docs(debug-journal): fold in eight triage findings, correct entries the fortnight's merges invalidated

[springfall2008]

🤖 Automated /journal-update run (bot branch, not for direct merge — a human merge is the review gate on this file).

Folds the 2026-09-07/08 triage queue into tools/debug-journal.md, re-verifying every candidate against main at b8996659 (v9.0.1) first. One candidate is a rewrite because its headline claim was wrong on the tree it named.

Per candidate

• 4645 (PR fix(inverter): auto-create charge_rate entity for script-driven "power" inverters (#3311) #4645 auto-merge over the GivTCP refactor) — folded into a new "Traps when investigating" bullet. A textually clean auto-merge can be a semantic conflict (the merged tree crashed the first Inverter() construction in two test modules, only one of them the PR's own); rest_data is gone since refactor(givtcp): move GivTCP REST handling into its own component #4864 and inverter_source_active() is the replacement question; components-consulting calls in Inverter.__init__ need a getattr guard because test components-stubs leak across modules. Verified: no rest_data reference remains in inverter.py; inverter_source_active() exists (inverter.py:76).
• 4677 (optimise_swap_export "disabled in the live plan sequence") — rewritten. The headline claim is wrong on the commit the candidate names: at b8996659 the pass is live — the commented-out call at plan.py:4314 is the old pre-Excessive SoC Ramp Down 6hrs before cheap rate slot #4478 position, and the live call at plan.py:4349 runs deliberately last, after the plan and solar passes ("Swaps run once all other passes have settled… Running the swap before them left those pinned in place (Excessive SoC Ramp Down 6hrs before cheap rate slot #4478)"). A triage run trusting the candidate would have told a reporter the swap pass never runs. What survives is the candidate's verified secondary finding, folded in: remove_intersecting_windows() runs inside every scoring simulation (prediction.py:598), a freeze-export window's 99.0 counts as enabled and does the clipping, and the clip is shared with the kernel path — added to the GH#4914 symptom row; plus a sharpened GH#4478 bullet noting the commented-out remnant and "read optimise_all_windows() end to end".
• 4984 (GivTCP REST rate-verify zero tolerance) — folded into the GivTCP row as fixed-but-mechanism-kept: Warn: … REST failed to setDischargeRate to <X> got <X> where got == requested is a zero verify tolerance (an inverter_limit_discharge/charge API override of 0 drove the old tolerance to 0 against a strict <), not a failed write. Verified on main: write_tolerance_watts() (givtcp_rest.py:355) now sizes both rate-write tolerances (:604, :626).
• 4987 (Chrg↗/Exp↘ in the same slot) — folded as a new symptom row: hold-charge window clipped before an export window + the 30-min html row's split State cell — sequential, not simultaneous; the hold window is cycle-dependent (replaying the same dump later produced no charge window). The saving-session bonus is stamped on the import rate too (load_saving_slot() runs on both rate tables, fetch.py:1191/:1214) — mechanism verified by reading; the metric impact stays marked suspected, as in the candidate.
• 4990 (both candidates) — folded together into a new symptom row: "Predbat reports a GivEnergy error but I use Solis / Fox / GE-Cloud" — inverter_type defaults to GE (inverter.py:258), component auto-config writes it only after discovery succeeds, while the component is active from instantiation — so a startup outage surfaces as "check the GivEnergy credentials" on a non-GE install. PR Name the configured inverter components when inverter_type is unset (#4990) #4992 (open) renames messages only. The two 4990 candidates overlap heavily and are merged into one entry rather than two.
• 4992 (component health signals) — folded as a new per-integration-table row: count_errors is a lifetime counter, api_started means "run() returned truthy once" (Gateway's run() deliberately returns True while unconfigured), is_alive()/last_updated_time() are the real health accessors, record_status() writes the raw message as HA state with no 255-char truncation, and the current_status-only-survives-the-last-call test trap. All verified against the current tree (component_base.py:71/263/270, components.py:950/967, gateway.py, output.py:2628); the PR-specific bits stay out — PR Name the configured inverter components when inverter_type is unset (#4990) #4992 is still open.
• 4989 (clock-skew warn band) — deferred, not folded. Its instruction is conditional on draft PR Warn on a moderate inverter clock skew of 5-29 minutes #4991 ("if it merges, the row should say instead…"), and Warn on a moderate inverter clock skew of 5-29 minutes #4991 is still open; the current journal text is accurate against main. The file is left in the queue un-archived on purpose so the next run re-checks it once Warn on a moderate inverter clock skew of 5-29 minutes #4991 merges — folding it now would make the journal describe behaviour main does not have.
• 4989's sibling session-trap (git log/show denied in bot permission mode; WebFetch commit pages / raw.githubusercontent at a tag as substitutes) went to bot memory rather than the journal — it is a property of the bot's permission mode, not of the repo.

