Closer Critique
Table of Contents
Related skills: Called by Editor in the structural pass. For series posts, enforces scoreboard non-negotiability (ties to style-guide.md's scoreboard template and each section-profile's operational rules).
Closing archetypes
| Archetype | Example | Verdict |
|---|---|---|
| Bolded maxim | "You don't have an AI problem. You have an eval problem." | PASS |
| Forward-looking question / statement | "Game 6 is tomorrow. The scoreboard will move." | PASS |
| Compressed mechanism | "The AI is not the bottleneck. The context you architect around it is." | PASS |
| Gratitude beat | "I'm grateful for it." (after a reader-correction post) | PASS |
| Scoreboard + restrained disclaimer | Running-tally block + one-paragraph financial disclaimer | PASS (and REQUIRED for series) |
| "In summary…" or "To conclude…" | prompt residue | FLAG tier-2 |
| Custom CTA | "If this resonated, subscribe!" | FLAG tier-1 |
| No close / dangling | Post ends mid-thought | FLAG tier-1 |
Scoreboard check (series posts)
If draft frontmatter has series: {slug}, the closer MUST include a scoreboard block per style-guide.md:
Running tally: P&L $NNN.NN, Brier 0.NN, W-L N-N
This week: +$N.NN or -$N.NN on {bet}.
Placed above the bolded maxim (if one exists).
Missing scoreboard = automatic tier-1 blocker. No-go on series posts until fixed.
Workflow
Evaluate closer:
- [ ] Step 1: Extract last paragraph (and bolded maxim if separate)
- [ ] Step 2: Classify archetype
- [ ] Step 3: If series, check scoreboard presence + format
- [ ] Step 4: Emit verdict + flags
Worked example
Series post closer (Kalshi Log) — missing scoreboard:
This concludes my Fed-meeting experiment. Thanks for reading.
The decision is not the news. The explanation is.
Flags:
- (Tier-1) Scoreboard missing. Series is
kalshi-log. Prior post ended with P&L +$127, Brier 0.18, W-L 4-3. This post must update. - (Tier-2) "Thanks for reading" — CTA residue. Cut.
Rewrite (scoreboard insertion above the maxim):
Running tally: P&L $134.50, Brier 0.19, W-L 5-3
This week: +$7.50 on the Fed decision not moving.
**The decision is not the news. The explanation is.**
Non-series post closer (good):
After training for two weeks, the model held. I do not know whether it would hold on a different dataset.
The context you architect around it is.
Classification: compressed mechanism + bolded maxim. PASS.
Guardrails
- Series posts without scoreboard = automatic tier-1. Non-negotiable.
- Scoreboard format follows
style-guide.mdexactly. Deviations (missing W-L, missing Brier) are tier-1. - Custom CTA ("subscribe if this resonated!") = tier-1. Substack's platform boilerplate is fine.
- Dangling close ("... and that's what I think about attention") = tier-1.
- Don't flag bolded maxims that are a full paragraph of prose — bolded-for-emphasis within a paragraph is different from a one-sentence maxim close. Accept both forms.
- The closer can be a scoreboard-only (no maxim) on series posts — acceptable if the scoreboard IS the close.
Quick reference
- Input: last paragraph + draft frontmatter.
- Output: classification + verdict + scoreboard-check result for series posts.
- Blocks: series posts without scoreboard, custom CTAs, dangling closes.