feat: crisis message after the test and non-blocking; weapon answer triggers nothing; ЕМИАС demo = Никита

- CRISIS_BLOCKING (default off for the demo): a crisis no longer locks the dialog; the message and the
  doctor's acute alert still happen. The clinical protocol (blocking, immediate stop) stays behind the flag and
  keeps its tests.
- Crisis signals inside a questionnaire (intake red flags, high C-SSRS, PHQ-9 item 9) no longer stop it: the
  test is finished and the crisis message comes at the end; the doctor's alert is raised at once/at the end.
- Weapon access: asked and stored, triggers nothing (rule disabled, no red flag, no safety block).
- ЕМИАС demo persona is «Никита В.»; the one-click demo stays «Андрей С.».

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

app/dialog/crisis.py

@@ -14,6 +14,7 @@ Decisions:
 
 from typing import Any
 
+from app.config import get_settings
 from app.core.safety import CrisisResource, CrisisScript, crisis_reply
 from app.dialog.contracts import Reply
 from app.dialog.fsm import Ctx, Step
@@ -37,6 +38,9 @@ def open_crisis(ctx: Ctx, categories: list[str], source: str) -> Step:
         text = "\n".join([ctx.texts.crisis.repeat, *_resource_lines(script.resources)])
         return Step(replies=[Reply(text=text, kind="crisis")])
     crisis = {"open": True, "opened_at": ctx.now.isoformat(), "source": source, "variant": script.variant}
+    if not get_settings().crisis_blocking:
+        # Non-blocking (demo default): the script is shown, the doctor is alerted, the patient keeps going.
+        return Step(replies=[Reply(text=script.text, kind="crisis")], data={"crisis": crisis})
     return Step(
         replies=[Reply(text=script.text, kind="crisis")],
         next_state="crisis",

app/dialog/engine.py

@@ -37,6 +37,7 @@ from weakref import WeakValueDictionary
 from sqlalchemy.exc import InterfaceError, OperationalError
 from sqlalchemy.ext.asyncio import AsyncSession
 
+from app.config import get_settings
 from app.core import clock, safety
 from app.core.intake_runtime import intake_enabled
 from app.core.privacy import scrub
@@ -68,6 +69,9 @@ from app.services.contracts import MonitoringService, get_monitoring
 log = logging.getLogger(__name__)
 
 LAST_PROMPT = "last_prompt"
+DEFERRED_CRISIS = "deferred_crisis"
+QUESTIONNAIRE_PREFIXES = ("scale.", "veteran.onboarding.", "intake.")
+DEFERRABLE_KINDS = frozenset({"crisis", "safety_plan"})
 MEDICINE_WORDS = re.compile(
     r"(лекарств|таблет|препарат|сертралин|пароксетин|флуоксетин|эсциталопрам|антидепрессант|побочк)", re.I
 )
@@ -145,6 +149,8 @@ class Engine:
 
         await self._apply(ctx, await self._start_pending(ctx), run)
 
+        self._defer_crisis_messages(ctx, event, run)
+
         if _consented(ctx):
             for reply in run.replies:
                 # Outgoing texts may quote the safety plan (names of relatives) — scrub before storing too.
@@ -177,6 +183,21 @@ class Engine:
         )
         return run.replies
 
+    def _defer_crisis_messages(self, ctx: Ctx, event: Event, run: _Run) -> None:
+        """Crisis messages triggered by test answers wait until the questionnaire is over (team decision
+        2026-10-04: let the patient finish the test, then show the crisis message; it blocks nothing). The doctor's
+        alert is not delayed — only the patient-facing message. A typed acute message is never deferred."""
+        if get_settings().crisis_blocking:
+            return  # clinical protocol: the crisis takes over immediately
+        if event.type == "button" and ctx.state.startswith(QUESTIONNAIRE_PREFIXES):
+            held = [r for r in run.replies if r.kind in DEFERRABLE_KINDS]
+            if held:
+                run.replies[:] = [r for r in run.replies if r.kind not in DEFERRABLE_KINDS]
+                ctx.data[DEFERRED_CRISIS] = [*(ctx.data.get(DEFERRED_CRISIS) or []), *(r.model_dump() for r in held)]
+            return
+        if ctx.data.get(DEFERRED_CRISIS) and not ctx.state.startswith(QUESTIONNAIRE_PREFIXES):
+            run.replies.extend(Reply.model_validate(r) for r in ctx.data.pop(DEFERRED_CRISIS))
+
     def _abandoned(self, ctx: Ctx) -> bool:
         """The current question step was left unanswered longer than `abandon_after_hours` (app.core.clock)."""
         if intake_enabled() or not _consented(ctx) or ctx.state in RESTING_STATES or crisis_open(ctx.data):

app/dialog/flows/intake.py

@@ -4,6 +4,7 @@ import math
 from dataclasses import asdict, replace
 from zoneinfo import ZoneInfo
 
+from app.config import get_settings
 from app.core import interpretation
 from app.core.intake_questions import (
     DIARY_KINDS,
@@ -15,6 +16,7 @@ from app.core.intake_questions import (
     group_answers,
     next_intake,
     red_flag_category,
+    red_flag_triggers,
     scale_question,
     texts,
 )
@@ -116,6 +118,39 @@ async def urgent(ctx, row=None, *, code="help_now", categories=None):
     return step
 
