fold §08’s rule-file resync into the same pass — arc.md/aims-rules.md’s predicate-stack narrative is now describing dead code.
  • Before believing ANY prose in this section’s body about “walk_dec.rs”, “the predicate stack”, “coexistence”, or “§09.2 activation is blocked” — re-verify against live compiler_repo source; the body predates the June 16 deletion by 2+ weeks and was never updated.
  • BUG-04-240 + siblings are NOT separate work — per the plan’s own Bug Fold-In directive they are this grind’s work, and BUG-04-240’s Phase-5 fixpoint soundness is a direct precondition of what Phase 7 lowers correctly.
  • Do not re-attempt any of the 109 ledger dead-ends (decisions/LEDGER.md §6) while doing this reconciliation — none of the graph/git findings above contradict them; they are orthogonal (root-cause-analysis dead-ends vs a workflow-tracking gap).

DIG DEEPER

cd compiler_repo && git show --stat 72126402d
cd compiler_repo && grep -n "burden_only\|ORI_DISABLE_PREDICATE_STACK_RC" \
  compiler/ori_arc/src/aims/emit_rc/edge_cleanup.rs compiler/ori_arc/src/aims/realize/emit_unified.rs
python -m scripts.plan_corpus.read plans/aims-burden-tracking --completion
scripts/intel-query.sh symbol-plans lower_burden_ops_to_rc --repo ori --human

(full dossier: phase-7-mechanical-lowering-burdeninc-rcinc-burdendec-rcdec—s-745db2fc.intel.md)