The worked examples

Seven loops, each shown actually running with the receipts: iteration logs, cost ledgers, before/after, charts, and (for two of them) real .xlsx output. Everything is synthetic and carries a "reconstruction for teaching" label; the dollar figures are illustrative — as of June 2026, verify before relying.

# Example Domain Primitive What it proves
1 The $217 Overnight Code Review coding /loop circuit-breaker + cost cap (ungoverned vs governed)
2 Two Loops, One Repo coding /loop ×2 multi-loop coordination (lease + merge-queue)
3 Claim-Ledger Security Fix coding /goal maker-checker on executable evidence
4 Reproduce-Before-You-Fix coding /goal reproduction-gate (prove the bug first)
5 Ad-Spend Reconciliation non-coding /goal validation-gate over messy data (.xlsx)
6 RFP / Security Questionnaire Pack non-coding /goal citation claim-ledger / cite-or-cut (.xlsx)
7 Deliverability Rescue (climb to 95) non-coding /goal metric-climb with an independent grader

Four coding, three non-coding; every example owns a distinct loop pattern.

The numbers, side by side

Cost per example, and the metric-climbers reaching their gate:

Cost per worked example

Coverage / score climb

Governance drives wasted work to zero

How to read an example

Each example folder follows the same shape:

Start with the overnight review — it's the clearest "ungoverned vs governed" contrast.

Back to the handbook