Reproducible framework intelligence
The EBA reporting framework — its Data Point Model, validation rules, and XBRL taxonomy — read directly from the published packages and measured, not commented on. Every figure below is reproducible from EBA's public files, and each was tested against EBA's own artefacts before it was believed.
Two findings
4,699validation rules — 62.2% of all rules
Across the DPM 2.0 transition (releases 4.1→4.2), these rules did not change their own expression, yet each read at least one data point whose meaning was silently redefined underneath it. A release diff shows nothing; the check now validates a different figure. Sixteen templates carry affected checks with no changed data points of their own.
Validated against EBA's official DPM 1.0→2.0 conversion file — 27 of 30 concept mappings reproduced exactly.
403genuine rule deactivations across the framework's history
Institutions track framework releases. But validation behaviour changes on its own schedule — 27 distinct change-dates against 9 releases, only two of them coinciding — concentrated in the own-funds module COREP_OF, with severity moving one way: 1,730 relaxations against 3 tightenings across the entire history. A submission that passed last quarter can fail this quarter with no framework change to point to.
Confirmed by EBA's own quarterly rule-update script, which performs exactly this mechanism.
Why it holds
Read it, or reproduce it
Correspondence is welcome from reporting, reg-change, and data functions at institutions and vendors.