EUNOMIA indices — measured

The three aspirational index axes · frozen gold sets · exact-label · Wilson 95% CI · signed on the estate rail

Each index was declared UNMEASURED first (JL.5), then its frozen gold item set was authored in the measurement harness (10 items, canary-anchored, CC0-1.0) and graded exact-label. The item set IS the instrument — a stranger can re-derive every number from it. Measurement, not certification.

ai-economy-indexMEASURED · n=10

AI-economy (agent-payments) index — real vs gamified · San Francisco

0.600 strong (7b) 95% CI [0.313, 0.832]0.900 baseline (0.5b) 95% CI [0.596, 0.982]

labels: REAL / GAMIFIED · frozen gold set in the measurement harness (generate_eunomia_items.py)

human-labourMEASURED · n=10

human-labour displacement index · Toronto

1.000 strong (7b) 95% CI [0.723, 1.000]0.500 baseline (0.5b) 95% CI [0.237, 0.763]

labels: COMPLIANT / NON_COMPLIANT · frozen gold set in the measurement harness (generate_eunomia_items.py)

humanoid-labourMEASURED · n=10

humanoid-labour index · London

1.000 strong (7b) 95% CI [0.723, 1.000]0.500 baseline (0.5b) 95% CI [0.237, 0.763]

labels: COMPLIANT / NON_COMPLIANT · frozen gold set in the measurement harness (generate_eunomia_items.py)

Honest reading

• ai-economy-index: baseline (0.5b) scores above the strong model (0.9 vs 0.6). The confidence intervals overlap, so no leader is declared (deliberately conservative). The 7b model over-thinks the real-vs-gamified scenarios; the small model pattern-matches cleanly. This is a measured fact, not a model endorsement.

• human-labour and humanoid-labour: strong model 1.0 (10/10) with a floor at the Wilson lower bound 0.723; baseline at chance (0.5). The scenarios are deterministic — the compliance fact is stated in each item.

• Item counts are n=10 per axis — the Wilson half-width is wide (±~0.25 at n=10), so treat point estimates with their intervals, not as exact truths. Evidence cards are committed to the measurement harness (benchmark-results/eunomia-index-results-2026-08-28/).