Continuous monitoring — NIST SP 800-53 SI-4

Streaming anomaly detection and incident investigation for federal zero trust architectures.

Three-model ensemble scoring, transparent per-model contributions, investigation-first analyst workflow, and OSCAL 1.1.3 continuous monitoring evidence emitted on demand. Built for FedRAMP Moderate and CMMC Level 2.

Anomaly score trend: 0.22 to 0.25, peak 0.70.
3Events per second on the demo deploymentVerify in the trust center
1,840Anomalies detected in the last 24 hoursAudit chain integrity
14NIST 800-53 control families coveredRead the KSI catalog
How it works

Six-tier architecture, end-to-end

Vector edge collectors normalize syslog, OTLP, and HEC payloads into ECS, route through Kafka, and land in a Postgres feature store with TimescaleDB hypertables. A three-model ONNX ensemble scores every event on the hot path. Analysts open every anomaly directly into an investigation panel with pgvectorscale similarity over 100M+ rows and a hash-chained audit log on every action. The reference deployment runs in AWS GovCloud or Azure Government; every tier scales to the platform’s 50,000 events-per-second target.

Reference

Read the deeper architecture

The technical specification at docs/architecture.md is the source of truth. The threat model lives at compliance/threat-models/stride.md and is also rendered at the trust center. Data-flow diagrams are published at compliance/data-flow-diagrams/.

Federal posture

FedRAMP Moderate baseline, CMMC Level 2 aligned

The platform produces an OSCAL 1.1.3 SSP, component definition, and POA&M that validate against the NIST schema. 61 FedRAMP Moderate Key Security Indicators emit continuously and are Cosign-signed with 10-year retention. Audit logs are hash-chained, AU-9 generated occurred-at-canonical, and shipped to S3 with Object Lock in COMPLIANCE mode for production deployments.

Control familyControlsHow SentryGrid implements it
Audit and accountabilityAU-2AU-3AU-6AU-9AU-12Forward-only hash-chained audit log; AU-9 row recompute on verifyChain; S3 Object Lock in COMPLIANCE mode for 10-year retention.
Continuous monitoringCA-7SI-4SI-4(2)SI-4(11)SI-4(13)SI-4(24)OSCAL 1.1.3 evidence emitted continuously; 61 FedRAMP Moderate KSIs Cosign-signed; ensemble traffic-anomaly analysis on the hot path.
Identification and authenticationIA-2AC-2AC-3AC-6Keycloak OIDC against the agency IdP; Casbin RBAC + ABAC at the API gateway; default-deny on every resource type.
System and communications protectionSC-7SC-8SC-13mTLS between services; default-deny egress; FIPS 140-3 validated cryptographic modules in the production overlay.
Configuration managementCM-3Two-party approval queue on rule promotion, model promotion, and retention changes. Enforced at the API layer, not the UI.
Incident responseIR-4IR-5Investigation-first workflow with annotation, escalation chains, and runbook attachment. Every state transition lands in the audit chain.

The public sandbox runs the demo profile — sample data only, magic-link auth, no FedRAMP claim. The production posture overlay (infra/helm/sentrygrid/values.govcloud.yaml) is the reference for customer engagements.

Verify in the trust center
Capability statement

Built for the FY26 cybersecurity priorities

Aligned to OMB / ONCD FY2026 cybersecurity investment priorities and CISA’s Continuous Diagnostics and Mitigation program data quality plan. Independently developed under FedRAMP Moderate baseline controls; FY26 readiness lives in the trust center.

Contact

Federal evaluation inquiries

Federal program managers, primes evaluating subs, and authorizing officials: contact [email protected]. Trust evidence is published continuously at the trust center.