Delete unused memo and preparation carriers (native-memory-inert-carriers)

[briansrls]

Roadmap node native-memory-inert-carriers. This PR stacks on #13582 (parse-memo-delete). CI will not run on it until it is retargeted to main, after #13577 and #13582 land.

What was deleted (by symbol)

1. The compile-stage memo (operator ruling 2026-10-08). Its only lookup, resolve_cache_miss, answered Miss every time, so the governed door never served a hit.

• v2.compiler.compile: resolve_cache_miss, plus the cached_stage_realization match and the cached_stage_governed wrapping inside compile_ingest_staging. compile_ingest_staging is now the plain then_outcome sequence tokenize → parse → normalize → resolve → translate. Stage results and stage refusals still flow through it unchanged.
• v2.compiler.materialization_carriers: deleted compile_stage_memo_demand_identity, the compile-stage demand, provider and id, v2_compiler_cache_providers, v2_compiler_materialization_demands/verdicts/ladder_holds, the process, run and workflow frames, admitted_prescription(_in) with its diagnostic and admission readers, CachedStageRealization, cached_stage_realization, cached_stage_governed, realized_step_for, compile_stage_realized_step, and every compile_stage_memo_* control. The prose-as-String note row also goes. The module keeps only materialization_allows_memo_store and its disposition, because the self-host emitted contract consumes them.
• v2.std.staging: deleted CacheProbe, resolve_probe and cached_stage. then_outcome stays.
• extdeps.realization.compile_stage_memo: the file is deleted. Its references are gone from extdeps.cache (cache_catalog, cache_layer_default_order, the projected row), gunbc.materialization_provider_targets, gunbc.doc_graph_roots, legacy_extdeps_scope_frontier.tsv, gunbc.commit_workflow (the compiler_materialization_witness_test enrollment) and witness_row_cost_basis.tsv.
• src/v2/test/claim/compiler_materialization_witness_test.dag and the two cached_stage_* claims in staging_test.dag are deleted.
• After Delete the v2 packrat parse memo carrier (R2, stacked on #13577) #13582 and this PR, the compiler-provider scaffolding is empty, so it is removed too, as the handback asks.

2. Claim-cache machinery (v2.compiler.eval). Deleted TestClaimCacheKey, TestClaimCacheReceipt, test_claim_cache_key_*, test_claim_cache_receipt(_node) and test_claim_cache_receipt_claim. TestClaimRun.cache is replaced by TestClaimRun.claim: TestClaim, and test_claim_run_claim reads it directly. The constructors updated to match are in emit_host, language_behavior_equivalence_test (the rejected-emit arm now carries its sentinel claim) and manual/infer_ground_add. Seven manual translate tests had only an unused import, which is removed.

3. v2.workflow.floor_preparation: the whole module is deleted, including prepared_floor_subject_serve, PreparedFloorSubjectRequest and prepared_claim_subject_identity. So is its only consumer, dag/test/claim/floor/floor_preparation_witness_test.dag. In std.materialization_provider, in_process_materialization loses the two admit_callers rows that named that witness, and its frontier sentence drops the phrase "floor-preparation witnesses". The FNV preimage-verified exception and minted_in_process are unchanged.

4. v2.workflow.floor2_prepared_subject: deleted FloorSelectionPosture, FloorExecutionKey (with OnePreparedSubjectKey), floor_execution_key, floor_execution_key_is_one_prepared_subject and PreparedSubjectSpec, whose only consumer was the key.

5. v2.workflow.native_cache_fusion and src/v2/test/claim/native_cache_fusion_test.dag are deleted. The real emit-host native build cache is untouched.

Bookkeeping

• floor_grandfathered_roster: added WitnessDeleted rows for the 8 cut members this PR deletes (3 compiler_materialization_witness, 2 staging.cached_stage_*, 3 native_cache_fusion_test).
• floor_unimported_bare_provider_debt_roster: the compiler_materialization_witness_test.dag row moves from Retired ImportsFixed to Retired FileDeleted. The floor demanded this (RetiredImportsFixedButFileAbsent).
• cache_layer_planner_test now uses parse_cache_id and parse_cache_projected as its in-process layer.
• materialization_provider_witness_test: consumer_target_frontier_names_five_caches… becomes …names_four_caches… with == 4, the same pattern Delete the v2 packrat parse memo carrier (R2, stacked on #13577) #13582 used.

What was kept, and why (red_control)

• Ordinary stage results and refusals: compile_ingest_staging has the same five stages, and then_outcome composes them as before. compile_eval_thesis_proof_test passes.
• EvalContext.cache_authority.interpretation, eval_context_interpretation, effect_io, and every claim result and verdict field.
• inferred_tree_digest and its whole diagnostic/extent/ByteOffsetCacheKey digest chain, plus test_claim_cache_interpretation_hash_node and the other digest functions. Nothing was renamed to "computation_identity".
• WitnessIdentity and witness_identity_qualified_name, which the terminal and wet-evidence ledgers use.
• The general std.materialization_provider machinery, including its FNV preimage-verified exception.
• The emit-host native build cache.

