| 1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680 |
- """Tests for the assistant service (Phase 2) using synthetic event streams."""
- import itertools
- import json
- import re
- import threading
- import time
- from typing import Any
- from stt_daemon import events
- from stt_daemon.assistant.inference import InferenceJob, InferenceQueue
- from stt_daemon.assistant.service import EXECUTE_BUSY, AssistantService
- from stt_daemon.assistant.triage import TriageAnalyzer
- from stt_daemon.assistant.triage_dump import TriageDumper
- from stt_daemon.config import Config
- _sequence = itertools.count(1)
- # A real analyzer (pure, stdlib-only helpers) backing the fake's retention
- # guard so guard behaviour is exercised end to end.
- _real_analyzer = TriageAnalyzer(Config().triage)
- class Clock:
- def __init__(self, start=1000.0):
- self.t = float(start)
- def __call__(self):
- return self.t
- def advance(self, dt):
- self.t += float(dt)
- class FakeSubscriber:
- def events(self):
- return iter(())
- def close(self):
- pass
- def _service():
- cfg = Config()
- clock = Clock()
- emitted: list[dict] = []
- def emit(type_, **fields):
- emitted.append({"type": type_, **fields})
- service = AssistantService(cfg, FakeSubscriber(), emit=emit, time_fn=clock)
- return service, emitted, clock
- def _ev(type_, clock, **fields):
- event = {"v": 1, "type": type_, "seq": next(_sequence), "ts": clock()}
- event.update(fields)
- return event
- def _transcript(clock, text, ts=None, session_dir=None):
- event = {
- "v": 1,
- "type": events.TRANSCRIPT,
- "seq": next(_sequence),
- "ts": clock() if ts is None else ts,
- "text": text,
- }
- if session_dir is not None:
- event["session_dir"] = session_dir
- return event
- def test_ready_requires_matching_segment_done():
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_START, clock))
- assert service.ready() is False
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- assert service.ready() is False # awaiting segment 1
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- assert service.ready() is True
- def test_stale_segment_done_does_not_satisfy_later_turn():
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- service.handle_event(_ev(events.SPEECH_START, clock))
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=2))
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- assert service.ready() is False # still awaiting segment 2
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=2, emitted=True))
- assert service.ready() is True
- def test_gated_segment_done_clears_readiness():
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=3))
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=3, emitted=False))
- assert service.ready() is True
- def test_speech_stop_without_seg_id_clears_readiness():
- """A discarded turn closes with speech_stop but no seg_id; readiness must recover."""
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_START, clock))
- assert service.ready() is False
- service.handle_event(_ev(events.SPEECH_STOP, clock)) # discarded: no seg_id
- assert service.speech_open is False
- assert service.awaited_seg_id is None
- assert service.ready() is True
- def test_status_resets_stale_speech_state_on_reconnect():
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_START, clock))
- assert service.speech_open is True
- service.handle_event(_ev(events.STATUS, clock, capture="x", segments_done=0))
- assert service.speech_open is False
- assert service.awaited_seg_id is None
- assert service.ready() is True
- def test_command_buffer_uses_arm_timestamp():
- service, _, clock = _service()
- clock.advance(10)
- assert service.arm() is True
- arm_ts = service.arm_ts
- assert arm_ts is not None
- # Replayed transcript from before arming is ignored.
- service.handle_event(_transcript(clock, "old", ts=arm_ts - 5))
- assert service.command_text == ""
- service.handle_event(_transcript(clock, "what time is it"))
- assert service.command_text == "what time is it"
- def test_mutual_exclusion_between_command_and_chat():
- service, _, _ = _service()
- assert service.toggle_chat() is True
- assert service.arm() is False # cannot arm in chat
- assert service.toggle_chat() is True # back to idle
- assert service.arm() is True
- assert service.toggle_chat() is False # cannot chat while armed
- def test_arm_timeout_disarms_and_emits_error():
- service, emitted, clock = _service()
- service.arm()
- clock.advance(service.cfg.assistant.command_arm_timeout_s + 0.1)
- service.tick()
- assert service.modes.mode.value == "idle"
- assert service.command_text == ""
- assert any(e["type"] == "error" and e["code"] == "arm_timeout" for e in emitted)
- def test_long_speech_does_not_disarm():
- service, emitted, clock = _service()
- service.arm()
- service.handle_event(_ev(events.SPEECH_START, clock))
- clock.advance(service.cfg.assistant.command_arm_timeout_s + 30.0)
- service.tick()
- assert service.modes.mode.value == "command_armed"
- assert not any(e["type"] == "error" for e in emitted)
- def test_captured_command_survives_arm_timeout():
- service, emitted, clock = _service()
- service.arm()
- service.handle_event(_transcript(clock, "what time is it"))
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- clock.advance(service.cfg.assistant.command_arm_timeout_s + 30.0)
- service.tick()
- assert service.modes.mode.value == "command_armed"
- assert service.command_text == "what time is it"
- assert not any(e["type"] == "error" for e in emitted)
- def test_stt_stall_cancels_armed_command():
- service, emitted, clock = _service()
- service.handle_event(_ev(events.HEARTBEAT, clock, worker_stall_s=0.0))
- service.arm()
- clock.advance(service.cfg.assistant.stall_timeout_ms / 1000.0 + 0.5)
- service.tick()
- assert service.modes.mode.value == "idle"
- assert any(e["type"] == "error" and e["code"] == "stt_stall" for e in emitted)
- def test_worker_stall_marks_stt_degraded():
- service, _, clock = _service()
- long_stall = service.cfg.assistant.worker_stall_ms / 1000.0 + 1.0
- service.handle_event(_ev(events.HEARTBEAT, clock, worker_stall_s=long_stall))
- assert service.stt_ok() is False
- assert service.ready() is False
- def test_chat_idle_timeout_exits_chat():
- service, _, clock = _service()
- assert service.toggle_chat() is True
- clock.advance(service.cfg.assistant.chat_idle_timeout_s + 0.1)
- service.tick()
- assert service.modes.mode.value == "idle"
- def test_chat_exit_phrase_exits_chat():
- service, _, clock = _service()
- assert service.toggle_chat() is True
- service.handle_event(_transcript(clock, "That's all."))
- assert service.modes.mode.value == "idle"
- assert service.chat_history == []
- def test_viewer_detached_exits_chat_and_forgets_history():
- service, _, clock = _service()
- assert service.toggle_chat() is True
- service.handle_event(_transcript(clock, "hello"))
- assert service.viewer_detached() is True
- assert service.modes.mode.value == "idle"
- assert service.chat_history == []
- def test_viewer_detached_is_noop_outside_chat():
- service, _, _ = _service()
- assert service.viewer_detached() is False
- assert service.modes.mode.value == "idle"
- def test_exit_chat_invokes_chat_reset():
- cfg = Config()
- resets: list[int] = []
- service = AssistantService(
- cfg, FakeSubscriber(), time_fn=Clock(), chat_reset=lambda: resets.append(1)
- )
- service.toggle_chat()
- service.viewer_detached()
- assert resets == [1]
- def test_chat_accumulates_turns():
- service, _, clock = _service()
- service.toggle_chat()
- for text in ("hello there", "how are you"):
- service.handle_event(_transcript(clock, text))
- assert service.chat_history == ["Unknown: hello there", "Unknown: how are you"]
- def test_execute_requires_readiness():
- service, _, clock = _service()
- service.arm()
- service.handle_event(_transcript(clock, "do it"))
- service.handle_event(_ev(events.SPEECH_START, clock))
- assert service.execute() is None # speech still open
- assert service.modes.mode.value == "command_armed"
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- assert service.execute() is None # awaiting segment 1
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- assert service.execute() == "do it"
- assert service.modes.mode.value == "executing"
- assert service.finish() is True
- assert service.modes.mode.value == "idle"
- def test_execute_reports_busy_when_queue_rejects():
- service, emitted, clock = _service()
- # A zero-capacity lane rejects the priority command job on submit.
- service._inference = InferenceQueue(size=0)
- service.executor = object()
- service.arm()
- service.handle_event(_transcript(clock, "do it"))
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- assert service.execute() is EXECUTE_BUSY
- assert service.modes.mode.value == "idle"
- assert not any(e.get("code") == "busy" for e in emitted)
- def test_turn_ended_requires_quiet_gap():
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- assert service.ready() is True
- assert service.turn_ended() is False
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- assert service.turn_ended() is True
- def test_emits_mode_and_command_text():
- service, emitted, clock = _service()
- service.arm()
- assert {"type": "mode", "mode": "command_armed"} in emitted
- service.handle_event(_transcript(clock, "hi"))
- assert any(e["type"] == "command_text" and e["text"] == "hi" for e in emitted)
- def test_cancel_clears_command_buffer():
- service, _, clock = _service()
- service.arm()
- service.handle_event(_transcript(clock, "hi"))
- assert service.cancel() is True
- assert service.modes.mode.value == "idle"
- assert service.command_text == ""
- assert service.arm_ts is None
- def test_pending_chat_is_not_lost_across_threads():
- """Transcripts appended on the reader thread while the ticker dispatches
- must all reach the model: the join-and-clear in _dispatch_chat is atomic,
- and the user bubble is emitted by the worker in _run_chat."""
