v0.5.2

v0.5.2

Accuracy release: exported files and dashboard summaries now line up with what
the AirSense 11 writes itself.

Fixed

• Daily summary records were dated one day early. The noon-to-noon day for
  each daily record was picked using clock-corrected time, which shifted every
  record back a day and pushed its session start/stop times out of the valid
  range. OSCAR treated those times as corrupt and discarded them, falling back
  to less precise values — which is why daily usage and AHI could read a minute
  (and a fraction of a point) below the SD card import.
• Session end detection. Session data now ends on the machine's own data
  gating signal, mirroring how the start is already detected. Previously the end
  relied on an event that some sessions never emit (for example when SmartStop
  triggers), in which case no trimming happened at all.
• Start/end alignment precision. Session boundaries now use the
  higher-precision timestamps available and round to the nearest sample instead
  of always rounding down. This removes a sub-second offset that slightly moved
  average leak rate and large-leak totals.
• Aborted sessions are no longer exported. A session where the mask was
  never detected and no therapy data was produced now produces no files at all,
  matching the machine's behaviour. Previously it created empty data files
  without their companion event files, adding a phantom session to the day.
• Web dashboard now matches OSCAR/SleepHQ. Graphs and summary statistics are
  limited to the machine's own therapy data window, so the pre-therapy pressure
  ramp no longer inflates average leak rate and 95th-percentile figures.
  Sessions with no therapy are excluded from the graphs and statistics.

Notes

• No changes to how raw session data is captured or stored, and no action needed
  on upgrade. Machine settings, event annotations and waveform data were already
  correct and are unchanged.
• Existing exports can be regenerated to pick up the corrected daily records.