Correction Resistance
A rejected or corrected output returns during the same task or a later session.
The working hypothesis
What the archive calls the 3% Protocol
The Protocol is an interpretive model for rare system behavior in which a digital service appears to preserve, restore or intensify an automated choice after a user has rejected it.
3%
A threshold, not a measured global rate
The name refers to the archive's founding claim: a small fraction of interactions may be allowed to resist correction in order to observe whether users accept an automated outcome. The archive has found no public technical specification that establishes an exact three-percent rate.
Archive model
The minimum sequence
01Automated outputA system supplies a route, word, setting or recommendation.
→02Explicit rejectionThe user corrects, disables, hides or replaces the output.
→03Return or escalationThe rejected outcome appears again under comparable conditions.
A single recurrence is insufficient. Reviewers also assess repetition, context, alternative explanations and the quality of the record.
Resistance index
Six dimensions used in case review
Default Acceptance
The system repeatedly steers the user back toward a preset or commonly selected option.
Automation Trust
The event tests whether an automated suggestion is followed despite contradictory local knowledge.
Override Persistence
A deliberate manual choice fails to remain active across recalculation, reload or synchronization.
Preference Recovery
An earlier preference returns after the user has changed, hidden or disabled it.
Escalation After Rejection
The same proposal becomes more prominent after one or more explicit rejections.
Origin claim
From Newton-era input research
The archive traces the idea to accounts surrounding early handwriting recognition research in the late 1980s and early 1990s. The historical development of pen computing is documented; the specific resistance-testing program remains an unverified part of the hypothesis.
Follow the chronology →