SCRIMED Governance + Learning Loop
Governance Is the New Competitive Advantage.
SCRIMED converts traces, reviews, policies, corrections, retests, regulatory watch, and value metrics into a governed operating loop. It remains synthetic/no-PHI and human-reviewed.
Memory Is Not Learning
Memory stores what happened; learning improves the next attempt only after review.
The trace stored what happened: an agent attempted a clinical-facing output after reading untrusted content.
The next attempt must require human review, evidence grounding, prompt-injection risk elevation, and retest before artifact promotion.
Artifact hash: scrimed-intel-8bbedab3
Governance Loop
Observe, evaluate, correct, approve, update, retest, and monitor before SCRIMED promotes changes.
Memory artifact
observe
Capture metadata-only traces, policy decisions, reviewer notes, route outcomes, and safety flags.
Human review and audit remain in the loop.
Evaluation packet
evaluate
Score schema fidelity, evidence quality, safety, workflow value, and reviewer disposition.
Human review and audit remain in the loop.
Correction artifact
correct
Create a correction artifact only after the failure mode is understood.
Human review and audit remain in the loop.
Approval record
approve
Require human approval before changing prompts, policies, workflows, demos, or buyer-facing claims.
Human review and audit remain in the loop.
Versioned artifact
update artifact
Update the prompt, policy, workflow, registry, benchmark, or documentation artifact.
Human review and audit remain in the loop.
Retest evidence
retest
Run contract, smoke, benchmark, and regression checks before promotion.
Human review and audit remain in the loop.
Monitoring signal
monitor
Track drift, cost, latency, reviewer outcomes, safety flags, and buyer-value metrics.
Human review and audit remain in the loop.
Agent Policy Engine
Every agent-facing action routes through risk, permission, review, and audit logic.
allow
public_safe
Synthetic, public-safe, no-PHI educational or governance content.
Audit hash and boundary statement.
require_review
review_required
Clinical-facing, confidential, external-action, or high-cost work that stays inside no-PHI limits.
Human approval before external use or artifact promotion.
deny
blocked
PHI export, payer submission, EHR writeback, diagnosis, treatment, prescribing, or irreversible production action.
Fail closed and route to boundary-release approvals.
A2A + MCP Interoperability
SCRIMED uses governed handoffs and scoped tool access instead of direct model-to-system control.
Allow SCRIMED agents to hand off tasks with identity, scope, risk, memory summary, and approval state.
Expose governed tool access through scoped middleware instead of direct agent-to-system access.
No production connector approval, raw schema exposure, payer submission, EHR writeback, or live PHI access.
Value-Based Pricing
SCRIMED prices by workflow value, governance readiness, and buyer outcomes, not token vanity metrics.
denial-risk-reduction
Denial risk reduction
Find missing documentation, policy-risk language, and appeal-readiness gaps before submission.
Price by recovered staff time, reduced rework, and prevented avoidable denials.
- Measured with: Synthetic prior-auth and documentation-before-authorization scorecards.
- Boundary: No payer submission or reimbursement guarantee.
documentation-time-saved
Documentation time saved
Reduce review, packet assembly, and handoff effort for human-reviewed workflows.
Price by workflow volume, role burden, and validated time-saved estimate.
- Measured with: No-PHI task timing, reviewer disposition, and acceptance criteria.
- Boundary: No autonomous chart filing, diagnosis, treatment, or EHR writeback.
workflow-throughput
Workflow throughput
Increase referral, scheduling, intake, and care-coordination visibility without live patient automation.
Price by routed workflow count, dashboard readiness, and governance overhead reduced.
- Measured with: Synthetic queue, status, and escalation metrics.
- Boundary: No patient outreach or live clinical action.
governance-readiness
Governance readiness
Give leadership an audit-ready operating model for AI review, safety, and evidence.
Price as recurring readiness retainer, evidence packet, or governance implementation work.
- Measured with: Policy coverage, smoke coverage, boundary register, and audit hash completeness.
- Boundary: No certification, legal approval, security assurance, or customer go-live claim.
Clinical Workflow Foundations
Radiology and wearables stay workflow-focused, specialist-reviewed, and synthetic-only.
lab_only_workflow_metadata
Radiology AI: Imaging Insight to Action
Convert synthetic imaging workflow signals into routing, follow-up, documentation, and specialist-review readiness without final interpretation.
No final medical interpretation, diagnosis, treatment, or autonomous patient outreach.
- flag delayed imaging turnaround
- route to human imaging reviewer
- prepare follow-up checklist
- bind provenance to synthetic DICOM metadata
- track closed-loop follow-up status
lab_only_synthetic_time_series
Wearables Intelligence Foundation
Prepare future RPM and wearable-data workflows around trends, access, adherence, and review queues.
No emergency triage replacement, diagnosis, prescribing, patient outreach, or live-monitoring claim.
- device connectivity gap
- trend anomaly
- adherence risk
- accessibility barrier
- review threshold reached
Regulatory Watch
SCRIMED tracks regulatory signals as readiness inputs, not approval claims.
watch
HIPAA privacy and security readiness
Readiness monitoring only; no certification claim.
BAA readiness, minimum necessary principle, audit trail, de-identification, retention
watch
HTI-6 and ONC interoperability signals
Product roadmap and evidence-gap tracking.
FHIR APIs, USCDI alignment, algorithm transparency, information blocking risk
watch
FHIR and identity rules
Synthetic validation and architecture planning.
