Code RoomBenchmark regression bisect
MediumPrep Room Coding #22

Benchmark regression bisect

Vibe & agenticAI agents & automationSenior–Staff~16 min

A benchmark regressed sometime in the last two hundred commits, and benchmark numbers are noisy. Design an agent that finds the culprit: how it searches, how it separates real regressions from noise, and what it does once it believes it has found the commit — remembering there's a person attached to every commit it blames. How would you validate the agent on regressions you've already root-caused, and what's the pitch for building it?

Implement
classify_bench_regression(baseline_runs: list[float], candidate_runs: list[float], noise_band_pct: float) → str
Examples
in[[100,101,99,100.5],[118,119,117.5],3]out"regression"
in[[200,210,195],[205,208,199],2]out"clean"
in[[98,100,102],[107,107,106],5]out"inconclusive"
What a strong answer looks like

Turn the fuzzy goal into a bounded agent task. Say what the agent reads, what it may do on its own versus draft for a human, and what it must never touch. Then earn the trust: how you’d verify it before rollout, and the pitch (the problem, what you built, why it matters).

0:00 of about 16 min

Agent build: design the agent in plain language (or narrate it). The coach grades the decomposition, the guardrails, the verification plan, and the pitch.

Which questions mattered is sealed until you submit. Telling you now would just be handing over the edge cases.