- cfg = Config()
- cfg.assistant.turn_gap_s = 0.01
- dispatched: list[str] = []
- lock = threading.Lock()
- def emit(type_, **fields):
- if type_ == "chat_message" and fields.get("role") == "user":
- with lock:
- dispatched.append(fields["text"])
- service = AssistantService(cfg, FakeSubscriber(), emit=emit, chat_fn=lambda text: "ok")
- service.toggle_chat()
- service.start()
- try:
- total = 200
- producer = threading.Thread(
- target=lambda: [
- service.handle_event(
- {
- "v": 1,
- "type": events.TRANSCRIPT,
- "seq": i,
- "ts": time.time(),
- "text": f"w{i}",
- }
- )
- for i in range(total)
- ],
- daemon=True,
- )
- producer.start()
- deadline = time.time() + 10.0
- while time.time() < deadline:
- service.tick()
- with lock:
- words = sum(len(text.split()) for text in dispatched)
- done = words == 2 * total
- if not producer.is_alive() and done:
- break
- time.sleep(0.002)
- producer.join(timeout=5)
- with lock:
- words = sum(len(text.split()) for text in dispatched)
- assert words == 2 * total
- finally:
- service.stop()
- def _full_queue_service():
- service, emitted, clock = _service()
- service.toggle_chat()
- service._inference = InferenceQueue(size=1)
- service._inference.submit(
- InferenceJob(name="blocker", run=lambda: None, evictable=False)
- )
- return service, emitted, clock
- def test_chat_refusal_rebuffers_and_emits_queue_full_once():
- """A refused chat submit re-buffers under lock and emits queue_full on flip."""
- service, emitted, _ = _full_queue_service()
- service._pending_chat = ["hello world"]
- service._dispatch_chat()
- assert service._pending_chat == ["hello world"]
- assert service._chat_queue_full is True
- errors = [e for e in emitted if e["type"] == "error" and e.get("code") == "queue_full"]
- assert len(errors) == 1
- assert [e for e in emitted if e["type"] == "chat_message"] == []
- service._dispatch_chat()
- assert service._pending_chat == ["hello world"]
- errors = [e for e in emitted if e["type"] == "error" and e.get("code") == "queue_full"]
- assert len(errors) == 1
- def test_chat_dispatch_success_clears_latch_without_user_bubble():
- """Success clears the latch; the user bubble comes from _run_chat, not dispatch."""
- service, emitted, _ = _full_queue_service()
- service._pending_chat = ["hello world"]
- service._chat_queue_full = True
- service._inference.drop(lambda j: True)
- service._dispatch_chat()
- assert service._pending_chat == []
- assert service._chat_queue_full is False
- assert [e for e in emitted if e["type"] == "chat_message"] == []
- def test_run_chat_emits_user_first_then_reply():
- service, emitted, _ = _service()
- service.chat_fn = lambda text: "reply!"
- service._run_chat("hello")
- assert emitted == [
- {"type": "chat_message", "role": "user", "text": "hello"},
- {"type": "chat_message", "role": "assistant", "text": "reply!"},
- ]
- def test_chat_notify_emits_compaction_note():
- service, emitted, _ = _service()
- service.chat_notify("compacted", turns=4)
- assert {
- "type": "chat_message",
- "role": "note",
- "text": "Compacted 4 earlier turns into a summary.",
- } in emitted
- def test_chat_notify_emits_truncated_error():
- service, emitted, _ = _service()
- service.chat_notify("truncated", message="cut off")
- assert {"type": "error", "code": "truncated", "message": "cut off"} in emitted
- def test_clear_empties_chat_history():
- service, emitted, clock = _service()
- service.toggle_chat()
- for text in ("hello", "how are you"):
- service.handle_event(_transcript(clock, text))
- assert service.chat_history == ["Unknown: hello", "Unknown: how are you"]
- assert len(service._pending_chat) == 2
- assert service.clear() is True
- assert service.chat_history == []
- assert service._pending_chat == []
- assert {"type": "cleared"} in emitted
- def test_clear_invokes_chat_reset():
- cfg = Config()
- resets: list[int] = []
- service = AssistantService(
- cfg, FakeSubscriber(), time_fn=Clock(), chat_reset=lambda: resets.append(1)
- )
- service.toggle_chat()
- service.clear()
- assert resets == [1]
- def test_clear_does_not_change_mode():
- service, _, clock = _service()
- service.toggle_chat()
- assert service.modes.mode.value == "chat"
- service.clear()
- assert service.modes.mode.value == "chat"
- def test_clear_always_returns_true():
- service, _, _ = _service()
- assert service.clear() is True
- service.toggle_chat()
- service.handle_event(_ev(events.TRANSCRIPT, Clock(), text="test"))
- assert service.clear() is True
- def test_heartbeat_emits_inference_depth_and_in_flight():
- """Calling tick() emits heartbeat with inference_depth (int) and inference_in_flight (bool)."""
- service, emitted, clock = _service()
- service.tick()
- heartbeats = [e for e in emitted if e["type"] == "heartbeat"]
- assert len(heartbeats) == 1
- heartbeat = heartbeats[0]
- assert "inference_depth" in heartbeat
- assert isinstance(heartbeat["inference_depth"], int)
- assert "inference_in_flight" in heartbeat
- assert isinstance(heartbeat["inference_in_flight"], bool)
- def test_status_contains_inference_depth_and_in_flight():
- """status() snapshot contains inference_depth (int) and inference_in_flight (bool)."""
- service, _, _ = _service()
- status = service.status()
- assert "inference_depth" in status
- assert isinstance(status["inference_depth"], int)
- assert "inference_in_flight" in status
- assert isinstance(status["inference_in_flight"], bool)
- _AUTO = object()
- class FakeTriage:
- def __init__(
- self,
- questions=_AUTO,
- state_pair=("ok", ""),
- escalate=False,
- parsed=_AUTO,
- normalized=_AUTO,
- gate=True,
- ):
- if questions is _AUTO:
- questions = {"urgency": {"type": "noul", "instructions": "how urgent"}}
- self._questions = questions
- self._state_pair = state_pair
- self._escalate = escalate
- self._parsed = parsed
- self._normalized = normalized
- self._gate = gate
- self.build_state_calls: list[tuple] = []
- self.normalize_calls: list = []
- self.prompt_calls: list[tuple] = []
- self.gate_calls: list = []
- @property
- def questions(self):
- return self._questions
- def state(self):
- return self._state_pair
- def build_state(self, window):
- self.build_state_calls.append(tuple(window))
- return "\n".join("- " + str(item) for item in (window or []))
- def normalize(self, payload):
- self.normalize_calls.append(payload)
- if self._normalized is not _AUTO:
- return self._normalized
- if not isinstance(payload, dict):
- return {}
- signals = {"order": [], "items": {}, "escalated": False, "input_tokens": 0}
- signals.update(payload)
- if self._escalate:
- signals["escalated"] = True
- return signals
- def build_brief_prompt(self, deltas, signals):
- self.prompt_calls.append(
- (tuple(deltas), dict(signals) if isinstance(signals, dict) else signals)
- )
- return "BRIEF-PROMPT"
- def build_pinned_brief_prompt(self, current_brief, deltas, signals):
- self.prompt_calls.append(
- (tuple(deltas), dict(signals) if isinstance(signals, dict) else signals)
- )
- body = "\n".join("- " + d for d in deltas)
- return (
- f"CURRENT BRIEF:\n{current_brief or '(none yet)'}\n\nNEW UTTERANCES:\n{body}"
- )
- def extract_bullets(self, brief):
- return _real_analyzer.extract_bullets(brief)
- def restore_missing_bullets(self, new_brief, prior_brief):
- return _real_analyzer.restore_missing_bullets(new_brief, prior_brief)
- def parse_brief(self, text):
- if self._parsed is _AUTO:
- return text if text else None
- return self._parsed
- def should_update_brief(self, signals):
- self.gate_calls.append(signals)
- return self._gate
- class FakeBriefFile:
- """Recording stand-in for BriefFile (write/clear/set_directory never raise)."""
