SERVICES
Services & Consulting
The research is free and stays free. What the Institute offers for hire is the discipline behind it: behavioral analysis of AI–human interaction — reading how extended, real conversations with a deployed model actually go, finding where they fail, and keeping them calibrated to the people they serve.
That practice has a long root. The founder spent twenty-five years in customer-experience operations — at DISH Network, at Wayfair, and in independent practice, in roles up to Senior Manager / Acting Director — reading transcripts at scale and owning quality, escalation, and communication analysis for operations from a few hundred to several thousand agents. Applied to the newest agent on the floor, that same discipline produced the CCD taxonomy, the preserved evidentiary record, and the Guardian Protocol in 2025. He is not trained in machine learning and does not pretend to be; what an engagement buys is the operations craft and a published, evidence-grounded method for assessing conversational failure — offered on exactly that basis.
The method transfers because the failure does. The content a drifting model latches onto changes from one user to the next; the shape of the failure — the markers, and the way they assemble — does not. Mapping a transcript to that structure, and separating a tone problem from a confidence problem from a confabulation because conflating them misdirects the fix, is the same work whether the subject is one person’s record or a production assistant serving thousands. That is what makes this a practice, not a case study.
What the Institute offers
Drift audits & monitoring
For organizations running a deployed assistant that cannot afford to drift: a structured read of real interaction transcripts against the CCD markers and the SCC diagnostic — convergence, flattening, inflation, confabulation — with ongoing monitoring to keep it calibrated as the model, the prompt, and the users change. If a knowledge-management, health, education, or finance team needs to know its AI is not quietly going off-course, this is the audit for it.
The research it applies →Calibrated-system design
Design or specification of an AI system tuned to your people and your domain — one that stays anchored to the user and resists drift by construction, rather than being patched after an incident. Built on the Guardian Protocol’s intervention architecture: cross-instance verification, fabrication checks, the user-words anchor, and escalation logic, fitted to how your system is actually deployed rather than to a slide about it.
The intervention architecture →SCC measurement at scale
Sycophantic-convergence measurement across a large deployment — in both directions, inflation and flattening — prioritized to the high-stakes conversations where a miscalibrated answer costs the most. A measurement service with a real, published methodology behind it, not a sentiment dashboard.
Litigation & regulatory advisory
Behavioral analysis applied to the AI-harm docket and to policy work: mapping the interaction record in a matter to the documented failure markers, and briefing counsel, regulators, and standards bodies on what extended-interaction failure looks like from the user’s side. Grounded in primary sources; delivered attributed and checkable.
Speaking & briefings
Plain-language briefings for teams, boards, policymakers, and press: what the documented record shows, what the available interventions are, and why the fix is narrow rather than anti-AI. Delivered in the same register as this site — attributed, checkable, de-dramatized.
How an engagement starts
Write to [email protected] with what you are deploying — or what already happened — and what you need to know. You get a scoping conversation, a written proposal with the method stated up front, and findings delivered with the same provenance discipline as the published research.
The line that doesn’t move. Engagements fund the public work; they never alter it. An assessment that finds nothing says so — that is the product. Consulting buys the Institute’s method and time, not its conclusions, and no engagement is accepted that would require otherwise.
Prefer to sustain the public work rather than engage it? See Support the Institute.