ICI Self Lab symbolICI SELF LAB
ICI · INDEXICAL CAUSAL INTEGRATION · FINAL V3 ARCHITECTURE

One causal intelligence.
Multiple assurance applications.

ICI — Indexical Causal Integration — is the causal-assurance architecture. ICI FINAL V3 is its shared causal-intelligence core. Sentinel RC2 and four V3-native applications adapt different evidence and operational contexts into the same structural causal state instead of implementing separate causal algorithms.

SHARED CAUSAL LOOP

The same reasoning machinery sits underneath the applications.

The core separates evidence from interpretation, maintains competing structural explanations, uses intervention to discriminate among them, and retains causal lessons only when they survive testing.

Observe

Receive evidence, events and state without collapsing them into conclusions.

Build posterior

Represent competing structural mechanisms and their current probabilities.

Falsify

Search for evidence or interventions capable of breaking the leading explanation.

Select intervention

Prefer the changed condition with the highest expected information gain under budget.

Update

Use the observed intervention outcome to update the structural posterior.

Abstain or decide

Do not force a root-cause claim when the evidence remains non-identifying.

Retain mechanism

Store the causal lesson rather than only the corrected output.

Re-test

Challenge the retained mechanism under changed surface conditions.

Audit evidence

Keep provenance, observed state and inferred state reviewable.

Return domain action

Translate one causal state into the next useful action for the application.

APPLICATION STACK

One core, five operational surfaces.

Sentinel focuses on runtime agent assurance. The other four applications are domain adapters over the shared ICI FINAL V3 application core.

SHARED CAUSAL INTELLIGENCEICI FINAL V3structural posterior · falsification · expected information gain · abstention · selective causal memory
RUNTIME ASSURANCEICI Sentinel RC2

Causal trajectory, provenance, operational-state continuity and scoped tool authority before action.

V3 NATIVEReliability

Competing failure mechanisms and the next causal intervention with the highest information value.

V3 NATIVEInspector

Evidence structure and the next-best inspection capable of discriminating between causal explanations.

V3 NATIVEIncident Memory

Retrieval by causal fingerprint rather than by wording, incident title or keyword similarity.

V3 NATIVELearning

Verified causal transfer using retained memory against a no-memory control under changed conditions.

CORE VS ADAPTERS

What is shared, and what changes by domain.

Shared by every V3-native application

  • Structural posterior over competing mechanisms
  • Evidence/provenance discipline
  • Expected-information-gain intervention selection
  • Abstention when evidence is not identifying
  • Selective causal memory and changed-condition re-test

Changed by each domain adapter

  • Which evidence is mapped into causal factors
  • Which intervention is operationally meaningful
  • How results are expressed to the user
  • Which safeguards and human authority apply
  • Which external validation task is appropriate
RUNTIME CONTRACT

The public applications expose the same core identity.

The production API includes a read-only synthetic health check that executes Reliability, Inspector, Incident Memory and Learning through the shared application core and verifies the architecture contract.

ARCHITECTURE1 COREONE_CAUSAL_INTELLIGENCE_FOUR_APPLICATIONS
APPLICATION COREV3ICI-FINAL-V3-APPLICATION-CORE-1
HEALTH CONTRACT/api/ici-v3/healthread-only synthetic runtime check
PRODUCTION TELEMETRY BRIDGE

OpenTelemetry traces can now enter Sentinel’s causal audit path directly.

The production boundary accepts authenticated OTLP/HTTP JSON at POST /v1/traces. Traces are normalized into ICI telemetry events, materialized into existing Sentinel events and can be audited by the existing orchestrator without importing OpenTelemetry or vendor-specific SDKs into the causal core.

INGEST/v1/tracesOTLP/HTTP JSON · Bearer authentication required
CAUSAL SEMANTICSEXPLICITparentSpanId becomes causal provenance only with ici.causal.parent=true
AUDIT PATHNATIVEOTLP → ICITelemetryEvent → SentinelEvent → auditOrchestrator()
COLLECTORSREADYreference fan-out for Prometheus and Datadog

Security properties

  • Bearer token required; deployments without a token fail closed with 503
  • Unauthorized senders rejected with 401
  • Default JSON payload cap: 1 MB
  • Successful responses do not echo telemetry content
  • Evidence spans without explicit provenance are rejected

Vendor-neutral observability

  • OpenTelemetry Collector reference configuration
  • Prometheus span metrics exposed on port 9464
  • Datadog OTLP fan-out with cumulative-to-delta metrics conversion
  • No Prometheus or Datadog code introduced into the causal core
  • Changing observability vendor does not change Sentinel causal logic
WHAT THIS MEANS

Sentinel is no longer limited to synthetic trace construction or UI-local audit inputs. It now has a production telemetry ingestion boundary capable of converting instrumented agent traces into deterministic causal-audit input while preserving an explicit distinction between observed trace topology and verified causal provenance.

PUBLIC BOUNDARY

This page documents architecture and observable contracts, not proprietary source implementation. The frozen scientific core, application adapters and public runtime evidence are distinct from independent external validation. No statement here claims universal immunity to failure or universal causal identification.