[iBobbyTS]

Problem

The bare two-tier thresholds misfire continuously on Apple Silicon: P-core peak sensors oscillate ±5°C on a 2-4s scale under normal load, so the detector logs an "Instant spike/drop" event — plus a 15-line process-history dump — every few seconds. Measured on this hardware: 11,700+ events/day, roughly 200k log lines/day, burying the genuine thermal events the detector exists to surface.

Fix

• Detection moves into a pure, hardware-free AnomalyGate: a genuine event still reports immediately, but samples inside the 60s cooldown feed the rolling history silently, filtering the threshold-edge oscillation echo. History stays warm through cooldown, so prev-tracking survives suppressed samples.
• Calibration now suppresses only reporting — the gate is fed every sample, so the first post-calibration reading compares against a fresh previous instead of a stale pre-calibration one.
• The pre-spike dump collapses consecutive identical snapshots to a single line (ts → ts: idle (×n)).

Testing

swift build + full swift test pass, including six new unit tests in AnomalyGateTests (thresholds are strict-greater, cooldown suppression, prev-tracking through suppressed samples, sustained-tier window reset, history cap).