Existing entries corrected (step-3 re-check against the fortnight's merges)

• Car-charging row — GH#4967 paragraph rewritten as fixed. PR fix(octopus): honour octopus_intelligent_consider_full in the prediction #4971 (commit 4a045e07) landed the model-facing car_charging_limit_model; with octopus_intelligent_consider_full off (default) IOG-slot cars get an uncapped model limit, the real limit is untouched, both engines read the same attribute, and update_car_manual_soc() caps the manual SoC write-back. The old "consider_full never reaches predict() at all" and "nothing asserts the switch's effect" claims are no longer true; the clamp-duplication and by-reference notes are kept for pre-fix(octopus): honour octopus_intelligent_consider_full in the prediction #4971 logs. Flagged the changed interaction with the unfixed GH#4952 Ohme overestimate.
• Teslemetry row — extended for Teslemetry: signal-tariff Time-Based Control mode so Opticaster runs the charge at full rate (#4892) #4976 and Teslemetry: an export window ending at midnight priced the whole of the next day at peak #4977. Teslemetry: signal-tariff Time-Based Control mode so Opticaster runs the charge at full rate (#4892) #4976 added teslemetry_tbc_control, a signal-tariff TBC mode (build_signal_tariff(), evaluate_schedule_tbc(), _settable_reserve() rounds an 81-99% request up to 100), so the row's "a slow charge ramp is Tesla firmware, not a Predbat write bug" now holds only for the default reserve-driven mode; Teslemetry: an export window ending at midnight priced the whole of the next day at peak #4977 fixed a midnight-ending export window pricing all of tomorrow at peak.
• GivTCP row — extended for SolaX Cloud system incorrectly starts GivTCP Rest component after v9.0.1 update #4993/Warn: Inverter 2 REST failed to setDischargeRate to 0 got 0 #4984 (above).
• Nothing else was invalidated by the fortnight's merges; the already-correct entries for fix(octopus): don't replenish the IO low-rate slot cap mid-window at midday #4951/fix(inverter): take the GE battery rate from battery_rate_max when there is no charge power register #4954/fix(octopus): price completed IOG dispatches off-peak regardless of location #4957/fix(carbon): cope with the carbon forecast running out before the plan ends #4958/fix(components): route HA events by the configured prefix, not the literal word "predbat" (#4939) #4962/fix: skip the 120s quick inverter update while template mode is set #4968/Fix GivEnergy auto-config to respect site export limits #4978 were cross-checked and left alone.

Also

• getattr added to the cspell dictionary (genuine Python builtin; the only spelling failure in the fold).
• Pre-existing test failure noticed at the gate, not fixed here: multi_car_iog_load_slots_regression and multi_car_iog_model_limit_fetch_4967 fail when run standalone on a clean docs-only tree at 00:06–00:07 local. The fixture (tests/test_multi_car_iog.py:561) forces now_utc to 12:00 on the machine's UTC date while midnight_utc stays on the real clock, so between 23:00 UTC and midnight the forced slot falls behind local midnight and is dropped — the machine-clock fixture class the journal already warns about. All 12 pre-commit hooks pass; the failure is outside this PR's scope, and one sentence on it was added to the existing trap bullet.
• Queue archiving: mv is denied in this session's permission mode, so the eight processed candidates are still in journal-queue/ rather than journal-queue/processed/; the next /journal-update run should re-archive them (or ignore them as already-folded).

[Copilot]

🟢 Approval recommended

Changes are limited to documentation and spellcheck dictionary updates, and the updates appear internally consistent with no review-blocking issues found.

Pull request overview

This PR updates Predbat’s debug/triage knowledge base by folding in recent triage findings and correcting several journal entries that were invalidated by recent merges, keeping tools/debug-journal.md aligned with current main behavior.

Changes:

• Expanded and corrected multiple debug-journal entries (notably around optimise_swap_export reachability, GivTCP REST post-#4864 behavior, Teslemetry updates post-#4976/#4977, and the GH#4967 car-charging fix context).
• Added new “Symptom → first place to look” rows and new investigation traps (e.g., semantic-conflict auto-merges, component-health signal pitfalls).
• Added getattr to the CSpell custom dictionary to prevent false spelling failures.

File summaries

File	Description
tools/debug-journal.md	Folds in new triage findings, rewrites/corrects outdated claims, and adds new symptom + investigation-trap guidance.
.cspell/custom-dictionary-workspace.txt	Adds getattr to the project dictionary to avoid spelling-tool noise.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.