feat: add worth-it rating rubric to /coding:self-improve

[bborbe]

What

Adds a gate-then-grade rating to /coding:self-improve Step 3, mirroring how /vault-cli:reflect rates learnings (a significance filter → scored tiers).

Why

Today Step 3 is a binary "Keep-forever test" (Recurrence AND Evidence → keep or discard). Borderline candidates — the common case — have no rating, so the call is a coin-flip. This surfaced live: a session weighed a git-flag footgun candidate, wrote "borderline but it clears," and the operator had to ask "is it worth it? I'm unsure." The ambiguity was structural.

Change

• Step 3 → hard evidence gate, then a weighted worth-it score:
  • Recurrence +2, Cost +1, Generality +1, Repair +1, Obvious-anyway −2
  • Tiers: ≥3 propose · 1–2 borderline (name under Rejected) · ≤0 discard
• Step 6 → per-proposal Worth-it: <score> line.

Under the rubric the triggering candidate scores 2 → borderline → don't propose, matching the conclusion reached manually.

Scope

Prose-only edit to the command definition + CHANGELOG ## Unreleased bullet. No code; /coding:pr-review has no code surface to review.

[ben-s-pull-request-reviewer]

Selector Mode Review — in-session

Files changed: CHANGELOG.md, commands/self-improve.md

---

Concern 1: Rubric scoring arithmetic ✓

The weights and thresholds are internally consistent:

Signal	Weight	Max contribution
Recurrence	+2	+2
Cost	+1	+1
Generality	+1	+1
Repair	+1	+1
Obvious-anyway	−2	−2

Max positive score = 2+1+1+1 = +5. The ≥3 "propose" threshold means you need ~60% of available signal. The Obvious-anyway −2 penalty can flip a borderline +1 or +2 into discard territory — which is the intended effect ("kills 'just do it in the natural order' non-improvements").

Tiers verified:

• ≥ 3 (propose): minimum viable signal — e.g. Recurrence+Cost = 3, or Recurrence+Generality+Repair = 4
• 1–2 (borderline): low signal, named but not proposed
• ≤ 0 (discard): Obvious-anyway alone clears the −2 penalty, no proposal

Verdict: Consistent.

---

Concern 2: Step 6 worth-it score display ✓

Step 6 output format (lines 88–92) correctly integrates Worth-it: <score> per proposal:

Per proposal:
- **Change:** what, and where (exact target file / artifact)
- **Worth-it:** <score> (e.g. recurrence +2, cost +1)
- **Evidence:** verbatim quote(s) + how often

The example in the rubric (Worth-it: <score> with component breakdown) matches the description in the CHANGELOG ("surfaces a Worth-it: score per proposal").

Verdict: Correctly integrated.

---

Concern 3: CHANGELOG.md unreleased section ✓

## Unreleased exists (line 11) and contains the expected bullet (line 13):

feat: add a worth-it rating rubric to /coding:self-improve Step 3 — gate-then-grade …

Verdict: Present and correct.

---

Selector Mode Traceability

Rule ID	Doc	Status
—	No applicable judgment rules for commands/ file changes	N/A (cmd file, not code)

No ast-grep rules target commands/*.md. This is a pure docs change — mechanical funnel returned zero findings (expected).

---

Final Verdict

No issues found. The PR correctly adds the worth-it rating rubric to Step 3 and displays it in Step 6 output.

{
  "verdict": "approve",
  "summary": "Docs-only change adds a correctly-arithmetic worth-it rating rubric to /coding:self-improve Step 3 and integrates the Worth-it score display into Step 6 output format. CHANGELOG Unreleased entry is present and accurate. No issues identified.",
  "comments": [],
  "concerns_addressed": [
    "correctness: rubric scoring arithmetic verified — weights (+2/+1/+1/+1/−2) and tiers (≥3 propose, 1-2 borderline, ≤0 discard) are internally consistent with max score +5",
    "correctness: Step 6 worth-it score display correctly integrated — per-proposal 'Worth-it: <score>' line present in output format",
    "correctness: CHANGELOG.md ## Unreleased section exists with the expected feat bullet for this change"
  ]
}