- def __init__(self):
- self.writes: list[str] = []
- self.clears = 0
- self.directories: list[str] = []
- def set_directory(self, directory):
- self.directories.append(directory)
- def write(self, text):
- self.writes.append(text)
- def clear(self):
- self.clears += 1
- class FakeLane:
- def __init__(self, ready=True):
- self._ready = ready
- self.submits: list[tuple] = []
- self.drops = 0
- self.predict_gate = threading.Event()
- def ready(self):
- return self._ready
- def state(self):
- return ("ok", "") if self._ready else ("loading", "")
- def submit(self, state, questions, lang, generation):
- self.submits.append((state, questions, lang, generation))
- return True
- def drop_pending(self):
- self.drops += 1
- class FakeDebugSink:
- def __init__(self):
- self.events: list[tuple] = []
- def emit(self, type_, **fields):
- self.events.append((type_, fields))
- def _ok_reply(text):
- return "ok"
- def _triage_service(
- triage=None,
- lane=None,
- brief_fn=None,
- chat_fn: Any = _AUTO,
- brief_file=None,
- brief_reset=None,
- chat_reset=None,
- debug_sink=None,
- ):
- cfg = Config()
- cfg.assistant.turn_gap_s = 0.5
- clock = Clock()
- emitted: list[dict] = []
- def emit(type_, **fields):
- emitted.append({"type": type_, **fields})
- if triage is None:
- triage = FakeTriage()
- if lane is None:
- lane = FakeLane()
- if chat_fn is _AUTO:
- chat_fn = _ok_reply
- factory = (lambda: debug_sink) if debug_sink is not None else None
- service = AssistantService(
- cfg,
- FakeSubscriber(),
- emit=emit,
- time_fn=clock,
- triage=triage,
- laya_lane=lane,
- brief_fn=brief_fn,
- brief_reset=brief_reset,
- brief_file=brief_file,
- chat_fn=chat_fn,
- chat_reset=chat_reset,
- debug_sink_factory=factory,
- )
- return service, emitted, clock, triage, lane
- def _triage_events(emitted):
- return [e for e in emitted if e["type"] == "triage"]
- def _escalated_payload():
- return {"order": [], "items": {}, "escalated": True, "input_tokens": 3}
- def test_triage_status_disabled_when_no_analyzer():
- service, _, _ = _service()
- status = service.status()
- assert status["triage_state"] == "disabled"
- assert status["triage_reason"] == ""
- assert status["triage"] is None
- def test_triage_status_reflects_analyzer():
- service, _, _, _, _ = _triage_service(
- triage=FakeTriage(state_pair=("loading", "warming up"))
- )
- status = service.status()
- assert status["triage_state"] == "loading"
- assert status["triage_reason"] == "warming up"
- assert status["triage"] is None
- def test_status_contains_questions_from_analyzer():
- questions = {
- "urgency": {"type": "noul", "instructions": "how urgent"},
- "sentiment": {"type": "score", "instructions": "tone", "criteria": ["low", "high"]},
- }
- service, _, _, triage, _ = _triage_service(triage=FakeTriage(questions=questions))
- status = service.status()
- assert status["questions"] == questions
- assert status["questions"] is triage.questions
- def test_status_questions_none_without_analyzer():
- service, _, _ = _service()
- assert service.status()["questions"] is None
- def test_status_event_carries_questions():
- service, emitted, _, triage, _ = _triage_service()
- service.tick()
- statuses = [e for e in emitted if e["type"] == "status"]
- assert len(statuses) == 1
- assert statuses[0]["questions"] is triage.questions
- def test_triage_captures_transcript_lang():
- service, _, clock, _, _ = _triage_service()
- assert service._last_transcript_lang is None
- event = _transcript(clock, "hola")
- event["lang"] = "es"
- service.handle_event(event)
- assert service._last_transcript_lang == "es"
- service.handle_event(_transcript(clock, "hello"))
- assert service._last_transcript_lang is None
- def test_maybe_triage_passes_lang_to_lane():
- service, _, clock, _, lane = _triage_service()
- event = _transcript(clock, "hola")
- event["lang"] = "es"
- service.handle_event(event)
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- _, _, lang, _ = lane.submits[0]
- assert lang == "es"
- def test_maybe_triage_runs_in_idle_mode():
- service, _, clock, triage, lane = _triage_service()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- state, questions, lang, gen = lane.submits[0]
- assert state == "- hello"
- assert questions is triage.questions
- assert lang is None
- assert gen == 0
- assert triage.build_state_calls == [("hello",)]
- assert service.triage_history == ["hello"]
- def test_maybe_triage_uses_transcript_buffer_not_chat_history():
- service, _, clock, triage, lane = _triage_service(chat_fn=None)
- service.chat_history = ["stale chat only"]
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert lane.submits == []
- assert triage.build_state_calls == []
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- assert triage.build_state_calls == [("hello",)]
- def test_maybe_triage_requires_settled_turn():
- service, _, clock, _, lane = _triage_service()
- service.handle_event(_transcript(clock, "hello"))
- service.tick()
- assert lane.submits == []
- def test_maybe_triage_periodic_fires_before_turn_settles():
- service, _, clock, _, lane = _triage_service()
- service.cfg.triage.periodic_s = 0.2
- service.handle_event(_transcript(clock, "hello"))
- assert service.turn_ended() is False
- clock.advance(0.25) # due (0.2) yet still quieter than turn_gap_s (0.5)
- service.tick()
- assert service.turn_ended() is False
- assert len(lane.submits) == 1
- assert lane.submits[0][0] == "- hello"
- def test_maybe_triage_periodic_disabled_ignores_accumulating_speech():
- service, _, clock, _, lane = _triage_service()
- service.cfg.triage.periodic_s = 0.0
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(0.25)
- service.tick()
- assert service.turn_ended() is False
- assert lane.submits == []
- def test_maybe_triage_periodic_rearms_after_submit():
- service, _, clock, _, lane = _triage_service()
- service.cfg.triage.periodic_s = 0.2
- service.handle_event(_transcript(clock, "first"))
- clock.advance(0.25)
- service.tick()
- assert len(lane.submits) == 1
- service.tick()
- service.tick()
- assert len(lane.submits) == 1 # identical window, no new speech
- service.handle_event(_transcript(clock, "second"))
- clock.advance(0.25)
- service.tick()
- assert len(lane.submits) == 2
- assert lane.submits[1][0] == "- first\n- second"
- def test_triage_debug_emits_submit_and_result():
- sink = FakeDebugSink()
- service, _, clock, _, lane = _triage_service(debug_sink=sink)
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- submits = [event for event in sink.events if event[0] == "triage-submit"]
- assert submits == [
- ("triage-submit", {"trigger": "settled", "window": 1, "generation": 0})
- ]
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- results = [event for event in sink.events if event[0] == "triage-result"]
- assert len(results) == 1
- assert results[0][1]["escalated"] is True
- assert results[0][1]["generation"] == 0
- def test_triage_debug_marks_periodic_trigger():
- sink = FakeDebugSink()
- service, _, clock, _, lane = _triage_service(debug_sink=sink)
- service.cfg.triage.periodic_s = 0.2
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(0.25)
- service.tick()
- submits = [event for event in sink.events if event[0] == "triage-submit"]
- assert submits[0][1]["trigger"] == "periodic"
- def test_triage_debug_emits_brief_dispatch():
- sink = FakeDebugSink()
- service, _, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: "RAW", debug_sink=sink
- )
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- brief = [event for event in sink.events if event[0] == "triage-brief"]
- assert brief == [("triage-brief", {"action": "dispatch", "generation": 0})]
- def test_triage_debug_emits_brief_gate_skip():
- sink = FakeDebugSink()
- service, _, clock, _, lane = _triage_service(
- triage=FakeTriage(gate=False), brief_fn=lambda prompt: "RAW", debug_sink=sink
- )
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- brief = [event for event in sink.events if event[0] == "triage-brief"]
- assert brief == [
- ("triage-brief", {"action": "skip", "reason": "gate", "generation": 0})
- ]
- def test_maybe_triage_skips_unready_lane():
- service, _, clock, _, lane = _triage_service(lane=FakeLane(ready=False))
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert lane.submits == []
- def test_maybe_triage_skips_without_questions():
- service, _, clock, triage, lane = _triage_service(
- triage=FakeTriage(questions=None)
- )
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert lane.submits == []
- assert triage.build_state_calls == []
- def test_maybe_triage_dedupes_identical_window():
- service, _, clock, triage, lane = _triage_service()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- state, questions, lang, gen = lane.submits[0]
- assert state == "- hello"
- assert questions is triage.questions
- assert lang is None
- assert gen == 0
- service.tick()
- service.tick()
- assert len(lane.submits) == 1
- service.handle_event(_transcript(clock, "second"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 2
- assert lane.submits[1][0] == "- hello\n- second"
- def test_second_transcript_produces_second_submit():
- service, _, clock, _, lane = _triage_service()
- service.handle_event(_transcript(clock, "first"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- service.handle_event(_transcript(clock, "second"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 2
- assert lane.submits[1][0] == "- first\n- second"
- def test_tick_under_50ms_with_blocked_lane():
- service, _, clock, _, lane = _triage_service()
- assert not lane.predict_gate.is_set()
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- start = time.perf_counter()
- service.tick()
- elapsed = time.perf_counter() - start
- assert len(lane.submits) == 1
- assert elapsed < 0.05
- def test_on_triage_result_emits_signals():
- service, emitted, clock, _, lane = _triage_service()
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- payload = {
- "order": ["urgency"],
- "items": {"urgency": {"kind": "noul"}},
- "escalated": False,
- "input_tokens": 12,
- }
- service._on_triage_result(gen, payload)
- staged = _triage_events(emitted)
- assert len(staged) == 1
- assert staged[0]["stage"] == "signals"
- assert staged[0]["triage"] == service.last_triage
- assert staged[0]["triage"]["escalated"] is False
- assert staged[0]["triage"]["input_tokens"] == 12
- assert service.status()["triage"] == service.last_triage
- def test_triage_events_carry_the_server_timestamp():
- service, emitted, clock, _, lane = _triage_service(brief_fn=lambda prompt: "RAW")
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- expected_ts = clock()
- service._on_triage_result(gen, _escalated_payload())
- signals = [e for e in _triage_events(emitted) if e.get("stage") == "signals"]
- assert len(signals) == 1
- assert isinstance(signals[0]["ts"], (int, float))
- assert signals[0]["ts"] == expected_ts
- job = [j for j in service._inference._queue if j.name == "brief-update"][0]
- job.run()
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert len(contents) == 1
- assert isinstance(contents[0]["ts"], (int, float))
- assert contents[0]["ts"] == expected_ts # content reuses the signals timestamp
- def test_on_triage_result_suppresses_stale_generation():
- service, emitted, clock, _, lane = _triage_service()
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service.clear()
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is None
- assert [e for e in _triage_events(emitted) if e.get("stage") == "signals"] == []
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 2
- def test_pending_brief_capped_by_char_budget():
- """The buffer keeps the newest deltas within budget, dropping the oldest."""
