Technical

A common plan hiding inside AI roadmaps: wait — the next generation will be good enough to trust. But capability and defensibility are different axes. Scaling moves one and leaves the other exactly where it was; a more capable model in an ungoverned harness produces the same indefensible decision, better written.

Problem-solving strategies find a path to a goal. The scientific family governs something harder: what you are entitled to believe, and how belief must move when evidence arrives. Two strategies, released as executable graph shapes: the hypothetico-deductive method and Bayesian updating.

You do not need the math. You need one accurate mental model: an assessor whose training is frozen into habits, whose entire knowledge of your case is the pile on the desk, and who answers by writing the most plausible next word. From that model, most of the governance questions answer themselves. This is lesson one of a curriculum.

The epistemic family asks what is true. The problem-solving family asks how to get from here to a defensible there — and its shared enemy is motion mistaken for progress. Five strategies, released together as executable graph shapes: decomposition, means-ends analysis, analogical reasoning, constraint satisfaction, and working backward.

The oldest information format still in production use is the Euclidean proof: definitions, assumptions, numbered propositions, each step warranted by something already established. We use it as an output contract for AI determinations — because it is the only prose format ever devised that makes missing justification structurally visible.

A natural experiment in regulatory architecture: take one model package, adjudicate it under SR 11-7 and under SR 26-2 on the same reasoning engine, and diff the dossiers. The evidence and math are identical; the classification, anchors, and conclusion vocabulary diverge. The diff is a map of where regulation actually lives.

Effective challenge is the load-bearing idea in model risk management, and it has always run on organizational willpower — which is exactly what erodes under deadline pressure and familiarity. Encoding challenge as an adversary node makes it structural: steelman the position first, then attack in both directions, at a rigor locked to the model's materiality.

The industry uses one word for two different things: the capability and the process. AI asks a model what it thinks; artificial reasoning defines how the system must think — as an explicit computation that records itself as it runs. Here is the distinction in full, what the term does not claim, and the falsifiable bet underneath it.

Every number an LLM produces is a prediction of what a number would look like. That's tolerable in a chatbot and disqualifying in a decision system. The engineering answer is deterministic tool-nodes — zero tokens, byte-reproducible — but the property that actually matters is authority: the model may cite the numbers, never recompute or override them.

The four-fifths rule is a fifty-year-old screening heuristic that AI decisioning has made load-bearing again. It is a tripwire, not a verdict — and both halves of that sentence matter. What the Adverse Impact Ratio measures, where practical and statistical significance diverge, and why the computation should never come from the model being tested.

A chatbot produces one reasoning event per interaction. An agent produces one per step — planning, tool selection, interpretation, revision — hundreds per task, invisible in the tool logs everyone mistakes for observability. The agent wave is a multiplication of exactly the thing enterprises never learned to govern.

A reasoning strategy is not a topology and not a prompt — it is the shape of thought itself, and it can be engineered. This series walks through the IRG strategy inventory a few strategies at a time, starting with the epistemic family: abduction, deduction, and induction as executable, gated graph shapes.

Accuracy benchmarks ask whether the system was right. That is the wrong question for production. The right question is whether the system behaved well given what it actually knew—and that is what epistemic integrity measures.

AI has already proven it can write, analyze, and classify. What it has not proven, in the institutional sense, is that its reasoning can be trusted for the decisions a regulated business is actually built on. The blocker is not capability. It is accountability.

Apply different prompt sets to the same reasoning graph and you get different convergence paths, different abstention rates, and different epistemic integrity scores. Prompt engineering isn’t cosmetic—it’s the configuration of an AI system’s epistemic personality.

Hallucination rates aren’t evenly distributed. They increase with document complexity and input ambiguity—which means they concentrate in the domains where accuracy matters most: medicine, law, finance. The cause is architectural, and so is the fix.