gardening-watering-tuning

To fix biased plant watering predictions, this Skill analyzes past prediction-vs-actual watering intervals and proposes signed weight deltas for base and climate attributes. You run the nightly reflect step to generate safe per-attribute tuning updates, then apply the accepted proposals.

You can manually override watering weight parameters by adjusting the four editable weights stored in plants/<id>/tuning.yml. The Skill enforces clamping to safe weight ranges and limits changes to ±0.10 per adjustment, writing all modifications to an auditable adjustment_log.

The editable watering prediction attributes you can tune are base, climate_temp, climate_rh, and moss_pole. These weights adjust how cadence forecasts respond to base scheduling and climate factors, and are managed strictly through the plants/<id>/tuning.yml file.

The nightly reflect loop stops proposing watering weight adjustments when there are not enough fresh watering cycles logged. The Skill requires sufficient recent prediction-vs-actual interval data and enforces per-attribute spent-cycle cutoffs from tuning.yml logs before proposing any deltas.

Yes, you can reset a plant's watering tuning parameters back to default values while keeping the watering formula fixed. The Skill safely reverts the four editable weights in plants/<id>/tuning.yml and records the reset action in the auditable adjustment_log.

Guardrails for adjusting plant watering prediction weights include clamping per-adjust deltas to ±0.10, enforcing declared safe weight bounds, and requiring fresh watering cycles. The Skill applies these constraints during the nightly reflect loop and manual overrides to ensure safe tuning.