Add Strategy Explorer, YAML config exposure, and calendar/drift fixes

.gitignore

@@ -30,6 +30,7 @@ ENV/
 .mypy_cache/
 .ruff_cache/
 .uv-*-cache/
+.uv-cache-codex-review/
 .coverage
 .coverage.*
 htmlcov/

README.md

@@ -166,12 +166,15 @@ a config's `robustness:` block in one pass. All five branch on
 `validation.method`: with `walk_forward`, each starts from the same
 walk-forward-stitched out-of-sample result rather than a single backtest, so
 the evidence never silently comes from a different validation method than
-the one configured. `stress-test` and `sensitivity` re-run the whole
-walk-forward selection process per scenario/candidate, since each one
-represents a different cost/methodology assumption or parameter to
-re-optimise under. `bootstrap` and `permutation-test` do not: they resample
-or permute the walk-forward's already-realised out-of-sample return series
-statistically, without re-running the selection process itself.
+the one configured. `sensitivity` re-runs the whole walk-forward selection process per candidate
+parameter pair. `stress-test`'s commission/slippage scenarios reuse the
+baseline's own cached per-fold candidate weights and only re-score them
+under the new costs (cheaper, since selection never depends on execution
+costs); its execution-delay and reduced-universe scenarios genuinely change
+the weights themselves, so each still re-runs selection end to end.
+`bootstrap` and `permutation-test` do neither: they resample or permute the
+walk-forward's already-realised out-of-sample return series statistically,
+without touching the selection process at all.
 
 `walk-forward`, `stress-test`, `sensitivity` and `robustness` show a live
 progress bar with an ETA in the terminal, and checkpoint their progress to
@@ -222,8 +225,10 @@ progress bar with an ETA while a run is in flight. Both modes' Robustness
 tab includes stress tests, block bootstrap, a Monte Carlo permutation test
 and a 2-parameter sensitivity heatmap,
 individually or via "Run all robustness tests" — in Walk-forward mode,
-stress tests and sensitivity re-run the whole selection process per
-scenario/cell rather than a single backtest.
+sensitivity re-runs the whole selection process per cell; stress tests do
+too for execution-delay/reduced-universe scenarios, but reuse the
+baseline's cached candidate weights (just re-scored under new costs) for
+commission/slippage scenarios.
 
 ![QuantLab dashboard results](reports/figures/dashboard_results.png)
 
@@ -281,13 +286,13 @@ produces the full 26-fold report the numbers below are drawn from.
 Real, non-cherry-picked results from real Yahoo Finance / Binance data (2008–2025
 for ETFs, 2018–2025 for BTC), net of modelled transaction costs:
 
-| Experiment | Universe | Period | CAGR | Sharpe | Max DD | Trades |
+| Experiment | Universe | Period | CAGR | Sharpe | Max DD | Fills |
 | --- | --- | --- | --- | --- | --- | --- |
-| Cross-sectional momentum (example above) | 8 multi-asset ETFs | 2008–2025 | 5.7% | 0.44 | −15.8% | 277 |
+| Cross-sectional momentum (example above) | 8 multi-asset ETFs | 2008–2025 | 5.5% | 0.43 | −15.8% | 1,514 |
 | ↳ walk-forward out-of-sample | same | 26 folds | 4.7% | 0.32 | −18.5% | — |
-| Mean reversion (z-score) | 5 equity ETFs | 2010–2025 | 4.4% | 0.25 | −34.6% | 1,180 |
+| Mean reversion (z-score) | 5 equity ETFs | 2010–2025 | 4.4% | 0.25 | −34.6% | 3,351 |
 | Pairs trading (EWA/EWC, vs SPY) | 2 country ETFs | 2010–2025 | 0.9% | −0.35 | −8.0% | 1,314 |
-| Trend following (BTC) | BTCUSDT | 2018–2025 | 30.3% | 0.93 | −43.9% | 1,560 |
+| Trend following (BTC) | BTCUSDT | 2018–2025 | 30.3% | 0.93 | −43.9% | 1,563 |
 
 Every number above — including the negative Sharpe on the pairs trade — is
 reported as computed; nothing here is filtered for looking good. **These are

configs/btc_trend.yaml

@@ -2,6 +2,15 @@
 # Crypto uses 365 periods/year and higher costs than ETFs.
 experiment_name: btc_trend_following
 
+research_question: >-
+  Does a simple moving-average trend-following rule on Bitcoin capture a
+  persistent risk-adjusted edge once volatility targeting and elevated
+  crypto transaction costs are accounted for?
+hypothesis: >-
+  H1: net of modelled transaction costs, long-only trend following on BTC
+  has a genuine positive risk-adjusted edge. H0: any apparent edge is
+  indistinguishable from noise or is eliminated by trading costs.
+
 data:
   instruments:
     - symbol: BTCUSDT
@@ -18,6 +27,9 @@ strategy:
     fast_window: 20
     slow_window: 100
     long_only: true
+  # Price series the strategy's own generate_signals() reads. Execution/
+  # costs always use the raw close regardless of this setting.
+  signal_price_type: adjusted_close
 
