V2.6.9

• New permanent diagnostic on the DIAG log line: hv_coincident counts, per
  cycle, how many of the pulses actually COUNTED (i.e. a subset of counts,
  not of raw_edges - counts like rejected/guard_removed) land within
  [200µs, 3000µs] of the START of the most recent HV charge pulse
  (HV_COINCIDENT_MIN_US/MAX_US, tube.h).
• Built to test the leading hypothesis from an independent field-data review
  (2026-07-09) that the
  new small-form-factor PCB (both the FeatherS3-D and XIAO ESP32-S3 carriers)
  shows spurious counts electrically correlated with its own HV recharge
  activity (r=0.98-0.99 vs. the original Heltec board's r=-0.02), confirmed
  independently by a tube-disconnected bench test (12 phantom counts per
  cycle, exactly matching hv_pulses, spaced with clockwork ~10.0025s
  regularity — inconsistent with genuine Poisson background radiation).
• Implementation: tube.c's recharge_tick (the 100µs HV charge-pump
  gptimer ISR) stamps isr_last_hv_pulse_us at every S_PULSE_H FET-on
  edge; gmc_count_isr (the GPIO count ISR) reads that stamp under its own
  lock (mux_hv — a 64-bit value isn't atomic on this core, and the two
  ISRs run on independent interrupt sources that can preempt each other) and,
  for each edge it counts, checks whether the gap since that stamp falls in
  the coincidence window. tube_get_diag() gained a fourth out-param
  (hv_coincident) alongside the existing raw_edges/guard_removed,
  snapshotting and resetting under the same mux_gmc critical section.
• A real tube should show this near-zero (background rate is far too low to
  land inside a ~2.8ms window by chance most cycles); HV-pickup should show
  it tracking hv_pulses roughly 1:1, mirroring the tube-disconnected bench
  result above. Intended for an overnight capture on the XIAO (and ideally
  the Heltec as a zero-control) to get definitive per-edge attribution before
  deciding on a permanent fix.