Skip to content
mentorship

Performance engineering work sample

Accelerator performance lab

Performance work starts with a verified bottleneck and ends with a measured speedup. Clever code without a trace or correctness gate is noise.

Two hours for a timed attempt, then open-endedPublic Anthropic challenge plus an original experiment worksheet

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. Clone the official public challenge and run its submission tests unchanged.
  2. Record baseline cycles and map execution units, dependencies, and scratchpad limits.
  3. Use the trace to choose one bottleneck hypothesis.
  4. Predict the effect before changing code.
  5. Measure, verify, and either keep or revert the change.
  6. Review agent changes for weakened tests or machine constraints.

Deliverables

  • A verified cycle count and unchanged official tests.
  • An experiment ledger with predictions and measurements.
  • One failed optimization and what it taught you.
  • A short explanation of simulator-to-hardware limits.

Graduation gates

01

The official verification command passes.

02

Every kept change has a measured effect.

03

Scratchpad and dependency constraints remain valid.

04

The candidate can distinguish instruction count, memory traffic, and critical-path depth.

Starter files

External practice artifact

Anthropic original performance take-home ↗

Official public challenge. Follow its current verification instructions.

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 performance rubric →

← All executable labs