Forecasts · 4 active · none resolved yet

Forecasts

What KPSGILL expects next — and a permanent record of whether it happened. A forecast here is a prediction, never a scheduled event; the decision calendar holds the known dates. Every forecast states its confidence, its assumptions, the competing scenario, and the condition that would prove it wrong.

ACTIVEModerateIssued · horizon 12 months · Future of medicine

Forecast: the first AI-adjacent board discipline cases arrive before the guidance does

Forecast. 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 board guidance is issued.
  • Boards receive complaints framed as documentation failures rather than as AI failures.

Competing scenario

A model framework issued to state boards during 2026 is adopted quickly enough that the first contested cases are decided under a published standard. This would require boards to move faster on guidance than they have on any comparable technology.

What would prove it wrong

Publication of adopted, risk-tiered AI guidance by a majority of large-state boards before the first contested case is decided.

Anchored to the documentary record behind the Policy Lab forecasts section.

ACTIVEModerate to highIssued · horizon 18 months · Future of medicine

Automation disclosure arrives through metrics, not through a prohibition

Forecast. The first enforceable constraint on algorithmic coverage denial 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 metrics follow rather than cause it.

What would prove it wrong

Published metrics showing no overturn-rate difference between automated and clinician determinations.

Anchored to the documentary record behind the Policy Lab forecasts section.

ACTIVELow to moderateIssued · horizon 3 years · Future of medicine

Continuing certification is unbundled by purchasers, not by regulators

Forecast. The single-board condition breaks first when a large employer or health system drops it for recruitment reasons, and only afterwards through statute or litigation.

Assumptions

  • Workforce scarcity in shortage specialties continues.
  • Recruitment pressure outweighs the 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.

What would prove it wrong

Large systems publishing privileging criteria that retain the single-body condition while carrying unfilled positions.

Anchored to the documentary record behind the Policy Lab forecasts section.

ACTIVEModerateIssued · horizon 5 years · Future of medicine

Provenance becomes a certification requirement, and the attestation problem is solved sideways

Forecast. Machine-readable provenance for generated clinical text arrives as an electronic-record certification criterion rather than as medical-record legislation, and the false-authorship attestation disappears with it.

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.

What would prove it wrong

A certification cycle passing with no provenance criterion while ambient documentation share continues to rise.

Anchored to the documentary record behind the Policy Lab forecasts section.

Method

Forecasts are issued only where the documentary record underneath them is published on this site, and each is anchored to that record. Confidence language is fixed — very likely, likely, moderate, low — and is never mixed with the separate vocabularies used for factual verification or evidence strength. A forecast that cannot name what would falsify it does not publish.

Register: data/proposals.json · statuses ACTIVE → CONFIRMED / PARTIALLY CONFIRMED / MISSED / INVALIDATED / SUPERSEDED.