- service, _, clock, _, _ = _triage_service(brief_fn=lambda prompt: "RAW")
- service.cfg.triage.brief_pending_char_budget = 30
- service.handle_event(_transcript(clock, "aaa"))
- service.handle_event(_transcript(clock, "bbbb"))
- assert service._pending_brief == ["Unknown: aaa", "Unknown: bbbb"]
- service.handle_event(_transcript(clock, "ccccc"))
- assert service._pending_brief == ["Unknown: bbbb", "Unknown: ccccc"]
- assert service.triage_history == ["aaa", "bbbb", "ccccc"]
- def test_pending_brief_untouched_without_triage():
- service, _, clock = _service()
- service.handle_event(_transcript(clock, "hello"))
- assert service._pending_brief == []
- assert service.triage_history == ["hello"]
- def test_on_triage_result_accepts_non_chat_mode():
- service, emitted, clock, _, lane = _triage_service()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- window = service._last_triage_window
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert service._last_triage_window == window
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals"]
- def test_on_triage_result_ignores_none_payload():
- service, emitted, clock, _, lane = _triage_service()
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, None)
- assert service.last_triage is None
- assert _triage_events(emitted) == []
- def test_none_payload_dispatches_nothing_and_keeps_deltas():
- """A Laya predict failure never dispatches: no job, no events, deltas stay."""
- calls: list[str] = []
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: calls.append(prompt) or "RAW"
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- assert service._pending_brief == ["Unknown: hello"]
- service._on_triage_result(gen, None)
- assert service.last_triage is None
- assert [e.get("stage") for e in _triage_events(emitted)] == []
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert calls == []
- assert service._pending_brief == ["Unknown: hello"]
- assert service._brief_outstanding is False
- def test_none_payload_without_brief_fn_enqueues_nothing():
- service, emitted, clock, _, lane = _triage_service()
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, None)
- assert service.last_triage is None
- assert _triage_events(emitted) == []
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert service._pending_brief == ["Unknown: hello"]
- def test_on_triage_result_ignores_empty_signals():
- service, emitted, clock, _, lane = _triage_service(
- triage=FakeTriage(normalized={})
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is None
- assert _triage_events(emitted) == []
- def test_substantive_turn_dispatches_one_brief_job():
- """A substantive settled turn calls the gate once and queues one brief job."""
- service, emitted, clock, triage, lane = _triage_service(
- triage=FakeTriage(escalate=True), brief_fn=lambda prompt: "RAW"
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "we might churn"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(
- gen, {"order": [], "items": {}, "escalated": False, "input_tokens": 0}
- )
- assert service.last_triage is not None
- assert service.last_triage["escalated"] is True
- assert service._last_brief_signals == service.last_triage
- assert service._inference.depth() == 2
- assert [j.name for j in service._inference._queue] == ["chat", "brief-update"]
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals"]
- assert len(triage.gate_calls) == 1
- assert service._brief_outstanding is True
- assert service._brief_token == 1
- def test_brief_dispatches_without_escalation():
- service, _, clock, _, lane = _triage_service(brief_fn=lambda prompt: "RAW")
- service.toggle_chat()
- service.handle_event(_transcript(clock, "all calm"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(
- gen, {"order": [], "items": {}, "escalated": False, "input_tokens": 0}
- )
- assert service.last_triage is not None
- assert service._inference.depth() == 2
- assert [j.name for j in service._inference._queue] == ["chat", "brief-update"]
- def test_gated_out_turn_skips_brief():
- """A gated-out turn stores signals but never calls brief_fn."""
- calls: list[str] = []
- service, emitted, clock, triage, lane = _triage_service(
- triage=FakeTriage(gate=False),
- brief_fn=lambda prompt: calls.append(prompt) or "RAW",
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "small talk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert "brief" not in service.last_triage # no prior brief to retain
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals"]
- assert calls == []
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert service._brief_outstanding is False
- assert len(triage.gate_calls) == 1
- def test_gated_out_turn_retains_previous_brief():
- """A gated-out settled turn keeps the last accepted brief in the panel."""
- service, emitted, clock, triage, lane = _triage_service(
- triage=FakeTriage(gate=False), brief_fn=lambda prompt: "RAW"
- )
- service.toggle_chat()
- signals = _escalated_payload()
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: first")
- service._run_brief_update(0, signals, service._now(), token)
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "RAW"
- emitted.clear()
- service.handle_event(_transcript(clock, "small talk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "RAW"
- assert "brief" not in service._last_brief_signals # raw signals stay raw
- staged = [e for e in _triage_events(emitted) if e.get("stage") == "signals"]
- assert len(staged) == 1
- assert staged[0]["triage"] == service.last_triage
- assert staged[0]["triage"]["brief"] == "RAW"
- assert len(triage.gate_calls) == 1
- def test_gated_out_turn_retains_brief_when_extract_disabled():
- """`extract_enabled = false` skips the dispatch but still retains the brief."""
- service, emitted, clock, _, lane = _triage_service(
- triage=FakeTriage(gate=False), brief_fn=lambda prompt: "RAW"
- )
- service.toggle_chat()
- signals = _escalated_payload()
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: first")
- service._run_brief_update(0, signals, service._now(), token)
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "RAW"
- service.cfg.triage.extract_enabled = False
- emitted.clear()
- service.handle_event(_transcript(clock, "small talk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "RAW"
- assert [j.name for j in service._inference._queue] == ["chat"]
- staged = [e for e in _triage_events(emitted) if e.get("stage") == "signals"]
- assert staged[0]["triage"]["brief"] == "RAW"
- def test_gated_out_turn_retains_brief_without_brief_fn():
- """No brief_fn means no dispatch, but a retained brief is still carried."""
- service, emitted, clock, _, lane = _triage_service(triage=FakeTriage(gate=False))
- service.toggle_chat()
- service.last_triage = {**_escalated_payload(), "brief": "KEEP"}
- emitted.clear()
- service.handle_event(_transcript(clock, "small talk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage["brief"] == "KEEP"
- staged = [e for e in _triage_events(emitted) if e.get("stage") == "signals"]
- assert staged[0]["triage"]["brief"] == "KEEP"
- def test_later_accepted_update_replaces_retained_brief():
- """A later accepted update overwrites the retained brief text."""
- texts = iter(["OLD", "NEW"])
- service, emitted, _, _, _ = _triage_service(brief_fn=lambda prompt: next(texts))
- service.toggle_chat()
- signals = _escalated_payload()
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: first")
- service._run_brief_update(0, signals, service._now(), token)
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "OLD"
- token = _seed_brief(service, "Unknown: second")
- service._run_brief_update(0, signals, service._now(), token)
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "NEW"
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert contents[-1]["triage"]["brief"] == "NEW"
- service.clear()
- assert service.last_triage is None
- def test_no_brief_when_disabled_in_config():
- calls: list[str] = []
- service, _, clock, _, lane = _triage_service(
- triage=FakeTriage(escalate=True),
- brief_fn=lambda prompt: calls.append(prompt) or "RAW",
- )
- service.cfg.triage.extract_enabled = False
- service.toggle_chat()
- service.handle_event(_transcript(clock, "churn risk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert service.last_triage["escalated"] is True
- assert calls == []
- assert [j.name for j in service._inference._queue] == ["chat"]
- def test_no_brief_without_brief_fn():
- service, _, clock, _, lane = _triage_service(triage=FakeTriage(escalate=True))
- service.toggle_chat()
- service.handle_event(_transcript(clock, "churn risk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert service._brief_outstanding is False
- def test_brief_refused_when_queue_full():
- """A refused submit leaves deltas buffered and the outstanding flag clear."""
- service, emitted, clock, _, lane = _triage_service(
- triage=FakeTriage(escalate=True), brief_fn=lambda prompt: "RAW"
- )
- service._inference = InferenceQueue(size=2)
- service.toggle_chat()
- service.handle_event(_transcript(clock, "churn risk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert service._inference.depth() == 1
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service._inference.depth() == 1
- assert service._inference._queue[0].name == "chat"
- assert service.last_triage is not None
- assert service.last_triage["escalated"] is True
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals"]
- assert service._pending_brief == ["Unknown: churn risk"]
- assert service._brief_outstanding is False
- def test_second_result_while_outstanding_is_refused():
- """At most one brief job is outstanding: the second turn only stores signals."""
- service, emitted, clock, _, lane = _triage_service(brief_fn=lambda prompt: "RAW")
- service.toggle_chat()
- service.handle_event(_transcript(clock, "first"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen0 = lane.submits[0]
- service._on_triage_result(gen0, _escalated_payload())
- assert [j.name for j in service._inference._queue] == ["chat", "brief-update"]
- service.handle_event(_transcript(clock, "second"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen1 = lane.submits[1]
- service._on_triage_result(gen1, _escalated_payload())
- jobs = [j for j in service._inference._queue if j.name == "brief-update"]
- assert len(jobs) == 1
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals", "signals"]
- assert service._brief_outstanding is True
- assert service._pending_brief == ["Unknown: first", "Unknown: second"]
- def _seed_brief(service, *deltas):
- service._pending_brief = list(deltas)
- service._brief_outstanding = True
- return service._brief_token
- def test_retention_guard_reappends_fact_dropped_by_model():
- """A prior bullet the model omitted is deterministically restored."""
