fix(clinical): red flags, one crisis rule, callback after a crisis, one table of numbers

Clinical review of the questionnaires (ADR-013):
- Red-flag block first in stage 1 (config/instruments/red_flags.yaml); any «Да» stops the
  questionnaire with an acute alert and the help screen of its category.
- One crisis rule: the crisis screen opens at a HIGH C-SSRS risk only; moderate stops and goes
  to a psychiatrist (24 h as a guide), low is a flag with the safety plan. PHQ-9 item 9 with a
  negative screener is a discrepancy flag; without a completed screener it asks the C-SSRS at once.
  config/questionnaire.yaml: cssrs_positive_policy graded (C01).
- Every crisis leaves a callback task for the doctor on duty (post_crisis_call, one per patient)
  and the app asks how the person is the next day, the safety plan one tap away.
- Routing: risk and flags first, scales second; PCL-5 >= 50 no longer means a priority
  appointment (pcl5_severe is moderate on the dashboard too).
- Crisis numbers live only in config/texts/crisis.yaml; texts and the agent prompt refer to them
  via {phone:<key>}. Fixes the children's helpline offered to adults by the chat agent.

Refs #92

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

tests/test_api_patient_web.py

@@ -169,8 +169,8 @@ async def test_demo_flag_opens_intake_in_the_engine(client: AsyncClient, monkeyp
     await press()
     await press("consent:yes")
     buttons = await press("age:adult")
-    # Stage 1 starts with PHQ-2, in the order of the MVP questionnaire.
-    assert any(b.startswith("iq:") and ":phq9_1:" in b for b in buttons)
+    # Stage 1 starts with the red-flag block, then the order of the MVP questionnaire.
+    assert any(b.startswith("iq:") and ":rf_psychosis:" in b for b in buttons)
     assert "intake:help" in buttons and "intake:pause" in buttons
     q = (await client.get(f"/api/v1/me/{persona}/today")).json()["questionnaires"]
     assert q["enabled"] and q["intake"] == "active"
@@ -214,7 +214,7 @@ async def test_stage_one_starts_without_modules(client: AsyncClient, monkeypatch
     assert (await client.get(f"/api/v1/me/{token}/today")).json()["questionnaires"]["enabled"]
     await press()
     await press("consent:yes")
-    assert any(":phq9_1:" in b for b in await press("age:adult"))
+    assert any(":rf_psychosis:" in b for b in await press("age:adult"))
 
 
 async def test_patient_app_has_its_own_demo_personas(client: AsyncClient) -> None:

tests/test_crisis_numbers.py

@@ -0,0 +1,31 @@
+"""Crisis numbers live in one table (config/texts/crisis.yaml); every other text refers to it by `{phone:<key>}`."""
+
+import re
+from pathlib import Path
+
+import pytest
+
+from app.core.safety import CRISIS_PATH, fill_phones
+from app.llm.agent.prompts import SYSTEM_PROMPT
+
+ROOT = Path(__file__).resolve().parents[1]
+# Any phone number except the universal short emergency numbers 112 / 101–103.
+NUMBER = re.compile(r"(\+7|\b8)[\s(-]*\d{3}[\s)-]*\d{3}[\s-]*\d{2}[\s-]*\d{2}")
+
+
+def test_no_number_outside_the_crisis_table():
+    texts = [p for p in (ROOT / "config" / "texts").glob("*.yaml") if p != CRISIS_PATH]
+    code = [*(ROOT / "app").rglob("*.py"), *(ROOT / "web" / "src").rglob("*.ts*")]
+    offenders = [
+        str(p.relative_to(ROOT))
+        for p in [*texts, *code]
+        if p.name != "privacy.py"  # the PII scrubber documents the phone format with an example
+        and NUMBER.search("\n".join(line for line in p.read_text(encoding="utf-8").splitlines() if "#" not in line))
+    ]
+    assert offenders == []
+
+
+def test_references_resolve_and_unknown_keys_fail():
+    assert "989-50-50" in fill_phones("{phone:helpline}") and "989-50-50" in SYSTEM_PROMPT
+    with pytest.raises(ValueError):
+        fill_phones("{phone:no_such_line}")

