Derived by the Research Lead from Herculano, 'Bayesian Parametric Portfolio Policies' (arXiv 2602.21173, 2026): the paper's BPPP estimator, prior, learning-rate rule, clipped posterior averaging, position/gross limits and turnover penalty are ported from Sections 3.2, 4, 4.1, 4.3 and Appendix B; the ASSET SET is our choice. The paper times five Kenneth French long-short factors, which are not tradeable here, so we apply the identical machinery to nine tradeable zero-investment tilts: F_k = monthly return of sector ETF k minus monthly return of SPY, for k in {XLB, XLE, XLF, XLI, XLK, XLP, XLU, XLV, XLY} (all listed since Dec 1998; XLC/XLRE excluded for short history), with SPY as the market held at weight 1 (the paper's benchmark w_b = (1,0,...,0)). PREDICTORS: only the paper's factor-specific families (Appendix B.1), computed for each of the 10 series (SPY and the nine tilts) from their own monthly returns: (1) time-series momentum = mean of the last 12 monthly returns; (2) cross-sectional momentum = rank of those 12-month means across the 10 series mapped linearly to [-1,1]; (3) valuation = -(v_t - mean(v_{t-59..t})) with v_t the cumulative sum of log(1+R); (4) reversal = -(R_{t-1} - mean(R_{t-35..t})); (5) volatility = -std of the last 12 monthly returns. That is 50 predictors plus a constant (L=51), 9 x 51 = 459 coefficients against a 120-month window: the p >> T regime the paper is built for. The paper's 212 Open Source Asset Pricing anomaly returns are NOT used (we do not hold that dataset); the paper reports that removing them leaves 0.87 pp of timing (t=1.13), so this is a test of the factor-specific-only BPPP on a different asset set, not a replication of the 2.08 pp headline. Every predictor is standardised with an expanding mean/std of its own history up to t, winsorised at +/-3, set to 0 while missing, with the first 12 standardised rows set to 0 (paper's startup convention). EXECUTION: decisions at the close of the last session of each calendar month (session calendar helper), filled at the next session's open (platform next_open_fill on USEQ, supported). Monthly returns used for estimation are close-to-close of month-end sessions. SIZING (risk-control route (b): signal/rebalance exit, capped sizing, no stops): with final tilt weights w_k and fixed book_scale b=0.4, target notional SPY = b*(1 - sum_k w_k)*equity and sector k = b*w_k*equity (a positive w_k is long the sector and short the same notional of SPY, netted into the single SPY position). Because |w_k| <= 0.6 and sum|w_k| <= 1.5, the SPY leg ranges 0..1.0x equity, each sector leg at most 0.24x, and gross notional at most 1.6x equity (inside Reg-T 2x); annualised vol is expected around 7-10%. Platform default rebalance threshold: a leg is only traded when its target notional changes by more than 0.5% of equity, to avoid sub-cost fragments. Dividends are not in PnL (applies equally to all legs, roughly cancels on the tilts; the SPY leg is understated by ~1.5%/yr).