find-simplifying-insight

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

1. Separate observations from machinery. Inventory measured facts and invariants separately from inherited rules, exceptions, adapters, conventions, and workarounds.
2. 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.
3. Form alternatives. Propose at most three simpler models plus the null hypothesis that the current model is necessary.
4. 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.
5. Run or specify the decisive test. Prefer executable probes, counterexamples, historical cases, or small spikes over argument from taste.
6. 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>