v1.0.5

Added

• WebSocket mode selector (ws_mode): replaced the boolean ws_all_sessions toggle with a 3-state dropdown — Auto (WS only during paced forced discharge), All smart sessions (any smart session or force op), Always connected (WS preferred at all times with watchdog recovery). Existing configurations migrate automatically.
• Data freshness sensor: sensor.foxess_data_freshness exposes the current data source (ws, api, or modbus) as its state, with last_update and age_seconds attributes for staleness detection.
• Data staleness indicator on Lovelace cards: both cards compute data age client-side; badge turns red with elapsed time (e.g. "API · 2m") when data exceeds 30 seconds old.
• Structured session logging: session context (ID, type, SoC, power) enriched via logging.Filter; debug log sensor exposes structured data for E2E tests and power users.
• Target power display: Lovelace card shows current vs target discharge rate when they differ during feed-in pacing.
• Entity-mode E2E tests: input helpers simulate modbus entities; connection_mode fixture parametrizes cloud vs entity modes with function-scoped containers for full isolation.
• Reconfigure flow: add or update web portal credentials without re-creating the config entry. Accepts both raw password and pre-computed MD5 hash.
• Structural tests: AST-based verification of synchronous cancel functions (C-016) and brand import boundary (C-021).

Changed

• SoC display precision matches confidence: Lovelace card shows integer SoC until the first confirmed integer change (e.g. 93→92), then switches to 2 decimal places. Before the first change, interpolation is just an estimate; after, the real SoC is known to be near X.5, making interpolation meaningful.
• Interpolated SoC stored at full float precision: rounding applied only for change detection (2dp gate to prevent entity update storms), not storage.
• Session construction via factory functions: create_charge_session() and create_discharge_session() ensure consistent field defaults and reduce duplication across callers.
• Min SoC floor lowered to 0%: min_soc and min_soc_on_grid now accept 0, removing the previous 5% floor.
• Unified cancel functions: _cancel_smart_charge and _cancel_smart_discharge replaced with delegates to brand-agnostic cancel_smart_session, ensuring the _on_session_cancel hook fires from all cancel paths.
• Simulator fidelity: charge taper above 90% SoC, discharge taper below 15% SoC, per-app state isolation, stale-stream behaviour matching FoxESS cloud.
• E2E config uses production defaults: tests needing non-default options set them explicitly via the options flow, matching real user setup.
• GitHub Actions updated: checkout v4→v6, setup-python v5→v6, upload-artifact v4→v7, download-artifact v4→v8.

Fixed

• WS linger race captured stale forced-discharge data (D-009): _on_session_cancel now returns the WS stop coroutine; callers await it after override removal completes, so the linger captures post-session self-use data. Also fixed the clear_overrides service path which used fire-and-forget.
• Entity-mode service domain detection: apply_mode() used hardcoded "select" and "number" domains, breaking input_select/input_number entities from foxess_modbus. Added _entity_service_domain() helper to derive the correct domain from the entity ID prefix.
• Charge fdSoc regression: listener must pass fd_soc=100 to prevent FoxESS API validation failure (C-008).
• Smart discharge starting before scheduled window: deferred discharge listener omitted the start= parameter, bypassing the floor clamp. The inverter received fdPwr=0 but ignored it and discharged at full power.
• Force operations not cancelling opposite smart session: both force operations now cancel both session types, preventing leftover listeners from fighting the schedule.
• WS not connecting after deferred discharge start: timer now fires the WS-aware wrapper, so WebSocket connects as soon as forced discharge begins.
• WS reconnect during smart charge: charge listener wrapper wasn't triggering _maybe_start_realtime_ws.
• Session sensors delayed by ~30s: now subscribe via coordinator.async_add_listener for instant state propagation.
• Stale work mode badge after failed cleanup: override removal retries on each successful REST poll; cancel paths clear _work_mode immediately.
• Smart sessions survive transient API errors: errors retried on next timer tick; only 3 consecutive failures trigger abort. Previously any transient cloud outage killed a multi-hour session.
• SoC interpolation overshooting entity value: clamp tightened to [tick − 0.5, tick + 0.44] so the rounded display always matches the entity.
• REST poll starvation from SoC extrapolation: now updates entity data directly without resetting the poll timer.
• Feed-in pacing stuck at initial power: switches to self-use when target is below min_power_change threshold, enabling ramp-up on the next tick.
• Feed-in energy inflated at session start: baseline deferred to the listener's first tick when fresh data is available.
• Schedule horizon not set on immediate discharge start: computed inline before state dict creation.
• WS data_source badge stuck during stale stream: coordinator notified immediately when reconnecting, so badge shows "API" until fresh WS data resumes.
• WS stream-stolen recovery: tracks last useful data timestamp; reconnects after 30s of only stale frames. Token invalidated on handshake error, forcing fresh login.
• SoC interpolation stuck between ticks: timestamp set after integration, made unconditional.
• SoC clamp rounding: upper bound changed to 0.94 to prevent displayed value exceeding authoritative tick.
• Progress bar start SoC wrong after deferral: updated to actual SoC when session begins.