docs(debug-journal): fold in the 2026-09-25 queue slice; correct the #5103/#5186/#5053-adjacent entries

[springfall2008]

🤖 Automated journal flush — this PR was produced by the daily /journal-update run against the queue at /Users/treforsouthwell/predbat-triage-bot/journal-queue (slice: first 15 candidates in filename sort order, 2026-09-25). Every claim below was re-verified against origin/main (130596bc) before it went in. 33 candidates remain queued beyond this slice.

Queue candidates folded in (13 entries, 15 candidates)

Candidate	What happened	Why
5112-ml-model-status-signals	Folded — extended the Load ML row with the model-status signals: is_valid() reports "active" for a never-trained model (model_initialized set before the first epoch; validation_mae/age skip None), training_timestamp is None is not a sound never-trained test (legacy saved models), and train() stamps at the end of every curriculum pass so an abort leaves fresh stamps that suppress the ml_max_model_age_hours retrain. The normalisation-statistics pairing stays marked suspected, not measured.	Verified on main: is_valid() at load_predictor.py:2042-2066 still has the gap; PR #5112 is still an unmerged draft, so the entry says so explicitly.
5144-octopus-payment-method-variant-rows	Folded as fixed — Octopus row now records the overlapping DIRECT_DEBIT/NON_DIRECT_DEBIT rows, minute_data()'s last-row-wins, the period-to-period flip ("always the higher rate" was API-ordering artefact), and the fix: filter_payment_method() (utils.py) from PR #5145, preferring DIRECT_DEBIT, keeping nulls.	Verified: filter_payment_method() exists (utils.py:1313) and is applied on the day/night, component, annual and minute-data paths (octopus.py:1774/1775/2028/2937); added by 60db7e1c via b4edced9 (PR #5145). Its docstring itself records "not stable across periods" and the Agile null case.
5146-hold-for-car-gate-divergence + 5151-car-hold-skipped-during-export	Folded together into the Car charging row: the two different "Hold for car" gates (live execute.py block vs prediction.py's (car_load_scale > 0) and (not car_charging_from_battery) and set_charge_window), the deliberate export-window exemption, and the diagnostic pointers.	Verified: prediction.py:865-866 gate unchanged; execute.py carHolding block has if not isExporting: and the reserve bump; discharge_car_full_bat2 still in test_execute.py (moved :2324→:3160).
5150-control-active-restart-gate	Folded into the Sunsynk/DEYE row: PR #5150's persistence + upgrade-cache inference, and the wrong "presses this on every cycle" comment (both inverter.py press sites are change-gated; the unconditional post-restart re-commit is is_hm_format-only, which excludes HH:MM:SS). The suspected #5138 link is not recorded as fact — the journal already credits #5142 with the confirmed mechanism.	Verified: time_button_press: True for both types (config.py:2318/:2350) but the press sits inside if schedule_changed: (inverter.py:2820-2823) / if current_soc != soc: (:2079-2090); is_hm_format net at :2813-2818; SUNSYNK_RESTORE_MAX_CONTROL/DEYE_RESTORE_MAX_CONTROL live.
5152-solis-hold-timed-only-current	Folded into the Solis row: the only discharge-suppression lever is set_current_from_power()'s timed current; the standing battery_discharge_current registers are never driven; GS has no reserve/pause alternative. Firmware-semantics claim stays marked not live-log confirmed.	Verified: inverter.py:2985-2995 writes only timed_{direction}_current; grep for battery_discharge_current across the repo is empty; INVERTER_DEF["GS"] flags as the candidate said.
5153-predheat-weather-unit	Folded into the Predheat row: no °F conversion in get_weather_data() (runaway-to-external mechanism), plus the second bug (minute_data_age includes optional heating_energy's age → target sensor ignored), plus the three investigation traps.	Verified: predheat.py:186-201 passes temperature raw; minute_data_age = min(age_external, age_internal, age_target, age_energy) at :553.
5156-deye-tou-filler-truncates-windows	Folded as fixed. The Deye filler truncation is repaired on main: build_tou_slots() now lives in the shared tou_schedule.py, and _padding_segments() places padding just after the slot it follows with the same state — "a real but inert interval" — with the docstring citing GH#5156 by name. Kept the pre-fix tell (stops at 00:00/04:00/08:00/12:00/16:00/20:00 while the plan shows charging).	Verified: tou_schedule.py:146-174, docstring cites GH#5156; TOU_FILLER_TIMES no longer exists anywhere. The candidate's own one-line fix was superseded by the shipped rework.
5157-teslemetry-silent-drift-blind-reassert	Folded as implemented into the Teslemetry row: the reporter's silent-drift evidence kept (read-back stays correct, restart recovered), and the fix — run() re-asserts the whole device tuple (export rule, grid charging, reserve, mode, not the tariff) every FORCED_ASSERT_SECONDS (2h) with the dedupe bypassed.	Verified: FORCED_ASSERT_SECONDS = 2*60*60 (teslemetry.py:57), comments cite GH#5157, Info line names the re-assert; commit 077f5040 (2026-09-19).
5162-manual-soc-floor-inert-at-zero-rates	Folded into the keep-mechanisms section: the floor penalty is weighted by the import rate at the deficit minute, so it is inert inside 0p windows; the ceiling is export-rate-weighted and survives; the ✍ marker is drawn regardless of whether the plan met the target; the manual_charge lookalike trap.	Verified: prediction.py:1323-1328 + kernel mirrors; manual_soc_max/all_active_keep_max at fetch.py:935-959.
5166-edf-kraken-day-night-400 + 5166-kraken-rest-400-day-night + 5167-kraken-fetch-rates-error-contract	Folded together as a new Kraken row (the journal had none): EDF 400-on-day/night vs Octopus 200-empty cross-provider trap; the fix shipped in PR #5167 (merged) — GraphQL fallback gate now (400, 404, 410) per direction, authenticated retry stays 404/410, standing-charge fallback stays 404/410 deliberately; the (None, None) tuple contract and the failures_total double-count are recorded as fixed-on-main with the mechanism kept. The "E-TOU on E.ON Next = 400" re-attribution the review reverted is not recorded.	Verified on main: KRAKEN_REST_RATES_UNAVAILABLE_STATUSES/..._PRODUCT_NOT_FOUND_STATUSES (kraken.py:219-220), the (None, None) stash at :876-884, the counter snapshot at :900-906, the deliberate standing-charge gate at :999-1005.
5168-manual-rates-day2-extension-write	Folded as fixed — new "Manual rates (fetch.py basic_rates())" row. The mechanism (last-rule-for-a-slice won on day+2; exactly 48h out) is kept with "fixed in PR #5173"; the final fix stamps the whole horizon per weekday (2139b063 from review), not just the day+2 gate the candidate described. Kept the end: "06:59:59" final-minute trap and the metric_future_rate_offset test pin.	Verified: whole-horizon stamping at fetch.py:2290-2323 with the batpred#5168 comment; tests 13/14 + third-day assertions in 6/7 now pin every horizon day (the candidate's test blind spot is closed on main, so the entry says so as history).
5169-5053-cache-now-on-main	Dropped as a duplicate — the journal already reflects #5053 merged (both the redaction entry and the GH#5063 premise-trap bullet), and this candidate added nothing beyond what is already there. The one durable nugget (the lock must be a shared class attribute, not lazily built) is now recorded on the redaction entry as a fact of the merged PR #5171 mixin.	Verified: class Hass(LogRedaction) (hass.py:187), mixin in log_secrets.py:25.

