ГЛОБАЛЬНЫЙ РЕФАКТОРИНГ АРХИТЕКТУРЫ - Рефакторинг этапов Stage 3.4: расширен soft-refusal в deep/living, чтобы при слабом/непокрытом домене не падать в шаблонный Коротко/Что сломано.

docs/TECH/1CLLMARCH-FACT.md

@@ -2401,6 +2401,21 @@ Implemented in current pass (Stage 3.3 soft-refusal + anti-template limited repl
    - Extended regression pack: `11` files / `344` tests passed.
    - Type build: `npm --prefix llm_normalizer/backend run build` passed.
 
+Implemented in current pass (Stage 3.4 deep/living soft-refusal boundary widening):
+1. Extended boundary fallback trigger in policy composer:
+   - `shouldUseBoundaryFallbackReply(...)` now supports `broad_partial` mode when domain coverage is weak/unconfirmed.
+   - Added domain-aware gate (`focus_domain == null` or `focusDomainGroundingBlocked`) + weak-evidence envelope checks.
+2. Reduced template leakage for off-domain partial replies:
+   - Cases with weak partial support but no confirmed domain grounding now return soft-refusal boundary reply instead of rigid `Коротко/Что сломано/...` template blocks.
+3. Improved boundary fallback UX:
+   - Added deterministic heading variants to reduce repetitive wording.
+   - Added quick actionable line (`Что могу сделать сейчас`) sourced from nearest supported capability set.
+4. Added targeted regression coverage:
+   - Rewrote/normalized `assistantBoundaryFallbackReply.test.ts` in UTF-8 and added case for `broad_partial + weak evidence + uncovered domain`.
+5. Validation snapshot:
+   - Stage 3 focused suite (+boundary): `12` files / `348` tests passed.
+   - Type build: `npm --prefix llm_normalizer/backend run build` passed.
+
 Acceptance (Stage 3):
 1. LLM outputs strictly validated schema for extraction/decomposition (no free-form).
 2. Deterministic guards can block or downgrade answers when evidence insufficient.