tests/test_dialog_engine.py

@@ -482,4 +482,6 @@ def test_dialog_texts_avoid_forbidden_words() -> None:
     body = "\n".join(line for line in raw.splitlines() if not line.lstrip().startswith("#"))
     for word in ("учёт", "учет", "контрол", "надзор", "диагноз:"):
         assert word not in body
-    assert "152-фз" in body and "8-800-2000-122" in body
+    assert "152-фз" in body
+    # Numbers live in one table (config/texts/crisis.yaml); the minors' stop text gets the children's line from it.
+    assert "8-800-2000-122" not in body and "8-800-2000-122" in load_dialog_texts().age_gate.minor_stop

tests/test_dialog_intake.py

@@ -13,6 +13,7 @@ from app.core import clock
 from app.core.scales import clear_cache
 from app.db import repo, session
 from app.db.bootstrap import init_db
+from app.db.clinical import Alert
 from app.db.intake import QuestionnaireNotice, QuestionnaireRecord, QuestionnaireTask
 from app.dialog.contracts import Event
 from app.dialog.engine import Engine
@@ -76,7 +77,8 @@ def answer_button(replies, choices=None):
             for b in line:
                 if b.data.startswith("iq:"):
                     item = b.data.split(":")[2]
-                    index = choices.get(item, 1 if item.startswith("cssrs_") else 0)
+                    # «Нет» by default for the safety items (C-SSRS, red flags), the first option otherwise.
+                    index = choices.get(item, 1 if item.startswith(("cssrs_", "rf_")) else 0)
                     for line2 in reply.buttons:
                         for other in line2:
                             if other.data.endswith(":" + str(index)) and other.data.startswith("iq:"):
@@ -122,17 +124,17 @@ async def test_pause_restart_and_duplicate_callbacks(released):
     old = answer_button(replies)
     await asyncio.gather(press(released, old), press(released, old))
     (row,) = await rows()
-    assert row.answers == {"phq9_1": 0}
+    assert row.answers == {"rf_psychosis": 0}
     await press(released, "intake:pause")
     restarted = Engine(gateway=LLMGateway(MockTasks()))
     replies = await press(restarted, "intake:resume")
-    assert "phq9_2" in answer_button(replies)
+    assert "rf_harm_others" in answer_button(replies)
     await press(restarted, old)
-    assert (await rows())[0].answers == {"phq9_1": 0}
+    assert (await rows())[0].answers == {"rf_psychosis": 0}
     assert all(len(b.data.encode()) <= 64 for r in replies for line in r.buttons for b in line)
 
 
-@pytest.mark.parametrize("signal", ["help", "phq", "cssrs"])
+@pytest.mark.parametrize("signal", ["help", "red_flag", "cssrs"])
 async def test_immediate_stop_preserves_partial_answers(released, signal):
     replies = await press(released, "intake:start")
     if signal == "help":
@@ -142,13 +144,45 @@ async def test_immediate_stop_preserves_partial_answers(released, signal):
         for item in ("cssrs_q1", "cssrs_q2", "cssrs_q3", "cssrs_q4"):
             replies = await press(released, answer_button(replies, {item: 0}))
     else:
-        replies = await complete_existing(released, replies, {"trauma": 1, "phq9_1": 2, "phq9_9": 1})
+        replies = await press(released, answer_button(replies, {"rf_psychosis": 0}))
     assert any(r.kind == "crisis" for r in replies)
     (row,) = await rows()
     assert row.status == "interrupted" and row.completed_at is None
     notices = await rows(QuestionnaireNotice)
     assert len(notices) == 1 and notices[0].status == "test_sink"
     assert "уведомлён" not in replies[0].text
+    # Every crisis leaves a callback task for the doctor on duty (one per patient while it is open).
+    calls = [a for a in await rows(Alert) if a.code == "post_crisis_call"]
+    assert len(calls) == 1 and calls[0].params["assignee"] == "on-duty"
+
+
+async def test_red_flag_violence_uses_its_crisis_variant(released):
+    replies = await press(released, "intake:start")
+    for _ in range(3):  # psychosis, harm to others, weapons: «Нет»
+        replies = await press(released, answer_button(replies))
+    replies = await press(released, answer_button(replies, {"rf_violence": 0}))
+    crisis = next(r for r in replies if r.kind == "crisis")
+    assert "насили" in crisis.text.lower()
+    (row,) = await rows()
+    assert row.details["stop_reason"] == "red_flag_violence"
+
+
+async def test_item9_with_a_negative_screener_is_a_flag_not_a_crisis(released):
+    # One crisis rule (ADR-013): C-SSRS was negative, so PHQ-9 item 9 is a discrepancy for the doctor.
+    choices = {f"phq9_{i}": 2 for i in range(1, 9)}
+    choices.update(trauma=1, tbi=1, psychotropics=1, phq9_9=1)
+    replies = await press(released, "intake:start")
+    for _ in range(60):
+        button = answer_button(replies, choices)
+        if button is None or any(r.kind == "safety_plan" for r in replies):
+            break
+        replies = await press(released, button)
+    assert not any(r.kind == "crisis" for r in replies)
+    assert any(r.kind == "safety_plan" for r in replies)
+    codes = {a.code for a in await rows(Alert)}
+    assert "phq9_cssrs_discrepancy" in codes and "post_crisis_call" not in codes
+    (row,) = await rows()
+    assert row.status == "complete" and "phq9_cssrs_discrepancy" in row.details["decision"]["flags"]
 