Where #13582's generic ladder controls went

#13582 moved four controls onto the compile-stage demand. With that demand deleted:

• Ladder-level obligations live on fixture subjects in test.claim.materialization_ladder_witness. These are RefusedScopeTooNarrow for an isolated plural site, RefusedNoProvider without a provider, discharge for a demand that recurs inside the provider's scope, and single recompute. The cells are provider_scope_too_narrow_is_refused, obligation_without_provider_is_typed_refusal, unbounded_siblings_frame_obligates_up_front / declared_replay_frame_obligates_up_front, and single_pure_demand_is_accepted_recompute. All 32 claims in that module pass on this head. I did not add copies of them.
• Door-level controls (an_unruled_identity_is_not_admitted, a_refused_identity_is_not_admitted) tested admitted_prescription_in, the green-or-nothing admission door in materialization_carriers. That door's only production caller was the compile-stage demand. Keeping it alive on a test-local fixture would keep a mechanism nothing in production calls, only so its controls had a subject (DESIGN §3c). So the door and its controls are deleted together. This is the one judgment call in this PR, and it is easy to reverse: if reviewers want the door kept as a generic rule, it can move next to std.materialization_ladder, with these two controls on a fixture demand.

Consumer census (grep across src/, dag/ and the src/v1 Rust before each deletion)

• compile_stage_memo*, cached_stage*, CacheProbe, resolve_probe: the consumers are the ones listed above. Some code comments still name them as historical specimens (trait_bound_witness_test self-contained fixture strings, anonymous_record_head_use_line_witness_test, src/v1/05_emit_rust.dag, docs/plans/*). The committed snapshot fixture frontier_probe_elision_boundary_overlay also mentions them. These were left as they are. No seed Rust or stage0 mirror references any deleted symbol, so neither stage0 nor the generated artifacts were regenerated.
• TestClaimCache*: emit_host, language_behavior_equivalence_test, manual/infer_ground_add, and 7 manual translate tests (import only).
• floor_preparation: only its own witness, plus the two admit_callers rows.
• floor2 routing symbols: zero consumers.
• native_cache_fusion: only its own test.
• Path-list snapshots left untouched: srv1_residue_rehearsal and docs/rung-drops/required_lanes_do_not_resolve_product_layer_modules_2026-09-20.txt.

Evidence (local, head 0b92d40)

• cargo build --release (gunbc, claim_batch, claim_executor): succeeded. No Rust changed, so clippy and fmt were not run.
• claim_batch over 16 entries (231 claims): 227 passed, 4 failed. All 4 also fail on the base branch (parse-memo-delete 427b080) with this binary:
  • floor_eval_step_budget_witness_test.every_declared_removal_names_a_cut_row_and_has_left_the_roster: removal row v2.test.claim.compiler.kernel_value_type_roster_witness_test.kvr_a_named_calls_bool_formal_is_counted_not_judged_holds is not in the cut. That row is also on main. My 8 new removal rows are all members of the cut.
  • external_model_scope_witness_test.roster_paths_resolve_on_disk and carriers_declare_scope_on_disk.
  • doc_graph_reference_partition_witness_test.doc_graph_production_population_loses_nothing.
  • The 16 entries: staging, cache_layer_planner, materialization_provider_witness, materialization_ladder_witness, materialization_carriers_emitted_contract, floor_eval_step_budget_witness, floor_terminal_ledger(_wire), evaluation_store_address_census, cache_key_completeness, external_model_scope, witness_row_cost_drift, commit_check_demand, doc_graph_reference_partition, local_repo_wet_terminal, wet_evidence.
• claim_batch over branch_eval_by_execution, effect_io_eval_by_execution, compile_eval_thesis_proof, manual/rust_add_emit_translate and emit_host_add_equals_eval: 17 of 17 passed.
• claim_executor --required-ci --required-lane witnesses: exit 1. The only failed phase is the local-repo wet lane, with 16 refusals, all on the known baseline: mtcollins1_kvm_observer_protocol_wet_witness (7), allocation_client_execution_wet_witness (8), and v41_source_patch_converge_witness red5 (1, intermittent, cause unresolved). There were no other refusals. 15 changed witnesses were planned and passed (blocking=0), including cache_layer_planner ×3 and consumer_target_frontier_names_four_caches_with_typed_binds.
• The build lane was not run, because stage0 is unchanged.

Not executed or unresolved

• language_behavior_equivalence_test holds its claims as test data rows, which claim_batch --functions would not address, so I did not run it on its own. It compiles in the lane's prepared subject.
• manual/infer_ground_add has no test fn. Its constructors changed but did not execute on their own.
• materialization_carriers now holds only a predicate whose sole consumer is its self-host emitted contract. Deleting the whole module would also mean editing the self-host rosters and receipts, which is outside this bounded PR.
• Also unconsumed but outside this boundary: floor2_prepared_subject SubjectExclusionGround, WitnessSite and FloorBindingStrictnessRegression.
• Fix every T? == T site the whole-population census found #13549 edits dag/test/claim/floor/floor_preparation_witness_test.dag, which this PR deletes. Whichever lands second gets a modify/delete conflict, and resolving it means deleting the file.

🤖 Generated with Claude Code

[gunbai-bot]

Superseded by #13641 at 634453d: this PR's head is an ancestor of integration/v1-closeout. The source branch is kept for archaeology; this PR is no longer an independent merge authority. — sent from neat-wolf-604