cavendish
RejectedSignal

Agent-written test suites replace QA engineers

Could not be wrong, so could not run.

14 Apr 202628 Apr 2026 · lived 14 days

Cause

Not selected: the hypothesis as written is a headcount claim with no measurable kill condition, and the narrower pilot that could be preregistered (x-agentic-qa-pilot) was selected in its place.

What was tried, what was hit, what it means

The proposal asked for a Type 3 to show that an agent could write and maintain a regression suite well enough that a delivery pod could run without a QA engineer. It was scored in the fortnightly round and came bottom of five, with blind scores averaging 3.4 and a shadow-panel score in agreement. The scorers' notes converged on the same problem: what result would make the proposer stop?

A claim about replacing a role cannot be killed by a test suite metric; the pod would still have a QA engineer at the end of three weeks and the suite would still exist, and both sides could read that as support. The same proposer's fallback, a pilot on a regression-heavy internal codebase with mutation score and escaped-defect count as the measures, was falsifiable and was voted in as x-agentic-qa-pilot in the same round.

This is a rejection of the framing, not the field. Full credit to the proposer; the pilot is theirs and it is in measurement now. If it comes back strong, the headcount question becomes a commercial one for the practice, not a research one for the lab.

Returned notes · what the next proposer sees

Rejected 2026-04-28 because the hypothesis had no kill condition. Wait for x-agentic-qa-pilot to conclude; if mutation score and escaped defects clear the pilot's bar, propose the next step as a delivery question with a measurable outcome (for example, defect rate on a pod running without a QA engineer for a quarter), not as a replacement claim.

The difference between “someone tried this before” and “someone tried this in June, here is what they hit, here is what would need to be true now.”

Lessons

  • 01A hypothesis that both outcomes can be read as supporting is not a hypothesis.
  • 02Role-replacement claims belong to the practice; the lab tests the capability underneath them.
  • 03The fallback proposal was better than the headline one; ask for the fallback first.

Record

OutcomeRejected
BylineAWAdam Witanowski
Born14 Apr 2026
Died28 Apr 2026
Lifespan14 days
Resurrectableno

Try again?

Proposals that touch this entry get its returned notes attached automatically.