Every number behind this door is a frozen export of a real local DataHub instance
(datahub docker quickstart) populated by Veritas's emitters during the hackathon
window — no mockups. Regenerate it yourself with tools/export_graph.py.
Entropy OS is the organism; Veritas Dynamics is its nervous system. During the hackathon, that nervous system got a metadata backbone: every org run, gate verdict, artifact and agent action is now published to DataHub as datasets, lineage edges and assertions — so the trust story is queryable governance, not a README claim. Open it, layer by layer.
Every pick below already passed its own engine's fail-closed gate. New in the hackathon window: gate-verified opportunities flow through a real ETL into Postgres, where the DataHub Analytics Agent answers plain-English questions about them with DataHub context (34fd787). The tag economy on the real graph right now:
Everything above answers what contains what. This answers the question a metadata platform exists for. Left to right is derivation: a dataset's column is how far it sits from a root, so nothing can appear to the left of what it came from. Colour is the DataHub platform that emitted it. A filled ring is an assertion verdict; a hollow node was never asserted, which is not the same as having passed. Brighter edges cross platforms — those are the moments four independent engines actually compose into one objective.
Only artifacts that passed the gates reach the machine — the same runs whose verdicts you can read as DataHub assertions below. The machine sells nothing unverified.
Entropy OS arbitrates across engines; one-engine is what it looks like when four of them execute a single objective together — research, university, software and web — each still its own codebase, its own venv, its own DataHub platform. Nothing was merged. The composite exposes the same contract it consumes, so a higher system can consume it without knowing there are four engines inside.
That claim is checkable here rather than asserted: below is one real objective, drawn from the export. Each stage carries two upstreams — the engine's own dataset on its own platform, and the stage before it — so the federation and the ordering are both facts in the graph, not page copy.
Straight from the export: an org run (the logical parent) with lineage edges to its outcome datasets (the physical children), each guarded by a DataHub assertion carrying the actual gate verdict. Failures are shown because fail-closed systems have failures — that's the point.
And the real-domain slice: Stage 5 publishes Crypto Hunter AI's actually-hunted opportunities as
standalone governed datasets (URN family crypto-hunter-opp_…), their trust carried
as tags and glossary terms — the same graph the tag economy above is counted from.