Skip to content

dev to main - #201

Merged
bvweerd merged 6 commits into
mainfrom
dev
Oct 11, 2026
Merged

bvweerd merged 6 commits into
mainfrom
dev

Conversation

@bvweerd

@bvweerd bvweerd commented Oct 11, 2026

Copy link
Copy Markdown
Owner

No description provided.

bvweerd and others added 6 commits September 28, 2026 09:45
Updates the requirements on [pymdown-extensions](https://github.com/facelessuser/pymdown-extensions) to permit the latest version.
- [Release notes](https://github.com/facelessuser/pymdown-extensions/releases)
- [Commits](facelessuser/pymdown-extensions@11.0.2...12.1)

---
updated-dependencies:
- dependency-name: pymdown-extensions
  dependency-version: '12.1'
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
_planned_first_step_was_executed compared the optimizer's planned power
against the committed power (after the commitment filter). When the
commitment filter locked a different power level within a price period,
the difference exceeded the 10% tolerance and every discharge step was
classified as plan_not_executed, preventing any calibration samples.

The power check is redundant at the fleet level: the direction check
already filters out mode-resolution overrides (hybrid → zero_grid,
manual), and the per-battery planned_delta already uses the committed
setpoint from _last_battery_setpoints — so a power difference caused
by the commitment filter does not invalidate the measurement.

Remove the power comparison; only the direction (effective_mode) check
is needed at the fleet level.
…p duration

_previous_step_hours() was reading step_durations[0] from the last
published optimizer run. When a mid-period correction run (e.g. at t=7.5
min into a 15-min price step) overwrites self.data, its first step covers
only the remaining ~7.5 minutes. Calibration at the period boundary then
computes planned_delta against 7.5 min instead of the full 15 min, pushing
it below the step_too_small_to_measure threshold even though the SoC
snapshot covers the complete price step.

Fix: when the SoC snapshot age exceeds the data-derived step duration,
use the snapshot age as the measurement window. The snapshot is refreshed
once per price step, so its age equals the full step the measurement covers.
The snapshot age is only preferred when larger than data_hours, so tests
and first-startup cases without a snapshot timestamp are unaffected.
- Type schemas against probatio (HA 2026.10 replaced voluptuous); keep the
  voluptuous import at runtime for older releases.
- Allow implicit re-export from homeassistant.components.binary_sensor so
  BinarySensorDeviceClass can still be imported from the package.
- Device migration: on the single config entry device model move a device
  still under the main entry into its subentry instead of removing the
  main-entry link, which would delete the device.
- Point .python-version at the Python 3.14 environment.
…s-gte-12.1

deps: update pymdown-extensions requirement from >=11.0.2 to >=12.1
@github-actions github-actions Bot added documentation Improvement or addition to documentation enhancement New feature or improvement dependencies Update dependencies labels Oct 11, 2026
@bvweerd
bvweerd merged commit 6684d68 into main Oct 11, 2026
17 checks passed

This branch was successfully deployed

1 active deployment
github-pages — fe86ad44 Deployed Oct 11, 2026 by bvweerd via Deploy to GitHub Pages #100
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

dependencies Update dependencies documentation Improvement or addition to documentation enhancement New feature or improvement

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant