Skip to content
mentorship

Research work sample

Black-box model behavior lab

The goal is not prompt volume. The goal is the smallest experiment that separates your leading explanations.

90 minutesQuery-only behavior API, experiment template, five-minute readout

Do this only if the format matches

Confirm the round with recruiting first. This is an original practice exercise, not a claim that a specific company uses this exact prompt.

Protocol

  1. Run one control and record the exact observation without interpretation.
  2. Write three competing hypotheses that predict different results.
  3. Change one factor at a time before testing interactions.
  4. Use controls or repeats where variance could produce a false pattern.
  5. Find one boundary where the leading claim stops holding.
  6. Present the claim, evidence, uncertainty, and next discriminating experiment in five minutes.

Deliverables

  • A probe table with predictions and observations.
  • One interaction effect and one negative result.
  • A bounded behavioral claim, not an architecture guess.
  • A five-minute research readout.

Graduation gates

01

Every claim has a probe that could have falsified it.

02

Observation and interpretation remain separate.

03

The readout exposes uncertainty and boundary conditions.

04

The next experiment is chosen by expected information gain.

Starter files

Answer key is a rubric, not a solution dump

The linked question explains what strong performance looks like, common failure modes, and senior follow-ups. Attempt the lab before reading it.

Read the research rubric →

← All executable labs