Drawer 1
Data
Sources, fields, capture and cutoff dates, row counts, missingness and transformations.
Platform
A governed path from raw data to a replayable analytic output
RISX Science turns the algorithms developed for each use case into bespoke, governed analytic agents—configured around each client's data, clinical priorities, workflows, and economic objectives. Each layer constrains the one above it. A workflow can only show what an engine produced, an engine can only run on the governed substrate, and the substrate only holds facts whose state and time are explicit.
Layer 1 — Evidence foundation
The evidence bases RISX works from today
Analytic work starts from published, inspectable evidence: guideline logic, population benchmarks, program and utilization context, screening eligibility frameworks and the peer-reviewed record.
Evidence foundation
Public evidence bases in use now, and what each contributes.
| Source | Role | How it is used |
|---|---|---|
| NCCN guidelines | Evidence-based care pathways and clinical review criteria | Pathway definitions, protocol-defined review reasons and permitted next steps are written against published guideline logic rather than invented rules. |
| SEER / NCI | Cancer incidence, stage distribution, survival and population benchmarks | Population structure, expected stage mix and survival benchmarks used to calibrate modeled cohorts and sanity-check analytic outputs. |
| CMS public data | Utilization, quality, coverage and program-design context | Utilization and cost context, quality-program structure and coverage design inform economic modeling and program feasibility work. |
| CDC / USPSTF | Prevention and screening eligibility frameworks | Eligibility logic for screening and prevention pathways, applied as explicit, inspectable criteria with their source recorded. |
| Published literature and registries | Technology assessment, evidence synthesis and outcome hypotheses | Effect sizes, comparator choices and outcome definitions are drawn from the published record and carried through as declared assumptions. |
Client-specific claims, clinical records, and laboratory feeds are connected only under the appropriate data-use, privacy, and governance agreements.
Layers 2–5
Governed evidence substrate, engines, workflows, outputs
- Governed evidence substrateBitemporal facts (valid time and transaction time), snapshots for reproducibility, protocol records, and de-identified analytic tokens isolated from identity crosswalks. Fact states are explicit: OBSERVED, INFERRED, CONTRADICTED, STALE, UNKNOWN, REFUSED. Missing is never converted to negative.
- Analytic enginesTyped, versioned contracts for phenotype detection, event prognosis, conditional intervention benefit and external clinical/control arm. Engines run server-side only; the browser performs no clinical calculation.
- Controlled workflowsPayer, clinical and research surfaces bind an engine output to a permitted use. Evidence and policy stay separate. No automated denial, no autonomous member contact, and no member alert originating from an ECA analysis.
- Replayable outputsArtifacts, model runs and audit records are written to the governance registry with environment gates and kill switches. Outputs are either replayable with a manifest or refused with a code.
Technical note: ECA-Agent. ECA-Agent is the RISX Science method and packaging standard by which a validated analytic use case becomes a bespoke, governed analytic agent inside a client environment: a locked, versioned artifact combining a diagnostic model, a prognostic model, a conditional causal component where justified, and an economic decision policy, with monitoring and controlled retraining. It is not a chatbot and not an autonomous clinical decision-maker. ECA-Agent: Analytic Methodology and Deployment Lifecycle is restricted; authorized technical, reviewer and administrator accounts may sign in to read it.
Inspectability
The same five drawers on every result screen
No one should have to ask where a number came from. Each analytic screen exposes the drawers in a fixed order.
Drawer 2
Method
Question, cohort definition, comparator, estimand, censoring and statistical method.
Drawer 3
Rules & models
Rule version, artifact version, environment, validation status and kill-switch state.
Drawer 4
Validity
Support, calibration, balance and overlap, positivity, uncertainty and refusal gates.
Drawer 5
Output & replay
Result or refusal, limitations, approvals, digest and reproducibility manifest.
Operations
Runtime posture
Health and readiness endpoints, structured logging, per-product kill switches and environment promotion gates.
Current execution posture
Engine availability
| Product | Availability | Restriction |
|---|---|---|
| Phenotype detection (p_osa) | Server-side reference engine available | Synthetic artifacts only |
| Event prognosis (r_18) | Server-side reference engine available | Synthetic artifacts only |
| Conditional intervention benefit | Contract implemented | Refuses — no validated indication artifact |
| External clinical/control arm (ECA) | Target-trial workbench implemented | Refuses — no canonical run output |
Validation restrictions
| Item | State |
|---|---|
| Validated artifacts | NONE — every registered artifact is NOT VALIDATED |
| Permitted environments | local and test only for synthetic artifacts |
| Staging / production execution | Refused by the registry environment gate |
| Member alerts from ECA | Never permitted |
| Automated denial or member contact | Disabled |
Every registered artifact in this build is NOT VALIDATED. Staging and production execution are refused by the registry environment gate.