Dashboards, Observability & Context Graph

The Dashboards and Explore sections of the Caber User Console provide real time operational visibility into corpus governance coverage, document version lineage, characterization throughput, platform health, and ontological graph relationships. Every dashboard synchronizes with the connected deployment and updates its charts, Cytoscape canvases, and telemetry tables across both dark obsidian and high contrast light modes.

Home Overview Dashboard (/admin/home)

The Home Overview page (/admin/home) serves as the primary operational landing screen after connecting to a deployment. It combines corpus level snippet coverage with live enforcement telemetry and expandable infrastructure operations cards.

CONSOLE FUNCTIONAL PROOF/admin/homeExecutive Overview & Four Core Governance Questions Dashboard

Assertion Verified: Proves that the Executive Overview dashboard synthesizes all four governance questions—data duplication, runtime policy verdicts, AI provenance traces, and platform health—into a single real-time operational command center.

Executive KPI Strip (Sources, Chunks, Clusters, Verdicts)Duplicate Cluster & Policy Feed PreviewsPlatform Infrastructure Health Summary

Caber Console Executive Overview (/admin/home) summarizing corpus duplication, active policy enforcement, recent session traces, and service health.

Corpus Snippet Hero and Interactive Policy Sunburst

The top region of /admin/home renders the Snippet Coverage Hero (SnippetHero) alongside the two ring Policy Sunburst (PolicySunburst):

1. Corpus Snippet Hero Metrics: Displays total characterized documents, total indexed subsentence snippets (LNKD equivalence classes), active policy boundary coverage, and fresh versus stale version counts fetched from the home snippets telemetry endpoint.
2. Two Ring Policy Sunburst:
   • Inner Ring (Repository Sources): Segments the corpus by storage source (such as minio-contracts, mna-dataroom, or sec-edgar-archive).
   • Outer Ring (Governance Policy Coverage): Subdivides each source by the active governance policy or sensitivity classification protecting its regions.
   • Interactive Slice Selection: Hovering or clicking any slice highlights the corresponding repository or policy breakdown in the adjacent legend and filters the summary metrics to that segment.
3. Four Core Telemetry Panels: Immediately below the hero, a four column summary grid surfaces:
   • Enforcement Verdicts (/api/audit-v2/verdicts): Breakdown of ALLOW, REDACT, and BLOCK dispositions over the active time window (15m, 1h, 24h, 7d, 30d).
   • Proxy & Enforcement Latency (/api/perf/latency): p50, p95, and p99 stream inspection and policy evaluation latencies in milliseconds.
   • On Behalf Of (OBO) Principal Activity (/api/audit-v2/obo/summary): Count of delegated agent invocations where pycaber minted or validated a narrowed OBO token.
   • Top Policy Hits (/api/audit-v2/policies/hits): Ranked list of authored and system policies triggered during agent retrieval and inference.

Expandable Operations and Flagged Activity Cards

At the bottom of /admin/home, operators can click the Operations & Activity Details toggle (data-testid="home-operations-toggle") to reveal the secondary operational deck without leaving the home screen.

CONSOLE FUNCTIONAL PROOF/admin/homeExecutive Overview with Expanded Operations & Pipeline Telemetry

Assertion Verified: Proves that operators can expand the inline Operations telemetry drawer on /admin/home to inspect real-time ingestion throughput, queue backlog, and subsystem health without leaving the executive dashboard.

Expandable Operations Telemetry SectionPipeline Stage Throughput & Latency BreakdownOne-Click Drill-Down to Subsystem Views

Expanded Operations section on /admin/home showing live ingestion throughput, queue depth, and subsystem status cards.

Expanding this drawer reveals three live operational cards:

• Deployment Health Matrix (/api/deployment-health): Shows real time reachability and response latency for caber_core, OpenSearch (csi_audit_v3, csi_api_map, csi_obj_versions), Neo4j, and configured object stores.
• Recent Flagged Traces (/api/audit-v2/flagged): Lists the most recent traces containing surgical redactions (R) or policy blocks, with one click drill down links that open /admin/audit directly on the flagged traceId.
• Connector & Source Activity (/api/connectors/activity): Summarizes recent object ingest events, identity synchronization runs, and active SDK service heartbeats.

Object Clusters and Version Lineage (/admin/overlapping)

The Object Clusters & Versions page (/admin/overlapping) visualizes how subsentence content (RKCQF / LNKD shingles) flows across document revisions, copies, and repositories. It exposes the two core failure modes of ungoverned retrieval: stale version leakage (where an agent retrieves a superseded draft) and cross boundary policy drift (where shared clauses appear inside documents governed by conflicting access policies).

CONSOLE FUNCTIONAL PROOF/admin/overlappingDuplicate Data Tracker & Side-by-Side Freshness Cluster Inspector

Assertion Verified: Proves that Caber detects 85% to 99% enterprise content duplication down to 40-byte WRL sub-sentence phrases (97% to 98.7% coverage), separates canonical fresh sources (version_sequence == 0) from stale replicas, and enforces origin trade-secret policies even when text is copied into unlabeled channels.