 
+def _defer_urgent(row, code, categories=None):
+    """Remember the first crisis signal of this questionnaire; the questionnaire goes on (team decision
+    2026-10-04: the crisis message comes at the end and blocks nothing)."""
+    if not row.details.get("pending_urgent"):
+        row.details = {**row.details, "pending_urgent": {"code": code, "categories": categories or []}}
+
+
+async def _flush_urgent(ctx, row, step):
+    """Questionnaire over: acute alert + callback for the doctor, then the (non-blocking) crisis message."""
+    pending = (row.details or {}).get("pending_urgent")
+    if not pending or step.next_state == "intake.collect":
+        return step
+    row.details = {**row.details, "pending_urgent": None}
+    code, categories = pending["code"], pending.get("categories") or []
+    profile = await ctx.profile()
+    if profile:
+        profile.enhanced_monitoring = True
+    await service.alert(
+        ctx.session,
+        ctx.token,
+        f"{row.id}:{code}",
+        code,
+        ctx.now,
+        level=AlertLevel.ACUTE,
+        params={"record_id": row.id, "deferred": True},
+    )
+    await schedule_post_crisis_call(ctx.session, ctx.token, ctx.now, source="questionnaire:" + code)
+    crisis = open_crisis(ctx, categories, source="questionnaire:" + code)
+    for reply in crisis.replies:
+        reply.buttons.append([Button(label=ctx.texts.safety_plan.title.rstrip(":"), data="safety_plan")])
+    return step.then(crisis)
+
+
 async def item9_discrepancy(ctx, row):
     profile = await ctx.profile()
     if profile:
@@ -170,18 +205,27 @@ async def manual_review(ctx, row, code):
 
 
 async def advance(ctx, row):
+    return await _flush_urgent(ctx, row, await _advance(ctx, row))
+
+
+async def _advance(ctx, row):
     a = row.answers
     c = score_cssrs_recent(cssrs_answers(a))
     policy = load_policy()
     # Red flags (asked first): any «Да» stops the questionnaire with the help screen of its category.
     for item_id, value in a.items():
-        if item_id.startswith("rf_") and value == 1:
+        if item_id.startswith("rf_") and value == 1 and red_flag_triggers(item_id):
             category = red_flag_category(item_id)
-            return await urgent(ctx, row, code="red_flag_" + item_id[3:], categories=[category] if category else [])
+            if get_settings().crisis_blocking:  # clinical protocol: stop at once
+                return await urgent(ctx, row, code="red_flag_" + item_id[3:], categories=[category] if category else [])
+            _defer_urgent(row, "red_flag_" + item_id[3:], [category] if category else [])
+            break
     # One crisis rule for the whole app (ADR-013): the crisis screen opens at a HIGH C-SSRS risk only. Moderate →
     # stop + psychiatrist (about 24 h); low → flag, safety plan, continue.
     if c.risk_lower_bound == "high":
-        return await urgent(ctx, row, code="suicide_signal")
+        if get_settings().crisis_blocking:
+            return await urgent(ctx, row, code="suicide_signal")
+        _defer_urgent(row, "suicide_signal")
     if a.get("dphq_9", 0) > 0 and row.kind in DIARY_KINDS:
         # Diary depression module, item 9 ≥1 → the C-SSRS right away («Этап 2»); its risk level decides.
         return await item9_assessment(ctx, row)

app/dialog/scale_session.py

@@ -33,6 +33,7 @@ from dataclasses import replace
 from datetime import timedelta
 from typing import Any
 
+from app.config import get_settings
 from app.core.privacy import scrub
 from app.core.scales import Item, ScaleDef, ScaleResult, load_scale, score
 from app.db import repo
@@ -237,7 +238,10 @@ async def _record(
     sex = profile.sex if profile is not None and profile.sex in ("m", "f") else None
     result = score(scale.id, answers, sex=sex)  # type: ignore[arg-type]
     crisis_flags = [f for f in result.flags if f in ctx.texts.settings.crisis_flags]
-    if crisis_flags or session["idx"] + 1 >= len(scale.items):
+    # Clinical protocol (crisis_blocking): a crisis flag ends the scale at once. Demo default: the scale is finished
+    # first and the crisis message comes at the end of the questionnaire (engine defers it), blocking nothing.
+    stop_now = bool(crisis_flags) and get_settings().crisis_blocking
+    if stop_now or session["idx"] + 1 >= len(scale.items):
         return await _finish(ctx, scale, session, answers, result, crisis_flags)
     updated = {**session, "answers": answers, "idx": session["idx"] + 1}
     sessions = {**(ctx.data.get("scales") or {}), scale.id: updated}
@@ -278,7 +282,7 @@ async def _finish(
     # Crisis path decided by the scale's own flags (deterministic), not by the LLM.
     risk_screen = ctx.texts.settings.risk_screen_scale
     if crisis_open(ctx.data):
-        crisis = Step(next_state="crisis")
+        crisis = Step(next_state="crisis") if get_settings().crisis_blocking else Step()
     else:
         crisis = open_crisis(ctx, categories=[], source=f"scale:{scale.id}")
     step = done.then(crisis)