| targets | ["*"] |
| name | find-simplifying-insight |
| description | Use when proliferating rules, exceptions, adapters, or workarounds suggest the underlying model may be wrong and a falsifiable simpler model is needed. |
| codexcli | {"interface":{"display_name":"Find Simplifying Insight","short_description":"Test simpler models against the evidence","default_prompt":"Use $find-simplifying-insight to test whether one simpler model can replace these rules and exceptions."}} |
Find Simplifying Insight
Look for the equivalent of a heliocentric shift: one model that explains the observations while making many inherited rules unnecessary. Treat elegance as a hypothesis, never as evidence, and do not force a revolution when the current model survives testing.
Method
- Separate observations from machinery. Inventory measured facts and invariants separately from inherited rules, exceptions, adapters, conventions, and workarounds.
- Expose the load-bearing assumptions. For each cluster of compensating rules, state the assumption they protect. Identify contradictions, repeated translations, and rules whose only justification is another rule.
- Form alternatives. Propose at most three simpler models plus the null hypothesis that the current model is necessary.
- Make each model falsifiable. For every candidate, state:
- facts and invariants it preserves;
- rules or exceptions it eliminates;
- a novel prediction or observable consequence;
- the cheapest decisive test that could disprove it.
- Run or specify the decisive test. Prefer executable probes, counterexamples, historical cases, or small spikes over argument from taste.
- Choose a verdict. Use exactly one:
adopt, spike, reject, or no simplifying insight.
Guardrails
- A shorter explanation that loses facts is not simpler; it is incomplete.
- Moving complexity behind a new abstraction does not eliminate it.
- Count migration cost, operational risk, and new failure modes.
- Preserve the null hypothesis throughout the comparison.
- If no candidate survives falsification, report that plainly and keep the existing model.
Output contract
Observed facts:
Current assumptions and compensating rules:
Candidate: <model or null hypothesis>
Preserves:
Eliminates:
Predicts:
Falsification test:
Result/evidence:
Verdict: adopt | spike | reject | no simplifying insight
Why:
Next test: <only when verdict is spike>
