What changed

• integrates the redesigned frozen-cohort dashboard into run-live-forward-daily
• refreshes only archived-cohort prices by default, extends immutable July 20–27 Ridge/XGBoost portfolios, and creates zero new model lists
• adds a holding-period laboratory for fixed cohorts and 1–60-session rebalance policies with Ridge/XGBoost toggles and an always-visible SPY benchmark
• publishes an allowlisted, schema-validated static Pages bundle and rejects private fields or malformed cohort data
• updates the README and deployment guide, including the public dashboard link
• adds pull-request validation for the Pages artifact while keeping production deployment restricted to main

Why

The old daily portfolios remain useful as frozen evidence for studying holding periods and replacement behavior, but they should not keep changing membership or be confused with the primary horizon-matched 5/10/20 experiment. The daily dashboard path now monitors those cohorts without silently retraining them.

Impact

python -m src.cli run-live-forward-daily --as-of YYYY-MM-DD now performs a lightweight after-close refresh, rebuilds both the private dashboard and sanitized site/ bundle, and reports new_model_lists: 0. Historical one-session retraining remains available only through --retrain.

Validation

• python -m pytest -q: 218 passed
• Ruff checks passed
• public bundle validated: 17 snapshots, 14 evaluations, 12 cohorts, 120 policy paths
• private and public dashboards rendered in Chrome with no JavaScript errors
• cohort/policy controls and the 1–20-session comparison table were exercised in-browser
• GitHub Actions workflow parsed successfully
• git diff --check passed