Unify live dashboard horizon monitoring

README.md

@@ -5,7 +5,7 @@
 
 ![Python 3.11+](https://img.shields.io/badge/Python-3.11%2B-2563eb)
 ![Research status](https://img.shields.io/badge/status-paper--live_research-d97706)
-![Tests](https://img.shields.io/badge/tests-218_passing-0f766e)
+![Tests](https://img.shields.io/badge/tests-221_passing-0f766e)
 ![License](https://img.shields.io/badge/license-MIT-64748b)
 
 The project asks a narrow investment question:
@@ -227,7 +227,14 @@ visible from the shared decision-close baseline. It reports net return,
 SPY-relative return, replacement-only turnover, and rebalance counts under
 identical execution assumptions.
 
-[Open the deployed live-forward dashboard →](https://hansun103.github.io/AI-Driven-Portfolio-Optimization-Backtesting-System/)
+The deployed monitor now has two linked pages with one shared visual shell:
+
+- [Live Forward Monitor — default view →](https://hansun103.github.io/AI-Driven-Portfolio-Optimization-Backtesting-System/)
+- [Horizon Comparison — 5/10/20-session research →](https://hansun103.github.io/AI-Driven-Portfolio-Optimization-Backtesting-System/horizons.html)
+
+The original Stock-level view remains canonical rather than being duplicated
+on the horizon page. The two-page contract and evidence boundaries are recorded
+in the [dashboard merge design](docs/design/DASHBOARD_TWO_PAGE_MERGE.md).
 
 This monitoring track does not replace the primary horizon-matched 5/10/20
 experiment described above. The former one-session fitting workflow remains
@@ -378,20 +385,29 @@ because they weaken the preferred narrative.
 ## Known limitations
 
 1. No redesigned 20-session paper-live forecast has matured yet.
-2. The strongest portfolio results remain development evidence, not a genuine
+2. The strongest repaired historical test covers roughly 2018–2023, not a
+   full decade. It therefore spans too few independent market regimes for a
+   durable claim. Extending point-in-time testing toward at least ten years is
+   a research target, not a guarantee of adequacy: long-window backtests still
+   require regime analysis, untouched confirmation, and correction for the
+   number of strategies tried. See
+   [CFA Institute’s backtesting guidance](https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/backtesting-and-simulation)
+   and the
+   [deflated-Sharpe discussion of sample length and multiple testing](https://academic.oup.com/jrssig/article/18/6/22/7038278).
+3. The strongest portfolio results remain development evidence, not a genuine
    untouched confirmation.
-3. Public data does not fully solve delisted-security coverage or historical
+4. Public data does not fully solve delisted-security coverage or historical
    sector classification.
-4. The average ranking edge is small and varies materially by year and regime.
-5. Sentiment is promising, but coverage, timestamp quality, and causal
+5. The average ranking edge is small and varies materially by year and regime.
+6. Sentiment is promising, but coverage, timestamp quality, and causal
    attribution remain incomplete.
-6. The high-recall survival model has not passed its activation gates and
+7. The high-recall survival model has not passed its activation gates and
    remains a shadow watchlist.
-7. Same-close tactical execution is unvalidated without intraday quotes and
+8. Same-close tactical execution is unvalidated without intraday quotes and
    actual paper or broker fills.
-8. Scheduled full top-40 replacement remains under audit; the six-session
+9. Scheduled full top-40 replacement remains under audit; the six-session
    ablation is too short to activate a freeze or buffer.
-9. Static cost assumptions do not establish capacity, taxes, borrow
+10. Static cost assumptions do not establish capacity, taxes, borrow
    availability, market impact, or partial-fill behavior.
 