- prior = (
- "Summary: rolling\n\nPrior Details:\n"
- "- The budget was $305 million.\n"
- "- Sense could find prey.\n"
- )
- new = "Summary: newer\n\nPrior Details:\n- Sense could find prey.\n"
- service, emitted, _, _, _ = _triage_service(brief_fn=lambda prompt: new)
- service.last_triage = {**_escalated_payload(), "brief": prior}
- service._pending_brief = ["Unknown: more"]
- service._brief_outstanding = True
- service._run_brief_update(0, _escalated_payload(), service._now(), service._brief_token)
- assert service.last_triage is not None
- brief = service.last_triage["brief"]
- assert "- Sense could find prey." in brief
- assert "- The budget was $305 million." in brief
- assert brief.count("- Sense could find prey.") == 1
- def test_retention_guard_leaves_complete_brief_unchanged():
- """A brief that already carries every prior bullet is stored verbatim."""
- prior = "Prior Details:\n- fact one\n- fact two\n"
- new = "Summary: newer\n\nPrior Details:\n- fact one\n- fact two\n"
- service, _, _, _, _ = _triage_service(brief_fn=lambda prompt: new)
- service.last_triage = {**_escalated_payload(), "brief": prior}
- service._pending_brief = ["Unknown: more"]
- service._brief_outstanding = True
- service._run_brief_update(0, _escalated_payload(), service._now(), service._brief_token)
- assert service.last_triage is not None
- assert service.last_triage["brief"] == new
- def test_pinned_prompt_carries_prior_brief_and_new_utterances():
- """The brief request hands the model the pinned current brief + new deltas."""
- seen: list[str] = []
- service, _, _, _, _ = _triage_service(
- brief_fn=lambda prompt: seen.append(prompt) or "UPDATED"
- )
- service.last_triage = {**_escalated_payload(), "brief": "PRIOR BRIEF TEXT"}
- service._pending_brief = ["Unknown: hello there"]
- service._brief_outstanding = True
- service._run_brief_update(0, _escalated_payload(), service._now(), service._brief_token)
- assert len(seen) == 1
- assert seen[0].startswith("CURRENT BRIEF:\nPRIOR BRIEF TEXT")
- assert "NEW UTTERANCES:\n- Unknown: hello there" in seen[0]
- def test_brief_merges_content_and_writes_file():
- brief_file = FakeBriefFile()
- service, emitted, _, triage, _ = _triage_service(
- brief_fn=lambda prompt: "the brief", brief_file=brief_file
- )
- service.toggle_chat()
- signals = _escalated_payload()
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: hello")
- ts = service._now()
- service._run_brief_update(0, signals, ts, token)
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert len(contents) == 1
- assert contents[0]["triage"] == {**signals, "brief": "the brief"}
- assert service.last_triage == contents[0]["triage"]
- assert triage.prompt_calls == [(("Unknown: hello",), signals)]
- assert brief_file.writes == ["the brief"]
- assert service._pending_brief == []
- assert service._brief_outstanding is False
- def test_empty_brief_requeues_deltas():
- """A `""` reply keeps signals, re-buffers the deltas, and clears the flag."""
- service, emitted, _, _, _ = _triage_service(brief_fn=lambda prompt: "")
- service.toggle_chat()
- signals = _escalated_payload()
- service.last_triage = signals
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: hello")
- ts = service._now()
- service._run_brief_update(0, signals, ts, token)
- assert [e for e in _triage_events(emitted) if e.get("stage") == "content"] == []
- assert service.last_triage == signals
- assert "brief" not in service.last_triage
- assert service._pending_brief == ["Unknown: hello"]
- assert service._brief_outstanding is False
- def test_brief_error_requeues_and_clears_flag():
- """A raising brief_fn leaves the flag clear and the deltas buffered."""
- def boom(prompt):
- raise RuntimeError("llm down")
- service, emitted, _, _, _ = _triage_service(brief_fn=boom)
- service.toggle_chat()
- token = _seed_brief(service, "Unknown: hello")
- ts = service._now()
- service._run_brief_update(0, _escalated_payload(), ts, token)
- assert _triage_events(emitted) == []
- assert service.last_triage is None
- assert service._pending_brief == ["Unknown: hello"]
- assert service._brief_outstanding is False
- def test_brief_empty_buffer_returns_early():
- """An empty buffer returns early: no call, no event, flag cleared."""
- calls: list[str] = []
- service, emitted, _, _, _ = _triage_service(
- brief_fn=lambda prompt: calls.append(prompt) or "RAW"
- )
- service.toggle_chat()
- service._brief_outstanding = True
- token = service._brief_token
- ts = service._now()
- assert service._pending_brief == []
- service._run_brief_update(0, _escalated_payload(), ts, token)
- assert calls == []
- assert _triage_events(emitted) == []
- assert service._brief_outstanding is False
- def test_brief_suppressed_after_reset():
- service, emitted, _, _, _ = _triage_service(brief_fn=lambda prompt: "RAW")
- service.toggle_chat()
- service._pending_brief = ["Unknown: hello"]
- service._reset_triage()
- emitted.clear()
- service._run_brief_update(0, _escalated_payload(), service._now(), 0)
- assert _triage_events(emitted) == []
- assert service.last_triage is None
- assert service._pending_brief == ["Unknown: hello"]
- def test_brief_works_after_leaving_chat():
- seen = []
- service, emitted, _, _, _ = _triage_service(
- brief_fn=lambda prompt: seen.append(prompt) or "RAW"
- )
- service.toggle_chat()
- assert service.modes.exit_chat() is True
- token = _seed_brief(service, "Unknown: hello")
- signals = _escalated_payload()
- service._last_brief_signals = signals
- service._run_brief_update(0, signals, service._now(), token)
- assert seen != []
- assert [e.get("stage") for e in _triage_events(emitted)] == ["content"]
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "RAW"
- def test_deltas_arriving_during_call_survive_and_flush():
- """Deltas appended mid-call are not consumed; the finally flushes them."""
- service, emitted, clock, _, _ = _triage_service()
- service.toggle_chat()
- signals = _escalated_payload()
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: first")
- appended: list[int] = []
- def _brief(prompt):
- if not appended:
- appended.append(1)
- service.handle_event(_transcript(clock, "during call"))
- return "first brief"
- service._brief_fn = _brief
- service._run_brief_update(0, signals, clock(), token)
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert len(contents) == 1
- assert contents[0]["triage"]["brief"] == "first brief"
- flushed = [j for j in service._inference._queue if j.name == "brief-update"]
- assert len(flushed) == 1
- assert service._brief_outstanding is True
- assert service._pending_brief == ["Unknown: during call"]
- flushed[0].run()
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert len(contents) == 2
- assert contents[1]["triage"]["brief"] == "first brief"
- assert service._pending_brief == []
- assert service._brief_outstanding is False
- def test_in_flight_brief_after_clear_emits_and_writes_nothing():
- """Stale in-flight job: no event, no file write, no requeue, flag untouched."""
- brief_file = FakeBriefFile()
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: "RAW", brief_file=brief_file
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- job = [j for j in service._inference._queue if j.name == "brief-update"][0]
- assert service._brief_outstanding is True
- service.clear()
- assert service._brief_outstanding is False
- token_after_clear = service._brief_token
- emitted.clear()
- job.run()
- assert [e for e in _triage_events(emitted)] == []
- assert brief_file.writes == []
- assert service._pending_brief == []
- assert service.last_triage is None
- assert service._brief_token == token_after_clear
- assert service._brief_outstanding is False
- def test_clear_resets_triage():
- resets: list[int] = []
- chats: list[int] = []
- brief_file = FakeBriefFile()
- service, emitted, clock, _, lane = _triage_service(
- triage=FakeTriage(escalate=True),
- brief_fn=lambda prompt: "RAW",
- brief_file=brief_file,
- brief_reset=lambda: resets.append(1),
- chat_reset=lambda: chats.append(1),
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "churn"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service._inference.depth() == 2
- assert service._pending_brief == ["Unknown: churn"]
- assert service._brief_outstanding is True
- token_before = service._brief_token
- assert service.clear() is True
- assert service._triage_generation == 1
- assert service.last_triage is None
- assert service._last_triage_window == ()
- assert service.triage_history == []
- assert service._pending_brief == []
- assert service._brief_token == token_before + 1
- assert service._brief_outstanding is False
- assert service._last_brief_signals is None
- assert resets == [1]
- assert brief_file.clears == 1
- assert chats == [1]
- assert lane.drops == 1
- assert service._inference.depth() == 0
- assert list(service._inference._queue) == []
- cleared = [e for e in _triage_events(emitted) if e.get("stage") == "cleared"]
- assert len(cleared) == 1
- assert cleared[0]["triage"] is None
- assert {"type": "cleared"} in emitted
- signals_before = len(
- [e for e in _triage_events(emitted) if e.get("stage") == "signals"]
- )
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is None
- assert (
- len([e for e in _triage_events(emitted) if e.get("stage") == "signals"])
- == signals_before
- )
- submits_before = len(lane.submits)
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == submits_before
- def test_exit_chat_does_not_reset_triage():
- chats: list[int] = []
- bresets: list[int] = []
- brief_file = FakeBriefFile()
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: "RAW",
- chat_reset=lambda: chats.append(1),
- brief_reset=lambda: bresets.append(1),
- brief_file=brief_file,
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert service._triage_generation == 0
- assert service.toggle_chat() is True
- assert service.modes.mode.value == "idle"
- assert service._triage_generation == 0
- assert service.last_triage is not None
- assert service._last_triage_window == ("hello",)
- assert service.triage_history == ["hello"]
- assert service._pending_brief == ["Unknown: hello"]
- assert service._last_brief_signals is not None
- assert service._brief_outstanding is True
- assert lane.drops == 0
- cleared = [e for e in _triage_events(emitted) if e.get("stage") == "cleared"]
- assert cleared == []
- assert chats == [1]
- assert bresets == []
- assert brief_file.clears == 0
- def test_toggle_chat_enter_does_not_touch_triage():
- """Entering chat resets only the chat buffers; the brief engine is independent."""
