V6 Release Note

Archived

V6 was the live structural release from 1 Apr 2026 to 7 Apr 2026. It has been superseded by V7, which adds task-concentration-weighted exposure and a demand-persistence proxy. V6 baseline scores are preserved in every V7 occupation record for auditability.

What V6 introduced

V6 replaced the single-axis market modifier with a two-axis structural formula separating displacement pressure from demand resilience:

headline_risk = displacement_pressure x (1 - demand_resilience)

displacement_pressure = exposure x (1 - bottleneck)

demand_resilience = min(1, base_resilience x 0.45 + demand_signal_bonus)

Key changes from V5:

Deterministic 4-source exposure ensemble (Felten AIOE, Anthropic, Eloundou, ILO) with reliability weighting per Frank et al. (2025)
Explicit demand resilience as a separate axis rather than collapsed into a market modifier
Occupation-level industry-footprint momentum where available, falling back to group prior
Pure confidence scoring with policy caps for fallback mappings and contested signals
Statistical uncertainty intervals via 5,000-sample Monte Carlo bootstrap
Task-primitives sidecar (computed but not used in headline score)

Validation

V6 passed 165 of 169 structural validation checks. Cluster-level backtesting achieved 2/4 directional accuracy with 100% pairwise directional accuracy. The BLS cross-country validation showed negative correlation between risk and employment projections, consistent with the structural pressure interpretation.

Superseded by V7

V7 builds on V6 by wiring task-concentration data (previously a sidecar) into the headline exposure formula and adding a demand-persistence proxy to demand resilience. V6 scores are preserved as baseline_v6 in every occupation record. See the V7 release note and data downloads for the archived V6 dataset.