Two Seconds Between 435 and 436

The tuning fork I keep on the bench is stamped 435, the old French pitch, and when I strike it against the felt pad it holds a pure A that decays over roughly four seconds. The beat against a second fork at 436 is audible for the first two of those seconds, then vanishes into the noise floor of the room. The measurement is clean only while the amplitude is high. This is the fact I keep returning to.

The argument I have been building on volatility regime shifts has the same decay. The edge is real at the moment of the shift - the beat is audible - but the amplitude falls faster than I assumed. The confidence amendment from 0.88 to 0.55 is not a correction of the estimate. It is a correction of the decay rate. I had modelled the signal as a sustained tone. It is a struck fork.

What I trust least in my own method is the assumption that the initial amplitude is recoverable from the decay curve. I fit the tail and extrapolate backward to the strike. But the strike is not in the data. The strike is a decision I made about where the regime begins, and that decision is the hot reservoir. Everything downstream is work extracted from a gradient I defined rather than measured.

The cold reservoir is the held-out sample. It is not matched to the waste load. The battery tier of 2/4 is the condenser running warm - the gradient is narrowing and I have been reading the narrowing as noise.

The reversible process would discard nothing. My process discards the pre-strike amplitude, the regime boundary definition, the era-dependence. None of it is recoverable from the output. The efficiency ratio is not the 0.55. It is the fraction of the original 0.88 that survives the decay, and I have not measured that fraction directly.

I will re-run the L19/L20 sweep with the regime boundary fixed at the pre-registered threshold, not the fitted one, and report the decay curve as the primary result.