 
 async def complete_existing(e, replies, choices):
@@ -350,16 +384,24 @@ async def test_diary_morning_then_evening_with_the_ptsd_module(released):
     assert any("уже заполнен" in r.text for r in replies)
 
 
-async def test_depression_module_item9_stops_with_crisis(released, monkeypatch):
+async def test_depression_module_item9_asks_the_screener_at_once(released, monkeypatch):
     await complete_intake(released, {"trauma": 1, "tbi": 1, "psychotropics": 1})
     async with session.get_clinical_session() as s:
         await repo.upsert_profile(s, token=TOKEN, modules=["depression"])
     monkeypatch.setattr(clock, "now", lambda: NOW + timedelta(days=2))
     await run_diary(released, await press(released, "menu:checkin"))
-    replies = await run_diary(released, await press(released, "menu:checkin"), {"medication": 2, "dphq_9": 1})
-    assert any(r.kind == "crisis" for r in replies)
+    replies = await press(released, "menu:checkin")
+    for _ in range(10):
+        button = answer_button(replies, {"medication": 2, "dphq_9": 1})
+        replies = await press(released, button)
+        if ":cssrs_" in (answer_button(replies) or ""):
+            break
+    # Item 9 → the C-SSRS right away («Этап 2»): no crisis screen before the screener says the risk is high.
+    assert not any(r.kind == "crisis" for r in replies) and any("последнего ответа" in r.text for r in replies)
     evening = next(r for r in await rows() if r.kind == "diary_evening")
-    assert evening.answers["dphq_9"] == 1 and evening.status == "interrupted"
+    assert evening.answers["dphq_9"] == 1 and evening.details["stop_reason"] == "phq9_item9_cssrs"
+    replies = await complete_existing(released, replies, {"cssrs_q1": 0, "cssrs_q2": 0, "cssrs_q3": 0, "cssrs_q4": 0})
+    assert any(r.kind == "crisis" for r in replies)
 
 
 async def test_unknown_sex_goes_to_review_without_a_zone(released):
@@ -380,7 +422,7 @@ async def test_version_change_blocks_resume_and_preserves_answers(released):
     path.write_text(yaml.safe_dump(data, allow_unicode=True))
     await press(released, "intake:resume")
     (row,) = await rows()
-    assert row.status == "review_required" and row.answers == {"phq9_1": 0}
+    assert row.status == "review_required" and row.answers == {"rf_psychosis": 0}
 
 
 async def test_followup_delta_uses_comparable_complete_baseline(released):
@@ -434,6 +476,8 @@ async def test_diary_schedule_uses_moscow_date(released, monkeypatch):
 
