Repricing Core Book v2 — Locality
Live top-3
Validated track record
Walk-forward — Period A vs Period B
Rotation behavior
Pillar representation in top-3
Methodology
The locality signal evaluates each candidate across multiple causal paths ((pillar, order, mechanism, locality_score, amplified_by_pillars)). Order weights {1: 1.0, 2: 0.7, 3: 0.4} reflect distance from first-order locality (asset IS the friendly side) to third-order (asset captures regime-induced vol). Cross-pillar amplification kicks in when amplifying pillars co-fire. Per-pillar regime amplifier (1.5× when pillar percentile ≥ 70, 1.0× mid, 0.5× < 30). Cycle / inflation / credit overlays applied on top.
The signal scoring is universe-agnostic — it can score any asset given its tags. The portfolio construction is "rank everything cross-pillar, hold top-3, monitor and rotate." Druckenmiller's "all eggs in one basket, watching that basket carefully" applied at the cross-pillar level.
Vol-target wrapper: Moreira-Muir 2017 inverse-vol scaling, 15% target, gross_floor 0.5×, gross_cap 2.5×. Cycle overlay: RECESSION × 0.5 gross.
Source code: bci_compute/locality_signal.py · bci_compute/locality_signal_loader.py · bci_compute/food_pillar.py · scripts/locality_signal_cheap_probe.py · scripts/locality_signal_walkforward.py. Pre-reg locked at docs/locality_signal_phase0_pre_reg.md.