Methods — how the two are told apart
read me
1 · Quantify the state component, then remove it
The measured threshold is treated as baseline + chronic_trend + state_effect + noise. Each record is tagged by exercise state. On chronic-phase data the mean threshold of every non-reference state is differenced against the rested reference to estimate a state offset; the adjusted series subtracts each record's offset. Strata with fewer than two chronic records are left unadjusted and flagged low-confidence rather than fit on one point.
2 · Establish a baseline only after maturation
The phase clock runs off implant date: <42 d acute, 42–90 d maturing, >90 d chronic. The early peak-and-settle (your 0.50 → 0.875 → plateau) is maturation, not drift, and is excluded from baseline statistics. Mean and SD come from the first N chronic adjusted anchors.
3 · Detect drift on the adjusted series
A standardized tabular CUSUM (k=0.5σ slack, h=5σ decision interval) accumulates small sustained departures — the right tool for slow drift that never trips a single-point limit. An EWMA (λ=0.2) corroborates. Because state variance has been removed first, an exercise-day spike no longer feeds the accumulator.
4 · Triangulate mechanism before acting
Rising threshold with rising impedance points to conductor/insulation change or microdislodgement; rising threshold with stable impedance points to interface fibrosis or a capture transition. The decisive LBBAP-specific check is conduction-system capture: a lengthening paced QRSd or Stim-LVAT signals loss of the conduction component even when the myocardial threshold the device reports is flat. That channel is manual because no automatic algorithm computes it.
Caveats
An n-of-1 framework; offsets are descriptive, not inferential, at small n. Automatic capture management reports myocardial capture — confirm conduction capture on 12-lead. HyperkAlemia, antiarrhythmic changes and electrolyte shifts are acute threshold modifiers that should be logged in notes and read as state, not drift.