Regression tests and settings history

Regression tests and settings history

See exactly which answers improved or got worse after a settings change or a re-crawl.

All plans (runs share the Simulation Lab allowance) Open in TR-AI5 min read

Every change that affects answers — thresholds, web search settings, assisted mode, escalation rules, a re-crawl, a history import — is recorded as a settings version with a plain summary (“Auto-send threshold 0.82 → 0.85; KB re-crawled: 2,517 → 2,690 chunks”).

  1. 1
    Create a test set
    “Create from recent tickets” picks up to 50 recent questions whose answers were sent or approved, removing near-duplicates. You can add cases by hand.
  2. 2
    Run it
    Each case is answered again in test mode (nothing is sent). The first run is your baseline.
  3. 3
    Compare
    Later runs mark each case improved, unchanged or regressed, show old and new answers side by side, and list the settings changes in between.
  • Regressed means the outcome got worse, confidence dropped by 0.05 or more, or the answer moved away from the expected one.
  • Automatic reruns after a change are on by default, except on Starter where they would use your only monthly run.
Tips & tricks
  • Create the test set when things are working well — that is the baseline you want to protect.
  • Run it before and after every big knowledge-base restructure.
  • Keep separate suites for sensitive areas (billing, security) so their results are easy to read.