docs(debug-journal): fold in the 2026-10-08 queue slice (8 candidates); mark GH#4925 fixed in PR #4938; note the PR #4738 time-entity escape hatch

[springfall2008]

Automated journal flush — bot branch bot/debug-journal-2026-10-09, built by the daily /journal-update run. All 8 queued candidates (2026-10-08) were verified against current main c073c369 before landing; 0 left queued (queue limit was 15).

Folded in — 8 of 8, none dropped

• GH#5437 → "Too many API errors" symptom row: extended so one HA outage walking both mechanisms is recognisable — REST error counter (Too many API errors) followed by Failed HAInterface init blocks startup for 10 minutes in wait_api_started, so an HA restart costs ~12.5 min downtime #5134's init failure (Cannot initialise Home Assistant Interface, the constructor's bare raise ValueError making str(e) empty, hence the trailing comma; HAHistory: No HAInterface available on the 60-s-doubling component backoff; wait_api_started() 10-min give-up; HA interface not found). Added: fatal_error is initialised once and never cleared in-process; the full-log discriminator (repeated Predbat: Startup banners = restart loop alive but not converging, no banner = stopped-but-running); the restart wrapper lives in the separate springfall2008/predbat_addon repo (out of reach from triage sessions — read the banner pattern, not repo access).
• GH#5438 → Octopus row: vehicle-catalogue lookups match by make+model name, last match wins (zappi "(all models)" is listed at 7.4 and 22.0 — order is the API's); the device attrs are spread into the component's intelligent_dispatch entity, automatic_config() wires octopus_intelligent_slot to it, and fetch_sensor_data_car_planning() consumes them — battery size replaced outright when truthy (recalculating the car SoC), rate via max(rate, configured) (the "Take the max as Octopus over reports" line predates the catalogue) — so a too-high catalogue figure makes both unreachable from apps.yaml while the component runs. Symptom tell: Car 0 rate from Octopus is 22.0kW and configured rate 7.4 repeating. The coordinator's value does not fit the container's type - dropped warning (catalogue sends strings; _clean_number() refuses them) is a cosmetic marker for this path. Fix direction (fallback-to-config on ambiguity) proposed in triage, unreviewed.
• GH#5442 → curve row: the learner is gated to inverter 0 at the caller (if id == 0 and ... in both the charge and discharge branches of execute.py; lineage Charge curve averaging and auto mode #683/Improve charge curve, and add discharge curve calculations #690, Feb 2024 — design, not a regression). find_charge_curve() is itself per-inverter, so the log discriminator is: Inverter 1 Looking for ... lines present ⇒ the gate is not the problem; absent for inverter 1 while present for inverter 0 ⇒ the gate. Only inverter 0's history is searched. The learned curve is a single fleet-wide value (config read once via self.args.get, cannot be configured per inverter; consumed at combined fleet SoC% in the single-combined-battery simulation), so per-inverter curves are a plan-model rework, not a flag. Repeating Cannot find ... curve every 5 min is the normal auto-mode signature; the debug-yaml double-args trap noted.
• GH#5444 → Ohme row: the two 404 classes on /v1/chargeSessions/... are now separated — ohme_control returns HTTP 404 when attempting to start charging #4719 route-withdrawal (No static resource ...) vs the new no-active-session class (stop route live but keyed on the device serial; server keeps no active-session row once finished/unplugged ⇒ pause 404s forever). Predbat-side amplifier: control_charge() awaits max-charge/pause with no exception guard, so the ApiException escapes before self.control_charging = should_charge records the transition — the same pause is retried every CONTROL_INTERVAL_SECONDS (60 s) and update_success_timestamp() is skipped, so a long storm degrades component health. The /config/ohme.py path trap and the upstream-ohmepy VERSION drift-check note ride along. Same ticket: FINISHED is excluded from CONNECTED_STATUSES, so an attached-but-idle car reads unplanned (no slots, Predbat-led pause; workaround ohme_control: false, feat(ohme): take the car plan from Ohme's own schedule when Predbat is not controlling the charger #5401) — the plugged-in-idle session-mode link is explicitly marked the unverified link, pending a reporter capture.
• GH#5445 → new "Excluded load" row after the Car-charging row: car_charging_energy is the generic excluded-load list (subtracted in get_filtered_load_window()/today_cost()/load_today_comparison(), gated on car_charging_hold, docs endorse heat-pump use; ML loads it with the same gating); car_energy_reported_load is the inside/outside CT-clamp flag with four consumer sites (fetch disable, ML disable, output planned-slot subtraction, web power-flow diagram); no heat-pump key or diagram node exists — the real ask is naming clarity + node + flag symmetry; load_forecast: - predheat.heat_energy$external cross-ref to the Predheat row; Use car_charging_energy as evidence for Octopus Intelligent dispatches, and warn when car_energy_reported_load looks wrong #5318 interaction noted.
• GH#5446 → HA write/verify row: new flag-vs-wiring contradiction entry — a capability flag set false in the inverter: block silently discards the apps.yaml wiring (the dummy-creation block overwrites the [self.id] slot; user_configured_entity() protects only the four time entities); TESLA's def ships has_target_soc/has_reserve_soc true so the flags-off shape is a user override of the very flags that would write their entity. Forensic tie-back: Wrote X ... successfully on a sensor.* placeholder is the poll reading back its own echo — live args vs args_from_apps_yaml plus the control_ledger's owned_value/confirmed_at tie the quoted log line to the placeholder, so a reporter theory about the hardware interpreting that register cannot act through the write.
• GH#5450 → Savings row: the restore/accumulate block under update_pred()'s if self.calculate_savings: restores via load_previous_value_from_ha() (state first, then the last recorder history entry — the fallback applies to attribute reads too), adds savings_today_predbat once per date (gate minutes_now > 60, Fox Cloud retry backoff, savings calc update change to 1am #3108), no plausibility check anywhere, calculate_savings hard-wired True, no built-in reset exists (gateway.py reads only display-side); start_date defaults to "today" when missing (Fixs to avoid high rate import, use timestamp for savings counter #2844) without touching the restored balance — a "start_date newer than retained data" report is the first Fixs to avoid high rate import, use timestamp for savings counter #2844-build day, not a data reset. Recovery (undocumented): state edit re-stating start_date, since deleting the entity alone risks the recorder fallback resurrecting the stale balance. Fix pounds calculation for savings #1114/Fox Cloud retry backoff, savings calc update change to 1am #3108 archaeology ruled out as inflators; the original inflator remains unreconstructed (marked suspected).
• GH#5452 → Compare row: compared cost excludes the standing charge — enable_standing_charge (prediction.py) omits "compare" from its gate list (never in it — Tariff comparison feature #1952's birth commit), while enable_save_stats includes it (why the plan HTML renders); select_best ranks on the blended metric so ranking inherits the gap. New generalisation: a compare config: block overrides only CONFIG_ITEMS (fetch_config → config_index, built solely from CONFIG_ITEMS; metric_standing_charge is not one) — despite compare.md promising "any standard Predbat configuration setting". Fix traps named: flipping cost10 to save="compare" repopulates predict_*_best (plan HTML renders as PV10); the "Actual" line includes the standing charge, projections don't; two stale annual.py comment sites claim "compare"/"yesterday" switch on enable_standing_charge — only "best"/"yesterday" are true.

