Corrections Log
This page documents every significant score adjustment, the evidence that triggered it, and whether the previous assessment was too high or too low. Accountability through transparency.
Our Principles
1. Every correction is public
No silent edits. Every score change above ±0.2 is logged here with full rationale.
2. Direction matters
We record whether we overestimated or underestimated. Systematic bias in either direction is itself a finding worth tracking.
3. Sources trigger changes
Every correction cites the specific evidence that prompted reassessment. No vibes-based adjustments.
Corrections
New domain added — Previous: not tracked. Current: 3.5/5
Previous assessment: Social manipulation was partially captured within Social Cohesion (SOC) and Info Integrity (DIS). This underweighted the distinct threat of engineered social disruption.
Correction: Evidence from EU DisinfoLab, Senate Intel Committee, and Cass Review demonstrated that state-sponsored social engineering operates through different mechanisms than platform disinformation. Separated into distinct domain with weight 0.10. All existing domain weights reduced by 0.01-0.02 to accommodate.
Direction: Previous model underestimated social manipulation risk by bundling it with adjacent domains.
Confidence adjusted — Previous: 0.60. Current: 0.65
Trigger: Multiple countries deploying AI in judicial sentencing (Reuters, March 2026) provided new institutional-quality evidence for AI sovereignty risk. Override rates below 2% confirmed across government transparency reports.
Direction: Previous confidence was too low. New evidence from government deployment data increased our confidence in the SOV score.
Score confirmed at 4.5 after SIPRI 2025 publication
Assessment: SIPRI Yearbook 2025 data confirmed the three-way arms race dynamics we scored in the initial model. No correction needed — original assessment validated by primary source.
Direction: Assessment was accurate. SIPRI data aligned with DoomTicker scoring within confidence interval.
This log will grow as DoomTicker's continuous assessment model generates score updates. The Bulletin publishes once per year and has never maintained a public corrections log in its 79-year history.