Source: task5-ashare-leader/output/streaming_hist_2012/task22-results-vs-predictions.md
Last modified: 2026-05-19 03:46:27 · 2,519 bytes

Task #22 Results vs Pre-Registered Hypotheses

Source prediction: RayDalio task22-pre-registered-hypotheses.md locked before Phase 1 results.

Baseline Predictions vs Actual

Item Pre-reg prediction Actual Result
Extended annualized 15.5-18.5%, median 16.8% Y2 16.21% from 2015-02-17; 17.59% from 2018-02-09 Match
Extended Sharpe 0.80-0.95, median 0.87 Y2 0.842 from 2015-02-17; 0.881 from 2018-02-09 Match
Extended MDD -22% to -28%, risk from 2015 H2 Y2 -29.25% from 2015-02-17 Slightly worse than predicted
2014 slow bull +5% to +15% expected Not evaluable under current rule: strict candidates start 2015-02-17 due 756d EPS CAGR lag Prediction not testable
2015 leveraged bull -5% to +20%, high variance +61.52% in 2015 H1 with avg exposure 45.5% Strong upside surprise
2016 0% to +3% 2016 full year -3.64%; Jan-Feb circuit-breaker window 0 exposure / 0 return Slight miss on full year, defense held in circuit-breaker
2017 blue chip +5% to +12% +19.53%, avg exposure 72.5% Upside surprise

A/B Flip Watch

Item Pre-reg expectation Actual Phase 1 Status
Hard stop -8% Margin shrinks but still reject Still strongly rejected: hard8 Sharpe 0.268, MDD -55.08% vs three-axis Sharpe 0.669, MDD -28.77% same true extended window Reaffirmed / stronger than caution
cap 18 vs cap 12 Possible reversal Not run in Phase 1; Phase 2 pending Pending
public-fund sponsorship Could weaken Direction remains positive, but market-value-only beats combined bonus Partial flip / needs Phase 2
chain_cap_40 Not expected to hurt conclusion Y2 with chain cap 40 is top v1.3 scenario Reaffirmed
B3 RPS dual-window Could weaken Still approved: +0.091 Sharpe vs v1 production and lower MDD Reaffirmed
M-filter shutoff Reaffirm / strengthen Always-on MDD -59.0%; raw open1/close1 lower Sharpe than open3/close5 but better MDD and better 2024 OOS-B Reaffirm defense, no production change yet
Smart capacity Reaffirm / strengthen Combined with early4 + chain40 is top scenario; smart alone not useful Reaffirmed as combo only

New Finding

- 2010/2011 sample feasibility smoke passed for price, factors, benchmark, suspension/ST, industry, and fund universe. This means a true 2014 trade evaluation is likely feasible if we rebuild raw cache from 2010-01-01 or 2011-01-01, rather than trying to evaluate 2014 from a 2012-only warm-up.