Capstone 02 — The credit decision table

Motto — The policy axis nobody hard-codes twice: score × amount × employment, owned by credit, deployed like code.

Part of Phase 11 — Capstone. Combines Phase 5.

The Project

outputs/loan-decision.dmn — Phase 5’s creditDecision grown a third input, which is exactly how real tables evolve:

Design notes, all cash-outs of earlier lessons:

• hitPolicy="FIRST", deliberately. The rows are exception-then-default (Phase 5, lesson 02’s criterion), and the self-employed tightening depends on more-specific-first ordering. A salaried 760/₹8L file hits row 1. A self-employed 760/₹8L file falls past row 3’s cap into manual review. Reordering these rows is a policy change, and it must be reviewed as one.
• The catch-all decline is the completeness guarantee. No input combination returns nothing, so the process’s default gateway path is belt-and-braces, not a load-bearing hole-plug.
• The employment axis came from governance, not engineering. The committee’s “self-employed needs tighter caps” lands as two rows and a new input. No Java, no BPMN change — just one .dmn redeploy. That’s the asymmetry that justified Phase 5.

Verify It

Transcribe the table into Phase 5’s toy engine and run the golden cases before ever deploying — the toy is the oracle:

# with lesson 5.02's DecisionTable/Rule and hit_policy="FIRST"
golden = [
    ({"score": 760, "amount": 800_000, "employment": "salaried"},      "auto-approve"),
    ({"score": 760, "amount": 800_000, "employment": "self-employed"}, "manual-review"),
    ({"score": 710, "amount": 400_000, "employment": "salaried"},      "auto-approve"),
    ({"score": 660, "amount": 200_000, "employment": "self-employed"}, "manual-review"),
    ({"score": 600, "amount": 100_000, "employment": "salaried"},      "decline"),
]

The second case is the one that catches transcription errors — it only passes if row order survived the copy.

Challenge. Fill in the governance card (Phase 5, lesson 04’s checklist) for this table: owner, ritual, the five golden cases above, and the impact-replay query for the next band change. The capstone isn’t done until the card is.