Existing entries corrected (4)

1. GE Cloud row, GH#5103 "still live" → corrected to fixed in PR fix(gecloud): re-read EMS battery inverter settings and warn on a slot 1 override #5109 (merged 2026-09-24): settings are re-read hourly (SETTINGS_SLOW_REFRESH_SECONDS), EMS/gateway devices every cycle, and check_ems_inverter_slots() now warns on a non-00:00–23:59 slot 1 — so the old entry's "nothing checks the per-inverter slot windows" and "there is no warning" are no longer true. Pre-fix(gecloud): re-read EMS battery inverter settings and warn on a slot 1 override #5109 signature kept, plus the settings_from_cache nuance.
2. Teslemetry row, teslemetry_tbc_control "trial setting" → corrected to default on since PR feat(teslemetry): enable teslemetry_tbc_control by default #5188 (single source of truth in teslemetry.py, registry default components.py:635); verified by test_teslemetry_tbc_control_defaults_on_via_the_registry.
3. Redaction entry + GH#5063 premise bullet → _invalidate_log_secret_pattern() no longer lives in hass.py; PR Move the log-redaction cache into a self-contained mixin (#5169) #5171 moved the cache into the LogRedaction mixin (log_secrets.py), shared class-attribute lock.
4. Solis row → added PR Solis: cap slot currents at the inverter rating and distrust implausible recovery SOC #5189 (merged 2026-09-24, GH#5187): per-slot currents now also capped at the inverter's rated current on top of the sysCommand.max clamp. (Correction by merge-scan, not from a queue candidate.)

Left queued

33 candidates beyond the 15-candidate slice (5169-per-test-fixture… through 5222-soc-max-calculated…) — untouched, to be offered on a later day.