docs(debug-journal): map the IOG supervision bounce symptom into the octopus row (GH#5390)

[springfall2008]

🤖 Automated journal flush (triage bot) — folding a 9-candidate slice (all of today's queue; 0 left queued) into tools/debug-journal.md, every claim re-verified against current main (ee401eca, v9.3.5).

Folded in

• GH#5381 — rate_tolerances()' 5% write-gate band swallows the 0 W → low-power move. Folded as a new symptom-table row ("battery charges nothing in a low power window straight after an export / freeze-export on a rate-register inverter"): the skip is the change-gate (within_fuzzy on adjust_charge_rate()/adjust_discharge_rate()), the band is 5% of the per-minute ceiling in watts, low power floors at max(400, ...) (find_charge_rate()), the register is left at 0 W by export with charge_discharge_with_rate (Huawei) and by export freeze without a timed pause, and the existing adjust_charge_rate4 case pins 100 W → 0 W as no-write — a fix needs a deliberate decision on the to-0W direction. Re-verified the band, the 400 W floor and the pinned test on main.
• GH#5382 — folded into the marginal-rate-sensors row: the cheap/moderate bands read event-contaminated rate_min/rate_max (free/saving/Axle loaders precede the final rate scan, so an Octopus free session zeroes rate_min, cheap_threshold = rate_min × 1.2 = 0 pins is_cheap false) while the sensor's own published attributes come from rate_min_base/rate_max_base — the attributes are the shape the band should follow. Verified on marginal.py (:153-154 vs :160-164).
• GH#5384 — folded into the car-charging row: plan_car_charging() ranks on import price only while each low_rates window already carries an export average (built with alt_rates=self.rate_export, fetch.py) that only plan_iboost_smart() consumes; the overlapping asks pull in opposite directions (Charge the car on forecast solar surplus #5195/Add opportunistic solar (sun-following) car charging model #4125/Using up excess solar to dynamically charge EV #632 divert-to-car vs Consider solar export value when scheduling EV charging #5384 weigh-foregone-export) — check which the reporter wants before pattern-matching.
• GH#5385 — folded into the Solis row: CID 499 (max export power) read scales ×100 only when raw < 200 while the write is unconditional int(value) // 100 (read/write asymmetry; _discovery_export_limit() duplicates the split), test_solis.py::test_publish_entities_export_power_unit_conversion pins the threshold split (case 6: raw 200 stays 200.0) so a suite green is not unit-correctness evidence; the "apps.yaml can't override" half is by design (overwrite=True), inverter_limit_charge/_discharge the exceptions.
• GH#5386 — folded into the GivTCP REST row: a read-back miss larger than the REST verify tolerance (write_tolerance_watts()/12 charge, /25 discharge) never converges through either layer; Predbat's REST mirror entities are the read-back, so the dump alone proves non-convergence; why the register sits at 5100 is open (stale cached Control read vs firmware/BMS cap). Added as the third instance of the rate-verify family; both theories marked open/untested.
• GH#5388 — the generalisable half of the SolaX-counter finding: minute_data() clamps dips but only zeroes recovery ramps steeper than MAX_INCREMENT = 1.2 kWh/min, so a dip whose recovery is gentler than 1.2 kWh/min is counted in full as the day's energy — folded into the "Energy totals jump or reset" symptom row (MAX_INCREMENT confirmed 1.2 at const.py:110; the diff > max_increment zeroing at utils.py:1552), with a 0-PV-day counter-argument ("just publish 0" trips pv_calibration()'s down-day rules, solcast.py) added to the Solcast row. The SolaX-specific half is NOT re-added: PR fix(solax): hold a lifetime counter that dips rather than read the recovery as energy #5389 (fc3354f, merged 2026-10-04) already implemented the proposed fix and folded the SolaX-specific account into the SolaX row itself — verified hold_counter_dip(), the 24 h reset rule, the X1-AC 0-PV publish and the M2-fields note are all already in the row.
• GH#5390 — two folds: (1) into the car-charging row (generic trap): re: config args are first-match-wins — resolve_arg_re() breaks on the first hit, so an alternation across two integrations consults exactly one detection sensor; (2) into the octopus row: the "battery flips Charge↔Demand with a pausing car" symptom mapped to the Partial charge from intelligent dispatch didn't treat full 30-min slot as cheap for home battery #5316/fix(octopus): keep a started IOG dispatch's half hour cheap after the car stops #5319 supervision family (3-minute not-charging trigger, 2-minute confirming grace, strip + resume), with the hourly car-kWh log lines as the ground-truth check; and the version-archaeology clause ("works on 9.x/broke on 9.y needs git tag --contains before being accepted as a regression — the 5390 claim was arithmetically impossible as a regression") went into the Version-drift trap bullet.
• GH#5394 — new symptom row: the plan-tab tooltip charge kW shows the pre-clamp rate while the prediction clamps at min(rate, inverter_limit × inverter_loss) (and a hybrid with PV can legitimately exceed via battery_rate_max_charge_dc, so a fix can't be a bare min()) — display-only.
• GH#5395 — folded into the Fox row: the device MaxSoc setting is published and write-back works, but automatic_config() does not bind it; no per-inverter max-SoC arg exists and the planner's only ceiling is global best_soc_max — "published ≠ bound", tree-wide maxsoc grep is the tell; siblings GE Cloud does not bind battery_min_soc, so the planner models a floor the device rejects #4832/input_number for Configurable Max SoC% #2128/Fox ESS Modbus - Setting Max SOC value #3837/Fox Cloud can overshoot the desired SoC on a forced charge #3398 noted; the device-side MaxSoc application on the write path stays marked unverified.

