V7 Release Note

Current release

V7 is the live structural release as of 2026-06-11. It scores 562 occupations and 88 modern roles. V6 baseline scores are preserved in every occupation record for comparison.

What V7 changes

V7 introduces two formula changes to the V6 structural model, both grounded in recent research and using data already available in the pipeline:

1. Task-concentration exposure buffer

Hampole et al. (2025) find that mean task exposure depresses labour demand, but exposure concentrated in a few tasks offsets those losses — workers reallocate effort to non-exposed tasks. V7 therefore applies the task-concentration signal (derived from Anthropic task penetration data matched to O*NET task statements) as a buffer that reduces effective exposure.

task_signal = task_exposure_concentration x task_effective_coverage

exposure_v7 = clamp01(exposure x (1 - 0.20 x task_signal))

Correction (7 Jun 2026): the original V7 release applied this term as an exposure amplifier. That inverted the sign of the cited finding — Hampole et al. show concentration buffers labour-demand losses. The term now reduces effective exposure; all published scores reflect the corrected formula.

492 of 562 occupations have task data; 361 end up with a non-zero task_signal. Occupations without task data (or with zero measured concentration) receive task_signal = 0, preserving V6-equivalent exposure.

2. Demand-persistence proxy

Motivated by the Imas price-elasticity critique: when AI reduces service costs, demand may increase enough to offset displacement. Occupation-level elasticity data does not exist, so V7 builds a ranked composite proxy from four existing labour-demand signals — a partial response, not an elasticity measure.

demand_persistence = 0.4 x momentum_rank + 0.3 x vacancy_rank + 0.2 x scarcity_rank + 0.1 x demand_bonus_rank

demand_resilience_v7 = clamp01(base_resilience x 0.45 + demand_signal_bonus + 0.10 x demand_persistence)

This is not a measure of price elasticity. It is a demand-side counterforce proxy using market momentum, vacancy trends, wage scarcity, and official demand signals.

Complete V7 formula

exposure = reliability-weighted 4-source ensemble (Felten AIOE, Anthropic, Eloundou, ILO)

task_signal = task_exposure_concentration x task_effective_coverage

exposure_v7 = clamp01(exposure x (1 - 0.20 x task_signal))

bottleneck = pctile_rank(Pizzinelli theta)

displacement_pressure = exposure_v7 x (1 - bottleneck)

demand_persistence = 0.4 x mm_rank + 0.3 x vacancy_rank + 0.2 x scarcity_rank + 0.1 x bonus_rank

demand_resilience = clamp01(base_resilience x 0.45 + demand_signal_bonus + 0.10 x demand_persistence)

headline_risk = displacement_pressure x (1 - demand_resilience)

augmentation = exposure_v7 x bottleneck x base_resilience

Stability metrics

Median |delta risk|

0.0092

Target: < 0.03

Band flips

53 / 562 (9.4%)

Target: < 15%

Spearman rank correlation

0.9996

Target: > 0.95

Directional accuracy

100% (2/2 pairwise)

Target: ≥ 50%

Validation

V7 passes 221 of 221 internal release and data-contract checks. All 6 anchor occupations pass (Software Developer, Data Entry Clerk, Surgeon, Telemarketer, Registered Nurse, Data Scientist). External calibration is reported separately: the direct BLS segment remains significantly negative, and the Singapore cluster labour backtest currently passes 2 of 3 discriminating checks (a no-variance retrenchment column is excluded as inconclusive).

Research grounding

Hampole et al. (2025, NBER w33509) — Mean task exposure depresses within-firm labour demand, but concentration of exposure in a few tasks offsets losses: workers reallocate effort toward non-exposed tasks. V7 applies concentration as an exposure buffer accordingly.
Imas & Shukla (2026) — Argue exposure alone cannot predict displacement without output-demand price elasticity and job dimensionality. V7's demand-persistence proxy is a partial, labour-demand-side response: it measures recent demand persistence (momentum, vacancies, scarcity), not output-price elasticity, and does not capture dimensionality.
Brookings (2026) — Career pathway erosion when gateway occupations are automated. Informs the transition-capacity layer (separate from headline score).

Data and downloads

V7 dataset is available at /data in JSON and CSV formats. V6 baseline scores are preserved in the baseline_v6 field of every occupation record. New V7 fields: task_signal, demand_persistence, exposure_v7. The non-promoted forecast-readiness matrix maps labour-outcome and adoption inputs to their existing source owners and validation gates.