US Insurance Movement Research Payload 2026-10-02
Research Atlas Observatory; Generated Research & Atlases
US Insurance Movement Registry Family
A source-resolved research instrument for one US health-insurance jurisdiction, payer or payer class, workflow, evidence date, and accountable decision, with explicit claim classes, coverage, authority, and reopening conditions.
v0.1.0
Research outcome. This payload produces a dated, source-resolved return for one specified US health-insurance jurisdiction, payer or payer class, workflow, and evidence date. Facts, supported transfers, practitioner interpretations, local hypotheses, and open questions remain visibly distinct so a named receiver can see what the evidence establishes and which local observation can change the decision.
payload_id: "@driftframe/research-payloads/evidence-bounded-us-health-insurance-workflow"
version: "0.1.0"
execution_mode: read_only_research
compiled_on: 2026-10-02
source_evidence_as_of: 2026-09-24
review_by: 2027-03-24Purpose
Use this payload to examine how a US health-insurance administrative workflow currently moves, where an AI or automation capability fits, what evidence makes a discrepancy actionable, and which local comparison should precede a deployment or workflow claim.
A complete run creates:
- a bound jurisdiction, payer, product, workflow, channel, and evidence date;
- a claim inventory with visible evidence classes;
- a source hierarchy matched to each claim type;
- a retrieval and coverage ledger;
- a workflow-state and administrative-AI layer map;
- an actionability analysis for material discrepancies;
- a bounded conclusion, smallest useful next move, and reopening conditions.
A named accountable person or institution retains legal interpretation, coverage, authorization, payment, operational, publication, and deployment authority. The research return supplies evidence and routes the fitting decision.
Run parameters
Fill every required field before retrieval begins.
A complete run begins when jurisdiction, payer, workflow, evidence cutoff, receiving decision, and accountable receiver are bound. A missing value that changes the governing evidence produces a scoping pause naming the value, its decision consequence, and the source or role capable of supplying it.
Claim classes
Every material claim receives one class and a stable claim ID.
Tag | Evidence class | Admission test |
CF | Current fact | A fitting primary or contemporaneous source supports the exact statement within the bound scope and as-of date. |
ST | Supported transfer | Sending evidence, receiving similarities, transfer boundary, and unresolved mismatch are explicit. |
PI | Practitioner interpretation | The source-anchored reasoning is visible and presented as synthesis rather than external-state fact. |
LH | Local hypothesis | Population, comparator, measure, owner, threshold, and disconfirming result are defined. |
OQ | Open question | The question routes to the source, observation, or accountable role capable of settling it. |
State probability as probability of a named event, for a named population or system, over a named horizon, under named conditions, with the evidence base and calibration method visible.
Source spine
The Registry's public synthesis and Field Watch entry supply the layered comparison, primary-source register, model distinctions, and current review horizon:
- How US Insurance Movement Changed Around Generative AI — Rules, Transactions, Vision, Language Models, and Human Review
- US Insurance Movement Before and After Widespread Generative AI — A Layered Shift Across a Continuing Substrate
Two reusable Registry instruments provide the architecture and discrepancy tests:
- Evidence-Bounded Administrative AI Stack — From Source State to Accountable Decision
- When Does a Discrepancy Become Actionable? — A Principal-Level Probe for Administrative AI
Fresh current-fact claims return to the direct source that owns the claim. Starting authorities include:
- CMS Interoperability and Prior Authorization Final Rule
- CMS Prior Authorization API FAQs
- NAIC Health Insurance AI/ML Survey Report
- CAQH CORE priority topics and administrative-data exchange
- NIST AI Risk Management Framework
The source set establishes several bounded orientation facts as of the source window: CMS requirements have payer- and provision-specific horizons beginning across 2026 and 2027; the NAIC survey includes 93 responding health insurers under the authority of 16 states; the cited CAQH measure describes one electronic prior-authorization pathway. Each later run reopens the direct authority and preserves its exact population, definition, effective date, and response method.
