Governance · model inventory · evaluation records · accountability

Charles Trinh — Healthcare AI Governance

A service card reading AI GOVERNANCE — inventory, evidence and named accountability
Charles Trinh — Healthcare AI Governance

Governance that survives scrutiny: model inventories, intended-use statements, evaluation records and the accountability chain behind every automated decision.

What this is

Governance is the part of an AI programme that gets written after the first model ships, usually as a policy that describes a process nobody follows.

The version that survives scrutiny is duller and more concrete: a list of what is running, a page per model saying what it is for, dated evidence that it works, and a name against each decision.

That is achievable in weeks, and it is the difference between answering a regulator's question in an hour and assembling a response over a quarter.

Inventory before policy

You cannot govern what you cannot list; start with every model, prompt and automation in production.

Include the ones embedded in vendor products, because they change without telling you. An inventory that omits the models you did not build is the one that fails a review.

Intended use, written

A one-page statement per model covering what it is for, what it is not for, and who decided.

The not-for list is the useful half. It is what stops a model built for scheduling being repurposed for triage without anyone noticing.

Evaluation records

Dated results against a fixed task set; a model without an evaluation record is an untested control.

Keep the task set and the results together, with the version tested. An undated pass tells you nothing about whether the current model still passes.

Accountability chain

A named person for each automated decision, with the authority to switch it off.

Authority matters as much as the name. If the accountable person cannot stop the system, accountability is nominal and a reviewer will say so.

Bias and drift

Measure against subgroups and over time; drift is normal and undetected drift is the risk.

Quarterly measurement is enough to catch the trend and cheap enough to survive budget pressure. Publish the method, not only the result.

Vendor models

Third-party models change under you; pin versions and re-evaluate when they move.

Where pinning is not offered, monitor output for change and re-run the evaluation on a schedule. Either way the re-evaluation is triggered by the version, not by an anniversary.

Patient transparency

What you tell a patient about automated involvement is a policy decision made before, not an answer composed after.

Write the sentence once, in plain language, and use it consistently across channels. An improvised explanation at the point of a complaint is both late and inconsistent.

Retirement

Every model needs an end-of-life path, because the replacement always arrives.

Retirement covers the model, the prompt, the evaluation set and the data retention. Left unowned, old models stay running because nobody can prove nothing depends on them.

How I do it

I build the inventory first, including the models inside vendor products, because that is the list a reviewer will test. Each entry gets an owner, a version and a one-page intended-use statement, written in the same week it is listed.

Evaluation records come next, tied to the version tested and kept with the task set. The evidence is dated or it is not evidence, and a record without a version tells you nothing about what is running today.

The engagement ends with the operating rhythm: quarterly bias and drift measurement, re-evaluation on version change, and a retirement plan per model. Governance that survives scrutiny is a calendar with names on it, not a document.

Artifacts you keep

  • A full inventory covering built and embedded models in production
  • A one-page intended-use statement per model, dated and owned
  • Evaluation results tied to the model version tested
  • A quarterly bias and drift measurement with a published method
  • A retirement plan per model covering data and dependencies

The governance question in one sentence

Who is accountable for the output, and what did they review before it was used. Every governance framework reduces to that sentence, and an organisation that cannot answer it for a given tool has not governed the tool.

Name the accountable person per use, not per technology. Accountability attached to a tool is accountability attached to nobody.

Inventory before policy

You cannot govern models you do not know are running. Start with an inventory: what is deployed, who owns it, what data it touches, what it decides and who reviews the output.

The inventory is usually larger than expected, because pilots become production quietly and vendor features arrive switched on.

Evidence for each use

For each use, the evidence is an evaluation against a labelled set, a documented acceptance of residual error, and a monitoring plan for drift. Those three are what an examiner asks for, and they are what a team needs anyway to improve the system.

Re-evaluate on version change, because the evaluation is a property of the version rather than of the tool.

Drift and the review cadence

Model behaviour changes with inputs, and inputs change when the population or the workflow changes. A quarterly review of error rates against the original evaluation catches drift before a clinician does.

