152,781
Repeated observations
Matches the frozen Paper 0 observation window.
Factory Evidence Program
This is a public evidence and reproducibility record from the Factory's Layer-1 Evidence capability. It is not a separate product, endorsement, certification, ranking, safety verdict or North Star claim.
Cite this record
Fukui, K. (2026). The Registry-Runtime Gap in Agentic AI: A 35-Day Measurement of 31,407 Public Model Context Protocol Endpoint Records (Version 1.0.0) [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.21513720
Correction status: open
The v1.0 publication title reports 31,407 discovered endpoint records. The retained latest-normalized-URL reproduction yields 31,408. Neither is presented here as a corrected denominator until the original frozen discovery or deduplication snapshot is recovered.
The exact state-changing classifier version and intermediate classified-tool table are also unrecovered, so the published 30,641 and 7,618 classifier-derived values remain publication claims rather than currently reproduced Factory facts.
What the retained ledger does reproduce
152,781
Matches the frozen Paper 0 observation window.
4,162
Matches the frozen Paper 0 observation window.
78,041
Matches the frozen Paper 0 observation window.
78.2%
Matches the frozen Paper 0 observation window.
Publication boundary
The publication exposes aggregate methodology, integrity and correction status. It does not expose raw operational ledgers, credentials, owner-control material, suppression data or private topology.
What the study does not claim
A follow-on Paper 1 draft exists only on an unmerged research branch; it is prospective and non-authoritative pending provenance and owner-authority reconciliation tracked in issue #2026.