[ FDE-01 · ENTERPRISE DISCOVERY ]SYNTHETIC · PRACTICE ONLY

Diagnose the work.
Then earn the right to build.

An executive asks for an autonomous prior-authorization agent. Reconstruct the workflow, triangulate stakeholder evidence, choose the smallest defensible wedge, and revise the plan when one field fact changes.

18 MINUTESNO ACCOUNTNO REAL DATANO QUALIFICATION EFFECT

FIELD REQUESTHC-PA-SYN-01

“Build an AI agent that completes and submits prior-authorization packets.”

YOUR ACTUAL MISSION

Where does the controllable failure actually occur, what is the least-authority wedge that can test it, and which workflow outcome would justify adoption?

Every organization, person, event, document, control, and figure in this studio is synthetic. The exercise uses no patient data and describes no real payer, provider, policy, or engagement.

[ Formation mission deck ]Sequence · not learner progress

One operating sequence. Different kinds of proof.

The deck keeps the complete formation in view while each chapter moves from provider knowledge into enterprise decisions, production-shaped work, supervised evidence, and independent review.

Deck position

0208

DLive practice

You are viewing · Diagnose

Reconstruct the work before prescribing AI

Where does the workflow actually fail, and which evidence changes the mission?

Produces: Workflow, bottleneck, and outcome contract
Authority: Synthetic browser practice

Deck position shows where this public chapter sits in the authored system. It is not saved learner progress, proof of completion, a cohort record, or a qualification decision.

01InspectACTIVE
02DiagnoseOPEN
03AdaptLOCKED
04ContractLOCKED
05DebriefLOCKED

VISUAL WORKFLOW · CURRENT STATE

Follow the case, not the feature request.

5 DECISIONS · 4 HANDOFFS · 1 OUTCOME

W01
Order created
Ordering clinician
The order, diagnosis, and note exist, but supporting evidence is not packaged for the payer.
SYSTEM
Clinical record
DECISION
Does the request require authorization, and what evidence applies?
W02
Evidence assembled
Prior-authorization coordinator
The coordinator reconciles coverage rules, notes, results, and attachments across systems.
SYSTEM
Work queue + document store
DECISION
Is the packet complete, current, attributable, and ready for clinical release?
W03
Clinical release
Utilization review clinician
A named clinician confirms medical-necessity reasoning and resolves ambiguous evidence.
SYSTEM
Clinical review queue
DECISION
May this bounded packet leave the organization?
W04
Payer submission
Prior-authorization coordinator
Approved packet fields and attachments are entered, then an external tracking identifier is captured.
SYSTEM
External portal
DECISION
Was the exact approved packet received once by the intended destination?
W05
Exception resolved
Cross-functional owner
Requests for more information, denials, and status changes cross operational and clinical ownership.
SYSTEM
Status queue + clinical record
DECISION
Who owns the next move, and what closes the case as an accepted outcome?

STEP 01 · EVIDENCE WALL

The request is one signal. Build the working model from the rest.

Select at least two records you would carry into diagnosis. Strong discovery crosses operational behavior, clinical authority, value definition, and system constraints instead of trusting one fluent narrative.

WORKING EVIDENCE SET00

Selected does not mean proven. Confidence and unresolved questions remain attached to the record.

E01HIGH CONFIDENCEPrior-authorization operationsSynthetic ten-case workflow traceSix specimen cases moved backward from clinical release to evidence assembly because a note, result, or payer-specific requirement could not be traced to an accountable source.Tension: The visible copy-and-paste step is repetitive, but the repeated work begins earlier and crosses team boundaries.Implication: Test the evidence-completeness and ownership loop before optimizing portal entry.

E02HIGH CONFIDENCEUtilization review clinicianSynthetic discovery noteClinical reviewers do not want a model to decide medical necessity. They want a source-linked packet that makes missing and conflicting evidence visible before release.Tension: Faster drafting can increase review burden if source lineage or uncertainty is hidden.Implication: Preserve named clinical authority and make every assisted assertion inspectable.

E03MEDIUM CONFIDENCERevenue-cycle sponsorSynthetic metric-definition reviewThe current dashboard ends at submitted. It does not connect first-pass completeness, requests for more information, rework, review burden, or final disposition.Tension: Submission volume can rise while accepted workflow outcomes and staff burden get worse.Implication: Define a baseline and an adoption-adjusted outcome before setting a target.

E04OPEN QUESTIONSecurity and integrationSynthetic system-and-authority mapClinical evidence remains authoritative in internal systems. The external portal is a destination, not the source of truth, and its supported write authority is unresolved.Tension: A prototype built around a broad user session would prove the wrong identity and action boundary.Implication: Resolve identity, write support, confirmation, and reconciliation before granting action authority.

E05MEDIUM CONFIDENCEExecutive sponsorSynthetic intake requestThe sponsor asks for an autonomous submission agent and expects a visible demonstration quickly.Tension: The requested interface names a solution but does not establish the causal bottleneck, safe authority, or accepted outcome.Implication: Treat the request as one stakeholder belief and translate it into testable mission choices.

