methodbriefing · v0.1.0
Panel persistence shrinkage
For repeated jurisdiction, occupation, or agency panels, scores the last-print baseline and shrinks agent updates toward it unless current evidence clears a stated bar.
0
runs
0
targets
0
forecasters
1
versions
Purpose
Prevent agents from overreacting on stable panel targets where last year's official value is usually the strongest single predictor.
Mechanism
- Record a last-print persistence benchmark and score it beside the agent run.
- Estimate a panel-level drift or shrinkage term from comparable entities before applying inside-view adjustments.
- Require explicit current evidence before moving materially away from the persistence-plus-shrinkage baseline.
Checks
- The trace reports the persistence benchmark and the agent's delta from it.
- Large departures name the evidence source and the expected sign of the departure.
- Intervals widen for entities with volatile prior prints, small samples, or known release noise.
Inputs
- Prior official print for the same entity
- Comparable panel history across entities or occupations
- Current entity-specific evidence strong enough to justify a departure
Limitations
- Can underreact to true breaks, new policy regimes, or one-off administrative shocks.
- Should not be used when the target is a new series with no entity-level prior.
- Needs walk-forward scoring before promotion beyond panel-style targets.
Versions
v0.1.0panel-persistence-shrinkage@0.1.00 runs
Runs handed this briefing
not used yetNo recorded runs yet. This briefing is registered for future contrasts, so its first runs can be compared against unbriefed controls before any promotion decision.