KPSGILL Policy Lab · where a recommendation becomes implementable
The Policy Lab
Anyone can say a system is broken. The useful output is the drafted provision, its cost, its losers, and the argument that would be made against it — written down before someone else makes it.
The template every position is built on
A recommendation that cannot be drafted, costed, and argued against is an opinion. Twenty-four fields, in this order, is what converts one into something a legislator, a regulator or a board could pick up and use.
Three fields are non-negotiable and are the ones most often missing elsewhere: documentary baseline (a position with no record under it is a preference), strongest arguments against (stated by us, not paraphrased into weakness), and sunset and review (a reform that cannot be withdrawn is a permanent experiment).
Worked position: prospective payment for primary care
MODEL CMS POLICYPUBLIC COMMENT PROPOSAL
Problem. Primary care is paid for encounters and is expected to deliver continuity. Every high-value activity in the discipline — the call that prevents an admission, the medication reconciliation, the message answered at nine at night, the conversation that ends a referral cascade — is either unbilled or billed at a fraction of a visit. The payment design and the clinical model are pulling in opposite directions, and the discipline absorbs the difference as unpaid labour.
Documentary baseline. The CY 2027 Physician Fee Schedule proposal carries two conversion factors and an efficiency adjustment to work RVUs, and comments close 14 September 2026 CURRENT LAW / PROPOSED RULE — the record. That open comment period is the mechanism this proposal is written for.
Recommendation
A prospective, risk-adjusted primary-care payment with a floor tied to panel complexity, an explicit continuity measure, and a technology term that pays for asynchronous work instead of adding documentation to it.
Fiscal
Scored against avoided utilisation rather than against visit volume; the honest statement is that the first two years cost money before they save any.
Physician impact
Revenue stops being a function of how many people sat in a room. Panel size becomes a clinical decision rather than a financial one.
Patient impact
The activities patients actually value become the activities that are paid for.
Administrative burden
Falls, deliberately: fewer encounter-level justifications, one panel-level attestation.
Unintended consequences
Under-service, panel cherry-picking, and consolidation pressure on practices that cannot bear prospective risk. Each needs a measured guard, not an assurance.
The strongest arguments against
Prospective payment invites under-service.
Risk adjustment will be gamed.
Small practices lack the infrastructure to take prospective risk.
Answers
Under-service is measurable and can be paid against. Unbilled work is not measurable at all, which is why the current design hides its own failure.
Gaming is a coding-integrity problem with existing enforcement tools — and the enforcement record shows those tools in use.
Which is exactly why the floor must be tied to panel complexity rather than to negotiated capacity. A design that only large groups can accept is a consolidation policy wearing a payment policy’s clothes.
Metrics: share of primary-care revenue not tied to a visit · continuity index at practice level · asynchronous work volume paid rather than absorbed. Open question: should the technology term be conditioned on provenance standards for AI-drafted patient messages? On the current draft, yes — see the record integrity standard.
Forecasts
A forecast on this site carries four things or it is not published: stated confidence, the assumptions it rests on, a competing scenario argued in good faith, and the observation that would falsify it. Four are published below, longest horizon last.
FUTURE-OF-MEDICINE FORECASTHorizon 18 months · confidence: Moderate to high
Automation disclosure arrives through metrics, not through a prohibition
The first enforceable constraint on algorithmic coverage denial will not be a ban. It will be a reporting requirement whose published overturn rates make automated denial commercially unattractive before any statute prohibits it.
Assumptions
Authorisation metrics are published at plan level and are comparable; overturn rates for automated determinations exceed those for clinician determinations.
Competing scenario
A state enacts an outright bar on automated medical-necessity denial first, and the metrics follow rather than cause it. Plausible in a session where a single denial becomes a news story.
What would falsify it
Published metrics showing no overturn-rate difference between automated and clinician determinations.
Why it matters
It sets the order of work: the drafted standard puts metrics and disclosure ahead of prohibition for exactly this reason.
FUTURE-OF-MEDICINE FORECASTHorizon 3 years · confidence: Low to moderate
Continuing certification is unbundled by purchasers, not by regulators
The single-board condition will break first when a large employer or health system drops it for recruitment reasons, and only afterwards through statute or litigation. The credential loses its function before it loses its legal position.
Assumptions
Workforce scarcity in shortage specialties continues; recruitment pressure outweighs the administrative convenience of a single privileging line.
Competing scenario
The requirement survives because purchasers treat it as liability protection, and change comes only through competition enforcement — the scenario the litigation question is written for.
What would falsify it
Large systems publishing privileging criteria that retain the single-body condition while carrying unfilled positions.
Confidence note
Low to moderate. Institutional inertia is the most reliable force in credentialing, and it argues against this forecast.
FUTURE-OF-MEDICINE FORECASTHorizon 5 years · confidence: Moderate
Provenance becomes a certification requirement, and the attestation problem is solved sideways
Machine-readable provenance for generated clinical text will arrive as an electronic-record certification criterion rather than as medical-record legislation, and the false-authorship attestation will disappear with it — not because anyone legislated on attestation, but because the record will finally be able to distinguish.
Assumptions
Certification bodies act because vendors want one uniform standard; no major litigation forces the question earlier.
Competing scenario
A malpractice verdict turning on an unattributable generated note forces the question into state medical-record law first, which is faster and messier.
What would falsify it
A certification cycle passing with no provenance criterion while ambient documentation share continues to rise.
Why it matters
It is why the model standard is drafted in both statutory and regulatory form, and sunsets if certification adopts an equivalent rule.
The first AI-adjacent board discipline cases arrive before the guidance does
Within twelve months, state boards will be deciding cases in which a generated note is the central evidence, under standards written for human authorship — and the earliest published guidance will be reactive to those cases rather than prior to them.
Assumptions
Ambient documentation continues to spread faster than guidance is issued; complaints arrive framed as documentation failures rather than as AI failures.
Competing scenario
A model framework reaches state boards during 2026 and is adopted quickly enough that the first contested cases are decided under a published standard. This requires boards to move faster on guidance than they have on any comparable technology — possible, and not the way to bet.
What would falsify it
Adopted, risk-tiered AI guidance published by a majority of large-state boards before the first contested case is decided.
Why it matters now
If the forecast holds, the standard will be set by the facts of whichever case happens to arrive first. Model guidance exists to make that outcome avoidable.
Where the record is strong enough to support a position
Sequencing is not arbitrary. A position is drafted where the documentary layer already carries the instruments it argues about. These are next, in this order:
Prior authorization duration and algorithmic denial — the record carries the interoperability final rule, three state parity audits and a pending California bill.
Corporate control of clinical judgement — the record carries an operative California statute with a named enforcer.
Laboratory quality as payment integrity — the record now carries two settlements and a $1.6 billion programme-integrity campaign.
Single-payer and coverage architecture — the weakest documentary base of the four, and therefore the one that opens as an EVIDENCE GAP rather than a recommendation.