- bresets: list[int] = []
- brief_file = FakeBriefFile()
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: "RAW",
- brief_reset=lambda: bresets.append(1),
- brief_file=brief_file,
- )
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- triage_before = service.last_triage
- pending_before = list(service._pending_brief)
- assert service.toggle_chat() is True
- assert service.modes.mode.value == "chat"
- assert service._triage_generation == 0
- assert service.last_triage is triage_before
- assert service._last_triage_window == ("hello",)
- assert service.triage_history == ["hello"]
- assert service._pending_brief == pending_before
- assert service._last_brief_signals is not None
- assert service._brief_outstanding is True
- assert lane.drops == 0
- assert bresets == []
- assert brief_file.clears == 0
- cleared = [e for e in _triage_events(emitted) if e.get("stage") == "cleared"]
- assert cleared == []
- def test_clear_without_triage_emits_no_triage_event():
- service, emitted, _ = _service()
- brief_file = FakeBriefFile()
- bresets: list[int] = []
- service._brief_file = brief_file
- service._brief_reset = lambda: bresets.append(1)
- service.clear()
- assert {"type": "cleared"} in emitted
- assert _triage_events(emitted) == []
- assert brief_file.clears == 1
- assert bresets == []
- def test_stale_brief_after_clear_does_not_clear_newer_dispatch_flag():
- """A stale in-flight job must not clear a newer dispatch's outstanding flag."""
- brief_file = FakeBriefFile()
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: "RAW", brief_file=brief_file
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- stale_job = [j for j in service._inference._queue if j.name == "brief-update"][0]
- assert service._brief_outstanding is True
- service.clear()
- # A fresh turn dispatches a new brief AFTER clear bumped the token.
- service.handle_event(_transcript(clock, "fresh"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen2 = lane.submits[1]
- service._on_triage_result(gen2, _escalated_payload())
- assert service._brief_outstanding is True
- newer_token = service._brief_token
- pending_before = list(service._pending_brief)
- triage_before = service.last_triage
- emitted.clear()
- stale_job.run()
- assert _triage_events(emitted) == []
- assert brief_file.writes == []
- assert service._pending_brief == pending_before
- assert service._brief_outstanding is True
- assert service._brief_token == newer_token
- assert service.last_triage is triage_before
- def test_clear_resets_chat_latch():
- service, emitted, _ = _service()
- service.toggle_chat()
- service._chat_queue_full = True
- service._pending_chat = ["stuck text"]
- service.clear()
- assert service._chat_queue_full is False
- assert service._pending_chat == []
- assert {"type": "cleared"} in emitted
- def test_triage_end_to_end_signals_content_cleared():
- """Two settled turns collapse into ONE brief job (single-outstanding guard).
- The second `_on_triage_result` is refused as `"outstanding"`, so the queue
- holds one `brief-update` that consumes both deltas and emits one `content`.
- The panel keeps the FIRST turn's signals until the next settled turn (lag).
- """
- prompts: list[str] = []
- service, emitted, clock, triage, lane = _triage_service(
- brief_fn=lambda prompt: prompts.append(prompt) or "the brief"
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "first thought"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- _, _, _, gen0 = lane.submits[0]
- service._on_triage_result(
- gen0, {"order": ["urgency"], "items": {}, "escalated": False, "input_tokens": 1}
- )
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals"]
- assert prompts == []
- assert service.last_triage is not None
- assert service.last_triage["escalated"] is False
- assert [j.name for j in service._inference._queue] == ["chat", "brief-update"]
- triage._escalate = True
- service.handle_event(_transcript(clock, "we might churn"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 2
- _, _, _, gen1 = lane.submits[1]
- service._on_triage_result(gen1, _escalated_payload())
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals", "signals"]
- assert [j.name for j in service._inference._queue] == [
- "chat",
- "brief-update",
- "chat",
- ]
- for job in [j for j in service._inference._queue if j.name == "brief-update"]:
- job.run()
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert len(contents) == 1
- assert len(prompts) == 1
- deltas, _ = triage.prompt_calls[0]
- assert deltas == ("Unknown: first thought", "Unknown: we might churn")
- assert contents[0]["triage"]["escalated"] is False
- assert contents[0]["triage"]["brief"] == "the brief"
- assert service.last_triage == contents[0]["triage"]
- assert service.status()["triage"] == contents[0]["triage"]
- service.clear()
- cleared = [e for e in _triage_events(emitted) if e.get("stage") == "cleared"]
- assert len(cleared) == 1
- assert cleared[0]["triage"] is None
- assert service.last_triage is None
- service._on_triage_result(gen1, _escalated_payload())
- assert [e.get("stage") for e in _triage_events(emitted)] == [
- "signals",
- "signals",
- "content",
- "cleared",
- ]
- assert service.toggle_chat() is True
- service.handle_event(_transcript(clock, "idle thought"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 3
- assert lane.submits[2][0].endswith("- idle thought")
- def test_tick_emits_status_on_triage_state_change():
- service, emitted, _, triage, _ = _triage_service(
- triage=FakeTriage(state_pair=("loading", "warming up"))
- )
- service.tick()
- statuses = [e for e in emitted if e["type"] == "status"]
- assert len(statuses) == 1
- assert statuses[0]["triage_state"] == "loading"
- assert statuses[0]["triage_reason"] == "warming up"
- assert "triage" in statuses[0]
- triage._state_pair = ("ok", "")
- service.tick()
- statuses = [e for e in emitted if e["type"] == "status"]
- assert len(statuses) == 2
- assert statuses[1]["triage_state"] == "ok"
- assert statuses[1]["triage_reason"] == ""
- assert "triage" in statuses[1]
- service.tick()
- assert len([e for e in emitted if e["type"] == "status"]) == 2
- def test_no_status_emitted_when_triage_disabled():
- service, emitted, _ = _service()
- service.tick()
- service.tick()
- assert [e for e in emitted if e["type"] == "status"] == []
- def test_tick_heals_triage_panel_after_lane_ready():
- import time as _time
- from pathlib import Path
- from stt_daemon.assistant.laya_lane import LayaLane
- from stt_daemon.assistant.triage import TriageAnalyzer
- from stt_daemon.tui.state import ViewState
- class _FakeRouter:
- def preload(self, names):
- return self
- def predict(self, state, questions, model=None, lang=None, **kwargs):
- if "_warmup" in questions:
- return {"model": model, "answers": {}, "usage": {}, "routing": {}}
- return {
- "model": model,
- "answers": {
- "frustration": {
- "noul": 0.1,
- "confidence": 0.9,
- "answer_confidence": 0.9,
- },
- "urgency": {
- "choice": "low",
- "confidence": 0.9,
- "answer_confidence": 0.9,
- },
- },
- "usage": {"input_tokens": 5},
- }
- def _factory(**kwargs):
- return _FakeRouter()
- cfg = Config()
- cfg.assistant.turn_gap_s = 0.5
- cfg.triage.questions_file = str(
- Path(__file__).resolve().parent.parent / "questions.example.json"
- )
- clock = Clock()
- emitted: list[dict] = []
- def emit(type_, **fields):
- emitted.append({"type": type_, **fields})
- analyzer = TriageAnalyzer(cfg.triage)
- assert analyzer.questions, "example questions must load"
- holder: dict = {}
- lane = LayaLane(
- cfg.triage,
- agent_factory=_factory,
- on_result=lambda gen, payload: holder["svc"]._on_triage_result(gen, payload),
- on_state=lambda state, reason: analyzer.refresh(state, reason),
- )
- service = AssistantService(
- cfg,
- FakeSubscriber(),
- emit=emit,
- time_fn=clock,
- triage=analyzer,
- laya_lane=lane,
- chat_fn=_ok_reply,
- )
- holder["svc"] = service
- try:
- lane.start()
- deadline = _time.monotonic() + 5.0
- while not lane.ready() and _time.monotonic() < deadline:
- _time.sleep(0.005)
- assert lane.ready(), "lane did not become ready"
- assert analyzer.state() == ("ok", "")
- service.tick()
- statuses = [e for e in emitted if e["type"] == "status"]
- assert statuses, "tick emitted no status after lane ready"
- assert statuses[-1]["triage_state"] == "ok"
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- deadline = _time.monotonic() + 5.0
- while service.last_triage is None and _time.monotonic() < deadline:
- _time.sleep(0.005)
- assert service.last_triage is not None, "no triage signals from lane"
- view = ViewState()
- for event in emitted:
- view.ingest(event)
- assert view.triage_state == "ok"
- lines = view.triage_lines()
- assert lines
- assert not any("unavailable" in line for line in lines)
- assert not any("loading" in line for line in lines)
- finally:
- lane.stop()
- def _dump_brief(prompt, on_io=None):
- """Brief fake that mimics LlmClient.chat's on_io phases."""
