The models I am asked to look at are not necessarily wrong. Where a calculation is off or an assumption has gone stale, that is remediable — correct the calculation, recut the assumption. The ones that truly fail are different: everything checks out, yet the model cannot answer the question in front of it.

‍

That is a structural problem, and it leaves symptoms you can check for yourself.

‍

How to Diagnose a Model

‍

Four tests, none of which need a modeller:

• Ask it a question it was not designed to answer. If answering means working outside the model, it is not informing the decision.
• Move one input by ten per cent. If the lines that should respond do not, the connections were never built.
• Take an output and trace it back. If reaching the source assumption takes more than a few steps, nobody will do it under time pressure.
• Hand it to a colleague and ask them to run a scenario.

‍

If only one person can run it, you have a dependency.

‍

‍

What to Fix and in What Order

‍

The order matters more than the effort.

‍

Purpose first. Write down the decision the model has to inform, in one sentence. If nobody in the room can, that is the finding, and no amount of rebuilding will change it.

‍

Structure second. Separate assumptions from calculations, label them, and make every output traceable to its source. This is the least interesting work there is, and it returns the most.

‍

Integration last, and only where it changes an answer. Components combined at a high level are not integrated — they are adjacent. Connect what actually propagates: mining rate to processing volume to unit cost to cash flow.

‍

The instinct is to add detail. Detail is often what made the model hard to follow in the first place.

‍

When Rebuilding Is the Cheaper Option

‍

Not every model is worth repairing.

‍

If the logic cannot be followed, repair means understanding it first — often more work than starting again. If the purpose has changed, the architecture is wrong rather than the numbers, and a screening model pressed into a funding process will not survive being patched.

‍

Rebuilding is not a verdict on the original. Most of these models did exactly what they were built to do. They are now being asked to do something else.

‍

A model that has stopped informing decisions is short of purpose, structure, or both. Those are the things worth spending on.