Czech Democracy Index
← Methodology

Validation report — 2026-Q2

Generated automatically 2026-04-29 by pipeline:validate. Per methodology/governance.md and the CLAUDE.md "Validation" section.

Baseline divergence

Our structural baseline (weighted overall) = 85.0.

Threshold for a methodology review: persistent divergence > 10 points across two consecutive quarters against a reference index (V-Dem or EIU). Short-term deviations are tolerated — our score weights pillars differently from external indices and includes a weekly event component.

External index External (0–100) Compared target Value Δ Above threshold?
V-Dem (2024) 81.7 baseline overall 85.0 +3.3
EIU (2024) 80.8 baseline overall 85.0 +4.2
FH-FitW (2025) 95.0 baseline overall 85.0 -10.0
RSF (2025) 84.0 pillar media 92.0 +8.0
TI-CPI (2024) 59.0 pillar corruption 59.0 +0.0
WJP (2024) 74.0 pillar judicial 83.9 +9.9

Single-dimension indices (RSF, TI CPI, WJP) are compared with a specific pillar, not with the overall score. RSF↔media, TI CPI↔corruption, WJP↔judicial. Multi-dimension (V-Dem LDI, EIU, FH FitW) are overall composites → compared to our weighted overall.

Conclusion: no external index shows divergence > 10 pts. The baseline is within normal variability.

Latest snapshot vs baseline

Latest snapshot: 2026-W17 — overall 84.3. Shift from baseline: -0.7 pts (from 8 active events).

Pillar Baseline Snapshot Δ
electoral 91.8 91.8 +0.0
governance 86.3 84.3 -2.0
judicial 83.9 83.4 -0.5
media 92.0 91.5 -0.5
civil 96.9 96.4 -0.5
corruption 59.0 58.5 -0.5

Per-pillar diagnostics

Mapping recap (from methodology/structural_mapping.md):

  • electoral (15 %) = 91.8
  • governance (20 %) = 86.3
  • judicial (20 %) = 83.9
  • media (15 %) = 92.0
  • civil (15 %) = 96.9
  • corruption (15 %) = 59.0

If divergence in section 1 crosses the threshold, investigate which pillar contributes most. Common sources of noise:

  • corruption has only TI CPI (single source), so our pillar = TI CPI exact value. Any divergence must come from elsewhere.
  • judicial uses WJP overall as a proxy (per-factor data unavailable). Divergence against WJP itself = 0.
  • FH uses a 4-point scale across 7 categories, so small differences in FH get amplified when normalised to 0–100.

This report is generated automatically via npm run pipeline:validate -- --quarter=<Q>. For new quarters a new file is created; existing ones are overwritten after a new run (versioning is kept by git).