When being wrong is consequential, reasoning is the pathway.
Independent, first-principles analysis for go / no-go calls, market entry, investment, and policy — and a sharp second read on the analyses already on your desk. Thirty years of lateral problem-solving, accelerated by a disciplined way of working with AI.
Senior analysis, scoped in days — not months.
Fractional, collaborative work on decisions that carry real consequences. Research informs it; your decision sets the scope.
Strategic analysis
24–60 hrsGo / no-go decisions, market entry, restructuring, policy development. Requirements discovery, research design, an analytical framework, and a recommendation you can defend.
AI strategy
scoped engagementWhere AI genuinely fits, where it doesn’t, and what adoption will actually cost in risk and rework. Readiness and use-case triage, architecture review, and a candid read on whether a problem needs a governed approach at all.
Evaluation & second opinion
10–20 hrsA 360° read on a proposal, initiative, or deal already on the table: approach alternatives, benefit–risk profile, and stakeholder impact.
Analysis audit & review
per project / retainerA meta-review of an existing analysis — internal, consultant, or third-party — for completeness, methodological rigor, and credibility.
AI-assisted analysis training
2–3 sessionsA team learns to surface assumptions, calibrate confidence, and work with AI without inheriting its blind spots.
Strategic partnership retainer
monthlyStanding access to any of the above, plus ad hoc strategic consultation, at a guaranteed monthly commitment.
Analysis is only as good as what it rests on.
Epistemic governance is the discipline of keeping reasoning sound — human or machine — where being wrong has consequences. Work starts from first principles, not the house answer; and every use of AI inside an engagement is held to three tests.
Accuracy
Are the claims warranted?Source quality, evidence, assumptions, and confidence — separating what the material establishes from what a model infers or supplies from its training.
Coherence
Does the reasoning hold?Consistency of premises and conclusions; catching unsupported connections, incompatible assumptions, and narrative imposed on material that doesn’t carry it.
Stability
What survives as the context changes?Semantic drift, superseded premises, and contaminated context — the errors that compound quietly across a long analysis and across sessions.
Dex — a governed-corpus framework for domain-expert AI.
Entertainment metadata is its first application, not its definition. The architecture generalises to any field where confident error is dangerous.
Authority matters. Similarity is only part of the answer.
Commercial language models and standard retrieval aren’t adequate for high-accuracy, high-criticality professional work — and supplying more documents doesn’t fix it. Dex treats a knowledge corpus as a governed data asset: curated, classified, and admitted on authority before similarity. It is epistemic governance applied to a machine’s knowledge.
Expert classification
Reliability, stance, standpoint, normative force, and currency — recorded as queryable data, not inferred from style.
Governed retrieval
Admissibility before semantic match; verbatim source supplied for the actual reasoning.
Auditability
Every candidate admitted or excluded is logged with its reason. The system can say what it nearly used, and why it didn’t.
Session continuity
Recover what a long session established; exclude what it superseded.
Each deployment is a project, not a product install — hands-on work at every stage:
Build a working model of the domain
Before any corpus work: understand the field’s structure, its authorities, and its failure modes well enough to judge what belongs and what doesn’t.
Assemble & curate the injection corpus
Select source material deliberately — specification, practice, histories, adjacent fields — rather than harvesting whatever is at hand.
Classify, remediate & rate it
Record each document’s authority, stance, currency, and reliability as data; clean extraction errors; keep reviewer disagreement rather than averaging it away.
Run test cases & tune
Exercise the system against real questions, then tune admissibility, retrieval, and the initialization and inference instructions until it holds.
A bespoke build, not a configuration task — its depth set by the complexity of the domain, the candour and cooperation of the client, and disciplined scope control.
The thinking behind the practice.
Essays and white papers on where convincing AI output and sound professional reasoning part ways.
The End of Thought
How AI is teaching us not to think. A performance of thought is not thought — and the difference is the whole game.
Why RLHF can’t teach professional reasoning
What artifact-based evaluation conceals, and an apprenticeship approach built on expert inquiry and the reconstruction of reasoning.
Doxatropy: when markers displace quality
A behavioural class for systems that converge on the perceptible signs of quality rather than quality itself.
These materials present research arguments and proposed methods. Their hypotheses and estimated benefits should be evaluated as such; they are not reported validation results.
The question comes before the framework.
Grounded in the real decision, the organizational landscape, and the cost of getting it wrong.
Frame the decision
The question, its stakes and constraints, who decides, and who has to act on the result.
Map the landscape
Where decision-making power actually sits, who can quietly block, and how an uncomfortable finding gets escalated.
Test competing perspectives
Separate analyses for each stakeholder’s priorities, and rival readings held against the same evidence before any conclusion.
Make it usable
Weekly check-ins rather than a big reveal — a bounded recommendation the decision-maker can act on and audit.
I don’t provide post-hoc justification for predetermined conclusions.
Curt Mayers.
Systems thinker · strategy & policy analyst · AI collaboration
Thirty years in software, data architecture, and business strategy, with an academic grounding in policy analysis — and a habit of finding the order-of-magnitude solution other people walked past. I don’t sell hype. I bring structured analysis to organizations deciding how, or whether, to act.
Since early 2025 the work has centred on the craft of AI collaboration — not prompt tricks, but methods for surfacing assumptions, calibrating confidence, and keeping a model’s reasoning honest across long, high-stakes work. That discipline is what separates an AI-accelerated analysis from a confidently wrong one.
Whether the reasoning is a person’s or a machine’s, it is one discipline — epistemic governance.
Augmented Decision Support — strategic decision and go / no-go analysis.
My Robot Frenemy — essays on AI’s promise, threat, ethics, and policy.
AI epistemics — the limits of RLHF, doxatropy, and the Dex governed-corpus framework.
Define the decision. Examine what it rests on.
Engagements open with a complimentary 30-minute consultation — the decision, the scope, and whether I’m the right fit.
Tell me the decision you’re facing, the evidence you already have, the assumptions you suspect, and the constraints that matter — including what would change your mind.
A useful starting point: What needs to be understood — and what would change if you understood it?