Evidence hierarchy by claim
Claim | First authority | Boundary to preserve |
Rule or compliance horizon | Current statute, rule text, agency guidance, or state regulator source | Jurisdiction, payer class, provision, effective date, and qualified interpretation |
Payer or product requirement | Current policy, contract, provider manual, or companion guide | Payer, product, delegation state, policy version, and service date |
Transaction or transport state | Governing standard plus the actual transaction and acknowledgement | Receipt and structural validity establish their own states; later business states receive their own evidence |
Coverage, authorization, or adjudication | Governing policy plus accountable determination record | Actor, facility, service, date, scope, and decision authority |
Payment, posting, or recovery | Remittance, EFT or payment record, contract, ledger, and reconciliation | Adjudicated amount, remittance, cash movement, posting, and recovery remain distinct |
Workflow behavior or outcome | Authorized local records, direct observation, independent review, and measured workflow data | Population, time window, operating conditions, workload, recourse, and data owner |
Model or extraction performance | Task-specific benchmark and local evaluation for the receiving population | Model version, task, population, threshold, subgroup, drift, and failure behavior |
Industry adoption | Method-described regulator study, survey, or index | Sample, respondents, measure definition, response frame, and observation period |
When sources disagree, compare jurisdiction, effective date, payer and product specificity, semantic ownership, and revision state. The unresolved decision routes to the accountable receiver.
Execution procedure
1. Bind the run
Record the full parameter block, decision, receiver authority, evidence cutoff, review date, data boundary, and reversible actions available.
2. Open the source spine
Read the public Registry sources, then follow their direct primary sources for the claims in scope. Record source version, effective date, retrieval date, and any material change since the source window.
3. Build the claim inventory
Separate transport, receipt, representation, policy, authorization, adjudication, remittance, payment, posting, appeal, and recovery. Each state receives its own claim and evidence route.
claim:
id: CF-01
question: ""
proposed_class: CF
jurisdiction: ""
payer_or_population: ""
workflow_stage: ""
effective_date: ""
fitting_authority: ""
evidence_needed: ""
decision_relevance: ""4. Retrieve by authority lane
For each claim:
- Retrieve the first fitting authority.
- Record issuer, title, locator, publication date, effective date, version, and retrieval date.
- Capture the exact jurisdiction, payer or population, product, workflow, and channel covered.
- Record whether the source directly establishes, partially supports, contradicts, or leaves the claim open.
- Search for amendments, superseding material, corrections, and implementation guidance.
- Preserve unavailable or access-limited sources as coverage gaps.
Credentials, protected health information, private locators, and raw restricted records remain with their authorized owner. The research return carries sanitized metadata, authorized aggregates, and the accountable route for conclusions that depend on protected evidence.
5. Map state and time
Clock | Events or effective states |
Clinical and scheduling | Considered, ordered, scheduled, performed, changed |
Coverage and policy | Eligibility, benefit, policy version, product, jurisdiction, effective date |
Authorization | Request, receipt, evidence request, review, determination, extension, appeal |
Transaction | Interchange receipt, acknowledgement, reported status, adjudication, remittance |
Contract | Agreement, amendment, fee schedule, delegation, network state |
Payment and reconciliation | EFT or check, posting, matching, variance, recovery |
For every transition, record actor, facility, service, date, source, identifier, observed state, inferred state, decision state, and unresolved gap.
6. Map the administrative-AI layer
Layer | Record for the target workflow |
Source-bearing state | Originating record, version, identity, time, and owner |
Deterministic rules and transactions | Validated rule, standard, transform, acknowledgement, and failure state |
Document perception | Source page or region, extraction method, confidence, correction, and reviewer |
Statistical prediction | Task, model, population, threshold, calibration, subgroup, drift, and fallback |
Language retrieval and synthesis | Corpus, retrieval evidence, citation span, uncertainty, review, and correction |
Agent-like orchestration | Tool authority, state boundary, idempotency, receipt, retry, escalation, and stop |
Accountable decision | Authorized role, decision basis, reason, reversible action, and recourse |
Receipt, recourse, and return | What changed, why, who or what changed it, next route, and owning system update |
Select the smallest layer or combination that supports the workplace outcome while preserving provenance, uncertainty, authority, correction, and recovery.
