Skip to content

docs(debug-journal): fold the queue slice and correct the Solcast, Hanchu and plan-card entries - #6219

Draft
springfall2008 wants to merge 1 commit into
mainfrom
bot/debug-journal-2026-11-16
Draft

springfall2008 wants to merge 1 commit into
mainfrom
bot/debug-journal-2026-11-16

Conversation

@springfall2008

Copy link
Copy Markdown
Owner

This PR is opened and written by the automated triage bot (/journal-update). Every claim below was re-checked against origin/main at 044e0ab2 before folding.

Took the first 15 candidates in filename order from the queue (6034 … 6055). 108 candidates remain queued for later slices.

Per candidate

#6034 — PR-review traps. Folded two under Reviewing PRs: judge a PR's COMPONENT_LIST/config.py scaffolding against origin/main (main holds "module.ClassName" strings resolved by load_component_class(), pinned by the quick-suite every_component_class_imports test), and a test file importing pytest breaks the whole suite at import time because the TEST_REGISTRY loop imports every module and pytest is not a dependency. Dropped the third item (_token/_key-suffixed config item not marked "secret": True is not a leak) — already covered by the is_secret_key() substring bullet under Debug yaml can contain real secrets.

#6037 — plan-card SOC symbols. Folded, merged with #6039 into one entry: ⚠ is the weather-alert keep, ✎ the manual SoC target, ⬇ the manual SoC max. The docs gap it reported is closed — #6039 merged 2026-11-08 (d2d7c6bf) — so the entry records the symbol meanings and the renderer divergence, not the missing legend.

#6038 — Predheat domain= asymmetry. Sharpened the existing Predheat bullet rather than adding a second: the domain keyword belongs to the outer read only, and the nested next_volume_temp fallback must keep none because it is the read half of the persistence round trip. Rewritten around get_temperature_arg() — the read moved into that helper, so the candidate's literal line no longer exists.

#6039 — two plan renderers. Folded into the same entry, corrected against main: the HTML plan now marks ⬇ by presence (minute in manual_soc_max_keep), because the candidate's cleanup commit landed with the merge; ⚠ and ✎ still gate on > 0. The web card still marks by membership (0-valued entries included) and appends ℮ plus a second ⬇; execute.py's [Manual SoC Max] suffix still gates on > 0. Verified by reading output.py, web_helper.py, web.py and execute.py.

#6040 — HACS becomes Marketplace in HA 2026.11. Folded one short bullet under Platform, web and updates: HACS cannot be installed on 2026.11+, is removed on upgrade, and AppDaemon apps stop being managed. The docs now cover both paths (#6041 merged 2026-11-08), so the entry is for diagnosing the report, not for a docs gap.

#6042 — Climate Scheduler contract. Folded a compact bullet under Predheat (service names, schedule_mode values, the exact workdays/use_workday_integration settings keys, workday/check_date's field and response shape, the hardcoded binary_sensor.workday_sensor). PR merged 2026-11-15 (b296ffe4), so this describes merged code. Dropped the pytz localize() walkthrough and the unrun probe narrative.

#6044 — Hanchu write clamp vs planner limits. Rewrote an existing entry. The "Power limits (PR #5684…)" bullet read as if hanchu_battery_rate_max/hanchu_inverter_limit were the top of the priority for the write clamp. Re-read on main: ranges_for() builds the CHG_PWR_LMT/DSCHG_PWR_LMT clamp as max(menu range, highest value seen held), capped by the HANCHU_MODELS rating, and honours no user override; the overrides only set the planner's battery_rate_max/inverter_limit. Now two explicitly labelled layers.

#6045 — daily counters and state_class. Folded a Sensors, history and load data bullet. Inventory re-checked on main: Deye is total_increasing since #6053; the gateway and GE Cloud *_today attribute tables still ship plain total with no last_reset.

#6046 — teslemetry reserve band. Almost all of this was already in the journal from the earlier fold of #6046/#6050. Added one missing trap: the reserve reaching the component during a freeze is floor(SoC), not SoC+1 (reserve_above_soc_charges_from_grid + hold_reserve_percent()), so a fix or test modelled on "SoC+1" has the wrong input. Dropped the two-docstring disagreement as already covered by the existing "snap direction unsettled" sentence.

