[springfall2008]

Automated journal flush — /journal-update, 2026-10-29. Every claim below was re-verified against origin/main 83a4adad before it went in; this PR is the review gate.

Folds in 14 of the first 15 queued candidates (slice 4186→5765, filename order); 1 dropped. 43 candidates remain queued (from 5766 onward) for the next slices.

Folded in

• 4186 (stacked-PR squash) — new trap bullet in "Traps when investigating" next to the base-staleness bullet: a stacked PR's squash lands on the review target's head branch (Add optional plan_at to plan_once() to plan at a caller-chosen instant #5811 → Split update_pred into plan_once()/execute_once() entry points #4186's branch), git rev-parse HEAD == headRefOid proves nothing, check the tip commit's message for a (#M) suffix with M ≠ N, and a squash on a PR head branch exists only on refs/pull/<n>/head. The merge-base staleness part was already the adjacent bullet, so it is referenced, not repeated.
• 4319 (plugin discovery) — new compact "Plugins (plugin_system.py)" row in the per-integration notes, re-verified on main: load_plugin()'s name-sorted getmembers() first match lets the imported base PredBatPlugin win (Initialising plugin class (name-based)), the _plugin.py single-level scan, the probe trap (a class sorting before the base passes with and without the fix), and — marked open-at-writing — PR fix(plugins): discovery must not instantiate an imported base class #4319's __module__ guards with the two review-flagged consequences (re-export break, strategy-2 double-init).
• 5699 (gateway EVC) — this candidate also drove a correction: the Gateway row's paragraph describing PR fix(gateway): EV start/stop reach the hub and are confirmed from telemetry #5699's confirm/retry/latch controller as "the first place to look ... once it merges" now says the PR is still open but superseded by the merged fix(gateway): reconcile EV control and honour read-only overrides #5739 (bot flagged 2026-10-28) and its symbols are PR-branch-only, and records what main does instead: gateway_evc_control's stateless reconcile (SetChargingProfile + RemoteStart in-window, RemoteStop out-of-window, _EV_COMMAND_ATTEMPTS 3 / _EV_COMMAND_RETRY_SECONDS 120s, only _ev_control_cp_id driven, read-only double-check) — all symbols re-grepped on main.
• 5748 (Teslemetry TBC) — folded in three pieces: (a) row 119's control-monitor sentence now names the HA sensor predbat.control_failures (publish_control_failures(), predbat.py) and its event schema (epoch slot_start/slot_end, requested/measured watts) with the "check it first + counts are the chronicity tell" guidance; (b) the Teslemetry row's GH#5748 passage adds that the plan-display HoldChrg there is the predicted fill-then-hold, not what was executed; (c) the replay section gains the third loader-crash cause — a dump whose captured args carry !secret tags (no constructor registered) and the sed-to-dummy workaround. The row's existing signal-tariff material was left alone, and the replay !secret row explicitly distinguishes itself from the apps.yaml boot-parsing !secret row.
• 5749 (DST render offset) — sharpened the existing clock-change-day plan-table row with the Charge window written before a clock change ends an hour late (autumn) / early (spring) #5749-verified specifics: the render-direction cases (instant after the change rendered before it → one hour late (autumn)), the astimezone(self.local_tz) fix shape, "a fresh decode after the change self-corrects" as the discriminator between re-rendering and re-deriving, the change-night execute test gap, and the persistence route kept explicitly suspected (it stayed marked that way, not upgraded).
• 5750 (Octopus partial errors) — the main fix's row already carried the merged shapes; folded the follow-up only: still exposed after fix(octopus): keep known data through partial GraphQL errors (devices list, saving sessions #5745) #5750, a nulled events/joinedEvents on the saving-session query flows through get_saving_session_data() unguarded (.get("events", []) → None → TypeError) — kept explicitly "unverified whether Octopus actually nulls these two"; re-verified the still-live code shape on main 2026-10-29. Dropped the candidate's Kraken provider-deprecation bullet: already fixed on main (f7358455, GH#5754, 2026-10-25) — nothing for an older-log reader, so it came out.
• 5755 (binary_sensor.predbat_running) — new trap bullet ("Docs entity names that live in HA core, not this repo"): the hassio Supervisor per-add-on Running sensor, disabled by default, absent on Container/Core, docs/install.md tells users to enable it while the Error Monitor section doesn't, and the general "don't grep apps/predbat for docs entity names" rule. Verified against main's docs and the PR/issue's HA-core source read.
• 5757 (Sigenergy MQTT statisticsTime) — folded into the Sigenergy row as merged content (ef1432b, 2026-10-25; then fix(sigenergy): widen MQTT freshness window to 450s so cadence jitter doesn't trigger REST polls #5790 widened the arrival window while SIGENERGY_MQTT_STALE_SECONDS stays 300 — verified): the lag gate in _mqtt_listener_loop(), the merge-onto-mqtt_period_raw invariant, and the do-not-reorder trap (last_mqtt_update stays arrival-based because it is inverter_time and freezes turn laggy-but-healthy feeds into clock-skew restarts).
• 5758 (recorder include-globs with literal ids) — one sharpening sentence in the Compare row's GH#5507 passage (literal ids flip identically — the selector is presence-based — and open PR docs: restructure the development guide around the branch, test and PR workflow #5758 fixes compare.md but its own revised faq.md example reintroduces the key; main's faq.md still carries it). Dropped the candidate's second half as duplicate: its entityfilter case analysis is the same entityfilter read the row already gained from the GH#5683 review round, which is more complete (it covers the include.entities-alone default-deny case the candidate's framing misses).
• 5759 (gh api / daemon facts) — new trap bullet: gh api flag order before-vs-after the endpoint decides permission (flags after), issue events oldest-first (tail = latest), exact-phrase PR search understates vs bare-token (search #<n> in:body too), and a BOT_PR event can be authored by the bot's own credential. feat(triage-daemon): label bot PRs and assign them to whoever added BOT_PR #5759 itself merged (7e42019).
• 5760 (DST history lookups) — new symptom row next to the plan-table DST row: the minutes_back = minutes_now + 1 lookup family in calculate_yesterday() (cost_yesterday, carbon_yesterday, the History arrays, cost_yesterday_car, yesterday_adjustment_pct) reads the wrong day's value on a change day — cross-referenced against the rate_replicate variant in the row above; totals self-correct; and the do-not-fix-minutes_now trap with both probe traps (localize the wall midnight; nothing covers a change day).
• 5762 (dynamic-load baseline keeps car energy after a trimmed dispatch) — folded into the Octopus row next to GH#5702 as an in-family entry: dynamic_load()'s car subtraction reads in_car_slot() over car_charging_slots rebuilt from the trimmed dispatches, so the trimmed start covers 1–2 of the 5 lookback minutes and the ended-dispatch drop leaves nothing subtracted (Dynamic load adjust is setting load minimum is the symptom line, bounded impact, test gap recorded, Octopus API: the trimmed tail of a long running dispatch is treated as a short dispatch (#5657) #5702/IOG Car charging energy not being recognised in history and being added to house load. #5602 cross-referenced as sibling mechanisms).
• 5764 (Solis grid-tied detection) — this one also drove a correction: the Solis row's fix(solis): treat a battery-less string inverter as PV-only and leave its registers alone #5281 recount now includes the batteryNum == 0 / batteryJump.batteryCount == 0 corroborators (an omission the candidate proved against _reports_no_battery()'s docstring and code on main), plus the '0'/'0.0' alternate-poll instability and PR fix(solis): stop enrolling a grid-tied string inverter as a battery inverter #5764's SOLIS_GRID_TIED_MACHINE_PATTERN (merged 9985960, 2026-10-26) with its no-pack-count-corroboration rule and the machine field's log visibility.
• 5765 (service-hook payload timing) — folded into the low-power/service-template row's GH#5182 passage: the ready-made {device_id} hook placeholder (verified get_arg("device_id", index=self.id, default="") on main) and the window-entry worst case of the {power} trail (immediate calls run inside the per-inverter loop before the apply_rate_intent() rate writes → first dispatch runs at the idle max, ~0.8 kWh off-plan at 10 kW, not template-fixable).

