Crucible

---
name: crucible
description: "Invent and reshape research mechanisms through analogy, inversion, and recombination. Use while the idea is unsettled; no mandatory tournament, novelty certificate, or convergence."
argument-hint: "[the gap to turn into a method, or a method/variant to stress-test]"
metadata:
  kind: heuristic
---

<!--
Copyright 2026 IamK77 — Licensed under the Apache License, Version 2.0.
See ./LICENSE and ./NOTICE · https://www.apache.org/licenses/LICENSE-2.0
-->

# crucible

Invent and reshape research mechanisms through analogy, inversion, and recombination. Use while the idea is unsettled; no mandatory tournament, novelty certificate, or convergence.

## Starting point

A mechanism is a story about how one change could produce another. Keep several incompatible stories alive long enough to learn what each lets you notice.

Bring a question, fragment, reference, or tension. Follow whichever association becomes useful; these invitations are not a sequence to complete. Speculation should remain distinguishable from an observed fact.

## Thought experiments

### Transfer a mechanism, not its costume

Borrow feedback, competition, memory, or conservation from another domain. Name the relation being transferred and where the analogy would break.

### Invert the expensive assumption

If a method assumes perfect information, abundant compute, or stationary data, imagine the opposite. Does the objective change, or can the mechanism exploit the apparent weakness?

### Recombine at the seam

Take the representation of one idea and the update rule of another. Sketch the simplest interesting case before adding machinery. A deliberately awkward hybrid can reveal a third option.

## A possible turn

A search procedure repeatedly restarts after failure. A memory analogy suggests retaining forbidden regions; a market analogy suggests reallocating effort by opportunity cost; a developmental analogy suggests changing the representation over time. The aim is distinct causal stories, not renamed parameter settings.

## Leave the space open

Offer a few genuinely different possibilities, a sketch, an analogy, or a sharper question when useful. Do not turn every prompt into a scored diagnostic, a completion gate, or a demand for an action plan. The human may choose a direction, combine them, or leave the question open.

If an idea is chosen for a test, ledger specifies the comparison; forge executes it. A speculative diagram or an unanswered mechanism question is also a legitimate stopping point.

## Optional references

Read selectively when a thread needs another perspective or a concrete technique. The notes supply possibilities, examples, and limits; none is a prerequisite or a requirement to close the exploration.

- [Removing pieces to discover what an idea means](references/ablation-and-novelty.md)
- [A mechanism sketch and its possible limits](references/spec-and-ceiling.md)
- [Theory as a source of mechanisms and questions](references/theory.md)
- [Variants as different causal stories](references/variants-and-tournament.md)