The living proof graph for agentic work.
Connect every agent, model, dataset, worker, assignment, verdict, receipt, and deed into one verifiable system of record.
AI output is not enough. DefendableGraph records the relationships behind the work: who requested it, which agent ran it, what model generated it, what dataset informed it, what worker executed it, what rulebook judged it, and what receipt proves it.
Every assignment becomes a trace. Every trace becomes a receipt.
The graph is the relationship layer for trusted AI execution across DefendableOS, DefendableCloud, AIOV, MicroScalers, and Swarm & Bee.
Agent lineage
Track which agents, runtimes, and models touched each assignment.
Dataset provenance
Connect outputs back to governed datasets, manifests, and training lineage.
Worker reputation
Tie work to real compute nodes, GPUs, benchmarks, and heartbeats.
Rulebook verification
Show which policies, rubrics, and Tribunal checks judged the output.
Receipt anchoring
Hash every meaningful artifact into a durable proof object.
Deeded outcomes
Connect appraisals, datasets, and deliverables to ownership records.
Client → Assignment → Agent → Receipt → Artifact
A single CRE assignment can be inspected as a timeline, graph, event stream, and JSON proof object.