Post-F: закрепить семантическую целостность MCP-цепочек
This commit is contained in:
+7
-1
@@ -154,7 +154,7 @@ This phase is only healthy when the following are true:
|
||||
5. exact route success cannot silently degrade into a semantically wrong user answer;
|
||||
6. replay verdict is based first on human business meaning, only then on technical internals.
|
||||
|
||||
## Current Status - 2026-04-23
|
||||
## Current Status - 2026-04-24
|
||||
|
||||
This phase is already active in runtime code and replay-backed.
|
||||
|
||||
@@ -164,6 +164,10 @@ Materially hardened in this pass:
|
||||
- explicit one-organization open value-flow asks now stop for organization clarification instead of answering a global total under missing scope;
|
||||
- incoming-vs-outgoing money comparisons now defer away from one-sided exact supplier/customer recipes into the bounded bidirectional value-flow discovery path;
|
||||
- document asks with numeric counterparty suffixes such as `Жуковка 51` no longer collapse into bank-operation routing only because the suffix looks like account `51`;
|
||||
- rejected LLM predecompose rewrites can no longer inject account scope or truncated semantic hints for numeric counterparty suffixes such as `Жуковке 51`; the session state now keeps the counterparty anchor and does not carry `account: 51`;
|
||||
- explicit raw entity search after a ranking/value-flow dialogue now resets stale assistant/followup scope: `Найди в 1С Группу СВК` becomes `source_signal=raw_text` with `explicit_entity_candidates=["Группа СВК"]`, no stale organization/date/ranking scope in the discovery turn input, and downstream follow-ups then carry the grounded SVK object intentionally;
|
||||
- open-organization ranked value-flow questions now stay subjectless by design: predicate-shaped entity pollution such as `принёс наибольшую выручку организации в` is discarded, referential organization placeholders such as `эта организация` are resolved through the active organization scope, and the MCP ranking answer overrides exact-lane year-summary replies when the user asked `кто`;
|
||||
- ranked value-flow answer drafts no longer leak raw English/internal evidence lines into the user-facing response;
|
||||
- VAT exact materialization now survives period-window filtering instead of collapsing into `recipe_visibility_gap`;
|
||||
- confirmed VAT tax-period turns now keep raw month extraction separate from runtime tax-quarter execution, so exact VAT routes do not lose the intended tax period;
|
||||
- post-pivot receivables recovery now preserves the selected open-items recipe without weakening strict account-scope guards for risk replies;
|
||||
@@ -183,6 +187,8 @@ Replay-backed anchors for the current layer include:
|
||||
- `address_truth_harness_phase72_document_to_contracts_year_switch_live_rerun3`
|
||||
- `address_truth_harness_phase80_payments_to_contracts_to_documents_all_time_live_rerun1`
|
||||
- `address_truth_harness_phase82_human_mixed_integrity_status_dialog_post_m23_rerun_documents_scope_bidirectional`
|
||||
- `address_truth_harness_phase82_human_mixed_integrity_status_dialog_post_f_account_injection_guard_clean_scope`
|
||||
- `address_truth_harness_post_f_cross_stage_canary_agent_20260424_live5`, accepted `24/24`, and saved into autoruns as `AGENT | Post-F cross-stage semantic integrity canary` (`gen-ag04241406-abe4d8`)
|
||||
|
||||
## Honest Remaining Risk
|
||||
|
||||
|
||||
@@ -36,7 +36,7 @@ This package answers the next question:
|
||||
16. [16 - data_need_graph_and_open_world_mcp_plan_2026-04-22.md](./16%20-%20data_need_graph_and_open_world_mcp_plan_2026-04-22.md)
|
||||
17. [17 - post_f_semantic_integrity_hardening_2026-04-23.md](./17%20-%20post_f_semantic_integrity_hardening_2026-04-23.md)
|
||||
|
||||
## Current Status Snapshot (2026-04-23)
|
||||
## Current Status Snapshot (2026-04-24)
|
||||
|
||||
This package is no longer planning-only.
|
||||
|
||||
@@ -65,7 +65,7 @@ Current honest status:
|
||||
- pre-multidomain readiness: `~88%`
|
||||
- bounded-autonomy foundation readiness: `~86%`
|
||||
- open-world bounded-autonomy readiness: `~71%`
|
||||
- graph snapshot after latest rebuild: `5886 nodes`, `12754 edges`, `136 communities`
|
||||
- graph snapshot after latest rebuild: `5891 nodes`, `12769 edges`, `135 communities`
|
||||
- current breakpoint:
|
||||
- the validated hot paths are no longer structurally broken;
|
||||
- flagship continuity collapse is no longer the primary risk;
|
||||
@@ -91,6 +91,8 @@ Latest live proof now includes:
|
||||
- `address_truth_harness_phase72_document_to_contracts_year_switch_live_rerun3` accepted
|
||||
- `address_truth_harness_phase80_payments_to_contracts_to_documents_all_time_live_rerun1` accepted
|
||||
- `address_truth_harness_phase82_human_mixed_integrity_status_dialog_post_m23_rerun_documents_scope_bidirectional` accepted `19/19`
|
||||
- `address_truth_harness_phase82_human_mixed_integrity_status_dialog_post_f_account_injection_guard_clean_scope` accepted `19/19`, with the `Жуковке 51` numeric counterparty suffix kept as counterparty scope instead of leaking as account `51`
|
||||
- `address_truth_harness_post_f_cross_stage_canary_agent_20260424_live5` accepted `24/24`, proving a saved cross-stage AGENT canary across VAT metadata, numeric counterparty suffixes, open-organization value-flow clarification, ranked value-flow year switches, and SVK grounded reset; the saved autorun is `AGENT | Post-F cross-stage semantic integrity canary` (`gen-ag04241406-abe4d8`)
|
||||
- `address_truth_harness_phase11_manual_followup_meta_quality_live_rerun_vatfix` accepted `10/10`
|
||||
- `address_truth_harness_phase20_continuity_stabilization_live_rerun_vatfix` accepted `6/6`
|
||||
- `addressQueryRuntimeM23.test.ts` full semantic/runtime slice accepted `403/403` after Post-F VAT/date-basis, scope-recovery, open value-flow organization clarification, document-vs-bank arbitration, and reply-shape hardening
|
||||
|
||||
Reference in New Issue
Block a user