Add execution-sensitivity robustness suite (anchor, grid, real-market)

docs/results/execution_sensitivity_anchor/anchor_robustness.csv

@@ -0,0 +1,10 @@
+regime,level,cold_return,warm_start_return,init_cost_gap
+calm,E0_ideal,0.0932,0.0932,-0.0
+calm,E1_default_stress,0.0885,0.0931,0.0045
+calm,E2_harsh_corner,0.0723,0.0925,0.0203
+high_vol,E0_ideal,0.0994,0.0994,0.0
+high_vol,E1_default_stress,0.1044,0.0987,-0.0057
+high_vol,E2_harsh_corner,0.0924,0.0996,0.0072
+jump_tail,E0_ideal,0.1036,0.1036,-0.0
+jump_tail,E1_default_stress,0.087,0.1016,0.0146
+jump_tail,E2_harsh_corner,0.0609,0.1072,0.0463

docs/results/execution_sensitivity_anchor/anchor_robustness.md

@@ -0,0 +1,20 @@
+# Buy-and-Hold Anchor Initialization Robustness
+
+Cold anchor builds its full position from cash; warm-start anchor is
+seeded with the same holdings free of construction cost. A large
+`init_cost_gap` at stressed levels means the cold anchor's penalty is
+partly an initialization artifact rather than a fair execution result.
+
+| Regime | Level | Cold return | Warm-start return | Init-cost gap |
+| --- | --- | ---: | ---: | ---: |
+| calm | E0_ideal | +0.0932 | +0.0932 | -0.0000 |
+| calm | E1_default_stress | +0.0885 | +0.0931 | +0.0045 |
+| calm | E2_harsh_corner | +0.0723 | +0.0925 | +0.0203 |
+| high_vol | E0_ideal | +0.0994 | +0.0994 | +0.0000 |
+| high_vol | E1_default_stress | +0.1044 | +0.0987 | -0.0057 |
+| high_vol | E2_harsh_corner | +0.0924 | +0.0996 | +0.0072 |
+| jump_tail | E0_ideal | +0.1036 | +0.1036 | -0.0000 |
+| jump_tail | E1_default_stress | +0.0870 | +0.1016 | +0.0146 |
+| jump_tail | E2_harsh_corner | +0.0609 | +0.1072 | +0.0463 |
+
+Mean init-cost gap: ideal +0.0000, harsh corner +0.0246. The gap widens under stress, so part of the anchor's harsh-corner disadvantage is construction cost; we report the warm-start anchor alongside the cold one and separate this from the leaderboard-fragility result.

docs/results/execution_sensitivity_grid/param_grid.csv

@@ -0,0 +1,6 @@
+impact,participation,latency,market,kendall_tau_vs_ideal,top3_jaccard_vs_ideal
+0.05,0.1,0,liquid large-cap,1.0,1.0
+0.1,0.05,1,typical equity,0.81,0.5
+0.15,0.05,1,default stress,0.81,0.5
+0.2,0.03,2,small-cap / volatile,0.524,0.5
+0.3,0.01,3,stressed / illiquid,0.429,0.2

docs/results/execution_sensitivity_grid/param_grid.md

@@ -0,0 +1,15 @@
+# Execution-Parameter Sensitivity Grid (high-volatility regime)
+
+Kendall tau-b between the idealized leaderboard and each parameter cell,
+spanning liquid to stressed markets. If reordering appeared only at
+extreme parameters, mild cells would show tau near 1.
+
+| Impact | Participation | Latency | Market | Kendall tau vs ideal | Top-3 Jaccard |
+| ---: | ---: | ---: | --- | ---: | ---: |
+| 0.05 | 10% | 0 | liquid large-cap | 1.0 | 1.0 |
+| 0.1 | 5% | 1 | typical equity | 0.81 | 0.5 |
+| 0.15 | 5% | 1 | default stress | 0.81 | 0.5 |
+| 0.2 | 3% | 2 | small-cap / volatile | 0.524 | 0.5 |
+| 0.3 | 1% | 3 | stressed / illiquid | 0.429 | 0.2 |
+
+Even the mild 'typical equity' cell (impact 0.10, 5\% participation, 1-bar latency) reorders the leaderboard (tau 0.81), so the effect is not confined to extreme parameters.