- if on_io is not None:
- on_io({"phase": "request", "url": "http://llm/v1", "body": {"prompt": prompt}})
- on_io({"phase": "response", "payload": {"content": "parsed"}, "error": None})
- return "BRIEF"
- def _run_triage_flow(service, emitted, clock, lane, payload):
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, payload)
- return gen
- def _dump_files(tmp_path):
- return list(tmp_path.rglob("*.json"))
- def test_dump_writes_full_laya_and_llm_io(tmp_path):
- service, emitted, clock, triage, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- _run_triage_flow(service, emitted, clock, lane, _escalated_payload())
- jobs = [j for j in service._inference._queue if j.name == "brief-update"]
- assert len(jobs) == 1
- jobs[0].run()
- files = _dump_files(tmp_path)
- assert len(files) == 1
- assert re.fullmatch(r"\d{6}-\d{3}\.json", files[0].name)
- data = json.loads(files[0].read_text(encoding="utf-8"))
- assert data["ts"] == clock()
- assert isinstance(data["iso"], str) and data["iso"]
- assert data["laya"]["request"]["state"] == "- hello"
- assert data["laya"]["request"]["questions"] == triage.questions
- assert data["laya"]["request"]["model"] == "multilingual"
- assert data["laya"]["request"]["max_len"] == 1024
- assert data["laya"]["response"] == _escalated_payload()
- assert data["laya"]["error"] is None
- assert data["llm"]["request"]["url"] == "http://llm/v1"
- assert data["llm"]["request"]["body"] == {
- "prompt": "CURRENT BRIEF:\n(none yet)\n\nNEW UTTERANCES:\n- Unknown: hello"
- }
- assert data["llm"]["response"] == {"content": "parsed"}
- assert data["llm"]["error"] is None
- assert service._triage_dump._pending == {}
- def test_dump_laya_error_writes_laya_only_and_keeps_deltas(tmp_path):
- """A Laya predict failure finalizes the Laya half with `llm: null`.
- Nothing is dispatched, so no LLM half is ever recorded and the buffered
- deltas survive for the next settled turn.
- """
- service, emitted, clock, _, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- _run_triage_flow(service, emitted, clock, lane, None)
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert service._pending_brief == ["Unknown: hello"]
- files = _dump_files(tmp_path)
- assert len(files) == 1
- data = json.loads(files[0].read_text(encoding="utf-8"))
- assert data["laya"]["response"] is None
- assert data["laya"]["error"] == "predict failed"
- assert data["llm"] is None
- assert service._triage_dump._pending == {}
- def test_dump_without_brief_has_laya_only(tmp_path):
- service, emitted, clock, _, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- service.cfg.triage.extract_enabled = False
- _run_triage_flow(service, emitted, clock, lane, _escalated_payload())
- files = _dump_files(tmp_path)
- assert len(files) == 1
- data = json.loads(files[0].read_text(encoding="utf-8"))
- assert data["laya"]["response"] == _escalated_payload()
- assert data["llm"] is None
- assert service._triage_dump._pending == {}
- def test_dump_queue_full_has_laya_only(tmp_path):
- service, emitted, clock, _, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- service._inference = InferenceQueue(size=2)
- _run_triage_flow(service, emitted, clock, lane, _escalated_payload())
- assert service._inference.depth() == 1
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert service._pending_brief == ["Unknown: hello"]
- files = _dump_files(tmp_path)
- assert len(files) == 1
- data = json.loads(files[0].read_text(encoding="utf-8"))
- assert data["laya"]["response"] == _escalated_payload()
- assert data["llm"] is None
- assert service._triage_dump._pending == {}
- def test_dump_disabled_writes_nothing(tmp_path):
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: "RAW"
- )
- assert service._triage_dump.enabled is False
- service._triage_dump = TriageDumper(False, directory=str(tmp_path))
- _run_triage_flow(service, emitted, clock, lane, _escalated_payload())
- jobs = [j for j in service._inference._queue if j.name == "brief-update"]
- assert len(jobs) == 1
- jobs[0].run()
- assert _dump_files(tmp_path) == []
- def test_dump_stale_result_discards_without_file(tmp_path):
- service, emitted, clock, triage, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- real_normalize = triage.normalize
- def stale_normalize(payload):
- # Simulate a clear landing between the first check/begin and the re-check.
- service._triage_generation += 1
- return real_normalize(payload)
- triage.normalize = stale_normalize
- service._on_triage_result(gen, _escalated_payload())
- assert _dump_files(tmp_path) == []
- assert service._triage_dump._pending == {}
- def test_dump_stale_brief_discards_without_file(tmp_path):
- service, _, clock, _, _ = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- ts = clock()
- service._triage_dump.begin(ts, laya={"request": {}, "response": {}, "error": None})
- service._pending_brief = ["Unknown: hello"]
- service._triage_generation += 1 # stale before the brief even starts
- service._run_brief_update(0, _escalated_payload(), ts, service._brief_token)
- assert _dump_files(tmp_path) == []
- assert service._triage_dump._pending == {}
- assert service._pending_brief == ["Unknown: hello"]
- def test_dump_brief_cleared_mid_flight_writes_nothing(tmp_path):
- service, _, clock, _, _ = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- def stale_brief(prompt, on_io=None):
- # Simulate a clear landing mid-LLM-call (after the request phase fired).
- service._triage_generation += 1
- if on_io is not None:
- on_io({"phase": "request", "url": "http://llm/v1", "body": {}})
- on_io({"phase": "response", "payload": {"c": 1}, "error": None})
- return "RAW"
- service._brief_fn = stale_brief
- service._pending_brief = ["Unknown: hello"]
- service._brief_outstanding = True
- token = service._brief_token
- ts = clock()
- service._triage_dump.begin(ts, laya={"request": {}, "response": {}, "error": None})
- service._run_brief_update(0, _escalated_payload(), ts, token)
- assert _dump_files(tmp_path) == []
- assert service._triage_dump._pending == {}
- assert service._pending_brief == []
- assert service._brief_outstanding is False
- def test_dump_gated_out_writes_laya_only_and_no_pending(tmp_path):
- """A turn the gate leaves out writes `llm: null` and leaks no pending record."""
- service, emitted, clock, triage, lane = _triage_service(
- triage=FakeTriage(escalate=True, gate=False), brief_fn=_dump_brief
- )
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- _run_triage_flow(service, emitted, clock, lane, _escalated_payload())
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert len(triage.gate_calls) == 1
- files = _dump_files(tmp_path)
- assert len(files) == 1
- data = json.loads(files[0].read_text(encoding="utf-8"))
- assert data["laya"]["response"] == _escalated_payload()
- assert data["llm"] is None
- assert service._triage_dump._pending == {}
- def test_dump_coalesced_update_writes_laya_only_then_llm_only_flush(tmp_path):
- """A coalesced turn writes a Laya-only record; the later flush a separate llm-only one."""
- service, emitted, clock, _, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- service.toggle_chat()
- service.handle_event(_transcript(clock, "first"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen0 = lane.submits[0]
- service._on_triage_result(gen0, _escalated_payload())
- service.handle_event(_transcript(clock, "second"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen1 = lane.submits[1]
- service._on_triage_result(gen1, _escalated_payload())
- assert service._brief_outstanding is True
- appended: list[int] = []
- def _brief(prompt, on_io=None):
- if not appended:
- appended.append(1)
- service.handle_event(_transcript(clock, "during call"))
- if on_io is not None:
- on_io({"phase": "request", "url": "http://llm/v1", "body": {"prompt": prompt}})
- on_io({"phase": "response", "payload": {"content": "parsed"}, "error": None})
- return "BRIEF"
- service._brief_fn = _brief
- clock.advance(1.0)
- job = [j for j in service._inference._queue if j.name == "brief-update"][0]
- service._inference._queue.remove(job)
- job.run()
- flushed = [j for j in service._inference._queue if j.name == "brief-update"]
- assert len(flushed) == 1
- flushed[0].run()
- assert service._triage_dump._pending == {}
- records = [
- json.loads(path.read_text(encoding="utf-8")) for path in _dump_files(tmp_path)
- ]
- assert len(records) == 3
- coalesced = [r for r in records if r["laya"] is not None and r["llm"] is None]
- flushed_records = [r for r in records if r["laya"] is None and r["llm"] is not None]
- both = [r for r in records if r["laya"] is not None and r["llm"] is not None]
- assert len(coalesced) == 1
- assert len(flushed_records) == 1
- assert len(both) == 1
- assert coalesced[0]["ts"] != flushed_records[0]["ts"]
- assert flushed_records[0]["llm"]["request"]["body"] == {
- "prompt": "CURRENT BRIEF:\nBRIEF\n\nNEW UTTERANCES:\n- Unknown: during call"
- }
- def test_chat_client_failure_emits_llm_error_and_service_survives():
- """Adversarial (a): a chat client raising on `chat` surfaces as an `llm_error`
- event and the service keeps functioning afterwards."""
- from stt_daemon.assistant.llm import ConversationEngine, LlmError
- class BoomClient:
- def chat(self, messages, max_tokens=None, timeout_s=None, **kwargs):
- raise LlmError("boom")
- engine = ConversationEngine(Config().llm, BoomClient(), system_prompt="SYS")
- service, emitted, clock = _service()
- service.chat_fn = engine.send
- assert service.toggle_chat() is True
- service._run_chat("hello")
- errors = [e for e in emitted if e["type"] == "error" and e.get("code") == "llm_error"]
- assert errors and "boom" in errors[-1]["message"]
- assert service.toggle_chat() is True
- assert service.modes.mode.value == "idle"
- # ---------------------------------------------------------------------------
- # Conversation session adoption (SESSION/TRANSCRIPT session_dir).
