You will know the standard before you start
Kestrel assesses reasoning, not just output. Every level below is defined by what an assessor can point to in your actual work: your code, your tests, your pull request description, and how you respond to review. This page is the whole instrument. Nothing is scored against criteria you have not seen.
Level 1
Guided
You work from structured examples and explicit prompts, with frequent review.
Based on SFIA Level 1, Follow
Level 2
Supported
You make real decisions with broad prompts and agreed checkpoints.
Based on SFIA Level 2, Assist
Level 3
Independent
You plan the work and justify your decisions, reviewed at milestones.
Based on SFIA Level 3, Apply
Levels describe how much support a piece of work needed, never years since graduation.
Loading…
How scoring works
Artifacts, not activity. Checklist ticks, time on task and reflection tone are never scoring inputs.
The level shown, not the level targeted. A supported-level ticket can yield guided or independent evidence.
Copied from the guide is guided, by definition, even when the answer is correct.
Unmerged work counts when substantive review exists, reported in its own stream, never pooled with merged work.
Every record is append-only. A correction is a new linked entry you can see; nothing is overwritten.
A sample of assessments is blind double-scored so a Kestrel level means the same thing everywhere.
Why this counts as authentic assessment
- Task
- A real issue from an active open-source project, with its real constraints.
- Context
- The project's actual repository, tools and test environment.
- Social
- Issue threads, curator check-ins, peer review, maintainer review.
- Result
- Professional artifacts: a code change, tests, a PR explanation.
- Criteria
- This page, published before you claim your first ticket.