The framework

How the work gets scored.

Every candidate is measured against a rubric written for the role — scored criterion by criterion by an AI that shows its evidence, then a person makes the call. No trivia, no black box, no automatic verdicts.

01 — Authored

A rubric for the role

Each scenario carries a weighted list of criteria written for that role — not a fixed, one-size trivia bank.

02 — Scored on evidence

Every score cites the transcript

An AI judge scores each criterion 0–100 with a written reason pointing back to what the candidate actually did.

03 — Decided by a person

A human makes the call

Scores are an input, not a verdict. A named reviewer marks Pass, Maybe, or No — Can You Build? never decides on its own.

The rubric

Written for the role, weighted on purpose.

When you author a campaign, the scenario comes with a set of criteria and weights that match what the job actually needs. Before the timer starts, the candidate is told the criteria they'll be scored on — the bar is explicit, not a secret. (The weighting behind them is yours; we author weights to add up to 100%, and they're always applied relative to each other.)

And they're fixed before the first candidate starts. The rubric is frozen onto each candidate's run, and there is no way to re-weight it afterwards until a preferred candidate rises to the top — that isn't evidence, it's picking a winner and working backwards.

Example — a "Deep debug" scenario for a backend hire:

Deep debug · evaluation criteria weight
Root cause identification 30%

Did they find the actual problem at each layer?

Investigation strategy 30%

Did they ask smart, targeted questions of the AI?

Fix quality 25%

Are the proposed fixes correct and production-ready?

Prompt efficiency 15%

How directly did they get there — signal over thrash?

The score

A number you can trace back to a sentence.

Each criterion is scored 0–100 with a short written reason for that score. The overall is a weighted aggregate across criteria and stages — and it reports its coverage, so a partial run never masquerades as a complete one. The reasons sit next to the full transcript, so you can check the judge's work against what the candidate actually did.

cand·7QF2 · scored report 88
Root cause identification92

Traced the timeout to the connection pool, not the query — confirmed with a targeted follow-up.

Investigation strategy85
Fix quality86
Tied to the full transcript · coverage 100%
The verdict

The call is a person's. Always.

A named reviewer reads the report and the transcript, then records a decision. The score informs it; it never makes it. Can You Build? never accepts, rejects, ranks, or shortlists a candidate on its own — and the rubric, scores, reasons, decision and notes export as one auditable trail.

✓  PASS ⚠  MAYBE ✗  NO ★  SHORTLIST

Reviewer decisions — the candidate is never told a score, and no verdict is automatic.

See the framework in motion.

Walk through one candidate, from the challenge to the scored report.