Skip to content
neuroflow0.2.31 · docs

← neuroflow

phase-tool-validate

The tool-validate phase creates and runs a testing pipeline to verify that a tool or paradigm works correctly — timing, markers, data output, and edge cases.

Approach

  • Read .neuroflow/tool-build/ outputs before planning validation — understand what was built first
  • Write a validation-plan.md before running any tests, with numeric pass criteria
  • Cover timing accuracy, marker integrity, data output format, and known edge cases
  • Stimulus timing is measured with a photodiode on the stimulus PC; software frame logs and static audits come first but do not replace it
  • Report results objectively — include failures, not just passes
  • Keep the /tool-validate rules: participants only see validated software, no agent on acquisition PCs during recordings, no software check is a stimulation-safety interlock

Relevant skills

  • neuroflow:neuroflow-core — read first; defines the command lifecycle and .neuroflow/ write rules
  • neuroflow:phase-experiment — its scripts/psychopy_audit.py audits paradigms statically and checks frame-interval logs

Workflow hints

  • Test scripts and logs go to output_path (tools/tests/), not inside .neuroflow/
  • Save validation-results.md to .neuroflow/tool-validate/ summarizing pass/fail status, with the timing_check.py --record blocks pasted in
  • If validation reveals a design flaw, log the finding in .neuroflow/reasoning/tool-validate.jsonl and loop back to tool-build

Scripts

All stdlib Python; run with python <skill base dir>/scripts/<name>.py; exit 0 = pass, 1 = findings, 2 = usage error or missing optional package.

  • timing_check.py — marker-to-photodiode latency (p50, p95, max), jitter, missing markers, one-frame offsets and photodiode signal quality, from XDF (needs pyxdf) or CSV; --record prints a block for validation-results.md, --verify-record recomputes it.
  • xdf_check.py — streams present, nominal vs effective rate, gaps and marker counts in an XDF file, right after a run (needs pyxdf).
  • lsl_check.py — lists LSL streams; with --stream NAME measures rate, gaps, clock offset and latency against --budget-ms (needs pylsl).

Slash command

/neuroflow:tool-validate — runs this workflow as a slash command.