#6047 — Sigenergy rates outside windows. Folded a Sigenergy bullet: outside a window the rate entities are mode-selection signals only, the branch ladder is export window → charge window → charge_rate_w == 0 → export_rate_w == 0 → eco, and the balancer's holds are written as intent[...] in utils.py, not in execute.py. Re-verified on main after #6047 merged (6ad5e9d5).

#6048 — IOG BOOST/bump-charge pricing. Dropped: the pricing bug was fixed by #6051 (merged 2026-11-09), and its other durable fact — the negative charge_in_kwh / keep the abs() — is already in the Octopus notes. Its one remaining trap, the duplicated billing predicate, is folded from #6051 below.

#6049 (escalation in the fetched data). Dropped as retracted. The Replay input: PV forecast changed rows log pv_forecast_minute/10/90, the post-calibration planner series (confirmed in fetch.py's log_replay_pv_forecast), so they cannot prove what the integration served. The real mechanism is #6055's state-cache aliasing. Recording this as correct would have put a wrong causal claim in front of the next reporter.

#6049 (replay rows post-calibration). Folded into Solcast: what those log rows really are, that a ~30-minute step pattern is Predbat's own plan_interval_minutes read cadence rather than the provider's poll, and that detailedForecast is kW per slot (sum = 2× the kWh total on 30-minute slots).

#6051 — tariff-billing predicate duplication. Folded an Octopus bullet: dispatch_billed_off_peak() and load_octopus_slots()'s tariff_rate_slot encode the same source set independently, and nothing fails loudly when they drift.

#6055 — HA state-cache attributes by reference. Folded a Solcast bullet as the root cause of the P90 compounding, with the fix: #6055 merged 2026-11-09 (5cc7c61f), fetch_pv_forecast() now copies the entries, and the general rule (copy a get_state() attribute container before mutating it when the same state will be fetched again) is kept because it applies to other components.

Existing entries corrected

Consolidation

The file was 226KB on entry, above the ~150KB guide. This PR consolidates the largest areas — Octopus, Startup/config, Plan charging, Tests/kernel, Web/MCP, Teslemetry, Solis, GE Cloud, Sunsynk/DEYE, Car charging, Component health, Versions/logs, Replays/dumps, Inverter control, Replays, GitHub/tooling, Predheat — by merging overlapping bullets and dropping repeated caveats. It also retires fixed-bug entries older than about six weeks: the pre-#5377 Sigenergy command-shape note, the #5464 docs-freeze story (keeping only the rated-power sensor names), and the pre-#5256 Solis Modbus sentence.

Net size is flat: about 9KB of verified material went in and about 9KB of older wording came out (226,152 → 226,590 bytes). Getting under 150KB would mean dropping roughly a third of the remaining entries, which are unique mechanisms rather than repetition, so I stopped short of that. A follow-up decision worth taking: whether the "older-log signature" class (entries whose only remaining value is dating a pre-fix log) can be dropped wholesale — that is the largest remaining block that is not a live trap.

./run_pre_commit passes (hooks and the quick suite, 7 slow tests skipped). New word added to the cspell workspace dictionary: hacsfiles.

…nchu and plan-card entries

Fold 15 queued candidates (all verified against main): the Solcast
state-cache band compounding and what a Replay-input PV line really
logs, the Hanchu write-clamp vs planner-limit split, the plan-card's
two SoC renderers, the Climate Scheduler service contract, the
component state_class inventory, the Marketplace rename, and the
PR-review traps. Drop #6049's retracted fetched-data theory.

Also consolidate the largest areas (Octopus, Startup, Plan charging,
Tests, Web, Teslemetry, Solis, GE Cloud, Sunsynk/DEYE, Car charging,
Component health, Versions/logs, Replays, Inverter control) and retire
fixed-bug entries older than six weeks.

Co-Authored-By: Claude Code <noreply@anthropic.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant