Goal A classification: every change from main to V18 (8958ae93 + docs)

Date: 2026-09-07. Diff base: local main = 6ce7648a. Branch tip: fix/insights-verified-correctness = b1b369ed.

Two segments. The goal-1 segment (main..e479e81) is already on origin/main (a922627d), deployed 2026-09-06. The post-goal-1 segment (e479e81..b1b369ed) is the V8–V18 work that this split lands or parks.

Rules applied

• Group 1 (eval-system): fixture, corpus, holdout, scorer, oracle, latency method, eval scripts, evidence, reports, benchmark dataset, and their tests. Zero runtime effect.
• Group 2 (invariant-fix): deterministic behaviour with no dependence on model output: R01–R09, N02, N03, N07, F01, F02, F08, F10, recheck-verified F-series.
• Group 3 (meaning-layer): semantic contracts, requested result shape, follow-up membership contract, intent classification, deictic carry, planner routing, refusal gating on intent, CT measure basis guard, and any V11–V18 change that alters which SQL is generated or which question is answered versus refused.
• Tie-break rule used for uncertain items: if the change alters the SQL text that reaches the database, or the prompt that produces it, it is group 3 even when deterministic. If it only classifies, renders, persists, authorizes, bounds, or proves an already-produced result, it is group 2.
• Dependency rule: a group-2 hunk that needs a group-3 helper was either inlined (the helper is a pure proof with no meaning decision) or moved to group 3 with the hunk. Each such decision is listed below.

Counts

File-level counts. Twelve mixed source files carry hunks from more than one group; their hunk splits are listed in the next section. Test files count as one path each; 60 test files are parked whole and 21 mixed test files had their group-3 tests removed (listed at the end).

Uncertain items and the rule applied

Mixed files: which hunks go where

src/institutional_kb/api/routers/bi_query.py

Kept: ClassifiedBiRoute, classify_terminal_error at every catch, error_class/outcome fields, refusal status, trace snapshot guard/code, list_ancestor_turns + exact rerun pinning (R04), display-label deferral and requested identity fill (R07), authorize_cached_sql on directory caches (F13), _result_proves_client_key / _saved_turn_proves_client_key / _require_customization_key / _merge_verified_customizations (F01), export terminal responses, cancelled.set() on worker exit. Parked: intent refusal, _answers_offered_clarification, _blocked_parent_continuation, released population, deictic can_carry/clarify turn, saved-count turn, accepted_cohort, decide_routing-first restructure, GOLD_ONLY leg, membership directives, definition route legs, live schema context, requested shape/CT basis post-passes, executed_turn_sql anchor route.

src/institutional_kb/bi/canonical_columns.py

Kept: GOLD_BOOKED_COLUMNS/GOLD_REALIZED_COLUMNS from the registry and booked basis detection. Parked: question-driven alias preservation (requested output shape).

src/institutional_kb/bi/deictic_key_carry.py

Kept: sql_proves_member_key and its lineage helpers, _normalize_key hardening, merge_cohort_key_state strips client-supplied __cohort_keys__ (F01), build_cohort_key_state requires proved key identity. Parked: detection regex changes, population_scope_released, ResolvedCohort.can_carry/truncated/keys_available, cohort_generation_directive, proof-or-refuse inject_cohort_predicate, blocked-parent resolve, empty-set semantics.

src/institutional_kb/bi/single_shot_post_pass.py

Kept: validate_calendar_syntax, column_types plumbing, preserve_empty_null coalesce, enforce_desc_nulls_last on gold. Parked: repair_pd_ct_identity_joins, repair_explicit_selection.

src/institutional_kb/services/bi_federated_execution_service.py

Kept: prior_route-aware CT-only routing and _gold_leg_can_be_discarded proof before any gold-leg discard or annihilated-cohort rescue (R01), join-key identity proof and merge key capture (R02/R01), display-cap key capture only with proved identity, terminal refusals returned as answers (N03), FederatedCtUnavailableError error_class (N02), JobStoppedError propagation and provider-refusal handling in self-heal (N07), executed_sql/connection_id/snapshot basis in leg metadata (F09/R05). Parked: membership_mode/intersection/union execution, ct_scalar/client_count aggregates, snapshot_column, requested-shape checks on merged or promoted results, is_merge_input/source_bindings/followup_context on leg requests, MembershipProofError, regenerable guard names.

src/institutional_kb/services/bi_federated_merge.py

Kept: SourceGrainError/SourceUnitError, _validate_source_grain, unit_column handling, snapshot_measure_classes, source_label, unit-aware dimension rollup. Parked: aggregate_client_count, aggregate_ct_scalar.

src/institutional_kb/services/bi_federated_merge_replay.py

Kept: merge_measure_class, saved dimension rollup replay. Parked: membership operation, snapshot column, scalar/count post-aggregation replay.

src/institutional_kb/services/bi_federated_planner.py

Kept: CT leg generation runs under copy_context (job control reaches the thread) and a failed CT future propagates its typed error when gold returns no SQL (N07). Parked: everything else (snapshot money routing, membership modes, _require_ct_data_input, GOLD_ONLY instruction, _final_gold_grain, tie-break/regex routing tweaks, calendar prompt text).

src/institutional_kb/services/bi_query_entrypoint.py

Kept: typed PERMISSION denial. Parked: intent refusal before dispatch.