- # ---------------------------------------------------------------------------
- def _session_event(clock, session_dir):
- event = _ev(events.SESSION, clock)
- event["session_id"] = session_dir.rsplit("/", 1)[-1]
- event["session_dir"] = session_dir
- return event
- def _ring(service, clock, *items):
- """Feed a daemon-seeded replay window: status -> items -> ring_end."""
- service.handle_event(_ev(events.STATUS, clock))
- for item in items:
- service.handle_event(item)
- service.handle_event(_ev(events.RING_END, clock))
- def _cleared(emitted):
- return [e for e in _triage_events(emitted) if e.get("stage") == "cleared"]
- def test_status_includes_session_dir_without_brief_file():
- service, _, clock = _service()
- assert service.status()["session_dir"] is None
- service.handle_event(_session_event(clock, "transcripts/2026-09-30-12-47-29"))
- assert service.status()["session_dir"] == "transcripts/2026-09-30-12-47-29"
- def test_live_session_event_sets_session_and_records_directory():
- brief_file = FakeBriefFile()
- service, _, clock, _, _ = _triage_service(brief_file=brief_file)
- session = "transcripts/2026-09-30-12-47-29"
- service.handle_event(_session_event(clock, session))
- assert service.status()["session_dir"] == session
- assert brief_file.directories == [session]
- assert service._triage_generation == 0
- def test_live_transcript_session_sets_session_without_wiping_utterance():
- brief_file = FakeBriefFile()
- service, _, clock, _, _ = _triage_service(brief_file=brief_file)
- session = "transcripts/s1"
- service.handle_event(_transcript(clock, "hello", session_dir=session))
- assert service.status()["session_dir"] == session
- assert brief_file.directories == [session]
- assert service.triage_history == ["hello"]
- assert service._pending_brief == ["Unknown: hello"]
- assert service._triage_generation == 0
- def test_brief_write_lands_under_session_dir(tmp_path):
- from stt_daemon.assistant.brief_file import BriefFile
- brief_file = BriefFile(True)
- service, _, clock, _, _ = _triage_service(brief_file=brief_file)
- session = str(tmp_path / "transcripts" / "s1")
- service.handle_event(_transcript(clock, "hello", session_dir=session))
- brief_file.write("the brief")
- path = tmp_path / "transcripts" / "s1" / "brief.md"
- assert path.read_text(encoding="utf-8") == "the brief"
- def test_live_session_change_resets_triage_and_keeps_chat():
- brief_file = FakeBriefFile()
- resets: list[int] = []
- service, emitted, clock, _, lane = _triage_service(
- brief_file=brief_file, brief_reset=lambda: resets.append(1)
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- service.handle_event(_session_event(clock, "transcripts/a"))
- service.last_triage = {"signals": 1}
- service._pending_brief = ["stale"]
- emitted.clear()
- service.handle_event(_session_event(clock, "transcripts/b"))
- assert service.status()["session_dir"] == "transcripts/b"
- assert service.last_triage is None
- assert service._pending_brief == []
- assert service._triage_generation == 1
- assert service.chat_history == ["Unknown: hello"]
- assert resets == [1]
- assert lane.drops == 1
- assert brief_file.directories == ["transcripts/a", "transcripts/b"]
- assert len(_cleared(emitted)) == 1
- def test_sessionless_transcript_does_not_reset():
- service, _, clock, _, _ = _triage_service()
- service.handle_event(_session_event(clock, "transcripts/a"))
- service.last_triage = {"signals": 1}
- service.handle_event(_transcript(clock, "hello"))
- assert service._triage_generation == 0
- assert service.last_triage == {"signals": 1}
- assert service.status()["session_dir"] == "transcripts/a"
- def test_same_dir_readoption_does_not_reset():
- brief_file = FakeBriefFile()
- service, _, clock, _, _ = _triage_service(brief_file=brief_file)
- session = "transcripts/a"
- service.handle_event(_session_event(clock, session))
- service.last_triage = {"signals": 1}
- service.handle_event(_transcript(clock, "again", session_dir=session))
- assert service._triage_generation == 0
- assert service.last_triage == {"signals": 1}
- assert brief_file.directories == [session]
- def test_session_change_without_brief_file_does_not_raise():
- service, _, clock, _, _ = _triage_service(brief_file=None)
- service.handle_event(_session_event(clock, "transcripts/a"))
- service.handle_event(_session_event(clock, "transcripts/b"))
- assert service.status()["session_dir"] == "transcripts/b"
- assert service._triage_generation == 1
- def test_replay_single_session_adopts_without_reset():
- brief_file = FakeBriefFile()
- service, emitted, clock, _, _ = _triage_service(brief_file=brief_file)
- session = "transcripts/a"
- _ring(service, clock, _transcript(clock, "hello", session_dir=session))
- assert service.status()["session_dir"] == session
- assert brief_file.directories == [session]
- assert service._triage_generation == 0
- assert _cleared(emitted) == []
- assert service.triage_history == ["hello"]
- def test_replay_two_sessions_commits_newest_and_resets_once():
- brief_file = FakeBriefFile()
- service, emitted, clock, _, _ = _triage_service(brief_file=brief_file)
- a, b = "transcripts/a", "transcripts/b"
- _ring(
- service,
- clock,
- _session_event(clock, a),
- _transcript(clock, "old turn", session_dir=a),
- _session_event(clock, b),
- _transcript(clock, "new turn", session_dir=b),
- )
- assert service.status()["session_dir"] == b
- assert brief_file.directories == [b]
- assert service._triage_generation == 1
- assert service.last_triage is None
- assert service._pending_brief == []
- assert len(_cleared(emitted)) == 1
- def test_reconnect_tailing_current_session_is_noop():
- brief_file = FakeBriefFile()
- service, emitted, clock, _, _ = _triage_service(brief_file=brief_file)
- session = "transcripts/a"
- service.handle_event(_session_event(clock, session))
- service.last_triage = {"signals": 1}
- emitted.clear()
- _ring(service, clock, _transcript(clock, "replayed", session_dir=session))
- assert service.status()["session_dir"] == session
- assert brief_file.directories == [session]
- assert service._triage_generation == 0
- assert service.last_triage == {"signals": 1}
- assert _cleared(emitted) == []
- def test_reconnect_older_to_current_resets_once_without_rewind():
- brief_file = FakeBriefFile()
- service, emitted, clock, _, _ = _triage_service(brief_file=brief_file)
- a, b = "transcripts/a", "transcripts/b"
- service.handle_event(_session_event(clock, a))
- service.last_triage = {"signals": 1}
- emitted.clear()
- _ring(
- service,
- clock,
- _session_event(clock, a),
- _transcript(clock, "old", session_dir=a),
- _session_event(clock, b),
- _transcript(clock, "new", session_dir=b),
- )
- assert service.status()["session_dir"] == b
- assert brief_file.directories == [a, b]
- assert service._triage_generation == 1
- assert service.last_triage is None
- assert len(_cleared(emitted)) == 1
- def test_reconnect_only_newer_session_resets_once():
- brief_file = FakeBriefFile()
- service, _, clock, _, _ = _triage_service(brief_file=brief_file)
- a, b = "transcripts/a", "transcripts/b"
- service.handle_event(_session_event(clock, a))
- service.last_triage = {"signals": 1}
- _ring(service, clock, _transcript(clock, "new", session_dir=b))
- assert service.status()["session_dir"] == b
- assert service._triage_generation == 1
- assert service.last_triage is None
- assert brief_file.directories == [a, b]
- def test_reconnect_span_older_current_clears_buffered_older_turns():
- """A RING_END whose ring spanned an older session resets even when the
- newest session already matches the adopted directory (no re-set_directory)."""
- brief_file = FakeBriefFile()
- service, emitted, clock, _, _ = _triage_service(brief_file=brief_file)
- a, b = "transcripts/a", "transcripts/b"
- _ring(service, clock, _session_event(clock, a))
- assert service._triage_generation == 0
- service.handle_event(_transcript(clock, "old turn", session_dir=a))
- assert service._pending_brief == ["Unknown: old turn"]
- service.handle_event(_session_event(clock, b))
- assert service._triage_generation == 1
- service.handle_event(_transcript(clock, "new turn", session_dir=b))
- assert service._pending_brief == ["Unknown: new turn"]
- emitted.clear()
- _ring(
- service,
- clock,
- _session_event(clock, a),
- _transcript(clock, "old turn", session_dir=a),
- _session_event(clock, b),
- _transcript(clock, "new turn", session_dir=b),
- )
- assert service.status()["session_dir"] == b
- assert brief_file.directories == [a, b]
- assert service._pending_brief == []
- assert service._triage_generation == 2
- assert len(_cleared(emitted)) == 1
- def test_daemon_restart_new_session_resets_once():
- brief_file = FakeBriefFile()
- service, _, clock, _, _ = _triage_service(brief_file=brief_file)
- service.handle_event(_session_event(clock, "transcripts/a"))
- service.last_triage = {"signals": 1}
- service.handle_event(_session_event(clock, "transcripts/b"))
- assert service.status()["session_dir"] == "transcripts/b"
- assert service._triage_generation == 1
- assert service.last_triage is None
- assert brief_file.directories == ["transcripts/a", "transcripts/b"]
|