Diagram Gallery

The system diagrams, collected from across the documentation. Each image links to the full-size SVG; the noted section carries the surrounding description.

The ePIC Computing Model

In Foundations — the Echelon model: E0 detector and DAQ, the butterfly raw-data fan-out to the E1 host labs, E2 global facilities, and E3.

The Streaming Latency Spectrum

In Foundations — result latencies across the streaming computing model, from real-time DAQ validation through fast results, prompt processing, validation, and calibration.

Lifecycles

In Concepts — the state machines of the campaign task (with PanDA tail retry and operator rerun controls), the campaign staging, and the tag lock.

The ePIC Workflow Management System

In Architecture — the platform anchor: workflow domains served by shared web/database, agent, AI, data and workflow management layers over the Echelon resources.

PanDA in ePIC

In WFMS Platform — the deployed workload management infrastructure: the two fronts driving PanDA at BNL, identity, the ePIC monitor, and the growing resource landscape.

ePIC AI Infrastructure Stack

In WFMS Platform — ePIC services, the MCP instrumentation over them, and the AI services built on top.

corun-ai — LLM Execution and Artifact Service

In WFMS Platform — the corun-ai system: an interactive research site and the REST AI back end of the WFMS, sharing one execution machinery and artifact store.

LLM and Automated Services in epicprod

In WFMS Platform — human-gated LLM assistance and credentialed automation.

E0-E1 Workflow Schematic

In Streaming Workflows — the E0-E1 interface: DAQ exit buffer, STF and TF streams to the E1 facilities, and the streaming workflow testbed scope.

Prompt Processing Workflow

In Streaming Workflows — full-sample processing of STFs as they arrive, from run signals through the filling run dataset to PanDA jobs at the E1s, with the conceptual decision box.

Fast Processing Pipeline

In Streaming Workflows — the testbed fast-processing pipeline from simulated DAQ through TF slices to the standing worker pool.

iDDS/PanDA/Harvester Detail

In Streaming Workflows — worker provisioning behind the fast-processing pool.

Testbed Agent Management

In Streaming Workflows — how testbed agents are configured, launched, and supervised: CLI and MCP control, the per-user agent manager, and supervisord.

Testbed Multi-User Isolation

In Streaming Workflows — shared infrastructure with independent per-user operation: namespaces, per-user agent identity, and filtered views.

epicprod — the ePIC Production System

In Production System — the production workflow stages, LLM services, and credentialed operations of epicprod.

ePIC Automated Production Workflow

In Production System — the production workflow and the task catalog staged by campaign lifecycle.

ePIC Production Dataflow

In Production System — job orchestration and the payload-handled science data path against JLab Rucio.

LLM and Distributed Computing Services Integration — Detailed Architecture

In Production System — the detailed LLM operations architecture: SSE relay, LLM executors, credentialed agent, and artifact store.

The Physics Configuration System

In Physics Configuration System — from the tag catalog through dataset composition and the composed name to the submit-ready campaign task.

The Validation Loop

In Validation — availability, Hydra validation, AI assessment, record, and expert signoff between the WFMS and ePIC validation; proposal-stage interfaces dashed.

The WFMS Timeline

In Timeline — the three-year plan and the decade arc to physics datataking.

Streaming Computing Planning — FY25–FY31

In Timeline — the joint DAQ, computing, and AI planning lanes, the context for the WFMS out-years.