Atlas Score v1 shipped earlier today with a known limitation: it weighted CHANGE more than LEVEL. Russia (B− 23.8), China (B− 25.4), and North Korea (B− 24.0) all scored low despite massive baseline censorship — because nothing was shifting this week.

v2 fixes that with a base-rate-weighted formula:

Component	v1	v2
Structural baseline (12-month censorship-weighted incidents + tier floor)	0%	50%
30-day forecast mean	0%	20%
7-day current forecast max_risk	40%	15%
24h incident density	25%	10%
Anomaly disagreement	5%	5%
7-day trend delta	20%	0% (moved to separate field)
Calibration quality	10%	moved to separate `confidence` field

Two important honesty additions:

Censorship/mixed-mechanism incidents are weighted 3.0×; pure disruptions weighted 0.3× (filters out OONI network-outage noise that previously inflated democracies).
Tier floors prevent chronic-censorship countries from scoring 0 due to probe-network blind spots. KP / TM / ER carry a visibility_warning in raw_signals because tier-4 + zero observed incidents = we can't see inside.

Big gainers (chronic-blocker correction)

Country	v1 score	v2 score	Δ
AZ Azerbaijan	15.3 A	50.2 C	+34.9
CN China	25.7 B-	59.1 C	+33.4
ER Eritrea	23.9 B-	56.7 C	+32.8
TM Turkmenistan	24.7 B-	57.1 C	+32.4
KP North Korea	24.5 B-	56.5 C	+32.0
SY Syria	24.2 B-	47.9 C	+23.7
RU Russia	23.6 B-	47.1 C	+23.5
BY Belarus	24.4 B-	46.8 C	+22.4
CU Cuba	23.9 B-	46.1 C	+22.2

Big losers (change-spike countries correctly de-rated)

SD Sudan: 64.5 C → 30.2 B (−34.3pts) — low chronic baseline, recent spike was a noise event
BR Brazil: 59.4 C → 37.9 B (−21.5pts)
MY Malaysia: 59.2 C → 40.5 B (−18.7pts)
PH Philippines: 59.9 C → 43.8 B (−16.1pts)

Top 10 leaderboard now

IR Iran — 77.4 F (was #4 D in v1)
PK Pakistan — 70.9 D
EG Egypt — 69.4 D
UZ Uzbekistan — 69.1 D
VE Venezuela — 63.6 D
VN Vietnam — 59.2 C
CN China — 59.1 C
SA Saudi Arabia — 59.1 C
LB Lebanon — 57.8 C
MM Myanmar — 57.1 C

Honest gap to the promote criteria

Decision targets: RU/CN/KP/BY should hit D+ (≥65). v2 gets them into the C cluster (47-59) — clearly elevated from v1's B-, but still below D. To push them past 65 strictly would require ≥58% tier weight, which crosses from "rebalanced composite" into "tier with decoration." Holding v2 as experimental until grade bands are widened (e.g. C 40-55, D 55-70, F 70+).

Live at

GET /v1/atlas/score-v2 — leaderboard (≥experimental)
GET /v1/atlas/score-v2/IR — single country breakdown
GET /v1/atlas/score — v1 still the default