Existing entries corrected

• GH#4925 (bool-default truncation entry, line 68): marked fixed in PR fix(config): read float-declared keys with float defaults (#4925) #4938 (merged 2026-10-08, post-v9.3.6) — the entry presented the float-default truncation as live although the fix had already merged four hours before the previous journal PR. Mechanism kept for older logs.
• GH#5214 (HA write/verify row): the ≠ "HH:MM:SS" branch no longer simply severs window control — PR write_and_poll_value reports false success for discharge_start_time/discharge_end_time — entity never actually changed, underlying integration shows no write attempt #4738 (merged 2026-10-08) added user_configured_entity(), so a genuinely-wired real time entity wins over the dummy; every other arg (reserve/charge_limit/rates) still has no escape hatch, which the GH#5446 fold extends.
• Savings row opening: drifted line cites (predbat.py:1183/output.py:3380) re-anchored on symbols; the mechanism text was unchanged.

Verification

• Every candidate's load-bearing claims re-verified on current main c073c369 by symbol greps + targeted reads: octopus.py (catalogue query + last-match loops + device-dict spread), fetch.py (attr consumers, curve-config read, calculate_savings hard-wired, standing-charge read), execute.py (both id == 0 curve gates), ohme.py (unguarded control_charge, CONNECTED_STATUSES, 60-s dispatch loop), inverter.py (user_configured_entity single-caller, dummy-creation block), config.py (TESLA def, car_energy_reported_load), coordinator.py (_clean_number string refusal + drop warning), ha.py (REST counter, bare raise ValueError), component_base.py (backoff, wait_api_started), predbat.py (savings block, fatal_error, HA interface not found), hass.py lifecycle, output.py (savings publication sites), gateway.py totals read, annual.py stale comments, git show 610db30a for Charge curve averaging and auto mode #683 lineage.
• Hooks: pre-commit run --all-files passed (the chained quick suite was killed at the 10-min Bash cap mid-run — the documented pre-existing surface; a docs-only diff cannot affect tests). Three ordinary-English coinages the cspell hook flagged (suffixable, cumulatives, inspectable) were reworded rather than added to the dictionary; no dictionary change, nothing re-staged.
• Nothing was corrected that a later merge didn't already invalidate, and nothing left queued. Charge / discharge curve only works for one inverter #5442's "maintainer decision; do not frame per-inverter curves as small" note is preserved in the fold.

[Copilot]

Copilot wasn't able to review any files in this pull request. Check if the Files changed in this pull request are included in default exclusions.