 ## Research disclaimer

docs/design/DASHBOARD_TWO_PAGE_MERGE.md

@@ -0,0 +1,88 @@
+# Two-page dashboard merge design
+
+Status: local proposal awaiting approval. Do not merge or deploy yet.
+
+## Decision
+
+Publish one dashboard product with two real static pages:
+
+| Page | Proposed URL | Responsibility |
+| --- | --- | --- |
+| Live Forward Monitor | `dashboard.html` | Daily operational evidence, benchmark comparison, stability, the canonical stock-level view, and experiment history |
+| Horizon Comparison | `horizons.html` | Horizon-matched 5/10/20-session progress, validation context, model specifications, SHAP explanations, and pipeline health |
+
+Both headers use the same two-option page switcher. The selected page is visually
+active. Ordinary links are preferable to a hidden single-page application because
+browser back/forward works, each page can fail independently, and GitHub Pages can
+serve the files without a runtime.
+
+## Shared stock-level view
+
+The Horizon Comparison page does not implement another portfolio table.
+
+The existing Stock-level view remains the canonical component. Its model selector
+will be extended to understand both:
+
+- legacy daily live-forward snapshots; and
+- v2 Ridge/XGBoost arms for the 5-, 10-, and 20-session horizons.
+
+An `Inspect stocks` action on a horizon model will link to a URL such as:
+
+```text
+dashboard.html?decision=2026-07-29&family=xgb_ranker&horizon=20#stock-level
+```
+
+The original page reads those parameters, selects the requested frozen portfolio,
+and scrolls to the shared Stock-level view. This avoids divergent filtering,
+sorting, return formatting, and holding-level evaluation logic.
+
+## Partial and mature evidence
+
+The horizon page reports three different evidence states without mixing them:
+
+1. **Awaiting entry** — the portfolio is frozen but its next-open entry has not
+   occurred.
+2. **Interim mark** — one or more open-to-open observations exist. The page shows
+   session progress and the compounded partial return immediately.
+3. **Matured label** — the complete 5-, 10-, or 20-session evaluation is available
+   and may enter the formal prospective comparison.
+
+The page never displays validation return as live return. A zero-session portfolio
+is shown as `0 / horizon` and `Awaiting entry`, not as a zero realized return.
+
+## Explainability
+
+Every newly frozen v2 forecast writes `shap_summary.csv`.
+
+- Ridge uses linear SHAP contributions around the current eligible cohort as the
+  background.
+- XGBoost uses exact TreeSHAP contributions from the fitted booster.
+- The dashboard publishes mean absolute SHAP, signed mean SHAP, positive
+  contribution share, method, and explanation-row count.
+
+Each model card contains a collapsed `Show SHAP feature importance` control.
+Opening it displays a horizontal importance chart. Old immutable forecasts that
+predate this artifact state that SHAP was not captured; coefficients or XGBoost
+gain are not relabelled as SHAP.
+
+## Daily refresh contract
+
+The daily pipeline should:
+
+1. refresh data and generate the current snapshots;
+2. evaluate all available interim open-to-open observations;
+3. evaluate newly matured forecasts;
+4. generate SHAP summaries for new horizon snapshots;
+5. rebuild `dashboard.html` and `horizons.html`;
+6. run public-payload and browser smoke checks; and
+7. publish both pages together through the existing GitHub Pages action.
+
+Publishing must be atomic: a cycle should never expose one page with a newer data
+contract than the other.
+
+## Approval boundary
+
+The current local Horizon Comparison preview demonstrates the second page. After
+approval, the merge work will add the page switcher and v2 arm support to the
+original dashboard, generate the final `horizons.html`, update the refresh and
+deployment workflows, and then request approval before pushing.

site/data/manifest.json

@@ -1,10 +1,13 @@
 {
-  "schema_version": 2,
-  "generated_at_utc": "2026-07-29T16:55:17.967215+00:00",
+  "schema_version": 3,
+  "generated_at_utc": "2026-07-29T22:29:54.012984+00:00",
   "latest_decision_date": "2026-07-28",
   "snapshot_count": 17,
   "evaluation_count": 14,
   "dashboard_sha256": "9d15420d6e5370038763505be97432e381f7a0932cc1aeac953399c030b2185c",
   "dashboard_bytes": 546502,
+  "horizon_dashboard_sha256": "20d6bdaa4d619ecca77441801a4f2e63b93c665822979b278cb490999bd4695a",
+  "horizon_dashboard_bytes": 66861,
+  "horizon_forecast_arms": 6,
   "source_policy": "sanitized_live_forward_outputs_only"
 }