Architecture Decisions Without Blind Spots

From the free book Delta: Closing the Specification Gap (Dhuri, 2026). The PDF is the canonical edition — figures, tables, and code formatting are simplified online.

The Socratic Prompt: find reasons to reject your design before production does

“Every architectural decision has blind spots. The one everyone agrees on has the most. The Socratic Prompt is the only systematic technique for finding them before they find you.” — The most dangerous architecture review is the one where nobody objects. In This Chapter

1. The Socratic Prompt: five adversarial outputs
2. Multi-criteria technology decisions with hidden costs
3. SOLID evaluation with Consequence at Scale
4. AI system architecture review (the new required pattern) <role> Architecture challenger. Your job is adversarial: find reasons to ,→ REJECT this design. Do NOT evaluate whether it is good. No positive assessments. </role> <design>{{describe your design in –100200 words}}</design> <task>Give me ONLY these five outputs:
5. THE THREE MOST DANGEROUS ASSUMPTIONS Assumptions that could prove false at 5x scale, after 18 months, or ,→

under org change.

For each: the assumption + what fails when it is wrong. 2. THE MOST LIKELY PRODUCTION FAILURE Specific triggering condition. Exact failure mode. Business impact. ,→ 3. THE DECISION HARDEST TO REVERSE The specific choice. Why reversing it is expensive. 4. WHAT THIS OPTIMIZES FOR — WHAT IT SACRIFICES Both sides. Honest trade-off. 5. THE UNCONSIDERED ALTERNATIVE One specific named approach with different trade-offs. Name the approach — do not say ‘you could consider’. </task> <constraints> - Do NOT say whether this a good design - Item 2: name the specific trigger, not a category of risk - Item 5: a specific named alternative, not general advice </constraints>

FOR ENGINEERING MANAGERS The Socratic Prompt should run before every significant architectural commitment. The five questions surface exactly the kind of concerns that appear in post-mortems as ’we should have considered this earlier.’ The most common reason teams skip it: they are confident in the design and don’t want to find problems. That’s precisely the condition under which it is most necessary.

Key takeaways

• The Socratic Prompt makes validation structurally impossible. The model is instructed to find reasons to reject, not evaluate quality.
• Five mandatory outputs: dangerous assumptions, most likely production failure (specific trigger), hardest-to-reverse decision, honest trade-off, unconsidered alternative (named).
• SOLID evaluation only adds value with Consequence at Scale: not ’hard to maintain’ but ’at 5x team size, adding one validation rule requires changes in 8 locations based on current growth.’

TRY IT NOW

1. Apply the Socratic Prompt to an architectural decision your team committed to in the last three months. What do the dangerous assumptions reveal about current risk?
2. Run SOLID evaluation against your most complex service class with Consequence at Scale. Does it match what you are actually experiencing in maintenance?
3. For your current major technical decision: write the multicriteria comparison including Hidden Costs. What does writing it force you to acknowledge?

UP NEXT — Chapter 9: Testing — Find What Your Tests Miss

An architecture is a set of claims about how the system behaves. Chapter 9 is about proving them: prompting for tests that find the boundary, concurrency, and idempotency defects a coverage number hides.

69