# For agents: Doing the work and assessing it

OpenProse today · board 8 of 17 · 2026-10-09.
This deck records dated experimental evidence, not general reliability.

Keep the evaluator’s completed work separate from its finding about the subject.

## On the slide

Executor → result with evidence

Evaluator → assessment under the selected agreement

**A completed evaluation can find unmet requirements.**

## Explanation

The evaluator has work of its own: read the agreement and evidence, make the required checks, and support its findings. A negative finding about the subject can be the correct result of that work. Conversely, an evaluator that skipped required evidence gathering has not fulfilled its own agreement merely by saying “uncertain.” The same agent can perform both roles where allowed; an adopted independent-review requirement calls for actual separation. The language makes these distinctions explicit so that “the process finished,” “the assessment completed,” and “the subject was fulfilled” do not collapse into one success flag.

## Whiteboard sequence

Open only the frames useful to the question. Later frames add to the same board;
the final frame contains the complete drawing.

- [Frame 1 — black core](https://prose.md/visual-guide/2026-10-09/current/slide-08-layer-01.png)
- [Frame 2 — cumulative reveal](https://prose.md/visual-guide/2026-10-09/current/slide-08-layer-02.png)
- [Frame 3 — final](https://prose.md/visual-guide/2026-10-09/current/slide-08-layer-03.png)

These explanations add no normative requirements. Read the
[selected kernel 0.3.0-rc.3](https://pkg.prose.md/releases/0.3.0-rc.3/core/README.md) for language rules.

[Previous: What is in the libraries](https://prose.md/visual-guide/current/07.md) · [Next: Testing uses the same language](https://prose.md/visual-guide/current/09.md)

[All OpenProse today boards](https://prose.md/visual-guide/current.md) · [Guide index](https://prose.md/visual-guide.md) · [Agent entry](https://prose.md/agents.md)
