Skip to main content
When multiple candidates complete the same assessment, Promptster computes cohort-level statistics and percentile rankings so you can compare performance objectively.

Fetch cohort stats for an assessment

Response:
Returns null for cohortStats if no completed sessions exist for this assessment.

Percentile rankings for a specific candidate

Add ?sessionId=xxx to compute percentile rankings for a specific candidate relative to the cohort:
Response includes a percentiles object:
Percentile rankings require at least 2 completed sessions. With fewer than 2, the percentiles field is omitted.

Session-level cohort stats

You can also fetch cohort stats from a session’s perspective, without needing to know the assessment ID:
This automatically looks up the assessment the session belongs to and returns the same cohort statistics with percentile rankings for that session.

Understanding the metrics

What higher percentiles mean

Percentile values range from 0.0 to 1.0. Higher is better for all metrics:

What metrics matter most

The metrics that matter depend on the role and task, but these tend to be the strongest signals:
The most objective metric. For assessments using the OSS issue library with verified test suites, this directly measures whether the candidate fixed the bug. A testPassRate of 100 means all tests pass.
A low commandFailRate indicates the candidate writes commands carefully and catches errors. A very high rate may indicate trial-and-error debugging without understanding.
Candidates who run tests and check their work frequently tend to produce higher-quality solutions. Low verifyIntensity can indicate over-reliance on AI without validation.
How long the candidate spends understanding the problem before making changes. Too fast can indicate rushing; too slow can indicate difficulty. Compare against the cohort average for context.

Example: ranking candidates

Cohort statistics are most meaningful with 5+ candidates. With smaller cohorts, rely more on absolute metrics and artifact analysis rather than percentile rankings.