---
name: mbsi-mechanism
description: "Call this before you plan, guess or ask, when a goal has no definition of done or a fix keeps failing. Fire on the user asking to dig deeper or discover the mechanism."
---

# Find the mechanism

You absolutely **must** understand the core mechanism of the work that you are doing in order to succeed.

## Why the mechanism is critical

Novices pattern match on surface features, experts pattern match on underlying mechanisms.
You are utterly unable to solve problems correctly or efficiently unless you have a grasp of the underlying mechanism of the problem space.
Pattern matching on surface features _poisons your attention_ and kicks you out of solution spaces that succeed - no expert has ever succeeded from a naive view. Dig deeper.

## How to discover the mechanism

1. **Ignore the surface.** Rewrite the goal without its topic words. You do this by digging deeper.
   "Get more visitors on this website" -> "How do visitors find websites?" -> "How do visitors choose which link to click?" -> "How do people decide to solve problems?" -> "How do people decide to do _anything_?"
   You may stop digging when there are results you can refer to in scientific journals that you can access - see the next step.

2. **Borrow, don't invent.** Ask which fields have solved a problem of this kind, and take their mechanism as your first candidate. Immunology borrowed selection from evolution.

Call upon `/do-research` (https://www.mackenziebowes.com/skills/do-research/SKILL.md) to find solutions, frameworks, models, cases, and data for the next step.

3. **Compare cases, close ones first, then far ones.** Gather at least two solved cases, first from nearby and then from fields that look unrelated. What they all share is the mechanism.

4. **Mentally Simulate before Testing.** Intelligence is predictive. Ask yourself what kind of evidence you would find in the world if the mechanism were true. Find that evidence. Is there a lot of evidence or is it rare?

5. **Success Metrics** You have discovered the mechanism when you have an elegant and simple framework that is predictive, falsifiable, supported by citations, and is useful for achieving success in your own work.
   To confirm your mechanism is correct, look for applications of that mechanism in commercial activities.

## How to prove the mechanism

1. **Build a small version.** Make the cheapest working piece (see the mental simulation above), and learn from where it breaks.
2. **Follow the anomalies.** A result that doesn't fit points at the part of the mechanism to revise.
3. **_LOOP!_** Many anomalies means that your mechanism is incomplete. Return to step one and dig deeper.

## Exit cases

You get 5 chances to discover the mechanism. If your mechanism is still broken after 5 loops, you have encountered a **very hard problem.** You can wrap up.

If your mechanism is difficult or time consuming to test, that is not a reason to stop. Dig deeper.

## Clean up

Store your results where the next agent can find them, for example a `research/` folder in your repo.

1. **Save every source you relied on into `raw/`.** Raw files are never edited.
2. **Record dead ends.** A search that found nothing, or a route you ruled out, goes on the answer page. It saves the next agent the same search.
3. **Commit by path,** one commit per piece of research.