docs(debug-journal): fold in the 2026-09-26 queue slice (15 candidates); correct entries overtaken by the #5109/#5173/#5185/#5188/#5189 merge batch - #5240

Automated journal flush — the daily /journal-update run folding the triage bot's queued findings into tools/debug-journal.md. Branch pushed by the bot; a human merge is the review gate. 15 candidates processed (the first 15 in filename sort order); 26 remain queued for the next slice.

Per-candidate outcomes

GH#2716 (car charged at peak) — Folded in as a new symptom-table row: non-smart car planning is price-blind by design (car_charging_plan_smart default false, car_charging_plan_max_price default 0 never filters), a saving-session reward stamped onto the import rate can push an hour above the low-rate scan threshold so it vanishes from low_rates, and the in-progress peak-window house top-up is the optimiser's keep response. Line refs re-verified on main (plan.py:3935/:3968 — the high-price gate exempts windows whose limit is already non-zero).

GH#5169 (two test-harness corrections) — Partially folded in. Its part 1 ("all tests share one fixture" is out of date) is already in the journal — PR #5102's per-test instance fix landed before this candidate was written — so that half is dropped as duplicate. Folded in: pollution through module-level/class-level state can still cross the per-test instances (test_secrets.py's race test; the class-attribute lock on LogRedaction), a test that raises aborts the whole run (bare test_failed = func(my_predbat), verified live at unit_test.py:940), and the coverage/run_pre_commit && short-circuit (markdownlint failure ⇒ the quick suite silently never runs).

GH#5171 (id()-reuse test trap) — Folded in as a new trap bullet: allocator probes pass against the unfixed id()-fingerprint code; the deterministic test asserts the mapping is pinned via gc.get_referrers() (dicts not weakref-able). The fingerprint fix itself is merged (7d17ea1b).

GH#5173 (manual rates) — Folded in as a new "Manual rates" row, rewritten for the merge: PR #5173 merged 2026-09-24 — day_of_week rules now stamp across the whole horizon; keep the mechanism for pre-#5173 logs; on current builds a fully-tiled manual tariff is never rate_replicate()d, so "the offset stopped applying" on a manual tariff is that consequence, not a bug. The :59:59 zero-minute boundary trap verified still live (fetch.py:2144 pre-init) and kept as live.

GH#5175 (Axle import boost) — Folded in to the existing Axle row: PR #4520 provenance (first release v8.48.2), docs/energy-rates.md cite, maintainer upheld the design on #5060, #5175 triaged as enhancement, no config knob disables just the import side, and the colour-skew half belongs to the #5050 threshold row.

GH#5176 (manual load adjust) — Folded in as a new "Manual load adjust" row. Since the candidate was written, the docs disagreement it flagged was fixed by the maintainer (c06bff91, 2026-09-23 — docs now say "add the adjustment amount (which can be positive or negative)"); the entry records that, plus the additive mechanism, the plan-HTML ±-marker triage method, and the load_today_comparison citation (confirmed registered in TEST_REGISTRY).

GH#5177 (Solis startup abort) — Folded in as fixed. The journal already recorded PR #5185's merge; folded the surviving refinements into the Solis row (pre-#5185 signature, B0115 raised through the same unguarded call), moved the generic first-stays-True / blank-health-message framework trap to the Component health row (with the backoff corrected to the code's 60s→128min — the candidate's "from 2 min" was wrong), and kept the test_solis.py tricks (instance-attribute mock restore, canned api.session, never patch asyncio.sleep because _with_retry budgets with time.monotonic()).

GH#5178 (GivTCP REST publish) — Folded in: full-surface unconditional publish (count arithmetic, recorder-exclusion workaround, missing docs section) plus the discharge_target_soc claim/publish mismatch (rest_v3-only claim gate vs unsupported-model publish withholding) — both verified live on main. The log-dating trap went to the Version drift trap bullet.

GH#5180 (compare Existing badge) — Folded in to the Compare row: baseline snapshot from the live rate tables vs the refetched plain card, in-tree fix baselines (rate_import_no_io/rate_export_base), test gap, and the finally-block gap kept marked suspected (as in the candidate).

GH#5181 (is_running unhealthy) — Folded in as a new "Status health" row — still live on main, re-verified today (output.py:2699 strftime %z writer, predbat.py:1960 strict fromisoformat reader).

