> ## Documentation Index
> Fetch the complete documentation index at: https://docs.meerietechnology.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Best practice

> When a quantitative decision model is worth building, and how to tell a good model from a bad one.

## When to build a model

> [card] A model is worth it when
>
> * The stakes are high
> * The future is genuinely uncertain
> * There are several real options
> * The choice is hard to reverse
> * People disagree about the numbers

> [card] A model is not worth it when
>
> * The choice is cheap or easy to undo
> * There is already a clear answer
> * You have no information to base estimates on
> * The decision is already made and the model would only justify it

**The middle ground:** when you are not sure, start with a rough model of a few boxes. If the answer is clear, stop. If it is close or the stakes are high, add detail where the [tornado](https://docs.meerietechnology.com/revise/sensitivity-voi.md) says it matters.

## A good model versus a bad one

| | Good | Bad |
| - | - | - |
| **Framing** | A clear decision and real alternatives, including the current approach | One option, or a foregone conclusion |
| **Structure** | Only what drives the outcome | Too sparse to explain anything, or sprawling with boxes that do not matter |
| **Estimates** | Honest ranges backed by notes and sources | Single guesses, or false precision |
| **Dependencies** | Linked inputs are [correlated](https://docs.meerietechnology.com/estimate/correlation.md) | Everything assumed independent |
| **Win criteria** | Stated before looking at results | Chosen afterwards to fit the answer |
| **Results** | Checked with sensitivity and a sanity check | Taken at face value |
| **Revision** | Updated as you learn | Built once and left |
| **Evidence** | Sources recorded, outside evidence widens rather than narrows | Numbers with no trail |

## Habits that help

1. Frame the decision in one sentence.
2. Always keep the current approach as the baseline.
3. Set the [win rule](https://docs.meerietechnology.com/analyze/win-criteria.md) before you run.
4. Write the low end of ranges first.
5. Research only the estimates that matter.
6. Re-run after every meaningful change.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.