7. Test discrepancy actionability
For every material difference, produce this chain:
source-bearing difference
→ typed comparison
→ claim classification
→ governing evidence
→ accountable role
→ reversible or authorized action
→ receipt and recourse
→ return to the owning system layerClassify the difference as syntactic, semantic, temporal, jurisdictional, contractual, evidentiary, representational, policy-bearing, workflow-state, or model-induced. State which source owns the disputed meaning, which later state remains open, who can inspect and decide, which action is reversible, and where a confirmed correction returns.
8. Test local hypotheses
Retain only hypotheses that change the receiving decision. Candidate comparisons include:
- full-denominator visibility and deterministic pass sampling versus an exception-only queue;
- source-linked discrepancy review versus the current correction path;
- explicit layer and authority labels versus the current state vocabulary;
- grounded language synthesis with source spans versus the current preparation method.
Each hypothesis records population, comparator, measure, expected direction, threshold, adverse signal, owner, and disconfirming result.
9. Complete the coverage ledger
Dimension | Required bound | Evidence | Status | Gap and next receiver |
Jurisdiction and payer class | covered / partial / open / out of scope | |||
Named payer, product, policy or contract version | ||||
Workflow stage, channel, and effective-date window | ||||
Actor, facility, source, and identifier continuity | ||||
Administrative-AI layer and model or transform version | ||||
Authority, correction, appeal, and recourse | ||||
Population, subgroup, workload, and outcome evidence |
10. Produce the contracted return
Return these sections in order:
- Research return title
- Executive return in five sentences or fewer
- Bound parameters
- Current facts
- Supported transfers
- Practitioner interpretations
- Local hypotheses
- Workflow and layer map
- Discrepancy actionability analysis
- Conflicts, gaps, and coverage ledger
- Decision support and smallest next move
- Stop and reopen receipt
- Sources
- Machine node
Every current fact cites a fitting direct source. Every supported transfer names its receiving conditions. Every hypothesis names its local test and owner.
Stop, pause, and reopen
A run settles when every material claim is classified, source-resolved or explicitly open; effective dates and versions are visible; conflicts reach the appropriate receiver; workflow and layer maps cover the decision-relevant states; the coverage ledger is complete; and the return identifies one bounded next move or bounded indeterminate result.
A pause receipt is fitting when a material scope value is missing, a controlling source remains unavailable or conflicted, protected evidence needs an authorized owner, accountable authority is required, or the next action exceeds the granted research authority. Each pause names the evidence or role that resumes the work.
Reopen at the earliest of:
- the stated review date;
- a federal or state rule change;
- a standard, implementation guide, companion guide, policy, contract, or product-version change;
- a payer, jurisdiction, workflow, channel, population, or accountable-role change;
- new primary or local evidence;
- a local hypothesis result;
- a model, corpus, prompt, tool contract, orchestration path, or authorization-envelope change;
- a correction, appeal, source correction, or provenance event.
run_receipt:
status: "settled | bounded_indeterminate | scoping_pause | authority_pause | evidence_pause"
stopped_on: YYYY-MM-DD
strongest_supported_claim: ""
material_open_claims: []
accountable_receiver: ""
smallest_next_move: ""
review_by: YYYY-MM-DD
earliest_reopen_trigger: ""Machine node
Corpus integrity and provenance
This payload precipitates public Registry sources into an executable research instrument. Industry-level evidence retains its surveyed population and date; architecture guidance retains its transfer boundary; operational claims require local evidence for the named payer, product, workflow, population, and decision.