[3.0.2] - 2026-05-05

Fixed

• Deadlock when turning filter off — the filter-off command could hang indefinitely
• Rapid polling storm during preheat — excessive API traffic during the preheat cycle
• Command confirmation timeout loop — unconfirmed commands could trigger infinite rapid-poll cycles
• 30-second command latency — duplicate confirmation mechanisms caused unnecessary delays
• Silent command failures — commands acknowledged by the cloud but ignored by the spa are now retried once
• Rapid command sequences — sending two commands quickly no longer causes stale retries or conflicting state

Added

• 3-tier adaptive polling — polling interval adjusts automatically based on spa activity:
  • Idle (120s): nothing running, state stable for 10+ minutes
  • Active (30s): heater, filter, bubble or jet is on
  • Rapid (1s for 15s): after sending a command from HA
  • External change (5s for 15s): when someone uses the physical panel or MSpa Link app
• Historical prediction bias correction — past prediction accuracy is used to correct future time-to-target estimates, fading naturally as the rate model improves
• Device Detail diagnostic sensor — extended device info fetched once on startup (product images, device UUID, warning code, etc.)
• Temperature-bucketed heating rates — heating rate tracked in three temperature ranges for more accurate long-run estimates
• Session condition scalar — ambient conditions (hot/cold day) are reflected in the estimate within the first few minutes of heating
• Prediction accuracy tracker — logs estimated vs actual heating times for each session (last 10 persisted)
• Time-decay on stored rates — stale seasonal data gradually loses influence over time
• New sensor attributes: heat_rate_cold/mid/hot_deg_per_hour, session_condition_scalar, prediction_bias