85%–99% Enterprise Duplication Tracked97%–98.7% WRL Sub-Sentence Coverage (40B min)99%-Capped Overlap Edges & Policy HullsFresh (seq=0) vs Stale Replica Resolution

Caber Console Duplicate Data Tracker (/admin/overlapping) comparing a canonical fresh source against an unlabeled stale replica across shared 40-byte WRL spans.

Version Cluster Summary and Lineage Cards

At the top of /admin/overlapping, three KPI counters summarize the corpus lineage state returned by /api/overlap-compare/list:

• Version Clusters: Total connected document lineages resolved by Inverse Document Frequency (IDF) content containment analysis (csi_obj_versions).
• Documents in Clusters: Total document versions participating across all multi version lineages.
• Version Relationships: Total directed parent to child derivation edges across the resolved directed acyclic graphs (DAGs).

When an operator navigates to /admin/overlapping by clicking a provenance badge inside /admin/audit, a blue telemetry banner (data-testid="overlap-selection-banner") appears at the top of the page listing the exact sources Caber used in that trace and automatically expands the matching cluster.

Expanded Cluster Timeline DAG and Policy Boundary Hulls

Clicking any cluster row expands the interactive Cluster Timeline Graph (ClusterTimelineGraph) rendered with Cytoscape and the ELK layered layout engine.

CONSOLE FUNCTIONAL PROOF/admin/overlappingDuplicate Data Tracker with Expanded Cluster Lineage Graph

Assertion Verified: Proves that expanding a duplicate cluster row in /admin/overlapping renders an interactive directed lineage graph distinguishing the canonical fresh origin (seq=0) from downstream stale replicas and partial forks.

Interactive Cytoscape Cluster Lineage DAGCanonical Origin (seq=0) vs Stale Replica Nodes99%-Capped Pairwise Overlap Edge Weights

Expanded duplicate cluster on /admin/overlapping displaying the interactive lineage graph and member file freshness states.

The expanded cluster view provides left hand lineage metadata alongside the interactive DAG canvas:

1. Cluster Metadata Rail: Lists total versions in the cluster, which version Caber used during the audited session, the Latest (head) version in green, and the Original root candidate in blue, plus a Date Check warning whenever file last_modified timestamps contradict content containment ordering.
2. Corrected versus Pre Time Corrections Toggle: When timestamp anomalies exist, operators can switch between Corrected (blended content plus timestamp ordering) and Pre-Time Corrections (pure content containment ordering).
3. Node Badges and Policy Boundary Hulls:
   • Original Candidate: Pinned to the leftmost layer of the DAG.
   • Latest Version (head): Highlighted with a green border and pinned to the rightmost layer.
   • Caber Used (★ Caber Used): Marks the exact document version retrieved by the agent.
   • Policy Boundary Hulls (PolicyLegend): When versions inside the same cluster fall under two or more distinct governance policies, translucent convex hulls surround each policy group so operators immediately see where a clause crossed an authorization boundary.

Side by Side Document Comparison and Shared Nmer Inspector

Clicking any directed edge (arrow) between two version nodes in the cluster DAG opens the Overlap Comparison Panel (OverlapComparisonPanel) directly below the cluster card.

CONSOLE FUNCTIONAL PROOF/admin/overlappingSide-by-Side Document Comparison & Shared WRL Span Inspector

Assertion Verified: Proves that clicking an overlap edge or replica node opens a side-by-side Document Comparison view that highlights exact shared 40-byte WRL sub-sentence spans and diverging clauses across both documents.

Side-by-Side Source vs Target Text DiffHighlighted 40-Byte WRL Sub-Sentence OverlapsCross-Repository Provenance & Policy Comparison

Side-by-side Document Comparison panel on /admin/overlapping highlighting shared WRL sub-sentence spans between origin and replica files.

The comparison panel queries /api/overlap-compare for the selected source and target document paths and renders a synchronized forensic diff:

• Source and Target Document Panes: Displays the full reconstructed text of the parent version on the left and the derived version on the right, complete with MIME type badges and byte counts.
• Shared Subsentence Highlighting (shared_nmers): Highlights every shared LNKD equivalence span across both panes. Clicking a highlighted span in either document scrolls the opposite pane to the matching clause and displays the underlying nmer_hash.
• Dismissing the Comparison: Clicking the Close comparison button (X) or clicking the same DAG edge a second time closes the diff panel.

Ingest Performance and Characterization Telemetry (/admin/scanning)

The Ingest Performance page (/admin/scanning) monitors the characterization pipeline as Caber scans object stores and SQL databases, extracts text and named entities (GLiNER), and populates the WRL Chunker (RKCQF) and RBIX roaring bitmap indexes.

CONSOLE FUNCTIONAL PROOF/admin/scanningScanning & Ingestion Pipeline Performance Telemetry

Assertion Verified: Proves that Caber tracks end-to-end document scanning, parser routing, WRL sub-sentence chunking, and RBIX roaring bitmap indexing performance with per-stage throughput and latency charts.

