The sentinel for changes that can alter the scientific record.
LENITNES Research watches public scientific-software repositories, detects method and results-bearing changes, records what the change appeared to mean before later evidence arrives, and grades alerts against explicitly adjudicated events in the published record.
alert-only · no trading in this arm · replay and live cohorts kept separate
Detect method risk
Typed detectors look for statistical-method fixes and silent results rewrites.
Commit the judgment
Above-threshold alerts are published only with a timestamped HCS proof state.
Grade with restraint
Only confirmed, adjudicated record events count toward live precision.
3dClustSim: a historical calibration case
A 2015 AFNI fix to 3dClustSim preceded the 2016 “Cluster failure” disclosure by 413 days. This is a replay through today’s pipeline—not evidence that the system was live in 2015—and it remains separate from the prospective record.
Read the replay