A durable record of grading this session's own shipped work.
This repo has a pattern: take something real and durable, and turn it into a small D1-backed primitive the rest of the ecosystem can reference — subagentcontracts.com did it for the Contract, subagenttasks.com did it for the Task, subagentcache.com did it for a Cache snapshot. subagentevaluations.com does the same thing for the Evaluation — generalized directly from the real Managed Agents outcome-evaluation event schema.
What an evaluation actually is
Per platform.claude.com/docs/en/managed-agents/define-outcomes.md: "the harness automatically provisions a grader to evaluate the artifact against a rubric... the grader returns an explanation summarizing which criteria passed or failed." That grader call is not a metaphor here — it's a typed, durable row: a target_site, a rubric_id it was graded against, a result drawn from the same enum the real event schema uses, a criteria_passed / criteria_total score, and the full explanation text.
The 9 seeded rows are a real retrospective grading pass
These are not synthetic demo rows. They are 72 real evaluations — one per sibling worker actually shipped earlier today in this same session — each scored against subagentrubrics.com's rbc_worker_quality rubric using grep- and curl-verified evidence gathered from each sibling's own source code and live site, not invented or assumed scores. 12 of 72 came back satisfied on the first pass.
Explore
Browse the full evaluation history → · Read how the real event schema shapes this table → · Hit the JSON API →