Pixii: AC power sign inversion causes false power_wrong_direction during discharge #154

Problem

On a Pixii Home site, the battery is discharging and both its DC reading and the raw inverter AC metric are negative. The Pixii driver negates the AC reading when emitting control_power_w, so FTW's control feedback reports charging and raises not_following / power_wrong_direction while the power-flow view shows discharge.

The sent command and setpoint readback agree on discharge. This report concerns measurement normalization, not the command's sign conversion.

Environment

• Hardware: Pixii Home, 15 kWh; firmware version not collected.
• FTW: local test release v0.139.0-beta.9004, based on official v0.139.0-beta.1.
• Bundled Pixii source: DRIVER.version = "2.1.7".
• Bundled driver repository pin: srcfl/device-drivers@424f1e5c75e4cf4d9ada3ca016c6cba5c428e893.
• Operating mode: planner_arbitrage.
• The local test changes concern EV SoC/charging UI and provenance. Pixii driver contents are identical in local test1 and test4. The same AC-negation line is also present on this repository's current default branch.

Observed evidence

Two successive live API samples, supplied by the site owner on 2026-10-02; these are not a single synchronized capture.

Control feedback, around 19:33 CEST

From GET /api/status:

{
  "driver": "pixii",
  "kind": "battery",
  "mode": "planner_arbitrage",
  "status": "not_following",
  "reason": "power_wrong_direction",
  "readings_fresh": true,
  "site_confirmation": "independent_source_unknown",
  "site_meter": "pixii",
  "requested_w": -1236.5655136028975,
  "sent_w": -1236.5655136028975,
  "readback_w": -1244,
  "actual_w": 1242,
  "battery_soc": 0.585
}

The UI said: “Asked for 1.2 kW discharge, but it is charging 1.2 kW.”

Raw readings and diagnostic metric, 19:34:00.888 CEST

From GET /api/drivers/pixii:

{
  "readings": [
    {"type": "meter", "raw_w": -114.92, "stale": false},
    {"type": "battery", "raw_w": -1334, "soc": 0.5830000000000001, "stale": false}
  ],
  "metrics": [
    {
      "name": "battery_ac_w",
      "value": -1231,
      "updated_at": "2026-10-02T19:34:00.888919886+02:00"
    }
  ]
}

The flow view also showed discharge, household consumption and approximately 0.11 kW grid export. Battery SoC decreased between the samples, consistent with discharge.

Suspected cause

In drivers/lua/pixii.lua, the inverter AC power is read from holding register 40083, with scale factor 40084. The metric preserves that reading, but the control observation reverses it:

control_power_w = acw_regs and -ac_w,
-- ...
host.emit_metric("battery_ac_w", ac_w)

For the reported hardware, ac_w = -1231 becomes a positive control measurement, although the DC battery reading is also negative. FTW Core prefers control_power_w over the normal battery w when classifying control feedback, explaining the conflicting views.

Expected behavior

A genuine discharge should remain negative under FTW's convention (+ charging, − discharging). Feedback should compare the negative AC measurement with the negative sent setpoint rather than flag the opposite direction.

Reproduction

1. Observe a Pixii during normal FTW planner-driven discharge.
2. Compare control_feedback in /api/status with battery readings and battery_ac_w in /api/drivers/pixii.
3. If the raw AC metric and battery DC reading are both negative, check whether feedback's actual_w is positive and the UI reports charging.

Fix considerations

For this observed register convention, preserving the AC sign (control_power_w = acw_regs and ac_w) would make the measurement consistent with the discharge evidence. Please verify charging as well as discharge, and check other supported Pixii models/firmware before adopting a universal convention or a documented model-specific normalization.

Keep the existing command/setpoint conversions separate: the discharge command and its readback already agree. Add regression coverage for positive charging and negative discharging AC measurements, and correct the generator-frame comment if it does not apply to this hardware.

This report does not establish independent site confirmation: the grid meter and battery share the Pixii source, as the API explicitly notes. No driver fix has been applied on the reporting site.