Policy happens somewhere. Pick the place.

Reskilling. Sector-focused RCTs lift earnings 14–40% and persist (Year Up; Project QUEST @14y; Per Scholas @7y) — but on selected participants, at limited scale; broad WIA/WIOA programs show ~null average effects. Each slider detent is a named package pairing reach with success: today’s programs ≈ 40/15/5% success on near/moderate/broad skill jumps; sectoral-at-scale ≈ 60/35/15%; a GI-Bill-scale effort (the only precedent at the required magnitude: ~7.8M trained over 12 years) caps the dial at 75/55/35%. Success falls with skill distance in every package — 100% is not on the dial. Katz et al. 2022.

Relocation. US geographic mobility has fallen 40+ years; interstate moves ~halved since the 1980s, and displaced / less-educated workers move least; TAA relocation allowances historically moved workers in the low thousands per year. The slider scales take-up per ring — in-state matches (which include staying put in your own labor market) vs cross-state moves: today’s programs ≈ 40/4%; subsidies + licensure portability ≈ 55/12%; historic (WWII-scale) mobilization caps the dial at 70/25%. Molloy–Smith–Wozniak.

What the caps mean. The dials stop at evidence, not at logic: at zero friction the matching engine could place ~all feasible matches (the possibility frontier shown under the bar) and nearly clear the shortage on paper — a statement about skill distance, not an achievable program. Track shading: solid = observed programs; tinted = evidence-backed ambition; hatched = historical precedent only. Package rates are drafted calibrations pending final review.

Migration. Two-sided by construction: migrants add labor supply (which fills shortage jobs) and consumer demand (which creates them — the local-demand channel of Hong & McLaren 2015; NAS 2017). The net effect fills part of the gap — not a free lunch, and it attenuates toward neutral under the most extreme AI scenario, where displacement has already loosened the markets migrants would fill.

All figures are 2034 comparative-static potential, not forecasts. Everything on this page — the tiles above and the levers here — reads the same certified state-level cube (metro + non-metro, full US coverage): shortage at transformative AI × 10% wage cap, displaced = AI-attributable (net of the no-AI corner). Relocation & reskilling are matching overlays; migration volume is a structural GE re-solve; the bar fills domestic reabsorption first, migration on the remainder. Occupation granularity: 22 SOC major groups at state resolution — coarse cells overstate matchability (credential gates inside a group are invisible), which is part of why the take-up dials cap where they do. Migration tick marks (≈’20 trough, ≈’23–24 surge) are qualitative: the model baseline is a gross inflow, historical comparisons are net.