Verdict: benign (by construction) — screened 2026-08 against the
dataset’s own documentation (Documentation.pdf, 55 pp.).
The documentation states the rule directly:
“We assume that accounting variables are publically available 4 months after the end of the accounting period.”
Applied uniformly to every accounting variable, this removes filing speed from arrival ordering — a fast filer and a slow filer become visible at the same time. Arrival order can then no longer select on the disturbance, which closes the incomplete-cross-section channel structurally.
Because the published data carry no per-record arrival field, the verdict is established on planted truth rather than measured on the panel: the same planted leakage reads ρ̂ = −0.324 under as-filed release and +0.017 under uniform four-month release (positive control 7/7). The convention works; what cannot be done from the outside is verify the assumption against actual filing dates, since none are documented.
Structural protection via a blanket lag is a timeliness tax paid by all 406 characteristics — including signals a per-signal screen would release months earlier. That trade-off (uniform lag vs. graded release) is the subject of the release-control method paper.
And as with OSAP: the protection belongs to the convention. Rebuilding JKP-style characteristics from as-filed sources at observed filing dates re-opens the channel — there, 7 of 14 standard signals flag.