Live run 2026-06-24: flash.py apply NAK'd BUCD480,000 on both inverters — the
firmware rejects stop_charge_voltage=0 ("Full"). flash.py aborts on first setter
failure, so nothing changed and the cluster stayed in sync (verified).
The field flash.py calls stop_charge_voltage is actually the inverter's
battery_re_discharge_voltage (HA: sensor.lvx6048_*_battery_re_discharge_voltage):
the V at which loads switch back to battery after grid charging. 54.0 tops grid
charge to ~54V; raising to 56.0 is the corrected (but UNVALIDATED) lever and may
band-oscillate rather than hold absorption.
- calibration profile: 0 -> 56.0, with the finding documented.
- skill: solar-only is now the RECOMMENDED/known-good method; grid-assist demoted
to advanced/unvalidated with a mandatory diff-preview gate.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- eg4-lp4-v2.yaml: rationale updated 300Ah->600Ah/~30.7kWh; 120A combined is
now ~0.2C. Charge current deliberately NOT raised to 80A: not clipping — at
solar noon on a clear day (2026-06-24, ~5.8kW PV) the bank took only 71A of
the 120A cap. Real harvest limiter is PV (5.8 of 14.4kW nameplate, suspected
down strings), not the ceiling.
- REFERENCE: pack HA entity_ids are triple-prefixed
(sensor.eg4_lifepower4_lifepower4_1_lifepower4_1_*) — discover, don't construct.
No setting values changed; documentation/accuracy only.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The everyday profile caps grid charging at 54V, so the bank can go weeks
without a full charge and the EG4 BMS coulomb counters drift (proven: pack 6
read 76% SoC while at the same 53.4V/3.337V-per-cell as packs reading 50-55%
— all paralleled, so physically equal charge; the spread is pure drift).
- profiles/eg4-lp4-v2-calibration.yaml: temporary profile, identical to
canonical except stop_charge_voltage 54.0 -> 0 (Full), so grid can finish a
full charge to the 56.4V absorption hold that re-anchors every pack to 100%.
- calibration-charge skill: guided runbook (pre-flight safety, two methods
solar-only / grid-assist, live monitoring with cell-voltage/temp aborts,
re-anchor verification, mandatory revert).
- REFERENCE: scoped action-policy exception (this skill alone may flip
stop_charge, both units, user-confirmed, must revert); corrected pack-6 /
SoC-drift notes to the verified equal-voltage-different-SoC signature.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Four Skill-tool skills under .claude/skills/ that let an agent monitor and
troubleshoot the install (2x LVX6048, 6x EG4 LifePower4, OpenEVSE), grounded
in the real MQTT/HA topology rather than generic advice:
- solar-health-check : whole-system sweep + cross-checks + R/Y/G verdict,
incl. cross-unit "silently-dead inverter" detection
- troubleshoot-inverter: FWS fault decode, parallel sync, USB link recovery
- troubleshoot-battery : per-pack imbalance vs SoC-counter-drift, RS485 silence
- power-usage : PV/load/grid/battery balance + EVSE sessions
Shared lib:
- solar-snapshot : live MQTT capture (creds from powermon.yaml, no hardcoding)
- ha-history : HA recorder lookback (token from ~/.config/ha/token)
REFERENCE.md documents topology, real HA entity_ids (doubled slug), known
issues, and a safe-remediation-only action policy (restarts yes; setters no).
Action boundary: diagnose + restart wedged daemons / recover USB links;
never touches inverter/battery setters or flash.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