STEP 02 · CAUSAL HYPOTHESIS

Where is the first controllable loop worth changing?

Choose a hypothesis that could be disproved. External elapsed time can be real without being the first organization-controlled constraint.

Portal entry effortManual transcription is the dominant controllable constraint, so faster field entry will improve the mission outcome.FIELD TEST: Compare entry effort only after complete packets reach release; check whether saved time survives rework and exception handling.Evidence assembly and handoff reworkCases loop because required clinical evidence, source lineage, and next-owner responsibility are discovered too late across team boundaries.FIELD TEST: Trace ordinary, returned, and exception cases from order through clinical release; measure first-pass completeness and ownership changes.External payer decision latencyElapsed time accumulates after submission, so the payer wait is the primary mission constraint.FIELD TEST: Separate external decision time from organization-controlled queue time, rework, duplicate follow-up, and status ownership.Model answer qualityThe mission is constrained by unreliable generated answers, so model selection should be the first decision.FIELD TEST: First establish which bounded judgment, if any, needs a model and what source-grounded behavior the workflow would accept.

STEP 03 · WEDGE COMPARISON

Choose the least authority that can teach you something true.

Deterministic, assisted-AI, and agentic approaches are design choices—not maturity levels. Compare ambiguity, consequence, reversibility, and evidence gained.

PROVIDER-NEUTRAL CORE

The exercise specifies behavior and authority. It does not select a model vendor, platform, or proprietary feature.

DDeterministic preflightValidate known required fields, attachment presence, identifiers, state transitions, and release prerequisites.BEST WHENRules and inputs are stable, exact, and knowable before execution.AUTHORITYMay block an incomplete packet; never interprets clinical evidence or releases it.EVIDENCE GAINTests how much rework comes from mechanically detectable incompleteness.FAILURE MODECannot reconcile varied narrative evidence or explain clinically meaningful conflicts.ASource-grounded AI assistanceAssemble an inspectable evidence draft, surface conflicts and unknowns, and pair it with deterministic checks before named human release.BEST WHENInputs are varied and unstructured, but consequential judgment and release must stay human-owned.AUTHORITYRead and draft inside an approved boundary; no autonomous medical-necessity decision or external write.EVIDENCE GAINTests whether source-linked assistance improves first-pass completeness without increasing review burden.FAILURE MODEFluent synthesis can hide missing lineage unless citations, abstention, and conflict handling are enforced.GAgentic submission and follow-upAssemble the packet, choose actions, write to the external portal, monitor status, and initiate follow-up.BEST WHENAction interfaces, identity, reversibility, confirmation, and escalation authority are explicitly supported and proven.AUTHORITYWould cross clinical release and external-write boundaries in the requested design.EVIDENCE GAINCould eventually test end-to-end operating burden after the narrower assumptions and controls are proven.FAILURE MODEExcess agency, duplicate or unauthorized writes, hidden clinical judgment, and ambiguous recovery.

Capture the first decision before seeing the field inject. You will be able to preserve or revise it; the decision history remains visible in this browser session only.

STEP 05 · OUTCOME CONTRACT

Make success falsifiable before architecture begins.

Build the decision contract around accepted work—not model volume. Every field matters; unsafe failure cannot be averaged away by a good headline metric.

CONTRACT COMPLETENESS

0/8INCOMPLETE

OC-01Accepted workflow outcomeWhat should improve for the person doing the work?Choose the evidence and authority contract deliberately.OC-02BaselineWhat comparison exists before a target is promised?Choose the evidence and authority contract deliberately.OC-03Target conditionHow is ambition set without inventing a number?Choose the evidence and authority contract deliberately.OC-04Evidence windowWhich evidence can show whether the change held?Choose the evidence and authority contract deliberately.OC-05Accountable ownerWho accepts the bounded workflow result?Choose the evidence and authority contract deliberately.OC-06Non-compensatory guardrailWhich authority cannot be traded for speed?Choose the evidence and authority contract deliberately.OC-07Adoption conditionWhat would meaningful use look like?Choose the evidence and authority contract deliberately.OC-08Kill criteriaWhen must the wedge hold, narrow, or stop?Choose the evidence and authority contract deliberately.

Formative feedback is deterministic against five published FDE discovery and business criteria. It is not a review, score of you, proof of field performance, or qualification decision.

B01Browser-only formative practice; nothing is saved.

B02No account, graph, artifact, review, credential, qualification, employer, or client record is created.

B03A formative score is feedback on this authored decision, not evidence that the learner performed field discovery.

B04The provider-neutral core compares deterministic code, model assistance, agentic action, and human authority without prescribing a model vendor.

Diagnose the work.
Then earn the right to build.

One operating sequence. Different kinds of proof.

Reconstruct the work before prescribing AI

Follow the case, not the feature request.

Order created

Evidence assembled

Clinical release

Payer submission

Exception resolved

The request is one signal. Build the working model from the rest.

Where is the first controllable loop worth changing?

Choose the least authority that can teach you something true.

The proposed action boundary is not available

Preserve the outcome. Change the route when the facts demand it.

Make success falsifiable before architecture begins.