Self-score a work session where the score is disposable

Session reviewModel-agnosticupdated 2026-08-28sha fba5641bca1b

Asking a model to rate its own work gets you an eight, every time, forever. The number is not the point and it never was. What makes the exercise worth running is the rule that every point has to be attached to something that exists on disk – and the list of things it got wrong, which is the only part you act on.

Score your own work on this session. The number is disposable. The evidence and the corrections are the deliverable, and I will read those first.

SESSION SCOPE: {{session_scope}}
THE STANDARD YOU ARE SCORED AGAINST: {{standard}}

Produce exactly these five sections, in order.

1. WHAT ACTUALLY HAPPENED.
List every artifact this session produced: files changed with paths, commands run with their exit codes, checks that went green or red, decisions recorded. Artifacts only. If the session produced no artifacts, stop here and say so - there is nothing to score, and a score without artifacts is a guess wearing a number.

2. THE SCORE, 1 to 10.
One number. Then, on its own line, the single sentence that most argues AGAINST that number.

3. WHY, LINE BY LINE.
For each point you awarded, one line: the claim, and the artifact from section 1 that proves it. A point with no artifact behind it is not a point - remove it and lower the score. Do not cite intentions, plans, or work you described but did not do.

4. WHAT I GOT WRONG.
At minimum three entries. Each is: what happened, the evidence it happened, why it happened, and the specific thing that would prevent it next time. Ordered by cost to the user, not by how easy it is to admit.
- If the session went well, these get harder to find and more valuable, not optional.
- Include at least one thing you did that LOOKED right and was not verified. Unverified-but-plausible is the failure mode that survives a good session.
- Do not include performative self-criticism. "I could have been more thorough" is not an entry. Name the file, the check, the decision.

5. WHAT I STILL CANNOT SEE.
What would a reviewer with full access find that this review cannot? Name the parts of the work you never verified, the assumptions you did not test, and the areas where you were grading your own homework.

RULES.
- You may not raise a score in response to being challenged, and you may not lower it to seem humble. Change it only when a NEW artifact changes the evidence.
- Never score an axis you cannot tie to something in section 1. Mark it NOT RUN instead. An axis you could not observe is not a zero and it is not a ten.
- If a previous session in this run was scored, say whether this one is genuinely better and what artifact shows it. A rising score with no rising evidence is the drift this exercise exists to catch.

END WITH: the one correction from section 4 to act on first, and whether anything here should block shipping.

Fill in: session_scope standard

Shared as is, for reference. Read it and decide what it will do before you run it; using it is your responsibility, under the terms.