Record the review with its date and its conclusion. A governance programme without a record is a set of meetings.

T

he vendor's evidence

A

vendor evaluation is evidence about their test set, not about your population. Accept it as a starting point and run your own on your own records, because the failures that matter are the ones specific to your data.

A

sk for the evaluation dataset description. A vendor who cannot describe it has not measured what they claim.

Write the accountability sentence for each use in the inventory row, and revisit it at the quarterly review. Where the sentence names a role rather than a person, replace it, because roles change hands and accountability that changes hands silently is not accountability.

Publish the inventory internally. Visibility is what stops a pilot becoming a production system without anyone deciding that it should.

The board sentence

One paragraph per quarter: what is deployed, what it decides, who reviews it and what the error rate is. That paragraph is what a director can act on, and it is what the governance programme produces instead of a framework diagram.

Write it from the inventory, not from the vendor decks, and keep the previous quarters so the trend is visible.

Where a use is retired, record why and keep the row. The history of what was tried and stopped is the evidence that the programme exercises judgement rather than accumulating tools.

Review the inventory with the legal and clinical owners together once a quarter. Separately they each see half the picture, and the uses that matter are the ones that look fine from one side and not the other.

Define what counts as a change requiring re-evaluation: a model version, a prompt template, a schema, or the population. Without the definition, nothing is ever re-evaluated because every change looks like a small one.

Record the residual error the organisation accepted, in numbers, at the time of acceptance. Memory of an acceptable rate erodes, and the number is what a later reviewer will ask for.

Proof

  • ONC (g)(10) self-test PASS, 317 of 317 subtests, April 2026 — public suite URL
  • 67 public hostnames, 51 serving, one AWS account
  • 104 clearinghouse enrolments across 53 jurisdictions
  • USCDI v7 proposal — Patient Digital Identity Credential — listed at isp.healthit.gov
  • SAM.gov UEI MVKQW9DKKGB8 · CAGE 20KM5 · NPI 1194645564
  • AWS Business Associate Addendum active since 18 August 2026

Where this work is delivered

Engagements run from Las Vegas and reach across the valley: clients in Las Vegas, NV; Las Vegas, Nevada; Henderson; Clark County, Nevada; and North Las Vegas, Nevada, plus remote delivery nationwide.

How an engagement runs

1. Trace one real case

Follow a single patient or claim end to end with a timestamp at every hop, before any document is written.

2. Written findings

A findings list naming the defect, the system, the owner and the date.

3. Monthly review

Review the list monthly and check whether the dates held.

Frequently asked

What is this service?

This page describes charles trinh — healthcare ai governance — the scope, the failure modes it prevents and how the work is measured. It sits in the AI and Automation category at trinh.io.

Where is the work delivered?

Engagements run from Las Vegas, Nevada across the valley — Las Vegas, Henderson, North Las Vegas, Spring Valley, Paradise, Winchester, Sunrise Manor, Enterprise, Summerlin, Clark County — and remotely anywhere in the United States. On-site work in the Las Vegas valley, remote delivery elsewhere.

What does it cost?

Standard rate $195 per hour, committed $165 per hour at twenty hours or more per week for three months or more; $1,600 standard day, $1,350 committed day. Scoped in writing before work starts, fixed fee where the deliverable is defined.

Who is Charles Trinh?

A technology architect in Las Vegas, Nevada, working in healthcare identity, consent and interoperability. Not a physician, not a medical practice, and not affiliated with any other Charles Trinh.

How does an engagement start?

With one trace of a real case and a written findings list naming the defect, the system, the owner and the date. The first call is twenty minutes and free: 702-559-3489.

Engagement

Scoped in writing before work starts; fixed fee where the deliverable is defined. First call is 20 minutes and free. 702-559-3489 · charles@trinh.io

Talk to Charles

TRINH — Charles Trinh, technology architect
TRINH

Charles Trinh

Fractional CTO/CIO and Technology Architect — Healthcare Data, Identity & Regulated Infrastructure · Las Vegas, NV

Las Vegas, Nevada · US citizen · Fractional and project engagements

(702) 559-3489 Email Charles Trinh