This pattern describes what happens when the tool consumes the project. It sits alongside No Fat Modelling — the discipline that prevents this from taking hold — and connects to Science Waits Its Turn, where the model has not waited its turn but taken over entirely. The structural dynamic mirrors Relief Crowds Out Prevention: the visible, measurable activity starves the less visible but more consequential one. The Crooked Incentives of Project Financing explains why this keeps happening — modelling is fundable and deliverable in ways that recommendations and governance design are not.

Modelling and data collection expand to consume the resources available to them. What begins as a tool for answering questions becomes an objective in itself, and the questions it was built to answer remain open.

The sequence is familiar. A water management challenge arrives — a flood problem, a groundwater crisis, a delta under stress. The first response is to reach for data and models. More data will reduce uncertainty. A better model will reveal the answer. The proposal is written, the budget allocated, and the modelling effort begins.

Then it expands. It always expands. Data gaps are discovered and must be filled. The model domain needs extending. Calibration requires more measurements. The technical team is talented and conscientious and the work is genuinely difficult. Before long, eighty or ninety percent of the budget and timeline has been consumed by the effort to describe the problem with sufficient precision.

What remains is almost nothing. The recommendations — the actual reason the project existed — are written in the last few weeks by an exhausted team working from a model that has grown so complex its outputs are difficult to interpret and impossible to explain to the people who need to act on them. The interventions are sketched rather than designed. The governance questions are deferred. Community engagement happens, if at all, as a final consultation rather than a shaping force throughout.

The model's completion is reported as a project achievement. The questions it was built to answer remain open.

This is not a story about incompetence. The modellers are doing exactly what their incentive structure rewards. A completed, calibrated, documented model is a defensible deliverable. Recommendations are judgements, and judgements can be contested. Governance design is messy and slow. The financing structure does not distinguish between a model that answered the question and a model that replaced it.

There is a further irony. The knowledge that would actually ground the recommendations — the farmer who knows this field, the operator who knows this canal — was never in the model to begin with. See On His Land. And a model that has eaten the project has no remaining resources to check its outputs against what the river is actually doing. See When the River Disagrees.

Set the budget for recommendations, interventions, and governance design before the modelling budget is fixed, not after. Treat the model as a means and hold it to that role throughout — explicitly limiting its scope to what is needed to answer the questions the project exists to answer. When the model begins to expand, ask not whether the expansion is technically justified but whether it serves the purpose the project was commissioned for.

Connected patterns: No Fat Modelling — Science Waits Its Turn — The Crooked Incentives of Project Financing — Relief Crowds Out Prevention — When the River Disagrees — On His Land