AI Engineering

Model Combination sequences 2–3 complementary mental models when one model leaves a material blind spot on a multi-domain or high-stakes problem. It is used to assign distinct roles, set a conflict rule, and synthesize once.

Model Router maps a task’s domain and problem type to NONE or one primary thinking skill, with up to three complementary skills only when needed. Use it when the right analytical frame is unclear or several catalog skills seem plausible.

OODA Loop helps you make reversible decisions under time pressure by cycling Observe, Orient, Decide, and Act at about 70% confidence. Use it for incidents, intermittent failures, or other changing situations where fast re-observation matters.

Opportunity Cost helps compare a commitment against the best alternative use of the same time, money, or attention. Use it when deciding whether a choice creates more value than what it displaces.

Pre-Mortem Analysis assumes a plan has already failed and works backward to identify concrete causes. Use it before kickoff, launch, or other high-risk commitments to turn risks into mitigations and verification gates.

Probabilistic Thinking helps you forecast, estimate, and size risk with base rates, ranges, and explicit prior-to-posterior updates. Use it when a single confident number would be misleading or when new evidence should shift an estimate.

Reversibility helps you classify a decision by how costly it is to undo, so cheap-to-reverse choices can move fast and high-lock-in choices get more deliberate treatment. Use it when a decision could be staged, piloted, or redesigned to preserve options.

Second-Order Consequences traces immediate and delayed effects of a change across actors, incentives, and scale before you commit. Use it for strategy, policy, architecture, or any choice where the obvious fix may backfire later.

Socratic Questioning asks the few load-bearing questions needed when a request is vague, assumption-laden, or obvious. Use it to expose hidden requirements before building, debugging, or committing to a plan.

Steel-Manning builds the strongest faithful case against a proposal before you accept or reject it. Use it in design reviews or decisions where trade-offs matter and you want agreement conditions before updating or reaffirming.

Theory Of Constraints identifies the single binding stage in a throughput-limited system and guides exploit, subordinate, elevate, then recheck. Use it when one step dominates latency or output and local fixes elsewhere won't help.

Thinking Cynefin classifies a problem by cause-effect domain before choosing how to respond. Use it when the right mode is unclear, a method keeps failing, or a mixed problem needs decomposition first.