 portfolio:
   allocator: volatility_targeting
@@ -57,12 +69,30 @@ reproducibility:
 robustness:
   stress_test:
     enabled: true
+    # Each list is a set of magnitudes evaluated independently (one
+    # scenario row per value); an empty list disables that scenario type
+    # entirely. Crypto's higher baseline costs (see execution: above) make
+    # a larger commission stress meaningful here too.
+    commission_multipliers: [2.0, 5.0]
+    slippage_multipliers: [2.0]
+    execution_delays: [1]
+    best_days_removed: [10]
   bootstrap:
     enabled: true
+    # Width of the reported p_lower/p_upper percentile band (0.90 -> the
+    # 5th/95th percentiles).
+    confidence_level: 0.90
   permutation_test:
     enabled: true
   sensitivity:
     enabled: true
     parameters:
       fast_window: [10, 20, 40]
       slow_window: [50, 100, 200]
+
+# Where and what a run saves. Omit entirely for the default: reports/
+# generated/<experiment_name>/, both artefact toggles enabled.
+output:
+  # directory: reports/generated/btc_trend_following
+  save_html_report: true
+  save_figures: true

configs/default.yaml

@@ -18,6 +18,9 @@ data:
 strategy:
   name: buy_and_hold
   parameters: {}
+  # Price series the strategy's own generate_signals() reads. Execution/
+  # costs always use the raw close regardless of this setting.
+  signal_price_type: adjusted_close
 
 portfolio:
   allocator: equal_weight
@@ -60,5 +63,15 @@ robustness:
     enabled: true
   bootstrap:
     enabled: true
+    # Width of the reported p_lower/p_upper percentile band (0.90 -> the
+    # 5th/95th percentiles).
+    confidence_level: 0.90
   permutation_test:
     enabled: true
+
+# Where and what a run saves. Omit entirely for the default: reports/
+# generated/<experiment_name>/, both artefact toggles enabled.
+output:
+  # directory: reports/generated/default_buy_and_hold
+  save_html_report: true
+  save_figures: true

configs/demo_offline.yaml

@@ -3,6 +3,16 @@
 # NOTE: the prices here are SYNTHETIC (generated), not real market data.
 experiment_name: demo_offline_momentum
 
+research_question: >-
+  Does cross-sectional momentum on this small offline demo universe
+  produce a stable out-of-sample signal, purely as a fast, no-network
+  illustration of the walk-forward pipeline?
+hypothesis: >-
+  H1: the walk-forward pipeline selects a consistent momentum parameter
+  across folds on this synthetic demo data. H0: fold-to-fold selection is
+  effectively random. (Demo data is synthetic -- not evidence about real
+  markets either way.)
+
 data:
   # The synthetic symbols model US equity ETFs.
   instruments:
@@ -34,6 +44,9 @@ strategy:
     bottom_fraction: 0.5
     long_short: false
     signal_scaling: binary
+  # Price series the strategy's own generate_signals() reads. Execution/
+  # costs always use the raw close regardless of this setting.
+  signal_price_type: adjusted_close
 
 portfolio:
   allocator: inverse_volatility
@@ -64,6 +77,9 @@ validation:
   train_window: 500
   validation_window: 126
   test_window: 126
+  # Advance between consecutive folds' train windows. Omit (or set equal to
+  # test_window) for the default: contiguous, non-overlapping test blocks.
+  step: 126
   expanding: true
   optimization_metric: sharpe
 
@@ -80,12 +96,30 @@ reproducibility:
 robustness:
   stress_test:
     enabled: true
+    # Each list is a set of magnitudes evaluated independently (one
+    # scenario row per value); an empty list disables that scenario type
+    # entirely.
+    commission_multipliers: [2.0, 5.0]
+    slippage_multipliers: [2.0]
+    execution_delays: [1]
+    best_days_removed: [10]
+    reduce_universe_by: [1]
   bootstrap:
     enabled: true
+    # Width of the reported p_lower/p_upper percentile band (0.90 -> the
+    # 5th/95th percentiles).
+    confidence_level: 0.90
   permutation_test:
     enabled: true
   sensitivity:
     enabled: true
     parameters:
       lookback_period: [126, 189, 252]
       top_fraction: [0.25, 0.50]
+
+# Where and what a run saves. Omit entirely for the default: reports/
+# generated/<experiment_name>/, both artefact toggles enabled.
+output:
+  # directory: reports/generated/demo_offline_momentum
+  save_html_report: true
+  save_figures: true

configs/mean_reversion_etfs.yaml

@@ -3,6 +3,15 @@
 # research notebook.
 experiment_name: mean_reversion_etfs
 
+research_question: >-
+  Does short-term price mean reversion across liquid ETFs generate a
+  positive risk-adjusted edge after transaction costs?
+hypothesis: >-
+  H1: net of modelled transaction costs, a rolling z-score mean-reversion
+  signal across this ETF universe has a genuine positive risk-adjusted
+  edge. H0: any apparent edge is indistinguishable from noise or is
+  eliminated by trading costs.
+
 data:
   instruments:
     - symbol: SPY
@@ -29,10 +38,14 @@ strategy:
   name: mean_reversion
   parameters:
     lookback_period: 20
-    entry_zscore: 2.0
-    exit_zscore: 0.5
-    stop_zscore: 4.0
+    indicator: zscore
+    entry_threshold: 2.0
+    exit_threshold: 0.5
+    stop_threshold: 4.0
     long_only: true
+  # Price series the strategy's own generate_signals() reads. Execution/
+  # costs always use the raw close regardless of this setting.
+  signal_price_type: adjusted_close
 
