KERALA RISING · FROM EXPORTING PEOPLE TO EXPORTING IDEAS fall in inference price ignored), and revenues are estimated low (cautious sign-on, conservative perunit figures, wide ranges). The conclusion is therefore deliberately pessimistic, not optimistic. The reason this is safe to do is that the gap the book relies on, between a rounding-error verification cost and budgets in the thousands of crores, is so wide that the case holds even when every number is pushed the wrong way. A model that needs optimistic inputs is fragile; this one is built to survive harsh ones. The one assumption that cannot be made conservative, that the behaviour change happens at all, is not a financial input and is tested by the pilot, not the spreadsheet. The cost side The institution’s cost has three layers. The one-time build: founding the institution and building the reusable stack, which is where the bulk of capitalisation goes. The operating cost: a small, Kerala-based core team, cloud and inference, field coordination, audit, and the awards and events. And the marginal cost of each new program: a configuration cost a fraction the size of the build, because the stack already exists. The single most important line on the cost side is also the smallest. Inference, the actual artificial 249