src/institutional_kb/services/bi_query_execution_service.py

Kept: classify_execution_error delegation, executed_sql, generated_sql on error results, AdditiveGrainError/NativeDealUnitError with one bounded corrective regeneration, sql-heal on schema/sql classes, approved-query lookahead limit, provider/transport errors surface instead of rule-based fallback, _require_source_query_permission refactor + authorize_cached_sql, async PostgreSQL connect with JobDeadlineCursor, calendar syntax check, static _federation_leg_for, typed calendar validation only when physical types are known. Parked: _load_schema_context rewrite (discovery + cold inventory), federation_schema_context, generate_leg_sql rewrite (per-leg definitions, grounding directives, repairs), DefinitionApplicationError / N01 compile rewrite, _qualify_authorized_tables, canonical alias preservation, accepted_cohort / is_merge_input / generation_directive, intent refusal, CT basis / membership / requested-shape / explicit-selection validators.

src/institutional_kb/static/js/bi.js

Kept: all N03/R06/R05/F11 hunks. Parked: isClarificationAnswer gating of the clarify prefix (changes the submitted question text).

Post-goal-1 segment: every non-evidence path

Post-goal-1 segment: evidence trees (all group 1)

Goal-1 segment (main..e479e81): already on origin/main

Parked test files (group 3, moved whole)

• tests/test_answer_contract_role_promotion.py
• tests/test_ct_measure_basis_guard.py
• tests/test_federated_scalar_aggregate.py
• tests/test_federated_semantic_refusal.py
• tests/test_followup_membership_contract.py
• tests/test_insights_answer_contract_execution.py
• tests/test_insights_booking_membership_routing.py
• tests/test_insights_cold_schema_context.py
• tests/test_insights_cross_source_membership.py
• tests/test_insights_direct_service_secret_refusal.py
• tests/test_insights_federated_definition_application.py
• tests/test_insights_federated_followup_contract.py
• tests/test_insights_federated_grounding.py
• tests/test_insights_generation_schema_qualification.py
• tests/test_insights_gold_final_contract.py
• tests/test_insights_independent_source_conditions.py
• tests/test_insights_membership_generation_contract.py
• tests/test_insights_membership_measure_role.py
• tests/test_insights_membership_projection_bound.py
• tests/test_insights_membership_replay_routes.py
• tests/test_insights_n01_definition_schema.py
• tests/test_insights_pd_ct_identity_bridge.py
• tests/test_insights_planner_refusal.py
• tests/test_insights_read_only_intent.py
• tests/test_insights_requested_scalar_alias.py
• tests/test_insights_snapshot_and_trade_grain.py
• tests/test_insights_v10_empty_sum_shape_review.py
• tests/test_insights_v10_independent_terminal_review.py
• tests/test_insights_v10_member_lineage.py
• tests/test_insights_v10_membership_review.py
• tests/test_insights_v10_result_shape.py
• tests/test_insights_v11_query_review.py
• tests/test_insights_v13_guard_refusal.py
• tests/test_insights_v13_identity_refusals.py
• tests/test_insights_v13_membership_sources.py
• tests/test_insights_v13_money_source.py
• tests/test_insights_v13_trade_output_contract.py
• tests/test_insights_v14_result_independent_review.py
• tests/test_insights_v14_selection_obligations.py
• tests/test_insights_v15_membership_obligations.py
• tests/test_insights_v16_aggregate_group_scope.py
• tests/test_insights_v16_identity_refusal.py
• tests/test_insights_v16_rank_words.py
• tests/test_insights_v16_referral_scope.py
• tests/test_insights_v17_distinct_snapshot_sum.py
• tests/test_insights_v17_explicit_output.py
• tests/test_insights_v17_membership_predicates.py
• tests/test_insights_v17_output_recovery_native.py
• tests/test_insights_v18_membership_output.py
• tests/test_insights_v18_money_guard_recovery.py
• tests/test_insights_v8_chain_recovery.py
• tests/test_insights_v8_context_contract.py
• tests/test_insights_v8_ct_input_regressions.py
• tests/test_insights_v8_federation_contract.py
• tests/test_insights_v8_intent_shape_diagnostics.py
• tests/test_insights_v8_refusal_entrypoints.py
• tests/test_insights_v9_contract_review.py
• tests/test_measure_binding_payload_parity.py
• tests/test_requested_result_shape.py
• tests/test_requested_total_operation.py

Mixed test files (group-3 tests removed or file restored to base)

• tests/test_aggregate_cohort_materialization.py
• tests/test_answer_contract_saved_rerun_context.py
• tests/test_bi008_query_execution_safety.py
• tests/test_bi_definition_resolution.py
• tests/test_bi_followup_router.py
• tests/test_bi_schema_registry_api.py
• tests/test_bic006_bi_conversations.py
• tests/test_deictic_key_carry.py
• tests/test_fact_date_scope_wiring.py
• tests/test_insights_auxiliary_authorization.py
• tests/test_insights_entrypoint_authorization.py
• tests/test_insights_failed_sql_receipt.py
• tests/test_insights_refusal_provenance.py
• tests/test_insights_snapshot_source_binding.py
• tests/test_insights_v10_integrated_review.py
• tests/test_intent_classifier.py
• tests/test_merge_replay_spec.py
• tests/test_prd003_federation_dispatch.py
• tests/test_prd008_federated_planner.py
• tests/test_rt011_bi_sql_llm_runtime.py
• tests/test_sql_anchor_turn_polluted_thread.py