Production VoiceDelivery END tail-loss / terminal ACK / hard expiry / STT reconnect / invalidation / offline cleanup PASS Voice production upload policy PASS Wearable preferences/NVS failure and reboot recovery PASS Battery percent/filter/invalid/full-calibration/charge-state and compact NET rates PASS voice_recording_store_test: ok Settings Network / Known Networks / Forget All confirmation PASS Wearable shell immediate navigation/draft and UTC clock PASS RX8130 UTC / fixed Demo UTC+8 / day, month, year, leap rollover PASS Wearable RAM structs: DevicePreferences=12 LoadedPreferences=24 UiState=280 NVSBlob=25 WiFiProfile=104 WiFiStoreMax=521 PASS Wearable renderer round-safe Watch/Agent/Help/Settings; glance rows, battery icon, date/time order; all sizes/themes PASS Agent Live == manual page 0; pages tile the body with no overlap/gap PASS Watch glance host thresholds/hysteresis, task mapping, Codex percent/reset, redraw policy PASS Transcript cache: hot path / CN-TW collision / caller metrics / missing / PSRAM failure PASS CJK subset generator coverage PASS (GB2312 Level 1 3755/3755, glyphs=5000, 中/这 covered) Wi-Fi provisioning secrets boundary: no password logs, no token exposure, Portal keeps passwords private, fixed AP password defined once PASS Production Wi-Fi binding: real USB context, checked driver returns, exact-length credential copy PASS Ownership guard lifetimes / both lock orders / transition wait boundaries / 7 negative mutations PASS Old station waiter cannot clear a successful B connection PASS Concurrent takeover cannot interleave resolve and durable commit PASS Approval invalidates queued Portal worker before any driver call PASS Old worker cannot clear or overwrite a newer request PASS Delayed Portal cleanup cannot cancel a newer candidate PASS Preempted active candidate's late result leaves new A untouched PASS Driver failures and associated-SSID mismatch never persist B PASS Storage failure releases the unpersisted link for recovery PASS Failed candidate releases ownership and restores A with Portal still active PASS Pending station A disconnect still falls back to known B PASS Station driver failure cleans the old link before trying B PASS GOT_IP during connect is observed (snapshot before connect) PASS Late disconnect cannot clear new association; genuine disconnect releases steady state PASS Approval snapshot is lock-free; UI reporting never holds Wi-Fi locks PASS Production hello / Gateway handshake / History menu binding PASS SSID is required SSID is required Saved. Connection test: PASS. Saved. Connection test: PASS. ok test_add_effective_timeout_exceeds_firmware ok test_add_never_prints_password ok test_add_verdict_wording_is_honest ok test_build_command_add_encodes_values ok test_build_command_simple_verbs ok test_cli_add_has_timeout_headroom_over_firmware ok test_cli_add_prompts_with_getpass ok test_cli_add_reports_rejected ok test_cli_has_no_password_flag ok test_cli_rejects_missing_ssid ok test_cli_status_roundtrip ok test_decode_preserves_invalid_escape ok test_encode_decode_round_trip ok test_encode_matches_firmware_rules ok test_format_list_is_human_readable ok test_format_missing_end_is_not_success ok test_parse_add_verdict ok test_parse_error ok test_parse_incomplete ok test_parse_list_decodes_ssid ok test_parse_status ok test_parse_status_decodes_utf8_ssid ok test_redaction_removes_password Mac Wi-Fi helper: encoding round-trip, verdicts, redaction, no-argv password, timeout headroom PASS Production raw button polling / all Settings Back paths / Settings Voice starts=0 / Approval ownership PASS (host) Double A/B top-level only / openAgentSessionPicker reuse / Speech Mode APPLY PASS (host) Touch virtual A/B / top-level swipe / Agent browse / History controls / Settings entries PASS (host) Touch contact epoch ownership / stale cancellation / Long release PASS (host) History multi-screen message paging / touch parity / reset PASS (host) History cross-page fetch/cache traversal PASS (host) Navigator stale-epoch ownership / legal Long finish PASS (host) Production Voice 999/1000 CANCEL, 1001 UPLOADING; transcript delivery never auto-sends; EXPIRED A no RETRY / B CLOSE PASS Known Networks: single forget precise, Forget All confirmed twice PASS (host) Provisioning modal: explicit entry only, gestures blocked, Hold B cancels PASS (host) Production shell gestures / offline settings / Apply failure / Voice / Approval modal priority PASS (host) Raw GPIO / 30ms debounce / >600ms Hold A / Settings Back / no Voice or trailing CLICK PASS (host) Bounded Navigator paging/empty creation/capabilities/native choice semantics PASS tap inside slop resolves exactly one Tap PASS tap slop boundary 18px is inclusive PASS swipe in four directions resolves the dominant axis PASS swipe requires >=1.5x dominant axis PASS long press resolves Long then LongRelease PASS movement cancels Long and never yields Tap PASS sub-debounce contact is ignored PASS one primary gesture per contact PASS left/right half splits into virtual B/A PASS TouchTransform