Skip to content
AI Work Index

V5 Roadmap

V5 is archived as a former live research program. Its sidecars and experimental comparison artifacts remain published for auditability beneath the current live release.

V5 is no longer the live model. This page is preserved as the archived roadmap for the V5 research program that fed into the later V6 release.

The V5 comparison trail is preserved.

The roadmap no longer stops at sidecars. A combined experimental model remains available with structural validation 2/2 and realized-risk validation 2/4.

Open V5 model note →

Current live

V7

the live version when this roadmap was authored

Retained V4.3 baseline

2/3

shadow validation gates that cleared

Task-native coverage

492

occupations carried into the retained pre-V6 baseline

Status

archived live release

next major scientific release

Published Sidecars

Augmentation heterogeneity

pilot sidecar

Workflow-sensitive augmentation priors published separately from the live score.

Download v5-augmentation-heterogeneity.json

Empirical mobility transitions

enriched sidecar

Hybrid mobility sidecar blending observed priors, destination quality, wage preservation, and training ease.

Download v5-empirical-mobility.json

Posterior uncertainty

pilot sidecar

Source-calibrated latent measurement sidecar built on persisted exposure-source percentiles.

Download v5-posterior-uncertainty.json

Realized-risk forecast

pilot sidecar

Conservative realized-risk proxy published alongside structural and near-term risk.

Download v5-realized-risk.json

Integrated Candidate

V5 Model Note

The integrated V5 model is no longer live. This note preserves the final promotion comparison against the retained V4.3 baseline and the published V5 adjunct layers for auditability under V6.

Open report →

Status

experimental only

Structural validation

2/2

Realized validation

2/4

Transition band flips

169

Release Thesis

V5 should be the first release that upgrades the science beyond exposure alone: augmentation heterogeneity, empirical mobility, richer uncertainty, and a clearer realized-risk forecast.

effective_coverage = Σ_t w_it · exposure_t · success_t

automation_pressure = Σ_t w_it · exposure_t · success_t · autonomy_t · (1 - bottleneck_t)

augmentation_upside = Σ_t w_it · exposure_t · success_t · (1 - autonomy_t) · bottleneck_t

realized_risk_proxy = structural_risk_i · adoption_t · capability_t · κ_short

Workstreams

Augmentation heterogeneity

Replace one universal augmentation rule with workflow-sensitive priors informed by empirical productivity studies.

Deliverables

  • Role/archetype-level augmentation priors
  • Updated task-to-augmentation mapping
  • Validation note on which occupations move most

Empirical mobility transitions

Move transition support from broad heuristic rankings toward observed occupational pathways and wage/training tradeoffs.

Deliverables

  • Occupation-to-occupation transition prior
  • Observed wage delta and training duration fields
  • Transition ranking override when empirical evidence exists

Posterior uncertainty

Upgrade bootstrap intervals into a latent measurement model across the exposure sources and propagate uncertainty through the final score.

Deliverables

  • Latent exposure posterior
  • 80/95% intervals on exposure and net risk
  • Separation of statistical uncertainty from provenance confidence

Realized-risk forecast

Calibrate short-run realized effects separately from structural overlap using adoption and capability evidence.

Deliverables

  • Adoption and capability scalars grounded in literature
  • Realized-risk proxy distinct from structural and near-term risk
  • Validation family for short-run labour signals

Promotion Discipline

  • Each workstream must ship as an auditable sidecar artifact before it touches the headline model.
  • Validation families must map cleanly to the construct they are testing: structural, transition, augmentation, or realized risk.
  • No release should absorb multiple unvalidated constructs into the headline score at once.