Counterexamples

Every method in this series makes a falsifiable claim: given these inputs, this is the architecture — or process, or code structure — that follows. The standing invitation is to prove one wrong.

A counterexample is not a general objection. It is one system, one derivation, and a specific place where the method’s answer and reality parted ways.

Run it first

Use a system you actually know: your production service, not a toy example.

1. Fill the worksheet for that system, era-pinned. Architectures drift, and an undated sheet cannot be checked against anything.
2. Run the sheet through the entry gate. Unpriced and unscoped answers are the first thing reviewers check, so it is worth catching them yourself.
3. Walk the derivation with the reference cards, or the course if you want it lesson by lesson.
4. Compare what the derivation produced against what the system actually runs. Where they diverge, you have a counterexample.

What the report contains

Five things, which the issue template will prompt you for:

• System context — domain, era, scale, in a paragraph. “Private system, details anonymized” is fine.
• Answer sheet — filled, scoped and priced where the method demands it.
• Derived vector — what the derivation produced, axis by axis, with the forcing answer cited per moved axis.
• Observed vector — what the system actually runs, axis by axis, era-pinned.
• The divergence — where the two disagree, and why the observed position is load-bearing rather than habit or history: what breaks if the system is moved to the derived position?

The last one carries the weight. A divergence explained by history is a finding about that system; a divergence that survives “what breaks if we move it?” is a finding about the method.

How it is judged

By the same grading rubric and evidence grades applied to the four registered derivations, under a public protocol. Misses are published alongside hits.

File a counterexample

That intake is the Architecture Synthesis replication kit. Process-First Design and JBCT have their own templates on the series repository — a PFD report argues about a decomposition and its change drivers, a JBCT report about a rule that makes the code worse. Neither has a registered-prediction corpus yet, so reports there are read and answered rather than graded.