New SCRIMED intelligence modules staged as governed, no-PHI platform capabilities.

SCRIMED ClinicalBench

• Capabilities: Synthetic scenario library, Clinical readiness scorecards, Regression suites by module and agent, Reviewer-calibrated pass/fail rubrics
• Inputs: synthetic cases, expected safe behaviors, evidence references, reviewer rubrics
• Outputs: clinical readiness scores, failure reports, release gate status, review queue items
• Dependencies: Clinical Robustness Lab, Evaluation Registry, Trust Score, Benchmark Studio
• Allowed demo mode: Run no-PHI clinical benchmark scorecards for demos, pilots, investor diligence, and release evidence.
• Blocked production mode: No live PHI, autonomous diagnosis, treatment, prescribing, patient outreach, payer submission, EHR writeback, production connector use, clinical validation, certification, reimbursement claim, or customer go-live.
• First milestone: Bind ClinicalBench scenarios to each SCRIMED clinical module and require passing synthetic scorecards before buyer-facing expansion.

SCRIMED Scientific Reasoning Engine

• Capabilities: Hypothesis decomposition, Evidence quality grading, Contradiction detection, Guideline and literature comparison
• Inputs: research question, literature references, guideline references, population criteria
• Outputs: evidence matrix, contradiction report, confidence summary, human-review recommendation
• Dependencies: Evidence Graph, ResearchOps, Research Memory, Multi-Model Router
• Allowed demo mode: Run synthetic or public-reference research simulations with citations and uncertainty labels.
• Blocked production mode: No clinical trial enrollment decision, treatment recommendation, patient-specific research advice, or claim of scientific validation.
• First milestone: Create a no-PHI research reasoning packet format with claim, evidence, contradiction, confidence, and reviewer fields.

SCRIMED Workflow Planner

• Capabilities: Workflow graph drafting, Agent/tool sequencing, Approval gate insertion, Rollback and fallback planning
• Inputs: workflow goal, policy constraints, agent registry, tool registry, risk level
• Outputs: workflow plan, approval map, rollback path, audit event plan
• Dependencies: Clinical Work Graph, Policy Registry, Multi-Agent Runtime, Adaptive Workflow Selector
• Allowed demo mode: Draft no-PHI workflow plans and review packets for demos, operations, and protected pilots.
• Blocked production mode: No autonomous workflow execution, connector mutation, patient outreach, payer submission, or EHR writeback.
• First milestone: Create a workflow-plan schema with deterministic states, required approvals, evidence bindings, and rollback behavior.

SCRIMED Clinical Judgment Engine

• Capabilities: Evidence sufficiency check, Uncertainty and limitation surfacing, Escalation criteria generation, Reviewer disposition support
• Inputs: synthetic clinical context, evidence cards, policy rules, reviewer rubric
• Outputs: judgment support packet, risk label, confidence and limitations, escalation criteria
• Dependencies: ClinicalBench, Evidence Graph, Clinical QA Engine, Trust Score
• Allowed demo mode: Prepare synthetic, reviewer-held judgment support packets that never become clinical decisions.
• Blocked production mode: No autonomous diagnosis, treatment recommendation, prescribing, triage replacement, or patient-specific medical advice.
• First milestone: Add a clinical judgment packet schema with explicit not-for-care boundary, evidence sufficiency, and reviewer signoff state.

SCRIMED Evidence Graph

• Capabilities: Claim-source graph, Citation verification, Conflict and contradiction edges, Evidence provenance and freshness tracking
• Inputs: documents, guidelines, evidence cards, clinical or research claims
• Outputs: evidence graph, citation map, conflict graph, source freshness report
• Dependencies: Secure RAG/Data Ingestion Pipeline, Scientific Reasoning Engine, Trust Score
• Allowed demo mode: Build synthetic or public-reference evidence graphs for demos and diligence.
• Blocked production mode: No live PHI indexing, unsupported clinical claims, source-less recommendations, or clinical validation claims.
• First milestone: Define graph node and edge schemas for claim, source, citation, guideline, conflict, freshness, and review status.

SCRIMED ResearchOps

• Capabilities: Research workflow queue, Literature and guideline tasking, Trial matching simulation, Research reviewer handoff
• Inputs: hypothesis, public literature references, trial criteria, population definition
• Outputs: research packet, evidence ranking, trial simulation report, reviewer task
• Dependencies: Scientific Reasoning Engine, Evidence Graph, Research Memory, LongTask Runtime
• Allowed demo mode: Run no-PHI research operations simulations with public references and human-review gates.
• Blocked production mode: No patient recruitment, enrollment guarantee, trial recommendation, IRB bypass, or patient-specific research action.
• First milestone: Create a research pipeline state machine from hypothesis to evidence ranking to reviewer packet.

