Capstone — modernise a
Galileo-class platform
Everything converges here. You are the architect advising an insurer whose custom core — ledger-first, event-driven, production-proven — is mid-replacement. Propose what the next platform should keep, drop, and add.
§1The scenario
Client: a PH life insurer running a Galileo-class core: C# microservices, commands through a 3-node private blockchain audit ledger, Kafka to MongoDB read models, ~1.1 M policies, distribution via super-apps (GCash-style), currently migrating to a COTS core (eBao-style) with an ESB translation layer and an ID-mapping table for coexistence.
Board asks: "As we replatform, which ledger capabilities do we keep or rebuild, what new products do they enable, and how do we make our coming AI claims automation regulator-proof? Give us a 3-year architecture."
§2Deliverables (the pack)
| # | Artifact | Contents · draws on |
|---|---|---|
| D1 | Target architecture (1–2 diagrams + narrative) | Core platform, audit-ledger decision (keep chain / signed log / ledger DB — argue it), event backbone, AI layer with anchored decisions, payment rails. · M01, 05, 07, 09–11 |
| D2 | Platform selection | Scored matrix (M03 worksheet) for the ledger + any consortium/public legs; one page defending the weights, not just the winner |
| D3 | Product proposal | One parametric product (typhoon rider or embedded flight cover) with trigger design, oracle governance, basis-risk disclosure, distribution — and its regulatory questions · M05–06 |
| D4 | Auditable-AI design | Decision-record schema, anchoring rule, escalation policy, regulator replay procedure (Exercise 11.1, matured) |
| D5 | Risk register | ≥10 risks with owner + mitigation — must include: oracle failure, consortium funding (B3i test), key custody, PII/erasure, vendor lock-in, migration coexistence, model drift · M03, 08, 09 |
| D6 | PoC plan | 90-day proof scope with success metrics and a kill criterion (what result cancels the programme — write it before you start) |
§3Reference architecture (starting point — improve on it)
§4Scoring rubric
| Criterion | 1 · weak | 2 · developing | 3 · solid | 4 · distinguished |
|---|---|---|---|---|
| Framework discipline | Blockchain everywhere | Framework cited, not applied | Each ledger use passes M05's three questions | …and one tempting use case is explicitly killed with reasons |
| Technical soundness | Hand-wavy boxes | Right shapes, wrong details | Consistency, custody, PII, oracle governance all addressed | …with failure runbooks and an outbox-grade integration design |
| Domain fluency | Generic tech talk | Some insurance terms | Lifecycle, reinsurance, compliance woven through | …with regulator-facing artifacts (replay procedure, disclosures) |
| Evidence use | Vendor forecasts as facts | Some cases cited | Choices justified by M08/M09 precedents incl. failures | …B3i/fizzy/Contour tests applied to its own weak points |
| Decision quality | No trade-offs shown | Options listed | Clear recommendation + rejected alternatives | …plus kill criterion and 90-day falsifiable PoC |
Self-score honestly, or better: swap packs with a colleague and score each other. 15+/20 with no criterion below 3 = capstone passed.
§5Submission checklist
- ☐ D1–D6 complete; diagrams labelled; every ledger use passes the three questions
- ☐ The B3i question answered for any multi-party element ("who pays in year 4?")
- ☐ PII/erasure design stated (nothing personal on-chain)
- ☐ Oracle governance specified for any parametric element
- ☐ AI decisions anchored + replay procedure written
- ☐ Kill criterion written before the PoC plan
You now hold an unusual combination: ledger literacy, insurance domain fluency, the real deployment map across two industries, and an auditable-AI design pattern. Keep the company tracker alive (statuses rot fast in this field), re-run the M03 worksheet whenever a vendor pitches, and be the person in the room who asks the B3i question.
Scenario details generalised from the Galileo dossier. This capstone produces advisory artifacts — real programmes add legal, actuarial and regulator engagement from day one.