 portfolio:
   allocator: equal_weight
@@ -65,12 +78,30 @@ reproducibility:
 robustness:
   stress_test:
     enabled: true
+    # Each list is a set of magnitudes evaluated independently (one
+    # scenario row per value); an empty list disables that scenario type
+    # entirely.
+    commission_multipliers: [2.0, 5.0]
+    slippage_multipliers: [2.0]
+    execution_delays: [1]
+    best_days_removed: [10]
+    reduce_universe_by: [1]
   bootstrap:
     enabled: true
+    # Width of the reported p_lower/p_upper percentile band (0.90 -> the
+    # 5th/95th percentiles).
+    confidence_level: 0.90
   permutation_test:
     enabled: true
   sensitivity:
     enabled: true
     parameters:
       lookback_period: [10, 20, 40]
-      entry_zscore: [1.5, 2.0, 2.5]
+      entry_threshold: [1.5, 2.0, 2.5]
+
+# Where and what a run saves. Omit entirely for the default: reports/
+# generated/<experiment_name>/, both artefact toggles enabled.
+output:
+  # directory: reports/generated/mean_reversion_etfs
+  save_html_report: true
+  save_figures: true

configs/momentum_sp500.yaml

@@ -1,11 +1,20 @@
 # Example experiment:
 # Robust Cross-Sectional Momentum Across Liquid Multi-Asset ETFs.
-#
-# Research question: Can cross-sectional momentum generate stable out-of-sample
-# risk-adjusted returns across liquid ETFs after transaction costs and
-# volatility targeting?
 experiment_name: cross_sectional_momentum_etfs
 
+# Overrides the auto-generated report text (see quantlab.reporting.
+# research_summary) with this experiment's own wording. Both are optional --
+# omit either to fall back to the auto-generated version.
+research_question: >-
+  Can cross-sectional momentum generate stable out-of-sample risk-adjusted
+  returns across liquid ETFs after transaction costs and volatility
+  targeting?
+hypothesis: >-
+  H1: net of modelled transaction costs, long-only cross-sectional momentum
+  across this ETF universe has a genuine positive risk-adjusted edge. H0: any
+  apparent edge is indistinguishable from noise or is eliminated by trading
+  costs and volatility targeting.
+
 data:
   instruments:
     - symbol: SPY   # US large cap
@@ -46,6 +55,11 @@ strategy:
     bottom_fraction: 0.25
     long_short: false        # long-only for the main version
     signal_scaling: binary
+  # Price series the strategy's own generate_signals() reads. Execution/
+  # costs always use the raw close regardless of this setting. Default is
+  # "adjusted_close" (recommended for equities/ETFs, avoids false signals
+  # from splits/dividends) -- shown explicitly here for illustration.
+  signal_price_type: adjusted_close
 
 portfolio:
   allocator: inverse_volatility
@@ -77,6 +91,13 @@ validation:
   train_window: 1000
   validation_window: 252
   test_window: 126
+  # Advance between consecutive folds' train windows. Omit (or set equal to
+  # test_window) for the default: contiguous, non-overlapping test blocks.
+  # A smaller step overlaps test blocks for denser evaluation at higher
+  # compute cost. step must not exceed test_window -- a larger value is
+  # rejected outright (it would skip dates between folds' test blocks
+  # entirely), not silently allowed to skip periods.
+  step: 126
   expanding: true
   optimization_metric: sharpe
   # Freeze the published search space: tune timing while keeping the 25%
@@ -91,10 +112,21 @@ reproducibility:
 robustness:
   stress_test:
     enabled: true
+    # Each list is a set of magnitudes evaluated independently (one
+    # scenario row per value); an empty list disables that scenario type
+    # entirely. Values shown here match this project's own defaults.
+    commission_multipliers: [2.0, 5.0]
+    slippage_multipliers: [2.0]
+    execution_delays: [1]
+    best_days_removed: [10]
+    reduce_universe_by: [1]
   bootstrap:
     enabled: true
     n_iterations: 1000
     block_size: 5
+    # Width of the reported p_lower/p_upper percentile band (0.90 -> the
+    # 5th/95th percentiles).
+    confidence_level: 0.90
   permutation_test:
     enabled: true
     n_iterations: 1000
@@ -105,3 +137,10 @@ robustness:
     parameters:
       lookback_period: [126, 189, 252]
       top_fraction: [0.10, 0.25, 0.50]
+
+# Where and what a run saves. Omit entirely for the default: reports/
+# generated/<experiment_name>/, both artefact toggles enabled.
+output:
+  # directory: reports/generated/cross_sectional_momentum_etfs
+  save_html_report: true
+  save_figures: true