AI-Native Insurance: Hands-On Analysis & Design: The Contract Core — Five Gates, One Record, and the End of the Four-Day Answer (The AI-Native Architect for Insurers Book 1)

Three questions. Four days. Five systems. One of them unanswerable. On 8 September 2026, after a flood puts 1,240 claims on Northgate General Insurance in one night, a policyholder writes a short letter. Why did my premium rise 40%? Which of your two reserve figures is real? When you accepted my warehouse, did you know how much you already had in that zone? The company has the answers somewhere — in a policy system, a rate spreadsheet, a month-end ledger, a claims system and an adjuster's inbox — and it takes four days to assemble two of them. The third was never recorded. The Contract Core is the specification for a company that answers all three in a minute. It rebuilds the insurer around five gates every contract passes through: Select — a scored, tagged decision that refuses to guess when a fact is missing Price — a technical premium and a floor computed from a signed, versioned rate book, never a spreadsheet Capacity — accumulation checked at the second of quoting, not at month-end Reserve — versioned estimates on two clocks, so that two different figures can both be right Pay — a decision per line item against the policy's sections, so an uncovered item cannot be paid by habit Each gate writes the same eight-field record: what was decided, from which facts, under which parameter version, for what coded reason, by whom, when. Written once, never edited. From those records the book derives everything else: the reinsurance recovery decided months before the claim, the leakage that becomes a count instead of an estimate, the override rate that tells you whether your rules are tuned, and the pricing–selection spiral that a loss ratio hides and a record reveals. Inside The inverted production cycle, five tensions, seven design errors, and six commitments — each with a test for "done", a baseline and an owner Seven playbooks that change numbers and never structure: motor, accumulating property, health, liability, commercial accounts, renewals with history, suspicious claims A case space of 25,920 combinations reduced by eight theorems to 1,637 cells — every one constructed and read back by the printed code, every impossible one refused Four boundaries: three lanes for models, people and the core; one door for parameters; two stores for records and identity; replay at three levels The complete source code — sixteen modules, 148 tests, one demo — printed in Appendix B and runnable on a laptop Every chapter opens with objectives and closes with a ten-question quiz, answers on separate pages. Every term is defined where it first appears, with a worked number. No metaphors; direct language; nothing assumed. For underwriters, actuaries, claims leaders, architects and boards who want to know exactly what "AI-native" has to mean before anyone buys a model. Parameters are illustrative until signed; the method is what you are buying. Book One of two. Book Two, Building It, takes this specification into a company: architecture, three rollout rounds, model lanes, migration, the desks, the data products and the committed numbers.