Skip to content

Battery-rate draft evaluation

This evaluates the draft for issue #5090 using the contributor's v2 CSV bundle. The replay calls BatteryRateStrategy.record_reading and calculate_power, the same methods used by the running strategy. It models battery/status/drying events, exact sample expiry and periodic recalculation; it does not run the Home Assistant energy sensor itself.

Reproduce from the repository root:

curl -L --fail https://github.com/user-attachments/files/34276366/replay_jarvis_bundle_v2.zip -o /tmp/powercalc-5090-replay.zip
uv run python -m utils.replay_battery_rate /tmp/powercalc-5090-replay.zip

Interpretation

The ten-minute default estimates 1218.45 Wh against 1218 Wh from the plug (+0.04%). The three holdout windows total 305.36 Wh against 308 Wh (-0.86%). The original issue's minute-based replay reported -0.2% overall and -0.8% on holdout at the same coefficient. These are different simulations: the draft uses a fresh initial percentage as the startup reference, reacts immediately to battery events, and holds its output until the next relevant event or timer tick. Percentage increments now expire at their exact window boundary. The previous 60-second-only timer held increments too long, producing +1.42% overall error and +1.26% holdout error. The original replay sampled at one-minute boundaries and suppressed the rate when it had no reading at the start of the trailing window. The coefficient has not been adjusted to fit this draft.

The steady charging power MAE is 3.64 W, with +0.26 W bias, across counter spans of at least five minutes. Startup and the last increments around 100% have larger errors. The power comparisons use the 1 Wh-resolution traces from Oct 22 onward; the earlier two traces have 10 Wh resolution.

A shorter window reduces lag but increases step fluctuations. A longer window smooths percentage steps but increases startup lag and losses on interrupted charges. The native ten-minute replay gives -1.66% for the recorded mid-run recharge; hypothetical departure at 30% has -17.77% aggregate error. Total error can be positive even when uncounted delayed charging energy is lost, because coefficient error and baseline contributions also affect the total.

The Cut at 100% Wh column removes the rate term at 100% for comparison. The implementation keeps it while the charging condition holds. Uncounted charging Wh measures the remaining positive difference between the gained percentage times the coefficient and the charging energy already emitted; no correction is applied. Counter values for hypothetical departures are linearly interpolated between ticks, so those results are estimates. A complete charge's plug energy uses the last reported counter values at its defined boundaries.

The tables below are a snapshot of the draft replay, without a power clamp, with independent drying power added. These results support further review and device testing; they do not establish a universal accuracy guarantee.

Replay results

Fixed coefficient: 0.96 Wh/%, baseline: 3.5 W, drying: +57.2 W. No power clamp or energy compensation. Fresh window at each charge start. Battery/status/drying events, exact sample expiry and a 60 s timer update power; integration holds power between events. The first four charges are calibration data; the last three are holdout data. No coefficient is refitted.

Window: 5 minutes

Charge Set Plug Wh Native Wh Error Cut at 100% Wh
Tue 20 Calibration 210.0 209.25 -0.36% 207.76
Wed 21 Calibration 260.0 263.29 +1.26% 261.75
Thu 22 Calibration 242.0 242.19 +0.08% 240.33
Fri 23 Calibration 198.0 198.36 +0.18% 196.46
Sat 24 Holdout 96.0 97.67 +1.74% 96.13
Mon 26 recharge Holdout 43.0 44.31 +3.05% 44.31
Mon 26 final Holdout 169.0 165.40 -2.13% 163.60
Total 1218.0 1220.47 +0.20%

Holdout: 307.38 / 308.0 Wh (-0.20%).

Power segment Counter spans >=5 min MAE W Bias W
First 15 min 12 8.99 -8.83
Steady 102 3.93 +0.08
100% +/-15 min 14 8.05 +5.26

Hypothetical departures (aggregated Oct 22 onward; counter interpolated between ticks):

Departure at Plug Wh Native Wh before departure Error Uncounted charging Wh
30% 62.43 56.00 -10.30% 6.38
50% 183.50 182.46 -0.57% 7.98
70% 334.28 328.49 -1.73% 6.45
90% 571.22 558.10 -2.30% 9.10

Window: 10 minutes

Charge Set Plug Wh Native Wh Error Cut at 100% Wh
Tue 20 Calibration 210.0 209.25 -0.36% 206.64
Wed 21 Calibration 260.0 263.29 +1.26% 260.65
Thu 22 Calibration 242.0 242.19 +0.08% 239.70
Fri 23 Calibration 198.0 198.36 +0.18% 194.75
Sat 24 Holdout 96.0 97.67 +1.74% 95.08
Mon 26 recharge Holdout 43.0 42.29 -1.66% 42.29
Mon 26 final Holdout 169.0 165.40 -2.13% 162.90
Total 1218.0 1218.45 +0.04%

Holdout: 305.36 / 308.0 Wh (-0.86%).

Power segment Counter spans >=5 min MAE W Bias W
First 15 min 12 14.27 -14.22
Steady 102 3.64 +0.26
100% +/-15 min 14 7.46 +4.58

Hypothetical departures (aggregated Oct 22 onward; counter interpolated between ticks):

Departure at Plug Wh Native Wh before departure Error Uncounted charging Wh
30% 62.43 51.33 -17.77% 11.04
50% 183.50 176.35 -3.89% 14.09
70% 334.28 324.17 -3.03% 10.78
90% 571.22 552.95 -3.20% 14.26

Window: 20 minutes

Charge Set Plug Wh Native Wh Error Cut at 100% Wh
Tue 20 Calibration 210.0 209.24 -0.36% 205.10
Wed 21 Calibration 260.0 263.29 +1.26% 259.38
Thu 22 Calibration 242.0 242.15 +0.06% 238.67
Fri 23 Calibration 198.0 198.35 +0.18% 191.54
Sat 24 Holdout 96.0 97.67 +1.74% 93.83
Mon 26 recharge Holdout 43.0 38.17 -11.24% 38.17
Mon 26 final Holdout 169.0 165.40 -2.13% 161.83
Total 1218.0 1214.27 -0.31%

Holdout: 301.24 / 308.0 Wh (-2.20%).

Power segment Counter spans >=5 min MAE W Bias W
First 15 min 12 20.35 -20.35
Steady 102 3.56 +0.07
100% +/-15 min 14 6.37 +3.49

Hypothetical departures (aggregated Oct 22 onward; counter interpolated between ticks):

Departure at Plug Wh Native Wh before departure Error Uncounted charging Wh
30% 62.43 43.83 -29.79% 18.54
50% 183.50 163.78 -10.75% 26.67
70% 334.28 314.13 -6.03% 20.81
90% 571.22 545.63 -4.48% 21.57