The Three Tracks

ReadyCheck captures three tracks simultaneously during every session. Each track provides a different lens on what happened, and together they give a complete picture.

Screen Track

What it captures: Frame-by-frame screen recording of your app.

Why it matters: Screenshots are evidence. When a user says “the button disappeared,” the screen track shows exactly what was (or wasn’t) visible at that moment.

Access:

ada query @latest screenshot --time 15.5 --output /tmp/screen.png

Voice Track

What it captures: Microphone audio, automatically transcribed to text.

Why it matters: Voice is how users naturally express intent. Instead of writing detailed bug reports, users simply narrate: “When I click save, nothing happens.” ReadyCheck’s analysis starts from these natural descriptions.

Access:

ada query @latest transcribe segments --format json

The voice track is the primary input for the /analyze skill. Intent extraction identifies issues, severity, and time windows from the transcript.

Function Trace Track

What it captures: Every function call in the target process, with sub-microsecond timestamps.

Why it matters: This is ground truth. No matter what the user saw or said, the trace shows exactly what the code did. Combined with voice intent and screen evidence, you get deterministic debugging.

Access:

ada query @latest events --since-ns <start> --until-ns <end> --limit 100

Cross-Track Correlation

The real power is in combining all three:

  1. Voice tells you what the user intended (“I expected the list to update”)
  2. Screen shows you what actually appeared (the list didn’t change)
  3. Traces reveal why (the updateList() function was never called after the API response)

This is deterministic debugging — every issue has an answer because you have the complete picture.

Track Independence

Each track is captured independently. If one track isn’t available (e.g., no microphone access), the other two still work. The /analyze skill adapts:

  • No voice → switches to trace-first analysis, scanning for anomalies
  • No screen → continues with voice + traces, notes that visual evidence is unavailable
  • No traces → suggests re-running capture with the correct target