Connector & Object Store Scan ThroughputDocument Parser & NER Enrichment LatencyWRL Chunker & RBIX Bitmap Indexing Rates

Caber Console Scanning & Ingestion Performance view (/admin/scanning) monitoring connector scan progress, parser throughput, and RBIX indexing.

Characterization KPIs and Time Range Controls

Operators can filter /admin/scanning across preset windows (15m, 1h, 24h, 7d, 30d) to inspect four primary characterization panels:

1. Scan Summary Ribbon (/api/scan/summary): Reports total objects scanned, total bytes characterized, average characterization time per document, and active queue depth.
2. Throughput Timeline (/api/scan/throughput): Plots documents per minute and megabytes per second processed by the characterization workers over time.
3. Documents by Format (/api/scan/documents-by-type): Breaks down ingested objects by MIME type and parser pipeline (PDF, DOCX, Markdown, XLSX, HTML, JSON).
4. Detected Region & Entity Categories (/api/scan/regions) and Source Health (/api/scan/sources): Shows the distribution of extracted GLiNER sensitivity labels (PERSON, ORGANIZATION, ACCOUNT_NUMBER, ROYALTY_RATE) and per bucket scan completion status.

Platform Status and Container Logs (/admin/platform)

The Platform Status page (/admin/platform) provides infrastructure health monitoring and live log inspection across all Caber runtime services.

CONSOLE FUNCTIONAL PROOF/admin/platformPlatform Infrastructure Status & Live Container Log Inspector

Assertion Verified: Proves that operators can monitor real-time health probes across all backing data stores (OpenSearch, Neo4j, S3/VersityGW, PostgreSQL) and inspect structured service logs directly from /admin/platform.

Backing Store & Service Health ProbesFilterable Real-Time Service Log StreamSeverity & Subsystem Log Filtering

Caber Console Platform Status & Logs page (/admin/platform) displaying live service health indicators and filterable operational logs.

• Service Health Cards (data-testid="health-card-*"): Displays status indicators (Healthy, Degraded, Offline), uptime, and version metadata for caber_core, OpenSearch, Neo4j, and connected storage endpoints. Hovering a health card reveals detailed probe diagnostics.
• Filterable Service Logs (/api/deployment-logs): Streams structured container logs with instant filtering by service (caber-core, opensearch, neo4j, scanner), severity level (INFO, WARN, ERROR), and free text search.

Context Graph Explorer and Analytical Views (/admin/contextGraph)

The Context Graph Explorer (/admin/contextGraph) allows operators to query and visualize the Neo4j ontological graph that links Object, Region, Organization, Person, and PolicyDecision nodes.

CONSOLE FUNCTIONAL PROOF/admin/contextGraphNeo4j Context Graph & Ontological Provenance Explorer

Assertion Verified: Proves that identical sentence bytes in two different documents carry distinct meaning and governance verdicts depending on their ontological neighborhood (:Region NER rollup, LNKD-{xor}-{count} equivalence class, and CORI provenance).

Content is Finite; Context is InfiniteMorgan Company Ambiguity DisambiguatedLNKD-{xor}-{count} Equivalence ClassesDecision E7 :Region NER Rollup

Caber Console Context Graph Explorer (/admin/contextGraph) proving how ontology and provenance disambiguate identical sentence content across Morgan's company divisions.

Cypher Query Bar, AI Graph Translation, and Node Inspector

The Context Graph workspace combines query authoring with interactive graph manipulation:

1. Schema Summary and Label Chips: On load, the explorer introspects db.labels() and db.relationshipTypes() via /api/neo4j-query, displaying clickable node label pills (Object, Region, Organization, Person, PolicyDecision) and relationship type pills (CONTAINS_REGION, SUPERSEDES, OWNED_BY, AUTHORIZED_FOR, GOVERNED_BY, REQUESTED_BY) with live entity counts.
2. Cypher and Natural Language Query Execution: Operators can type a raw Cypher query (for example, MATCH (o:Object)-[r]->(t) RETURN o, r, t LIMIT 50) or use the natural language query mode to translate an English question into Cypher before executing it against Neo4j.
3. Interactive Cytoscape Canvas and Property Drawer: Clicking any node or relationship edge on the canvas opens the right hand property inspector showing all ontological attributes (such as lnkd_class, byte_span, version_sequence, freshness, or disposition), while double clicking a node expands its one hop neighborhood.

Access Authorization (/admin/accessAuth) and Data Flow (/admin/dataFlow)

Two specialized analytical views complement the Context Graph for targeted security reviews:

• Access & Authorization View (/admin/accessAuth): Aggregates principal to resource access decisions across identity sources, highlighting where delegated OBO scopes narrowed an agent's effective permissions compared to its service account identity.
• Data Flow & Lineage View (/admin/dataFlow): Traces how characterized regions move from upstream object stores (Object_Sources) through vector database queries (qdrant, opensearch), MCP tool responses, and outbound LLM completions.