Coverage and status

These figures describe the corpus LROHub holds today. They are a snapshot of one deployment, not a measure of what the platform does — that is on the main page, and it is proved by code paths and the gates that hold them, not by counts.

Coverage is expected to be partial and to move. Ingestion grows deliberately — by domain, by subject area and by demand — so a new research area starts thin by design. A low number here means a corpus that has not reached that area yet, not a capability that is missing.

The corpus, counted now

live · this request

Papers held as LROs

2,900

Claims extracted

Papers with a reproducibility manifest

36 / 37

Papers with agent output recorded

38 / 259

Agent/paper pairs an agent has reported on

2,897 / 2,900

Claims traceable to a source span

940 / 2,900

Claims carrying ontology concepts

50 / 4,099

Causal variables with an ontology id

3,602 / 4,099

Causal variables in only one relationship

Papers checked against the retraction registry

Per-pillar coverage

mixed · snapshot taken 2026-10-15 · 5 days ago

Pillar	Claim tier	Actually runs	Coverage
Executable arithmetic checks ESR	shipping	unattended	102 / 106 · 96% reported figures recomputed and reproduced snapshot · 2026-10-15
Causal Knowledge Engine CKE	partial	unattended	346 / 1,560 · 22% asserted relationships in the corpus graph snapshot · 2026-10-15
Autonomous Research Agents ARA	partial	unattended	38 / 259 · 15% agent/paper pairs an agent has reported on counted this request
Traceable-or-refused answers ZHA	shipping	on-ingest	2,897 / 2,900 · 100% claims anchored to a verified source span counted this request
Federated Knowledge Mesh FKM	partial	on-request	129 / 746 · 17% canonical concepts appearing in more than one LRO snapshot · 2026-10-15
Living Loop LL	shipping	unattended	37 / 37 · 100% papers re-checked against the retraction registry counted this request

Claim tier

How well the pillar’s own wording matches what it does.

shipping: Runs on every ingested LRO today. The evidence figure is measured from the catalog.
partial: Part of this runs in production; the rest is described below as not yet built.
roadmap: Designed and not yet running. Named here because it is where the product is going, not what it does now.

Actually runs

Whether anything turns it. A tier says nothing about this, and reporting the tier alone made working software read as half-built.

unattended: A named schedule runs it with no human in the path. The schedule must exist, be enabled and not be at fault, or `make capabilities` fails.
on-ingest: Runs inside the ingest path, on every article, as a precondition of publishing it.
on-request: Served by a named API route. Correct for a query surface: an endpoint answering a peer does not want a schedule.
manual: A human starts it. Named here so 'it runs' is never read as 'it runs itself'.
not-running: Implemented and nothing invokes it.

The corpus figures above are counted on every request, and so is every per-pillar row marked counted this request — those metrics are derivable from catalog rows, so ingesting a paper moves them on the next page load with nobody refreshing anything. The rows marked snapshot re-execute arithmetic checks, read the relationship graph or join ontology concepts: that work belongs to make capabilities and cannot run inside a page render. Every row says which it is, so neither can be mistaken for the other.