Examples
The measure of a methodology is whether it helps practitioners think more clearly. These examples are evidence, not instruction.
Four worked ICR cycle examples across different domains and delivery contexts. The fourth is a real production system — not illustrative.
What Each Example Shows
Each example follows the same structure:
- Domain and stakes — what the project is and what goes wrong if an assumption is wrong
- ICR cycle walkthrough — FORMALIZE → STRESS → CHECK → SURFACE → GATE
- What changed — how the project was different because the cycle ran
The first three examples are illustrative but grounded in realistic delivery scenarios. The fourth is from a live production system — real incidents, real constraints, real decisions.
The Four Examples
Legacy Banking Rule Extraction
Domain: Credit approval logic in a 40-year-old COBOL core banking system
Stakes: Regulatory compliance, incorrect approvals or denials at scale, audit exposure
Context: Phoenix Runtime pipeline being prepared to modernize the legacy system
The ICR cycle surfaces a regulatory vs. policy ambiguity in the DTI threshold, a self-employed applicant branch with 1987 logic, and reason codes mapped to a table not updated since 2011 — none of which were visible from the COBOL code alone.
Multi-Agent Enterprise Procurement
Domain: Purchase order approval workflow across a mid-size manufacturing enterprise
Stakes: Financial exposure from incorrect approvals, audit trail gaps, vendor relationship risk
Context: Five OCTO agents being configured to automate procurement routing
The ICR cycle surfaces a race condition in the confirmation webhook, a budget double-approval exposure from batch-updated committed spend, and a probationary vendor status that the binary agent model silently misroutes.
AI-First Feature Development
Domain: Loan origination feature in a fintech product
Stakes: Regulatory compliance, incorrect lending decisions, user trust
Context: Development team preparing to hand a feature spec to an AI coding agent
The ICR cycle surfaces four decisions the two-paragraph spec did not make: the self-employed path, adverse action notice requirements, product-specific score thresholds, and variable income calculation — all of which the agent would have had to silently decide on its own.
AI-Driven Publishing Pipeline ★
Domain: Cascade intelligence case study research — from raw idea to live publication
Stakes: Factual accuracy at public scale, editorial integrity, permanent DOI-archived record
Context: A real production pipeline. 247 published case studies. One post-publish incident that changed the architecture.
The ICR cycle surfaces that "cited" ≠ "correct" — a conflict that was invisible until UC-236 shipped with three factual errors despite passing every prior check. The unresolvable residue in a pipeline that automates everything else: is this thesis original enough to earn a DOI? That decision is human, permanent, and designed-in.
Contributing an example
If GDD helped you surface an assumption or catch a conflict in a real project, the most valuable contribution is a documented example. A good example has at least one [CONFLICT] that surprised you and an honest account of what changed because the cycle ran. Open a PR →