Architecture 01
Source-linked synthesis
Every statement in a summary resolves to the specific document, section and version it was drawn from. Unsupported sentences do not render.
Athena / Clinical decision support
Every summary keeps its sources attached.
A development-stage program applying the same accountability architecture as HERA Nexus to clinical review. Athena is not a medical device, is not FDA cleared, and is not used in patient care.
PrototypeDevelopmentProduction
Current stage: Development. Athena does not claim FDA clearance or any regulatory submission, medical-device status, clinical validation, diagnostic or treatment capability, deployment in a healthcare facility, HIPAA certification, patient-data access, a clinical partner, or any accuracy, sensitivity or specificity result. No output of this program should inform patient care.
Technical overview
Clinical AI fails on provenance long before it fails on accuracy.
A summary a clinician cannot trace is a summary a clinician has to redo. Athena treats the chain from source document to stated conclusion as the product: every line of a synthesis carries the record it came from, the transformation applied, and the boundary beyond which it should not be relied on. Accuracy work follows provenance work, not the other way round.
Architecture 02
Named accountability
A clinician owns each output before it exists. The system produces a draft for a person, never a conclusion on its own authority.
Architecture 03
Explicit uncertainty
Contradictions between sources, missing records and low-confidence extractions surface in the summary rather than being resolved silently.
Architecture 04
De-identified by construction
Development runs on de-identified retrospective data. No live patient record, identifiable data or care-setting integration is in scope at this stage.
Architecture 05
Refusal boundaries
Diagnosis, treatment recommendation, triage and dosing are out of scope by design, not by policy note. The system declines them.
Architecture 06
Reviewable error analysis
Failures are catalogued by type and cause with the clinician review that caught them, so the limitation set is public before any performance figure is.
Evidence status
What exists—and what must still be proven.
A claim advances only when it has a dated method, test conditions, limitations and a reviewable result.
| Area | Current status | Public interpretation |
|---|---|---|
| Regulatory status | No submission | Athena is not a medical device. No FDA clearance, CE mark, or regulatory submission of any kind exists or is claimed. |
| Clinical use | None | Not used in patient care, not deployed in any healthcare facility, and not evaluated on live patients. |
| Performance | Not validated | No accuracy, sensitivity, specificity, time-saving or error-reduction result is published. |
| Data | De-identified only | No identifiable patient data is processed. HIPAA compliance is not claimed, because no protected health information is handled. |
| Clinical partners | None disclosed | No hospital, health system, clinician group or research institution is represented as a partner. |
| Relationship to HERA | Shared architecture | Athena reuses the evidence-graph and authority-graph patterns. It is a separate program with its own gates, not a HERA feature. |
Development roadmap
Advance through evidence gates.
Scope expands only when technical evidence, safety review, resources and governance support the next stage.
Problem definition
Identify the review tasks where provenance loss is the actual failure, and the record types that carry it.
Evidence gate · CompleteSource-linked synthesis
Implement document ingestion, span-level attribution, contradiction surfacing and refusal boundaries on de-identified retrospective records.
Evidence gate · In progressClinician-in-the-loop error analysis
Have qualified clinicians review outputs against source records, and publish the resulting failure taxonomy and limitation set before any performance figure.
Evidence gateIndependent methodology review
Submit the evaluation protocol, dataset construction and error analysis for external methodological critique.
Evidence gateRegulatory pathway assessment
Determine whether any intended use would constitute a medical device, and what evidence a submission would require, before scoping one.
Evidence gateQualified collaboration
Turn the next technical question into evidence.
Planckchron is seeking clinical informaticists, health-system research groups and regulatory advisors willing to critique the evaluation protocol and the failure taxonomy. The useful conversation at this stage is about method, not about deployment.