Method

Two pillars, 50/50 headline. v1.1 adds the watching machine (CS6). v1.2 adds electoral integrity as CS7 inside Pillar 1. It is not a third pillar. Poverty stays in Pillar 2. Older P1 formulas stay on the page.

Pillars

P1 — Attitudes toward civil society

Higher = more closed / repressed civic space. Laws and behavior (de jure + de facto), not speeches.

  • CS1 Civic space / CSO
  • CS2 Expression & media
  • CS3 Association / assembly / NGOs
  • CS4 Organized labor (ITUC)
  • CS5 Voice / contestation (WGI VA)
  • CS6 Domestic control apparatus (v1.1, kept in v1.2)
  • CS7 Electoral integrity (v1.2) — not a third pillar

P2 — Socioeconomic distress

Higher = more deprivation, inequality, exclusion, shock. Unchanged in v1.1 and v1.2.

  • SE1 Deprivation (24%)
  • SE2 Human development (24%)
  • SE3 Distribution (20%)
  • SE4 Exclusion / youth (16%)
  • SE5 Shock & level (16%)

P1 formulas — v1.0, v1.1, and v1.2

The codebook rule is to always publish prior P1 formulas. v1.2 does not overwrite v1.0 or v1.1.

v1.0 baseline

P1_v1.0 = 0.40×CS1 + 0.20×CS2 + 0.16×CS3 + 0.12×CS4 + 0.12×CS5

Keep computing forever. Do not overwrite in place.

v1.1 with CS6 18% of P1

P1_v1.1 = 0.32×CS1 + 0.16×CS2 + 0.14×CS3 + 0.10×CS4 + 0.10×CS5 + 0.18×CS6

Kept beside v1.0 and v1.2.

v1.2 with CS7 16% of P1

P1 v1.2 = 0.26×CS1 + 0.14×CS2 + 0.12×CS3 + 0.08×CS4 + 0.08×CS5 + 0.16×CS6 + 0.16×CS7

Headline for new releases. Weights sum to 1. Electoral integrity is inside Pillar 1, not a third pillar.

GLII_ADD = 0.5 × P1 + 0.5 × P2 P2 weights unchanged. v1.1: ΔP1 ≈ 0.18 × (CS6 − CS_old) · ΔGLII_ADD ≈ 0.09 × (CS6 − CS_old) v1.2: ΔGLII_ADD ≈ 0.08 × (CS7 − old civic blend) A 40-point gap moves the headline by about 3 points. Enough to cross 50. Not a new democracy index.
Subindexv1.0 of P1v1.1 of P1v1.2 of P1Δ pp v1.1→v1.2
CS1 civic space40%32%26%−6
CS2 media20%16%14%−2
CS3 assembly / NGO16%14%12%−2
CS4 ITUC12%10%8%−2
CS5 WGI Voice12%10%8%−2
CS6 control apparatus0%18%16%−2
CS7 electoral integrity0%0%16%+16

v1.2 electoral integrity (CS7)

v1.2 adds electoral integrity as CS7 inside Pillar 1. It is not a third pillar. A cooked or missing count is how insecure leadership treats the one civic act that can remove it. Poverty stays in Pillar 2.

P1 v1.2 = 0.26 CS1 + 0.14 CS2 + 0.12 CS3 + 0.08 CS4 + 0.08 CS5 + 0.16 CS6 + 0.16 CS7. Those weights sum to 1. The headline stays 50/50: GLII_ADD = 0.5 P1 + 0.5 P2. Voice is trimmed so the ballot is not counted twice.

What is coded

CS7 is the mean of whichever of these exist, after inversion (high = worse integrity).

IDFieldRule
EI1 Electoral Integrity Project PEI 12.0 overall / concept index, latest national contest Score as 100 minus the PEI score. Carry forward at most 5 years, then missing. PEI 12.0 adds 41 contests from 2025.
EI2 V-Dem v2xel_frefair (clean elections, 0–1) Score as 100 times (1 minus x). If v2x_elecreg = 0, set 100.
EI3 Freedom House A1+A2+A3 only (electoral process, 0–12) Invert the sum. Do not add the political-rights total.

No electoral regime is scored 100, with flag_no_electoral_regime = 1. That means there is no count, not that the count was stolen. A Gulf monarchy and a rigged multi-party election must not share a fraud label.

Left out on purpose

Those swallow civic space and would turn this index into a second democracy ranking.

How far the headline moves

Approximate movement: delta GLII_ADD ≈ 0.08 times (CS7 minus the old civic blend). A 40-point gap moves the headline by about 3 points. Enough to cross 50. Not a new democracy index.

Four demo cases

Demo cases, not country statistics. The before column is the prior civic blend; the after column is v1.2. Not a full statistical merge.

