[stevevanhooser]

Why

ndi_daq_system_mfdaq.readchannels_epochsamples read a raw epoch one sample later than the same epoch ingested.

In this port, sample positions are 0-based: epochtimes2samples, epochsamples2times and the ingested reader all use them (see docs/developer_notes/ndi_xlang_principles.md, NDI-python#313). A raw reader's readchannels_epochsamples, though, takes 1-based samples, as in MATLAB. The DAQ system passed its numbers to both readers unchanged.

For the example Intan session, readchannels_epochsamples("analog_in", [1], 1, 1, 5):

Epoch	Values
raw	[0, -15.015, -10.92, -30.81, -22.425]
ingested	[-15.015, -10.92, -30.81, -22.425, -12.48]

The probe layer is affected too. ndi_probe_timeseries_mfdaq.readtimeseriesepoch turns times into 0-based samples with epochtimes2samples, reads the data through the DAQ system, and computes the returned times with epochsamples2times. On a raw epoch, the data were therefore one sample off from their own time vector. Ingested epochs were right.

What

• ndi_daq_system_mfdaq.readchannels_epochsamples adds 1 to s0/s1 before calling a raw reader. The ingested path is unchanged.
• Docstrings that called these samples 1-indexed are corrected: ndi_daq_system_mfdaq.readchannels_epochsamples and ndi_probe_timeseries_mfdaq.read_epochsamples.
• The system's bridge entry for readchannels_epochsamples is corrected the same way.
• CHANGELOG entry under Fixed.

The raw readers' own 1-based readchannels_epochsamples is not changed: their bridge entries record it, tests pin it, and ingestion depends on it. Only the DAQ system, which callers use, now converts between the two.

Testing

• New tests/test_daq_system_sample_indexing.py, on the example Intan session:
  • sample 0 through the DAQ system is the raw reader's first sample;
  • the raw time channel at samples 10..19 matches epochsamples2times(10..19), the pairing a probe returns;
  • raw and ingested reads agree, for both epochs, for ai and time, at the start and mid-epoch.
• All three tests fail on main and pass with the fix.
• Full suite locally: 6528 passed, 352 skipped.
• Lint (black, ruff) is clean.

---

Generated by Claude Code