Existing entries corrected (each named with the merge that moved it)

• Car-charging row, GH#5366 bullet — PR fix(car): accept car_charging_rate as a per-car list in apps.yaml #5383 (84c0119, merged 2026-10-04, post-v9.3.5) made car_charging_rate accept a list: car_list_arg() splits the base-key list across per-car slots, a per-car suffixed scalar wins over its slice, and a list on a per-car suffix is ignored (falls back to the base list's slice). The row now keeps the pre-fix(car): accept car_charging_rate as a per-car list in apps.yaml #5383 float(list) TypeError mechanism for older logs and cites test_fetch_config_options.py pinning the three new rules.
• Sigenergy row, GH#5376 bullet — PR fix(sigenergy): send the optional command fields inside commands[] #5377 (4aa2564) merged on 2026-10-03 before the last fold; the row said "fix in open PR fix(sigenergy): send the optional command fields inside commands[] #5377". Corrected to "fixed in PR fix(sigenergy): send the optional command fields inside commands[] #5377, merged 2026-10-03, first shipped v9.3.5", the test-locking sentence put in past tense (its "optional fields omitted" cases are the in-tree example of the fixed test shape), and the "until such a fix lands" sentence adjusted — the fixed surface was never hardware-tested, so a post-v9.3.5 holds/exports regression report may be that surface going live for the first time.
• Gateway row, GH#5227 bullet — the candidate_aios or list(all_inverters) last-resort branch was removed by eaef6e61 (2026-10-04, "defer auto-config until battery telemetry"): a fleet with no battery-capable telemetry now defers auto-config with a warn (_auto_configured stays False so it retries), while an empty-serial slot that does report battery telemetry is still bound. The empty-serial trap itself survives on the battery-telemetry shape.

Dropped

• GH#5388's SolaX-specific half — covered by the SolaX row as of PR fix(solax): hold a lifetime counter that dips rather than read the recovery as energy #5389 (above). Only the generalisable minute_data dip/recovery asymmetry and the 0-PV down-day note survived as folds, where not yet covered.

Verification

• Pre-commit hooks: all pass (cspell reworded "upstep" → "single-read upward jump" rather than adding a dictionary entry; no dictionary changes needed — maxsoc/Wallbox/zappi already present).
• Quick suite: all modules green except teslemetry, which aborts 00:00–01:00 local via the previously documented UTC-vs-local bug in test_teslemetry_local_weekday_follows_the_base_clock (test_teslemetry.py:2789) — reproduced again on this docs-only tree at 00:15 BST 2026-10-05, and unrelated to this diff. Every other module ran green (secrets, export_encoding, export_more_solar_warning, perf, plot, predict_pv_power, dashboard_device_class, inverter_config_sensor, model, model_kernel, kernel_parity, prediction_batch, inverter, execute, all web_/chat/agent_tools, all component suites, ge_cloud, solax, sigenergy, fox_api, all manual_, all minute_data*, rate_replicate, download, …).
• The journal file's "Adding to this file" section asks for verified claims: each fold says what was verified and on which tree; hypotheses stay marked as such.

🤖 Generated with Claude Code

[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.