← AI features

Layer-conditional weighting (LS6) — model card

v1.0 · Tier 1-4 (orchestrator)Live Regime + class-conditional factor weights.

Purpose

The same factor (e.g. F05_mom3m) carries different signal value in different regimes and asset classes. Risk-on rewards momentum; risk-off rewards quality. Crypto rewards funding-rate signals; equities reward insider buying. v76 LS6 implements the conditional weighting so the orchestrator's composite score adapts to context.

Decision flow

InputsWeight modulation
regime = risk-on, class = StocksF05/F06 momentum × 1.4; F22/F26 quality × 0.7
regime = risk-off, class = StocksF05/F06 momentum × 0.6; F22/F26 quality × 1.5
regime = transition, any classTier 4 sentiment × 0.4 (suppress whipsaw); Tier 1 macro × 1.3
class = CryptoPerpF31 funding × 1.6; F32 OI × 1.4 (crypto-only)
class = VolatilityF09 VIX-driven factors × 2.0; F22 quality × 0.2 (irrelevant)
class = REITsF03 yield curve × 1.5 (rates-sensitive)

Outputs

Modulates the weight passed to calculateScore(factors, weights) in the orchestrator. Final score still uses the same composite math; LS6 just changes which factors' contributions matter most.

Why this matters (Bible §1 Law 1)

The named dimension — probability-first forecast + per-action playbook + calibration — depends on accurate scoring across regimes. A class-blind, regime-blind orchestrator (the pre-LS6 state) produces materially wrong scores in transition regimes and crypto/vol classes. LS6 + the 14-class profile (in prescience) is the third structural piece of the moat (per ADR-0008).

Source

engine/v76-layer-conditional.js, fed by engine/regime.js + engine/factors.js

Reviewed: 2026-04-27 · Next: 2026-07-27 · Per Bible §15.61