Methodology // ContextGraphs

Reconciling
Context Graphs.

Enterprise agents inherit messy, contradictory knowledge. This is how domain experts clean it up - resolving conflicts in tribal knowledge and turning SOPs into a single source of truth.

The Missing Layer

Context Graphs Need Curators.

The industry is building context graphs—capturing decision traces, surfacing patterns, encoding institutional memory. This is necessary. It is not sufficient.

Knowledge foundations aren't assembled. They're *curated*. Every policy conflict, every ambiguous guideline, every edge case requires a human decision: What is true? What takes precedence? What do we codify?

Domain expertise doesn't sit alongside the graph. It sits *inside* it—at every node where judgment was required.

Human-in-the-Loop

The Three Gates

Reconciliation requires human judgment at three distinct points. These are not optional. They are the gates that ensure quality.

Gate 0

Knowledge Validation

Is the book internally consistent?

Expert
Domain / Clinical
Output
Gold Standard KB
Gate 1

Pattern Curation

Is this discrepancy signal or noise?

Expert
Clinical / Ops
Output
Curated Insights
Gate 2

Policy Decision

Should this update the Gold Standard?

Expert
Policy / Compliance
Output
Reconciled Policy
Note: These gates cannot be fully automated. They represent liability decisions.
System Overview

The Reconciliation Architecture

Layer 1: Construction

Ingest & Reconcile

Phases 1-3. We take conflicting policies and tribal knowledge, identify the delta, and merge them into a unified base.

Layer 2: Verification

Regulatory Check

Phase 4. The unified base is stress-tested against CMS guidelines and state mandates to create the Gold Standard.

Layer3: Recursion

The Feedback Loop

Phase 5. Agents execute the knowledge. Exceptions are audited by experts and fed back into Layer 1 to refine the rules.

System Architecture:
V5.0 Active
SYS.ARCH // RECURSIVE_KNOWLEDGE_LOOP_v5.1
FEEDBACK_ACTIVE
Structured
RAW INPUT
Conflicting Policies
Unstructured
SLACK / EMAIL
Tribal Knowledge
PHASE 1
CLEANING
SOP Resolution
PHASE 2
VALIDATION
Tribal Curation
PHASE 3
MERGE & CODIFY
Formalization Engine
Both streams unite here
PHASE 4
GOLD STANDARD
Regulatory Overlay
Runtime & Feedback Loop
LIVE OPERATIONS
Agent Execution
Loop Back
EXCEPTIONS & AUDIT SIGNALS
Phase 5
CLOSES LOOP
Expert Refinement
Refined knowledge feeds back to Phase 1
Interactive Demo

The Truth Engine

Walk through the 5-phase pipeline. Click to process each phase.

Phase 01
Phase 02
Phase 03
Phase 04
Phase 05
Domain Expert Review

PHASE 1: SOP RESOLUTION

Internal documentation often contradicts itself. The engine flags conflicts for expert resolution.

INPUTS
SOP v1 (Pg. 12)

Auth REQUIRED for CPT 77002.

SOP v1 (Pg. 45)

No Auth required for CPT 77002.

CONFLICT RESOLUTION
OUTPUT
AWAITING PROCESSING
ENGINE_STATUS: IDLE
The Flywheel

Knowledge Compounds Every Cycle

Each reconciliation cycle encodes domain expertise directly into your knowledge foundation. Conflicts resolutions become guardrails.

Cycle 1

Agent Interactions

Every decision trace surfaces how knowledge is actually used in the field.

Cycle 2

Discrepancies Surface

Contradictions between policy and practice become visible and actionable.

Cycle 3

Expert Resolution

Domain experts curate the resolution. Their judgment is encoded, not discarded.

Cycle 4

Gold Standard Evolves

The foundation grows richer. Next cycle starts from a stronger base.

Compounding Knowledge Infrastructure

Ready to clean your knowledge foundation?