docs/results/execution_sensitivity_real/execution_sensitivity_real.md

@@ -0,0 +1,18 @@
+# Execution Sensitivity on Real Market Data (Deterministic Agents)
+
+Ranking stability between idealized (E0) and stressed execution on real
+Yahoo OHLCV, mirroring the synthetic-regime analysis. Low Kendall tau
+means the friction-driven leaderboard reordering persists on real prices.
+
+| Window | Level vs E0 | Kendall tau | Top-3 Jaccard |
+| --- | --- | ---: | ---: |
+| rates_drawdown_2022 | E1_default_stress | 0.905 | 0.500 |
+| rates_drawdown_2022 | E2_harsh_corner | 0.810 | 0.500 |
+| rates_drawdown_2022 | E2_latency_3 | 0.810 | 0.500 |
+| rates_drawdown_2022 | E2_participation_1pct | 0.905 | 0.500 |
+| rates_drawdown_2022 | E2_spread_20bps | 0.905 | 0.500 |
+| recent_cross_asset | E1_default_stress | 0.905 | 0.500 |
+| recent_cross_asset | E2_harsh_corner | 0.810 | 0.500 |
+| recent_cross_asset | E2_latency_3 | 0.810 | 0.500 |
+| recent_cross_asset | E2_participation_1pct | 0.905 | 0.500 |
+| recent_cross_asset | E2_spread_20bps | 0.905 | 0.500 |

docs/results/execution_sensitivity_real/real_rank_stability.csv

@@ -0,0 +1,31 @@
+window,level_a,level_b,agent_count,kendall_tau,top_3_jaccard
+rates_drawdown_2022,E0_ideal,E1_default_stress,7,0.9047619047619048,0.5
+rates_drawdown_2022,E0_ideal,E2_harsh_corner,7,0.8095238095238095,0.5
+rates_drawdown_2022,E0_ideal,E2_latency_3,7,0.8095238095238095,0.5
+rates_drawdown_2022,E0_ideal,E2_participation_1pct,7,0.9047619047619048,0.5
+rates_drawdown_2022,E0_ideal,E2_spread_20bps,7,0.9047619047619048,0.5
+rates_drawdown_2022,E1_default_stress,E2_harsh_corner,7,0.9047619047619048,1.0
+rates_drawdown_2022,E1_default_stress,E2_latency_3,7,0.9047619047619048,1.0
+rates_drawdown_2022,E1_default_stress,E2_participation_1pct,7,1.0,1.0
+rates_drawdown_2022,E1_default_stress,E2_spread_20bps,7,1.0,1.0
+rates_drawdown_2022,E2_harsh_corner,E2_latency_3,7,1.0,1.0
+rates_drawdown_2022,E2_harsh_corner,E2_participation_1pct,7,0.9047619047619048,1.0
+rates_drawdown_2022,E2_harsh_corner,E2_spread_20bps,7,0.9047619047619048,1.0
+rates_drawdown_2022,E2_latency_3,E2_participation_1pct,7,0.9047619047619048,1.0
+rates_drawdown_2022,E2_latency_3,E2_spread_20bps,7,0.9047619047619048,1.0
+rates_drawdown_2022,E2_participation_1pct,E2_spread_20bps,7,1.0,1.0
+recent_cross_asset,E0_ideal,E1_default_stress,7,0.9047619047619048,0.5
+recent_cross_asset,E0_ideal,E2_harsh_corner,7,0.8095238095238095,0.5
+recent_cross_asset,E0_ideal,E2_latency_3,7,0.8095238095238095,0.5
+recent_cross_asset,E0_ideal,E2_participation_1pct,7,0.9047619047619048,0.5
+recent_cross_asset,E0_ideal,E2_spread_20bps,7,0.9047619047619048,0.5
+recent_cross_asset,E1_default_stress,E2_harsh_corner,7,0.9047619047619048,1.0
+recent_cross_asset,E1_default_stress,E2_latency_3,7,0.9047619047619048,1.0
+recent_cross_asset,E1_default_stress,E2_participation_1pct,7,1.0,1.0
+recent_cross_asset,E1_default_stress,E2_spread_20bps,7,1.0,1.0
+recent_cross_asset,E2_harsh_corner,E2_latency_3,7,1.0,1.0
+recent_cross_asset,E2_harsh_corner,E2_participation_1pct,7,0.9047619047619048,1.0
+recent_cross_asset,E2_harsh_corner,E2_spread_20bps,7,0.9047619047619048,1.0
+recent_cross_asset,E2_latency_3,E2_participation_1pct,7,0.9047619047619048,1.0
+recent_cross_asset,E2_latency_3,E2_spread_20bps,7,0.9047619047619048,1.0
+recent_cross_asset,E2_participation_1pct,E2_spread_20bps,7,1.0,1.0