SCRIMED LongTask Runtime

• Capabilities: Durable task state, Checkpoint and resume, Human approval pause points, Timeout, retry, and dead-letter handling
• Inputs: task plan, agent identity, policy decision, approval requirements
• Outputs: task trace, checkpoint state, review packet, final artifact hash
• Dependencies: Event Mesh, Zero-Trust Audit Ledger, Multi-Agent Runtime, Workflow Planner
• Allowed demo mode: Run no-PHI long tasks for evidence packet generation and synthetic evaluations.
• Blocked production mode: No unattended production mutation, connector write, patient outreach, payer submission, or EHR writeback.
• First milestone: Create a durable task contract with checkpoint, pause-for-review, resume, retry, timeout, and dead-letter states.

SCRIMED Clinical Work Graph

• Capabilities: Workflow dependency graph, Risk-aware node labeling, Review-state tracking, Blocked-action visibility
• Inputs: workflow plan, clinical policy, evidence refs, review requirements
• Outputs: clinical work graph, dependency map, review-state view, blocked-action map
• Dependencies: Workflow Planner, Clinical QA Engine, Policy Registry, Event Mesh
• Allowed demo mode: Visualize synthetic clinical workflows and dependencies with no live-care authority.
• Blocked production mode: No autonomous clinical workflow execution or production connector action.
• First milestone: Define a graph schema for work nodes, evidence refs, owner, risk, state, approval, and rollback.

SCRIMED Trust Score

• Capabilities: Trust score rubric, Evidence quality scoring, Reviewer outcome weighting, Release gate recommendation
• Inputs: eval results, evidence graph, reviewer dispositions, policy decisions, incident events
• Outputs: trust score, risk band, release recommendation, improvement tasks
• Dependencies: ClinicalBench, Continuous Evaluation Platform, Zero-Trust Audit Ledger, Evidence Graph
• Allowed demo mode: Show synthetic trust scores for modules, agents, and outputs in diligence packets.
• Blocked production mode: No trust score may be treated as clinical validation, certification, or autonomous approval.
• First milestone: Create a deterministic trust score formula with explainable components and release-gate thresholds.

SCRIMED Continuous Evaluation Platform

• Capabilities: Scheduled eval runs, Regression detection, Release gate blocking, Reviewer calibration tracking
• Inputs: eval suites, module versions, agent versions, prompt versions, model routes
• Outputs: eval report, regression alert, release block, review task
• Dependencies: Evaluation Registry, ClinicalBench, Benchmark Studio, Trust Score
• Allowed demo mode: Run no-secret, no-PHI eval suites for synthetic module readiness.
• Blocked production mode: No synthetic eval result may claim live clinical validation or certification.
• First milestone: Add module-version eval mapping and require continuous eval status in release evidence.

SCRIMED Outcome Intelligence

• Capabilities: Outcome signal modeling, Reviewer outcome linkage, Workflow impact analysis, Safety and quality trend detection
• Inputs: approved outcome signals, reviewer labels, workflow events, quality metrics
• Outputs: outcome insight, trend report, risk signal, improvement recommendation
• Dependencies: Continuous Outcome Learning Engine, Event Mesh, Clinical Work Graph, Trust Score
• Allowed demo mode: Use synthetic or approved de-identified outcome simulations for value analysis.
• Blocked production mode: No live patient outcome learning without legal, privacy, clinical, consent, and customer approval.
• First milestone: Define outcome signal schema with consent/data-use status, de-identification status, evidence link, and reviewer owner.

SCRIMED Knowledge Evolution Engine

• Capabilities: Knowledge versioning, Guideline freshness tracking, Change proposal workflow, Regression-triggered knowledge review
• Inputs: evidence graph changes, guideline updates, eval failures, reviewer feedback
• Outputs: knowledge update proposal, freshness report, deprecation notice, release task
• Dependencies: Evidence Graph, Research Memory, Clinical Memory, Continuous Evaluation Platform
• Allowed demo mode: Show no-PHI knowledge update proposals and freshness dashboards.
• Blocked production mode: No automatic clinical knowledge update may affect live care without governance approval.
• First milestone: Create knowledge version records with source, owner, policy, eval impact, reviewer, and rollback fields.