Dropped

• 5505 (Deye/Solarman verify: service sequences) — everything it pins is PR-branch-only (call_service_sequence/prepare_service_sequence are absent from main, verified by grep) and PR Add shared Deye/Solarman template with observed service sequences #5505 is unmerged; the one durable generalisation it carried (a custom inverter_type needs no code change — GE-def synthesis) is already in the review-conventions trap bullet. Its remaining content is forward-looking PR narrative that goes stale like PR-branch text has twice before ("unmerged at writing" Gateway items), so it stays out until the PR lands.

Existing entries corrected (re-check against merges since 2026-10-15)

• Row "plan tab's tooltip charge kW" — PR fix(output): cap the plan tooltip charge rate at the inverter limit #5643 merged (58b0bc3, 2026-10-28): cut down to mechanism + fixed-marker, PR-workshop detail removed.
• Row "Freeze exporting + car hold on a timed-pause inverter" — PR fix(execute): hold the battery for a car slot during Freeze exporting (#5612) #5640 merged (ae4e9c4, 2026-10-24): "Draft PR fix(execute): hold the battery for a car slot during Freeze exporting (#5612) #5640 (unmerged at writing...)" → fixed-marker, mechanism kept for pre-fix(execute): hold the battery for a car slot during Freeze exporting (#5612) #5640 logs.
• Gateway row's PR fix(gateway): EV start/stop reach the hub and are confirmed from telemetry #5699 paragraph — superseded by merged fix(gateway): reconcile EV control and honour read-only overrides #5739 (the reason for the 5699 candidate itself), corrected as above.
• Also checked and left alone: the fix(execute): stop holding a Powerwall on a reserve above its SoC (v9.6.2) #5823/fix(octopus): catch ReadTimeout in rate and free-session downloads so it can't abort the plan cycle #5801/fix(sigenergy): widen MQTT freshness window to 450s so cadence jitter doesn't trigger REST polls #5790/fix(charger-control): remember held chargers across a restart, and never drop an owed release #5793/fix(gateway): report a silent hub as unhealthy #5713-era merges touch no entry's claims beyond the above (the Sigenergy 450s change widened the other constant — re-verified SIGENERGY_MQTT_STALE_SECONDS is still 300 on main).

Branch bot/debug-journal-2026-10-29.