Trajectory-weighted calibration
One weight per full trajectory, with entry and exit markers, so multi-period calibration cannot silently destroy the panel structure a cross-sectional reweight would ignore.
strategies / dynamics
strategy 06 · dynamics
Time.
Populace's certified layer is cross-sectional: one snapshot, calibrated to one year of administrative targets. Retirement and social-insurance policy analysis needs a person aged forward through earnings, family structure, disability, and mortality — the benchmark U.S. models that do this (DYNASIM, MINT, CBOLT) are closed: internal to government, tied to restricted administrative records, or reachable only through institutional relationships. Dynamics specifies an open alternative, built as an extension of the certified cross-sectional layer rather than a replacement for it.
01 — the 30-second version
A design paper, not a validated run.
This page describes what the dynamics design paper specifies, not a completed backtest — no trajectory has been scored against administrative outcomes yet. The design commits to four elements before any implementation ships:
Transitions as conditional models
A Dynamics operator treats state transitions — earnings, family structure, disability, mortality, program participation — as conditional models, mixing deterministic demographic hazards with machine-learned earnings processes rather than a single monolithic projection.
Domains of validity, not point forecasts
Where parameter uncertainty dominates — including the 75-year actuarial balance — the design refuses a single point estimate and publishes sensitivity surfaces instead.
A scoring protocol before merge
Every claim is required to resolve against administrative publications, backtest with leakage control, or compute exactly from statute — and a contribution merges only when it improves a held-out score, not on review judgment alone.
From the populace-dynamics design paper abstract (preprint). The layer builds on Populace's certified cross-sectional foundation — 0.038 holdout loss against 0.317 for the enhanced CPS it replaced — but that number describes the cross-sectional layer being extended, not a result the dynamics layer has produced.
02 — scope
One validation domain first, country-agnostic by construction.
U.S. Social Security is the first validation domain the design targets — a single program with published long-range projections to score against. The layer itself is specified to be country-agnostic: nothing in the trajectory-weighting or transition-model design is U.S.-specific, but no other country's implementation has started.