Walk-forward confirmation of UPRO ensemble frontier

Objective

Confirm whether the leveraged-sp500-indicator-ensemble feasible CAGR/drawdown frontier survives when parameters are frozen from expanding training prefixes only.

Approach

1. Preregistered local 33-variant c5 neighborhood, focused-grid sensitivity, and -22% training buffer family.
2. Expanding annual January cutoffs from 2010-01-01 with one-bar execution and 5 bps costs.
3. Ranked training-only by parent objective; stitched OOS path is the primary verdict.
4. Published notebook, analysis HTML, and Research Library campaign iteration 3.

Files changed

• research/leveraged_sp500_walk_forward.py, tests/test_leveraged_sp500_walk_forward.py
• research/campaigns/leveraged-sp500-indicator-ensemble/walk-forward-trials.json, walk-forward-diagnostics.json, campaign.json
• research/experiments/1312.ipynb, docs/experiments/1312-analysis.html
• docs/research/leveraged-sp500-indicator-ensemble.html, docs/site-data/research/leveraged-sp500-indicator-ensemble.json

Results

• Primary local stitched OOS: CAGR 12.23%, max DD -25.45%, clears -25%: no.
• All 17 primary annual OOS segments individually clear -25%.
• Safety-margin stitched: CAGR 11.01%, max DD -21.53%, clears: yes.
• Focused-grid stitched: CAGR 12.41%, max DD -24.26%, clears: yes.
• Full-sample parameter choice was not used.

Conclusions

Frontier survival is mixed under the preregistered primary protocol. Keep exploratory; do not promote a strategy or indicator from this confirmation alone.

Links

• Research page: docs/research/leveraged-sp500-indicator-ensemble.html
• Notebook analysis: docs/experiments/1312-analysis.html
• Issue: #1312