Never automate decisions with genuine per-instance novelty — recurring categories with unique instances require deliberation

Why This Is a Rule

The automation criteria (Only automate decisions that are frequent, stable, and low-stakes — all three must hold or automation introduces risk) require stability — the optimal answer being the same each time. This rule addresses the subtle trap of decisions where the category recurs but the optimal response genuinely varies per instance. "Interpersonal conflict" is a recurring category, but each conflict involves different people, different stakes, different histories, and different optimal responses. Automating with a rule like "always address conflict directly within 24 hours" produces efficiency — you never agonize over timing — while producing rigidity: some conflicts need immediate address, some need cooling-off time, and some should be strategically ignored.

The danger is that category recurrence masquerades as decision stability. "I deal with this kind of thing all the time" feels like expertise supporting automation, but in novel-per-instance domains, what feels like expertise is actually pattern recognition that should inform deliberation, not replace it. The experienced doctor recognizes patterns across patients but deliberates for each diagnosis because each patient's constellation of symptoms is unique.

Automating novel decisions doesn't just produce occasional bad outcomes — it atrophies the deliberative capacity needed for the domain. If you stop deliberating about interpersonal conflicts, you lose the sensitivity to context that distinguishes good conflict resolution from formulaic responses.

When This Fires

• When tempted to create a "rule" for how you handle interpersonal situations, creative decisions, or complex diagnoses
• When someone says "let's just standardize how we handle X" where X has genuine per-instance variation
• When Only automate decisions that are frequent, stable, and low-stakes — all three must hold or automation introduces risk's criteria suggest automation but something feels off — check for hidden per-instance novelty
• When an automated response produces a bad outcome because the specific instance differed from the category norm

Common Failure Mode

Confusing expertise with automation readiness: "I've handled 50 team conflicts, so I can automate my response." Your 50 instances of experience should make you better at deliberating each new conflict, not make you stop deliberating. The expertise goes into richer pattern recognition and better contextual judgment — both of which require active deliberation to deploy.

The Protocol

(1) Before automating a recurring decision, ask: "Would the optimal response genuinely be the same for the next 10 instances?" (2) If yes (dress code, email filing, morning routine order) → safe to automate (Only automate decisions that are frequent, stable, and low-stakes — all three must hold or automation introduces risk). (3) If the response should vary by instance (conflict resolution, creative direction, strategic advice) → keep deliberative. (4) For borderline cases, try the automation for one week and track: how often does the automated response feel wrong for the specific instance? If >20% → the decision has too much per-instance novelty for automation. (5) In novel-per-instance domains, invest in deliberation quality (better frameworks, richer pattern libraries) rather than automation.