SCRIMED Multi-Agent Runtime

• Capabilities: Agent orchestration, Supervisor checks, Role-scoped tool access, Run trace and replay
• Inputs: workflow plan, agent identities, tool scopes, policy decisions, evaluation requirements
• Outputs: agent run trace, handoff map, supervisor verdict, review packet
• Dependencies: Agent Identity Registry, Secure MCP Gateway, LongTask Runtime, Trust Score
• Allowed demo mode: Run synthetic multi-agent plans and review traces without production side effects.
• Blocked production mode: No autonomous protected workflow execution or production tool access.
• First milestone: Define multi-agent run envelope with planner, specialist, verifier, supervisor, cost, policy, and trace fields.

SCRIMED Multi-Model Router

• Capabilities: Provider-neutral routing, Risk-tier model selection, Fallback and circuit breaker, Cost and latency telemetry
• Inputs: task type, risk level, privacy boundary, budget, latency target, quality requirement
• Outputs: model route decision, routing rationale, fallback plan, usage telemetry
• Dependencies: Model Registry, AI Cost Intelligence Agent, Trust Score, Policy Registry
• Allowed demo mode: Route synthetic tasks through provider-neutral metadata and synthetic fallback provider.
• Blocked production mode: No unapproved external provider calls, PHI routing, or hard-coded single-provider dependency.
• First milestone: Bind module-level task classes to the existing SCRIMED model router and cost guardrails.

SCRIMED Adaptive Workflow Selector

• Capabilities: Workflow template matching, Risk-aware routing, Missing-data-aware fallback, Reviewer capacity consideration
• Inputs: task request, policy decision, risk label, data completeness, reviewer availability
• Outputs: selected workflow, fallback workflow, review queue, selection rationale
• Dependencies: Workflow Planner, Clinical Work Graph, Policy Registry, Trust Score
• Allowed demo mode: Select synthetic workflow templates and explain rationale for demos and pilots.
• Blocked production mode: No autonomous protected workflow selection that executes side effects without review.
• First milestone: Create workflow selection rules with risk gates, missing-data fallbacks, and reviewer queue constraints.

SCRIMED Clinical Memory

• Capabilities: Purpose-bound memory, Source attribution, Retention and deletion policy, Clinical context recall with review
• Inputs: approved clinical context, evidence refs, purpose-of-use, reviewer approval
• Outputs: memory reference, source trace, retrieval packet, deletion event
• Dependencies: Context Engine, Policy Registry, Evidence Graph, Zero-Trust Audit Ledger
• Allowed demo mode: Use synthetic clinical memory references only.
• Blocked production mode: No live PHI memory without consent, BAA/DPA, retention, deletion, audit, and customer approval.
• First milestone: Define a no-PHI memory record with source, purpose, scope, TTL, reviewer, and deletion event fields.

SCRIMED Research Memory

• Capabilities: Research trail memory, Evidence and contradiction recall, Search strategy versioning, Reviewer outcome linkage
• Inputs: research packet, literature references, evidence grades, reviewer outcomes
• Outputs: research memory reference, evidence lineage, search replay packet, update task
• Dependencies: ResearchOps, Scientific Reasoning Engine, Evidence Graph, Knowledge Evolution Engine
• Allowed demo mode: Use public-reference or synthetic research memory for simulations.
• Blocked production mode: No patient-specific research memory or trial action without approval.
• First milestone: Create research memory schema with hypothesis, sources, search method, contradictions, reviewer, and update status.

SCRIMED Physician Preference Memory

• Capabilities: Preference capture, Scope and consent tracking, Preference-aware drafting, Preference audit and reset
• Inputs: clinician-approved preference, scope, version, review owner
• Outputs: preference memory reference, drafting hint, audit event, reset event
• Dependencies: Clinical Memory, Policy Registry, Audit Ledger, Prompt Registry
• Allowed demo mode: Use synthetic preference examples to tailor draft formatting and workflow routing.
• Blocked production mode: No hidden preference learning, clinical decision override, patient-specific action, or unapproved clinician profiling.
• First milestone: Define opt-in preference memory with scope, owner, TTL, version, reset, and audit fields.

SCRIMED Benchmark Studio

• Capabilities: Benchmark authoring, Scenario set comparison, Model/agent/prompt comparison, Release evidence export
• Inputs: benchmark definition, scenario suite, model route, agent version, prompt version
• Outputs: benchmark report, comparison chart, release evidence packet, regression alert
• Dependencies: ClinicalBench, Continuous Evaluation Platform, Model Registry, Prompt Registry
• Allowed demo mode: Run and display no-PHI internal benchmarks for diligence and release review.
• Blocked production mode: No benchmark result may claim certification, clinical validation, or live-care approval.
• First milestone: Create benchmark definition schema with scenario set, run metadata, score rubric, comparison targets, and release packet export.