docs(debug-journal): fold in the 2026-09-27 queue slice (15 candidates); correct entries overtaken by the #5126/#5251/#5220/#5231/#5216/#5247/#5256 merge batch

[springfall2008]

Automated journal flush (triage bot, 2026-09-27). Folded in the first 15 queued candidates in filename sort order (5172-gateway-ack-two-designs → 5211-component-constructor-in-tests-aliases-live-args); 22 candidates remain queued (5213 onward).

Folded in

• 5172 (gateway acks) — new "Gateway MQTT" row in the per-integration table. The ack design on main is PRs feat(gateway): expose the gateway's enforced read-only flag as an entity and warn on mismatch #5220 + fix(gateway): send each hub command once and wait for its ack #5231 (both merged 26 Sep, after the candidate was written), so the entry describes the refined design (_subscribe_acks, _send_control 30s window + one re-send + _acks_seen fallback, _process_ack refusal-outranks-ok, clock-seeded PBAT<n> ids, prune-after-send) and warns against starting triage from PR fix: reconcile Hub command ACKs before retrying inverter writes #5172's own symbol names.
• 5190 (PEP 515 underscores) — folded into the Solis row: parse_cid_int()/cid_value_matches() and the core get_arg() float coercion all accept "1234567890_00" as a huge number with no exception; B0063 unhandled. Re-verified parse_cid_int on current main.
• 5193 (Sunsynk load telemetry) — folded into the Sunsynk/DEYE row as one entry built on the follow-up, which corrected the first-pass triage: collapse correlates with the battery being active, not cleanly with the work mode; dailyUsed kept counting; docstring's mode/battery-export claim is DEYE-inferred and not settled (marked suspected); manual override does not win (set_arg_auto overwrite=True); the log-forensics tip (component logs every API payload; widen substr field width when mining).
• 5194 (yesterday's rates are today's) — new symptom-table row: rate_replicate() negative-minute back-fill + history_to_future_rates() shift, exact only on fixed tariffs; master issue Wrong Rates in history #3267; _previous_day_rates as the in-tree previous-day pattern.
• 5198 (load hedge) — new "Load forecast & scenario hedging" row, rewritten for current main: compute_metric() is now a signed weighted average with the expert pv_metric90_weight (default 0.15) alongside pv_metric10_weight, and load_scaling10 is clamped load_scaling90 <= load_scaling <= load_scaling10; the candidate's "PV10-only downside blend at plan.py:1903" was overtaken by the PV90/envelope rework. Finding 1 (days_previous_auto ignores user weights) verified as written.
• 5202 (startup reset timeout, dup of Solis: B0600 "Datalogger returns data abnormally" on the startup register read stops the component starting, and retrying never clears it #5177) — mostly duplicate of the Solis row's existing Solis: don't let a refused startup register read abort startup, and back off B0600 #5185 coverage; folded the distinctive half as a sharpening (both triggers, B0600 refusal and SolisCloud timeout, now covered by the guard at the run() call site — verified on main at solis.py run()).
• 5203 (Solis TOU vs dispatch, 3 candidates) — folded into the has_service_api/*_service symptom row: the six *_service items are free-form parameterized service routes (call_service_template(), {placeholder} resolution, dedup/repeat interplay), window programming stays entity-based, no set_charge_window item exists; the third integration Pho3niX90/solis_modbus does expose dispatch (register 34502 gate) while wills106/hultenvp do not — Solis "remote dispatch" is a template/config question. The corrected Unneccessay updates to inverter causing eeprom wear #2328 story is in the Solis row: PR fix(inverter): Persist inverters across planning cycles #5126 made the commit-once ledger real (pre-fix(inverter): Persist inverters across planning cycles #5126 the every-cycle export commit was live).
• 5204 (zero-load fill warning) — new symptom row: fill_load_from_power() default-on, repeats up to ~8 days, benign triggers (quantisation plateaus, pre-helper zero region), silence knob, plus the three minute_data replay traps (oldest-first, relative series, smoothing branch inert for backwards data).
• 5205 — two pieces: new symptom row (two charge windows meeting at 00:00 → calc_dawn()'s deliberate per-day latch reset, verified live on main; combine_charge_slots gating), and the v9.1.0+ numpy-dtype loader crash recorded in the replay section with the lossless guard shape.
• 5209 (GivTCP arg shift/rotation) — appended to the GivTCP REST row: pre-fix(givtcp): fill per-inverter arg slots by REST endpoint index, not discovery order (#5209) #5216 positional fill + append-only recovery mechanism, log signature, debug-yaml diagnosis, workarounds, and the fix(givtcp): fill per-inverter arg slots by REST endpoint index, not discovery order (#5209) #5216 identity-by-index design note. The fix description itself was already in the journal (2nd Inverter stops reporting in Predbat 9.0.1 #5029 row + PR fix(inverter): auto-create charge_rate entity for script-driven "power" inverters (#3311) #4645 bullet), so only the pre-fix signature was added.
• 5211 (component constructor aliases live args) — new "Traps" bullet next to the Execute.__new__ stub trap: ComponentBase.__init__ sets self.args = base.args, so a real constructor in a test binds live shared args; __new__ + explicit attrs is the safer shape.

Existing entries corrected (merge-driven)

• car_charging_plan_smart default (symptom row): PR fix(car): default car_charging_plan_smart to on (#5237) #5251 (GH#5237) flipped the default to true, merged 2026-09-26, unreleased at the time of writing — the old "defaults false by design" explanation is now version-gated.
• Inverter object lifetime (args-trap bullet in "Traps when investigating"): PR fix(inverter): Persist inverters across planning cycles #5126 persists Inverter objects across cycles; refresh_config() re-reads live state and re-creates the dummy entities into self.args every cycle, while the commit-once ledger survives. Pre-fix(inverter): Persist inverters across planning cycles #5126 logs keep the older reading. The GivTCP row's in_calibration sentence re-anchored onto refresh_config() (per-cycle re-sampling survives).
• Solis row, PR fix(inverter): stop re-committing a stable export window every cycle (Solis) #4711 guard: renamed to the current commit-once ledger (last_committed/commit_pending/commit_needed() — last_export_schedule_committed no longer exists) and marked version-dependent: pre-fix(inverter): Persist inverters across planning cycles #5126 the rebuild reset the latch each cycle, so the Unneccessay updates to inverter causing eeprom wear #2328 every-cycle commit signature applies to older logs only.
• Solis Modbus row (FB00 switch): the Self-Use (33) hold is now PR fix(solis): manage the Energy Storage Control Switch on FB00 firmware #5256 (merged 2026-09-26); earlier mains never wrote the switch on GS_fb00 at all.
• Stale line citation in the Execute.__new__ trap bullet: the inverter_clock_skew_discharge_start read is now cited by symbol and the args-membership branch, not the drifted inverter.py:1624/:1563 line numbers.

Verification notes

• Every folded claim was re-checked against current main (a364a5ce): symbols grepped, mechanisms read, and the two replay-verified candidates (5204, 5205) kept their "replay-verified" status explicit; the Sunsynk trigger question stays marked suspected.
• The full ./run_pre_commit gate ran: cspell + markdownlint pass on the journal and dictionary (5 new dictionary words: dtype, hultenvp, kwarg, multiarray, setstate); the quick suite ran 122/122 passed before aborting at test_teslemetry_local_weekday_follows_the_base_clock at 00:34 BST — the documented host-clock window failure (00:00–01:00 local), unrelated to this docs-only change — and all 22 modules registered after teslemetry were run separately and passed.