Added
- Pacing-transparency UI on the control card (UX #4 / #6 / #8) — four new data-surface attributes on
sensor.foxess_smart_operations(discharge_deferred_reason+charge_deferred_reason,discharge_safety_floor_w+discharge_peak_consumption_kw+discharge_paced_target_w,discharge_grid_export_limit_w+discharge_clamp_active) are rendered directly as rows onfoxess-control-card:- Deferred reason (UX #4) — both charge and discharge sections render a wide row with the explanatory text produced by
_explain_*_deferral()whenever the deferred phase is active. Users finally see why pacing is holding back (e.g. "Feed-in limit reached — holding until more solar is available", "Export clamp slack 5.5 kW exceeds projected peak 3.0 kW; single headroom applied"). Longer strings wrap via new.detail-row-wide/.detail-value-wrapstyling. - Safety floor (UX #6) — a
safety_floorrow appears wheneverdischarge_safety_floor_w > 0. An upward-arrow icon (mdi:arrow-up-bold) surfaces when the paced target is below the floor (the C-001 floor is actively raising paced power). The row is click-expandable: tapping it reveals a translation-aware explainer with the actual peak value interpolated (e.g. "Minimum discharge power. Computed as peak household load (1.0 kW) × 1.5, so the battery can cover sudden load spikes without pulling from the grid."). Mobile-friendly — no hover needed. - Export clamp split (UX #8) — when
grid_export_limitis configured, the discharge power row splits into an inverter / export pair separated by/. The export side takes the warning colour and shows amdi:fenceicon whendischarge_clamp_activeis true, so the user can see immediately that hardware clamping is capping their export. Unchanged for sites without an export limit.
- Deferred reason (UX #4) — both charge and discharge sections render a wide row with the explanatory text produced by
- Standalone
foxess-taper-card(UX #5) rendering the BMS acceptance-ratio histogram from the newtaper_profileattribute onsensor.foxess_smart_operations(both charge and discharge histograms as chart-friendly{soc, ratio, count}lists). The card shows per-SoC-bin horizontal bars with observation count annotation and a low-confidence marker (·) for bins with fewer than 3 observations. Users opt in by addingtype: custom:foxess-taper-cardto their dashboard; an ApexCharts variant is also covered as a user template indocs/lovelace-examples.md. - Card translation coverage test (
tests/test_card_translations.py): parses each card'sTRANSLATIONStable and asserts every non-English locale carries every English key. Structurally prevents the class of bug that left theinfo_logsensor ID broken in beta.7 — a locale missing a key silently falls back to the raw key name. - E2E card-wiring tests: 7 new
TestControlCardtests (split power row, clamp-active toggle, safety-floor row + expandable explainer, charge/discharge deferred-reason rows, charge section title phase distinction) and 3 newTestTaperCardtests (card mount, empty state, seeded-profile bar widths). Taper card added to the E2E Lovelace dashboard. - D-051 "Transparency attributes surfaced via card rows, not tooltips" captures the design decision — visible rows over hover tooltips / debug panels / consolidated toggles — in
docs/knowledge/04-design/lovelace-cards.md, tracing to C-020 and C-034.
Fixed
- Structured events silently dropped when a user set a per-module log level:
emit_eventusedlogger.info(message, extra=...), which Python's logging framework evaluates againstLogger.isEnabledFor()before propagation to ancestor handlers. On a live HA (v1.0.12) with alogger:YAML config pinningcustom_components.foxess_control.foxess.inverterabove INFO, everySCHEDULE_WRITEevent emitted fromInverter._post_schedule()was dropped at the child's level check — never reaching the debug-log sensor attached to the parentcustom_components.foxess_controllogger. Fixed by changingemit_eventto build theLogRecordvialogger.makeRecord()and dispatch vialogger.handle(), bypassingisEnabledFor()while still running the logger's filter chain and every handler's level/filter chain. Structured events are telemetry the integration emits on its own behalf; visibility is controlled at the handler level, not at the logger level. Applies to all six event types (ALGO_DECISION,TICK_SNAPSHOT,SCHEDULE_WRITE,TAPER_UPDATE,SERVICE_CALL,SESSION_TRANSITION). Three-test regression suiteTestInverterScheduleWriteReachesParentHandlercovers the exact production symptom (child at WARNING) + default-levels baseline + session-context-survives-child-override neighbourhood. - Page-fixture flake in E2E CI caused by HA's housekeeping navigation (Flaky Test Detection surfaced this on four consecutive tag runs v1.0.11–v1.0.13-beta.2; live-diagnosed by observing a real HA container). Two compounding bugs in the staged Lovelace-panel wait (
tests/e2e/conftest.py): (a) the final stage's predicate was a bare!!panelattach check, which could return truthy moments before HA's service-worker/auth-refresh navigation at t≈12s detached the panel — surfacing either as a 30swait_for_functiontimeout or, in the test body, asLocator.screenshot: Element is not attached to the DOM; (b)_wait_for_stagere-capped every retry atmax_stage_ms=30000even when 50+ seconds of overall budget remained, so a post-navigation rebuild under slow-shard CPU contention had insufficient time to complete. Stage-3 predicate now requires three settled signals (hass.connected === true,ha-panel-lovelacemounted,panel.hasswired) — proving the panel is past HA's initial navigation churn, not merely attached; and a new_LOVELACE_STAGE_TIMEOUTS_MSmapping lets the final stage use the full remaining overall budget on retries while keeping the tight 30s cap on stages 1+2 (shadow-root attachment is synchronous and catastrophic failures must still time out fast). - SoakRecorder dropped invariant-violation detail when persisting run results:
save_runintests/soak/results_db.pywrote aviolationscount to therunsrow but never inserted correspondingeventsrows, so the rule name and detail message were lost. Observed on a 2026-04-23 nightly whentest_charge_solar_then_spikerecordedviolations=3with zero persisted detail — making post-mortem diagnosis impossible and hollowing out the C-020 observability value of the recorder. EachInvariantViolationnow emits anevent_type='violation'row with{rule}: {detail}as the event detail, anchored to the nearest sample's SoC/power/state so cross-run queries (SELECT * FROM events WHERE event_type='violation') can correlate violations back to system state.