Constant PSR_FWHM_PIN_CEILING_US 

pub const PSR_FWHM_PIN_CEILING_US: f64 = _; // 10f64

Sanity ceiling (µs) on the configured PSR triangle FWHM: one decade above the PSR_FWHM_US_MAX fit bound. CalibrationConfig::psr_fwhm_ns is in NANOSECONDS (the VENUS FTS header convention: “folded triang FWHM 350 ns PSR”), and kernel-synthesis cost grows QUADRATICALLY with a wide fold’s width. The mechanism is NOT τ-step refinement — that applies only to folds FINER than the prompt design step, and a 50–350 µs fold’s FWHM/3 resolution floor is far coarser, leaving the step unchanged — it is the convolution itself: the ±FWHM fold-reach margin adds O(FWHM/step) τ-samples, each folded in convolve_same against a sampled triangle itself O(FWHM/step) long. Measured ~12 ms at 0.35 µs but ~1.3 s at 50 µs and ~28 s at 350 µs per single kernel-table synthesis at the default grid. A µs-as-ns unit slip (passing 350 meaning µs → interpreted as a 350 µs pin) would therefore turn a calibration into a multi-hour silent hang behind a physically fictitious fold. Any genuine width sits inside the fitted box; one decade of headroom keeps deliberate sensitivity studies possible while still catching the 1000× ns↔µs slip. 0.0 (fold disabled) is always accepted. pub for parity with the Python binding’s mirrored validation.