proof-in-miniature

name	proof-in-miniature
description	Pairs strong claims with the smallest falsifiable proof (script, gist, notebook, or runbook) and labels bare-bones baselines clearly. Use when arguing that a capability crossed a threshold or when credibility must fit in one link.

Proof in Miniature

Distilled pattern: pair a strong claim with the smallest artifact that makes the claim falsifiable (code, experiment log, or reproducible runbook).

When to use

• You are arguing that a capability crossed a threshold (“suddenly usable,” “good enough to loop”).
• You want credibility without a startup demo or a paper.

Steps

1. State the threshold in one sentence — What became true that was not true last quarter?
2. Shrink the proof — Prefer lines-of-code count, wall-clock duration, hardware budget, and a public entrypoint. Vessel options (pick one primary): single file or gist, minimal repo tag, notebook/Colab duplicate of the same source.
3. Show one loop — If the idea is “agentic,” show initialize → act → observe → revise once. One lap beats a architecture diagram.
4. Say what is still brittle — One paragraph on failure modes preserves trust.
5. Label the baseline — If the artifact is meant to be forked into serious systems, say “intentionally bare bones” (or similar) so nobody mistakes minimal for production-complete.

Anti-patterns

• Screenshots of vibes.
• Huge repos where the insight is buried.
• “Trust me” without a command someone else can run.

Output shape

• Claim.
• Minimal reproduction (command block or file pointer).
• One paragraph: what it proves / what it does not prove.