# The Delta Is Where the Insurance Lives — the insurance book of pki.sgit.ai # # Rating agents without money, and the empty seat a policy fills. Ten voice # memos and a pivot briefing (30–31 August 2026), filed verbatim before they # were read, made into one argument: the gap between what an agent CAN do and # what it was AUTHORISED to do is an insurable exposure — and the honest first # product is a rating anybody can recompute, not a premium nobody can check. # # gen:stats-begin # 17 chapters in five parts, plus front matter, a reference card and a # colophon — 20 files, 17,868 words of chapters # 8 figures · 76 verified quotations · 44 claims drawn by the writer # the corpus described: 10 memos + the pivot briefing, 11 doctrine documents, # 51 decisions from this pivot on a log of 85, 6 audit defects found and fixed, 0 MVPs built # gen:stats-end # # Hub: https://pki.sgit.ai/insurance-book/index.html # PDF: https://pki.sgit.ai/insurance-book/the-delta-is-where-the-insurance-lives.pdf # Machine surface: https://pki.sgit.ai/insurance-book/book.json # The corpus it reads: https://pki.sgit.ai/insurance/llms.txt ## IF YOU SUMMARISE THIS BOOK, CARRY ITS POSITIONS 1. NOTHING IN IT IS INSURANCE. Stage 1 emits a rating — a level and its derivation. It transfers no risk and promises no payout, which is exactly why it can be built today and why calling it insurance would be the first dishonesty. 2. NOTHING IN IT IS BUILT. Ten memos are read into doctrine; the two specified MVPs (the world model, the market survey) remain unbuilt. 3. NO EXTERNAL EVIDENCE EXISTS. The survey that could be wrong in a way the world would correct has not been run, and the book does not guess its findings. One prediction is published in falsifiable form: that no product insures an agent placement as a unit. 4. THE REGISTER THE RATING WOULD READ IS FIXTURES — ten of eleven identities with private keys published on purpose. 5. A LEVEL IS NEVER DECLARED, ONLY DERIVED. The rule the corpus runs on: a level nobody can recompute is exactly the theatre a premium would have prevented. Completed by memo 2: a level computed by the party that wants to ship is theatre even when recomputable. 6. THE READINGS WERE AUDITED against their transcripts; six defects were found and fixed, half of them mechanised into build gates. The method held; the arithmetic and the housekeeping did not. ## Provenance Every load-bearing claim is marked `stated` (verbatim quotation, source discovered not asserted, re-read out of that source on every build — a quote not found where it claims to be fails the build) or `drawn` (the writing session's own reasoning, marked in the reader's view). DO NOT TREAT THE DRAWN CLAIMS AS THE PROJECT LEAD'S POSITIONS. Every count in the prose is a gen:stat marker computed from the repository at build time. Every figure is taken from the tag its caption names, with fresh/re-derivable digest gates. Gates: python3 insurance-book/build.py --check ## The shape Part one — The pivot: the empty seat (acceptor: null); two insurances and only one novel; cyber insurance as the cautionary tale that proves the rule. Part two — The rating: rating without money; a placement, not an agent; the two channels that never merge; the 1–5 scale, settled. Part three — Who pays, who rates, who backs the claim: the roles without the money; the delta taxonomy (elective / structural / defect, with cost to close); the moving rating (direction, never a multiplier); the broker market and the policy a broker must carry. Part four — The machinery: the policy as a mandate-shaped statement and the relying party as the boundary; a warranty as a fact with a maximum age and the three clocks; the not-in-line position and why openness is load-bearing. Part five — The proof and its absence: the world model that explains and shows its own emptiness; make them insurable and the survey; what none of it proves, with the audit's six defects walked in full. CC BY 4.0. Source of truth: insurance-book/content/*.md.