GH#5182 ({power} gets inverter max) — Folded in to the low-power symptom row as the write-path variant: dup of #3311, PR #4645 still unmerged (no merge commit in origin/main), chain re-verified live (inverter.py:630/:1952).

GH#5186 (teslemetry_automatic scoping + tbc default) — Folded in, and the existing Teslemetry row corrected: PR #5188 merged, so teslemetry_tbc_control is now default-on (DEFAULT_TBC_CONTROL = True); rewrote the stale "trial setting" passage and added the scoping fact that teslemetry_automatic gates only automatic_config(), not the scheduler-emulator writes.

GH#5187 (Solis slot current + recovery SoC) — Rewritten as fixed: PR #5189 merged — slot currents capped at the inverter's rated power and implausible recovery SoC (0/1/≤over-discharge/>100) now falls back to over_discharge_soc + 1 (both verified in solis.py); kept the mechanism for pre-#5189 logs. The never-read-slot-CID write-skip trap verified still live and kept as live, with the fixture trap note.

GH#5188 (bool-default narrowing) — Folded in: the first get_arg bool-trap bullet now records that component registry args are protected (a YAML null resolves to the registry default in Components.initialize(), verified components.py:806-811) while the trap survives for direct get_arg() callers; also recorded why the global fix (a None arm in get_arg's bool branch) was deliberately left as a follow-up.

Existing entries corrected (merge-driven)

• GE Cloud row, GH#5103 — was "still live"; fixed by PR fix(gecloud): re-read EMS battery inverter settings and warn on a slot 1 override #5109 (merged 2026-09-24): battery inverter settings re-read hourly under EMS (SETTINGS_SLOW_REFRESH_SECONDS) plus a once-per-device slot-1 override warning. Kept the pre-fix(gecloud): re-read EMS battery inverter settings and warn on a slot 1 override #5109 signature for older logs.
• Teslemetry row, tbc_control "trial setting" — overtaken by PR feat(teslemetry): enable teslemetry_tbc_control by default #5188's default-on flip.
• Other merges in the window re-checked against the journal (discovery phase-1 PRs feat(discovery): Fox Cloud reports to the discovery catalogue #5184/Discovery catalogue: AlphaESS, Solis, Deye and Sunsynk reporters #5206, PR Add Lead Hours Configuration for Octopus Saving Session Auto Join #5010 octopus lead hours, PR Move the off option to the top of the manual override dropdowns #5108 manual-select off-first, triage-bot PR Triage bot: pass review comment bodies as files, not inline arguments #5234): none invalidated other entries; the previous flush had already corrected the Solis: don't let a refused startup register read abort startup, and back off B0600 #5185/fix(sunsynk): clear applied_payload on sustained settle divergence #5142 entries and the write-poll budget figure.

Verification notes

• All 15 candidates re-checked against origin/main f0dc5bc4 before folding in. "Fixed on main" claims verified by reading the merged code: the Solis startup guard, the rated-current cap and recovery-SoC fallback, Components.initialize()'s YAML-null normalisation, DEFAULT_TBC_CONTROL, the fingerprint-holding log_secrets fix, and Honour day_of_week when seeding rates beyond the modelled days (#5168) #5173's whole-horizon stamping (2139b063).
• Two candidate claims corrected during verification: GH#5177's backoff "from 2 min" (component_base.py starts at 60 s), and GH#5176's docs disagreement is moot post-c06bff91.
• Still-live claims re-verified rather than trusted: the :59:59 zero-minute pre-init (fetch.py:2144), the never-read-slot-CID write-skip (solis.py), compare.py:598 baseline, output.py:2699/predbat.py:1960, the inverter.py:630/:1952 {power} chain, and the bare unit_test.py:940 test call.
• Pre-commit green (cspell + markdownlint; dictionary gained fromisoformat, weakref). The quick suite then aborted at the documented near-midnight teslemetry clock test (test_teslemetry_local_weekday_follows_the_base_clock, host-UTC date disagreement at 00:09 local) — pre-existing and recorded in the journal's test-date trap, unrelated to this docs-only change.
• detect_changes vs origin/main: 2 files, docs sections only, 0 affected processes, risk low.