FHIR R4/R5 readiness, SMART launch, patient matching, HHS identity policy
watch
OCR and HHS enforcement signals
Governance watch only; legal review required before claims or production use.
privacy enforcement, AI guidance, cybersecurity guidance, patient access rules
Activated SCRIMED Skills
Operational skills convert strategy into repeatable, audited build motion.
active
Governance Skill
Route every sensitive SCRIMED action through policy, review, audit, and retained boundaries.
Bind governance decisions to every future clinical-facing route and demo artifact.
- Maturity: review_ready
- Outputs: policy decision, risk score, audit event
- Guardrails: No live PHI or source patient records. No autonomous clinical decisioning. No diagnosis, treatment, prescribing, or final triage authority. No EHR writeback, order entry, chart mutation, or payer submission. Clinical-facing outputs stay decision support, education, workflow support, or administrative automation. Human review, audit evidence, and escalation remain required for clinical-facing and irreversible workflows.
active
Agent Learning Skill
Turn reviewed failures into correction artifacts, retests, and monitored improvements.
Promote failed synthetic traces into reusable regression cases.
- Maturity: lab_ready
- Outputs: correction artifact, regression test candidate, learning update
- Guardrails: Memory is not learning until evaluated, approved, retested, and monitored.
active
Clinical Safety Skill
Keep clinical-facing content in decision-support and human-review lanes.
Map every clinical module to reviewer role, escalation reason, and benchmark domain.
- Maturity: review_ready
- Outputs: review requirement, red flag, boundary statement
- Guardrails: No live PHI or source patient records. No autonomous clinical decisioning. No diagnosis, treatment, prescribing, or final triage authority. No EHR writeback, order entry, chart mutation, or payer submission. Clinical-facing outputs stay decision support, education, workflow support, or administrative automation. Human review, audit evidence, and escalation remain required for clinical-facing and irreversible workflows.
lab_only
Regulatory Watch Skill
Track HIPAA, HTI-6, FHIR, OCR, HHS identity, and adjacent AI governance signals.
Attach owners and update cadence for each watch domain.
- Maturity: defined
- Outputs: watch item, owner, roadmap implication
- Guardrails: Legal review required before claims, policy changes, or production release.
active
AI Visibility Skill
Help AI search engines describe SCRIMED accurately using public boundaries and canonical routes.
Add structured page metadata recommendations to high-value public routes.
- Maturity: lab_ready
- Outputs: AI-safe overview, product map, description guardrails
- Guardrails: No autonomous clinical care framing, no certification claim, no live PHI claim.
active
Value Pricing Skill
Tie price to measurable workflow value rather than token usage or generic chatbot access.
Create audience-specific pricing proof packets for clinics, health systems, and strategic partners.
- Maturity: review_ready
- Outputs: pricing signal, pilot offer, margin guardrail
- Guardrails: No ROI guarantee, revenue guarantee, reimbursement assurance, or audited financial claim.
lab_only
Radiology Workflow Skill
Convert imaging workflow metadata into follow-up, routing, documentation, and review readiness.
Bind imaging-to-action scenarios to DICOM metadata, specialist review, and closed-loop follow-up status.
- Maturity: defined
- Outputs: workflow action recommendation, review requirement, audit event
- Guardrails: No final imaging interpretation or diagnostic authority.
lab_only
Wearables Intelligence Skill
Prepare no-PHI trend, adherence, access, and escalation workflows for wearable and remote monitoring data.
Define synthetic time-series scenarios and human escalation thresholds.
- Maturity: defined
- Outputs: review signal, accessibility note, workflow escalation
- Guardrails: No diagnosis, emergency triage replacement, or patient outreach automation.
active
Interoperability Skill
Keep A2A, MCP, FHIR, HIE, and identity integration plans governed and connector-safe.
Map MCP tools to explicit per-agent authorization scopes.
- Maturity: lab_ready
- Outputs: readiness status, blocked action, integration boundary
- Guardrails: No production connector activation or raw system-of-record access.
active
Observability Skill
Trace cost, latency, tool calls, policy events, reviewer outcomes, and safety signals.
Attach learning-loop stages to AI Flight Recorder outputs.
- Maturity: review_ready
- Outputs: trace summary, audit hash, monitoring signal
- Guardrails: No token, secret, PHI, or raw connector payload logging.
active
DevSecOps Skill
Protect the build with contract checks, smoke tests, boundary scans, and route integrity.
Add governance-learning loop to nonsecret and public smoke coverage.
- Maturity: review_ready
- Outputs: pass/fail evidence, blocker, release note
- Guardrails: No deployment promotion without explicit human approval.
active
Investor Narrative Skill
Translate governance, safety, workflow value, and platform architecture into a diligence-ready story.
Build a founder run-of-show connecting value pricing, governance, and proof routes.
- Maturity: review_ready
- Outputs: pitch point, demo path, safe claim
- Guardrails: No securities advice, valuation assurance, certification claim, or customer go-live claim.
Clinical Boundaries
Research/demo use only. SCRIMED supports human-reviewed decision support, education, workflow support, and administrative automation only.
- No live PHI or source patient records.
- No autonomous clinical decisioning.
- No diagnosis, treatment, prescribing, or final triage authority.
- No EHR writeback, order entry, chart mutation, or payer submission.
- Clinical-facing outputs stay decision support, education, workflow support, or administrative automation.
- Human review, audit evidence, and escalation remain required for clinical-facing and irreversible workflows.