maps, mirrors and clamps in one place PASS Raw TouchGesture reducer / transform PASS (host) history cache never exceeds kCachePages PASS history append_older makes the fetched page current PASS history cache eviction keeps traversal linear and in range PASS history cache older/newer boundaries PASS AgentBrowse manual page / live tail restore / reset PASS HistoryScreen linear paging and position PASS HistoryScreen multi-screen message paging is linear and reversible PASS history_bounded_text truncates on a UTF-8 boundary PASS Bounded Agent history cache / view / UTF-8 PASS (host) Wi-Fi profile store, validation, selection, backoff, no-auto-SoftAP, candidate-first, initialized-marker/first-run migration, Missing/Valid/Corrupt policy, candidate-disconnect ordering, exact-length credential copy, persistent recency sequence, SSID/QR PASS T1 steady connected: 100 ticks, no reconnect PASS T2 candidate takeover: old station waiter STALE, B wins PASS T3 candidate failure: C discarded, station reconnects A PASS T4 set_config failure: B never persisted PASS T5 actual SSID mismatch: rejected, not persisted PASS T6 candidate success: B persists and holds, no 2s reconnect PASS T7 approval preemption: late candidate STALE, never commits PASS reverse: late candidate cannot touch a new station generation PASS generations are unique across ownership changes PASS concurrency: candidate takeover always invalidates a captured station attempt (200 runs) PASS concurrency: 4-thread hammer keeps ownership invariants (50 runs) PASS Wi-Fi connection ownership: T1-T7 + interleavings + concurrency PASS (host) B4 portal candidate failure: not persisted, AP stays up, station auto-reconnects A PASS B3 steady connection: 100 ticks, no reconnect/mark_success PASS item 12 approval preemption: AP stopped, late candidate ignored PASS preemption beats a candidate that would otherwise succeed PASS Wi-Fi network lifecycle: B3/B4 + Approval/Provisioning ownership PASS (host) USB production binding: real context reaches live store (status/list/add/remove/clear) PASS (host) Network menu: renderer order == reducer mapping, Wi-Fi Setup requires explicit confirmation PASS (host) QR encoder byte-mode matrix match, finder invariants, growth, bounds PASS USB Wi-Fi protocol encoding, candidate-first verdicts, redaction PASS Portal HTML escaping, hostile-SSID safety, body bound and form parse PASS Production hard expiry clears voice identity / keeps B CLOSE across disconnect PASS (host) Network callbacks / new hello / generation / authoritative DONE / retained Voice on weak network / stale operation PASS (host) Terminal notification exactly-once (DONE/FAILED, no periodic or reconnect replay) PASS (host) Production WebSocket -> Cloud -> board History page/error dispatch PASS (host) ..... ---------------------------------------------------------------------- Ran 5 tests in 1.684s OK Approval modal arbitration and exactly-once notification PASS single press waits the full gap then emits exactly one CLICK PASS double press emits DOUBLE_CLICK with zero leading CLICK PASS 299 -> DOUBLE; 301 / exact gap -> first SINGLE then next sequence PASS long press emits one LONG and one LONG_RELEASE with zero CLICK PASS first short + second long -> LONG only, then LONG_RELEASE PASS A->B / B->A never produce a DOUBLE_CLICK PASS contact bounce produces no spurious gesture PASS stale epoch / explicit invalidation discard pending CLICK PASS allow_double=false keeps immediate CLICK and never doubles PASS Raw ButtonGesture single/double/long state machine PASS (host) Raw GPIO / 30ms debounce / >600ms Hold A / Settings Back / no Voice or trailing CLICK PASS (host) Agent render / Read Screen / UTF-8 / redraw selection PASS (host only) TAP version 13 # (node:3022759) ExperimentalWarning: The MockTimers API is an experimental feature and might change at any time # (Use `node --trace-warnings ...` to show where the warning was created) # Subtest: archived inspection grammar; mutations, composition and unknown tools need approval ok 1 - archived inspection grammar; mutations, composition and unknown tools need approval --- duration_ms: 1.818494 type: 'test' ... # Subtest: production Pi pager never registers synthetic tool-call approval ok 2 - production Pi pager never registers synthetic tool-call approval --- duration_ms: 0.390062 type: 'test' ... # Subtest: archived approval channel waits past the old limit using a controlled clock ok 3 - archived approval channel waits past the old limit using a controlled clock --- duration_ms: 11.2769 type: 'test' ... # Subtest: a session has one owner and an existing runtime receives input ok 4 - a session has one owner and an existing runtime receives input --- duration_ms: 20.527434 type: 'test' ... # Subtest: text activity batches by character threshold without splitting Unicode ok 5 - text activity batches by character threshold without splitting Unicode --- duration_ms: 3.875676 type: 'test' ... # Subtest: text activity flushes on the throttle timer ok 6 - text activity flushes on the throttle timer --- duration_ms: 30.68892 type: 'test' ... # Subtest: plain text removes noisy markdown and redacts credentials ok 7 - plain text removes noisy markdown and redacts credentials --- duration_ms: 0.417564 type: 'test' ... # Subtest: tool summaries keep only a short safe action ok 8 - tool summaries keep only a short safe action --- duration_ms: 0.706637 type: 'test' ... # Subtest: final reply excludes commentary, tools and prior assistant messages ok 9 - final reply excludes commentary, tools and prior assistant messages --- duration_ms: 0.422764 type: 'test' ... # Subtest: registered Pi handlers emit live activity and settled final independently ok 10 - registered Pi handlers emit live activity and settled final independently --- duration_ms: 446.935858 type: 'test' ... # Subtest: existing-session snapshot restores native final/recent text after overlay restart ok 11 - existing-session snapshot restores native final/recent text after overlay restart --- duration_ms: 0.555193 type: 'test' ... # Subtest: queued user requests each complete before the native agent loop settles ok 12 - queued user requests each complete before the native agent loop settles --- duration_ms: 9.390939 type: 'test' ... # Subtest: low-level agent restarts inside one user turn keep the running timer anchored ok 13 - low-level agent restarts inside one user turn keep the running timer anchored --- duration_ms: 6.43486 type: 'test' ... # Subtest: official confirm/select/input/editor preserve native answers ok 14 - official confirm/select/input/editor preserve native answers --- duration_ms: 2.121513 type: 'test' ... # Subtest: native abort/timeout/session close cannot replay replies ok 15 - native abort/timeout/session close cannot replay replies --- duration_ms: 6.688637 type: 'test' ... # Subtest: target polling preserves focused select and open options ok 16 - target polling preserves focused select and open options --- duration_ms: 4.188753 type: 'test' ... # Subtest: pending target request is not overwritten by a stale poll ok 17 - pending target request is not overwritten by a stale poll --- duration_ms: 1.126575 type: 'test' ... # Subtest: approval polling preserves buttons, focus and in-flight decisions ok 18 - approval polling preserves buttons, focus and in-flight decisions --- duration_ms: 1.375182 type: 'test' ... # Subtest: approval buttons submit their own ID and binary decision ok 19 - approval buttons submit their own ID and binary decision --- duration_ms: 2.906426 type: 'test' ... # Subtest: unified target selector includes Pi and Codex ok 20 - unified target selector includes Pi and Codex --- duration_ms: 1.929853 type: 'test' ... # Subtest: effort options follow selected model metadata and hide for Pi ok 21 - effort options follow selected model metadata and hide for Pi --- duration_ms: 0.978748 type: 'test' ... # Subtest: an old pending approval stays actionable without a human decision ok 22 - an old pending approval stays actionable without a human decision --- duration_ms: 1.691174 type: 'test' ... # Subtest: speech mode binds the selected device target and test has no arbitrary text ok 23 - speech mode binds the selected device target and test has no arbitrary text --- duration_ms: 1.361477 type: 'test' ... # Subtest: native approval buttons preserve all offered decisions and labels ok 24 - native approval buttons preserve all offered decisions and labels --- duration_ms: 1.250316 type: 'test' ... 1..24 # tests 24 # suites 0 # pass 24 # fail 0 # cancelled 0 # skipped 0 # todo 0 # duration_ms 718.178938 ........................................................................ [ 21%] ........................................................................ [ 42%] ........................................................................ [ 63%] ........................................................................ [ 84%] ..................................................... [100%] =============================== warnings summary =============================== .venv/lib64/python3.11/site-packages/starlette/testclient.py:51 /home/lighthouse/esp32-dev/.cache/main-promotion-test-20261009/server/.venv/lib64/python3.11/site-packages/starlette/testclient.py:51: DeprecationWarning: The anyio.abc.BlockingPortal alias is deprecated, use anyio.from_thread.BlockingPortal instead. _PortalFactoryType = Callable[[], AbstractContextManager[anyio.abc.BlockingPortal]] -- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html 341 passed, 1 warning in 6.75s HTTP PASS: http://127.0.0.1:40093/health Admin PASS: HTML and API status WSS PASS: hello/hello_ack via ws://127.0.0.1:40093/device/ws Gateway -> Device PASS: server_snapshot received Device -> Cloud PASS: heartbeat and button_event recorded Cloud -> Device PASS: display_message received Pi -> Gateway -> Device PASS: RUNNING and DONE target states gateway TCP/WebSocket smoke test passed PI authenticated TLS WSS voice transport + FAKE STT + confirmed input: PASS CODEX authenticated TLS WSS voice transport + FAKE STT + confirmed input: PASS