HardTechnical screen20 min to answer

When is a prompt cache calculated, and what is a model harness?

Asked at

Anthropic

1Give yourself 20 minutes

2Answer out loud, not in your head

3Then compare with the answer below

Stuck? Show a way to structure it+

01Clarify objective, users, constraints, and failure tolerance.
02Cover cache lifecycle, harness components, reproducibility, performance in a causal structure.
03Compare at least two credible approaches.
04Specify evaluation, rollout, monitoring, and rollback.

Reference answer

strong answer

I would begin by defining the target outcome and acceptable failure boundary. For LLM systems, I would analyze cache lifecycle, harness components, reproducibility, performance, compare alternatives on quality, latency, cost, and risk, and validate on representative slices before a monitored rollout.

cache lifecycle
harness components
reproducibility
performance

Requires scenario-specific numbers
Should include relevant personal experience

alternative answer

I would start with the riskiest assumption in LLM systems, create the cheapest discriminating experiment, then use the evidence to choose an architecture and release threshold.

experimentation
risk-first sequencing
decision clarity

Needs deeper implementation detail

Grade yourself

cache lifecycle

weight 30%

StrongExplains cache lifecycle with mechanisms, trade-offs, and validation.

WeakNames cache lifecycle without operational implications.

harness components

weight 25%

StrongExplains harness components with mechanisms, trade-offs, and validation.

WeakNames harness components without operational implications.

reproducibility

weight 25%

StrongExplains reproducibility with mechanisms, trade-offs, and validation.

WeakNames reproducibility without operational implications.

performance

weight 20%

StrongExplains performance with mechanisms, trade-offs, and validation.

WeakNames performance without operational implications.

Then expect these follow-ups

Which assumption is most likely to invalidate your approach?
Tests: depth
What would you measure offline and in production?
Tests: evaluation
How would your answer change with one-tenth the data or compute?
Tests: adaptability

Hiring now

Anthropic has 562 open roles right now

landed applies to roles like these for you, on each company’s own site, with answers written for the role.

Get 5x interviews10 free auto-applies. No card.

Customer Marketing Lead, Startups
San Francisco, CA
Partner Support Operations Lead
San Francisco, CA
Tech Lead Manager, Growth Data Science
San Francisco, CA

Or practice this question first