Demo case P1 before → v1.2 Headline What changed
Closed-poor, no real count (CS7 = 92) 78.6 → 80.8 71.9 → 73.0, still Q4 Already at the floor. The ballot confirms it.
Closed-rich, managed election (CS7 = 80) 73.1 → 74.3 47.7 → 48.3, still Q2 Control state. The managed count is why Q2 is not just “rich and rude to NGOs.”
Open-poor, messy but real count (CS7 = 40) 27.9 → 29.8 43.0 → 43.9, still Q3 A 40 is still worse than their very open civic scores, so P1 ticks up. A genuinely clean count (CS7 near 15) would pull P1 down.
Mid civic, cooked count (CS7 = 85) 47.7 → 53.7 43.8 → 46.8, crosses into Q2 v1.1 left them under 50. The ballot is the exit they will not allow. In the demo, P2 is 40, so they become a control case, not a powder keg.

If that same cooked count sat on a distressed Pillar 2, the cross would be into Q4, not Q2. v1.2 can say the leadership will not allow a count that can remove it, or that there is no count. It still cannot say the leader is illegitimate, or that a high CS7 means the next uprising. A high CS7 with a low P2 is often a managed electoral autocracy that is stable. A high CS7 with a high P2 is the powder keg with no exit.

Episodic press

The press score is the press system, not a famous case. WikiLeaks and Snowden are flag_episodic_press. They can add at most 4 points. They cannot set the band. A constitutionally aligned press that fights the security state and still publishes is not coded as a structurally broken press.

United States and Nigeria Illustrative v1.2 inputs from published proxies, not a full statistical merge and not a court finding: United States, grade AA, Q1, lower, score 20.9, civic hostility 24.1, everyday hardship 17.7, rank 51. Nigeria, grade CCC, Q4, high, score 58.6, civic hostility 57.6, everyday hardship 59.5, rank 24. Their grades come from their rank, like every other country (see Letter grades). Nigeria’s nearest country type is the powder-keg type, because both pillars are above 50 (Q4). The United States is Q1 because both pillars are under 50; Q1 has no type of its own, so its nearest listed type is the open type. The US score is not built from Snowden or WikiLeaks. Bulk collection stays as contested oversight, not as a mercenary-spyware state. Nigeria’s press constraint is structural; known surveillance contracts and a weak electoral-integrity record sit on high hardship.

Letter grades (AAA to D)

Every country on the ranking gets a credit-rating-style letter: AAA is the least insecure end of this list and D the most insecure. The letter is not new data and not a new score. It is read straight off the site’s demo rank (1 = most watched, most insecure) with the fixed cut points below. The cut points sit on the band boundaries, so a more insecure rank never gets a better letter, a worse band never gets a better letter, and no band holds letters that are not next to each other. The site checks this every time it is built and refuses to publish if rank, band and letter disagree.

The letter is read straight off the site’s demo rank (1 = most insecure). It adds no new data. Rank, band and letter always agree. These are illustrative demo values, not an official ranking.
GradeDemo rankBandReads as
AAA52–54LowerLeast insecure on this list
AA50–51LowerVery low insecurity
A46–49LowerLow insecurity
BBB40–45LowerLower band, upper end
BB31–39ModerateModerate insecurity
B25–30ElevatedElevated insecurity
CCC17–24High (Iraq, rank 23: Elevated)High insecurity, lower end
CC10–16HighHigh insecurity
C5–9HighHigh insecurity, top end
D1–4Very highMost insecure on this list

One quirk in the source data: Iraq (rank 23) is in the Elevated band while Nigeria (rank 24, an illustrative v1.2 score) is in the High band. Both are CCC, so neither the rank nor the band is contradicted. Grades move only when a rank moves. Rank, grade and band are illustrative demo values, not an official ranking.

Before 24 October 2026 the site used AAA, BBB and CCC as names for its three computed country types. Those are now called the powder-keg type (closed-poor, Q4), the control type (closed-rich, Q2) and the open type. They are profiles, not grades; their old page ids (country.html?id=AAA and so on) still work.

CS6 children (equal 25% when all present)

CS6A — Legalized / unchecked surveillance
FOTN / surveillance items; proxy 100−FOTN
CS6B — Spyware / intercept + civic targeting
Carnegie / Citizen Lab; 0 ≠ innocent
CS6C — Interior-budget shape + intel opacity
IMF GFS / OBS — never SIPRI milex
CS6D — Digital repression operations
KeepItOn / OONI / NetBlocks user-hours

Need ≥2 children. Do not regional-mean impute. CS6 = mean(available A–D).

Forbidden inputs (R13–R20)

Missing-data summary (R1–R11)

Sensitivity — CS6 weight of P1

Publish rank correlations at CS6 weights 0% / 10% / 18% / 25% of P1 so 18% is not a sacred single choice.

0%
= v1.0
10%
light tilt
18%
headline v1.1
25%
hot sensitivity

Worked demo — the three computed country types

Already-scaled 0–100 inputs from sheet 04_Demo_formulas.

TypeLabel Civic v1.0Civic v1.1Δ civic Hardship Score v1.0Score v1.1 Type
Interpretation The powder-keg type (closed-poor + apparatus) barely moves — already maxed. The control type stays Q2 — CS6 is why you trust the control label. The open type’s P1 falls when the watching machine is thin. CS6 is not a one-way “add points to countries we dislike” knob. This table is still v1.0 beside v1.1. The four v1.2 ballot cases above are separate demo cases, not these types, not letter grades, and not country statistics.