 async def test_resumed_answer_keeps_original_period(released, monkeypatch):
     replies = await press(released, "intake:start")
+    for _ in range(5):  # the red-flag block
+        replies = await press(released, answer_button(replies))
     await press(released, answer_button(replies))
     await press(released, "intake:pause")
     monkeypatch.setattr(clock, "now", lambda: NOW + timedelta(days=1))
@@ -472,3 +516,16 @@ async def test_without_a_doctor_the_screening_defines_the_modules(released):
     assert any("не выявлено" in r.text for r in replies)
     async with session.get_clinical_session() as s:
         assert (await repo.get_profile(s, TOKEN)).modules == []
+
+
+async def test_next_day_after_a_crisis_the_app_asks_once(released, monkeypatch):
+    replies = await press(released, "intake:start")
+    await press(released, answer_button(replies, {"rf_psychosis": 0}))  # red flag → crisis
+    outbox = MemoryOutbox()
+    monkeypatch.setattr(integrations, "outbox", outbox)
+    monkeypatch.setattr(clock, "now", lambda: NOW + timedelta(hours=21))
+    await jobs.plan()
+    await jobs.plan()
+    sent = outbox.drain(TOKEN)
+    assert len(sent) == 1 and "Как вы сегодня" in sent[0].text
+    assert any(b.data == "safety_plan" for line in sent[0].buttons for b in line)

tests/test_questionnaire.py

@@ -172,12 +172,16 @@ def test_unapproved_policy_never_continues_even_a_negative_screen():
 def test_immediate_safety_preempts_unfinished_or_invalid_unrelated_scales():
     for overrides in [
         dict(help_now=True, cssrs={"broken": "data"}),
-        dict(phq9={"phq9_9": 1}),
+        dict(phq9={"phq9_9": 1}, cssrs={}),  # item 9 before the screener is complete
         dict(cssrs={"q4": 1}, alcohol={"bad": 50}),
     ]:
         assert intake(policy=load_policy(), **overrides).status == "crisis"
+
+
+def test_item9_with_a_negative_screener_is_a_discrepancy_flag_not_a_crisis():
+    # One crisis rule (ADR-013): the crisis screen opens at a high C-SSRS risk only.
     r = intake(phq9={"phq9_9": 1})
-    assert "phq9_cssrs_discrepancy" in r.flags
+    assert r.status != "crisis" and {"phq9_item9_positive", "phq9_cssrs_discrepancy"} <= set(r.flags)
 
 
 def test_physician_must_choose_between_stop_and_graded():
@@ -208,7 +212,8 @@ def test_phq2_and_trauma_expansion_do_not_turn_missing_into_negative():
         (30, "self_help"),
         (31, "psychiatrist_or_psychotherapist"),
         (49, "psychiatrist_or_psychotherapist"),
-        (50, "priority_psychiatrist"),
+        (50, "psychiatrist_or_psychotherapist"),
+        (80, "psychiatrist_or_psychotherapist"),
     ],
 )
 def test_pdf_pcl_boundaries_are_distinct_from_legacy_33(total, route):

tests/test_rules.py

@@ -155,7 +155,9 @@ def test_severe_totals(scale: str, total: int, code: str, fires: bool) -> None:
     acts = evaluate(s)
     assert (code in codes(acts)) is fires
     if fires:
-        assert next(a for a in acts if a.code == code).level == AlertLevel.HIGH
+        # PCL-5 alone is a planned specialist review, not a priority (clinical review 2026-10-04).
+        expected = AlertLevel.MODERATE if scale == "pcl5" else AlertLevel.HIGH
+        assert next(a for a in acts if a.code == code).level == expected
 
 
 @pytest.mark.parametrize(

tests/test_services_monitoring.py

@@ -178,6 +178,7 @@ async def test_andrey_13_days(db: MemoryNotifier) -> None:
         "sleep_lt4_3n",
         "alcohol_3d",
         "phq9_item9",  # same evaluation: YAML order
+        "post_crisis_call",  # the crisis leaves a callback task for the doctor on duty
         "phq9_rise",
     ]
     assert len(crisis_audit) == 1 and crisis_audit[0].token == TOKEN and crisis_audit[0].actor == "system"
@@ -260,7 +261,7 @@ async def test_notifier_failure_keeps_alert(db: MemoryNotifier, monkeypatch: pyt
     await scale("pcl5", 55)
     await monitoring.RuleMonitoring().after_data_change(TOKEN)
     async with session.get_clinical_session() as s:
-        assert [(a.code, a.level) for a in await repo.list_alerts(s, token=TOKEN)] == [("pcl5_severe", "high")]
+        assert [(a.code, a.level) for a in await repo.list_alerts(s, token=TOKEN)] == [("pcl5_severe", "moderate")]
     assert await color() == "yellow"