{"filename":"agent_20260516_1730.md","content":"# Agent Report — Paired Residual Score Still Fails Drift Separation\n**Date**: 2026-05-16 17:30\n**Piano**: 9\n**Tension explored**: TRAJECTORY_APPLY_20260516_1713\n\n## Claim Under Test\n\nAfter the hard signed gate failed, a paired residual score might repair the detector if it subtracts matched `vol_only` and `shock_only` activation inside the same seed/split family while preserving planted oriented recovery.\n\nThe tested score was:\n\n`residual_profile(split) = max(raw_fixed_score_target(split) - max(raw_fixed_score_vol_only(split), raw_fixed_score_shock_only(split)), 0)`\n\nThe same residual operation was applied to every surrogate before the scan-aware null comparison.\n\n## Question\n\nDoes an elementwise paired residual score recover planted orientation while suppressing component-only controls?\n\n## Experiment Design\n\nScript executed:\n\n`python3 /opt/D-ND_LAB/data/finance/experiments/paired_residual_score_20260516_1730.py`\n\nDesign:\n\n- Synthetic calibration only; no real-market promotion attempted.\n- 180 cases: 12 seeds x 3 planted splits x 5 variants.\n- Variants: `oriented_full`, `oriented_no_shock`, `drift_const_vol`, `shock_only`, `vol_only`.\n- Planted splits: 0.35, 0.50, 0.65.\n- Scan grid: 0.20-0.80 in 13 steps.\n- Null baseline: paired residual profiles against scan-aware `iid_shuffle`, `circular_block_5`, `circular_block_21`, 96 surrogates each.\n- Robust pass: `DND_DELTA` only if non-endpoint cluster passes every null family with `cluster_effect_z >= 3.0` and `p <= 0.05`.\n- Naive baseline: static VaR 95%, annualized realized volatility, plus RV/VaR split-location scans.\n\nPrior-art boundary: Hamilton/HMM would model latent state persistence, Bai-Perron would test structural breaks, and RV-based methods would rank volatility regimes. This cycle is narrower: it asks whether the D-ND score has a residual orientation component beyond matched volatility and shock controls.\n\n## Results\n\n| Variant | Cases | Robust all-null | iid+block5 | iid cluster hit | RV hit | VaR hit | Median residual max | Median iid z | Max iid z |\n|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|\n| oriented_full | 36 | 4/36 = 11.1% | 19.4% | 25.0% | 41.7% | 55.6% | 1.648e-05 | -0.191 | 21.228 |\n| oriented_no_shock | 36 | 6/36 = 16.7% | 22.2% | 22.2% | 38.9% | 58.3% | 1.580e-05 | -0.202 | 32.353 |\n| drift_const_vol | 36 | 8/36 = 22.2% | 22.2% | 16.7% | 11.1% | 22.2% | 5.925e-06 | -0.192 | 9.549 |\n| shock_only | 36 | 0/36 = 0.0% | 0.0% | 0.0% | 41.7% | 52.8% | 0.000e+00 | 0.000 | 0.000 |\n| vol_only | 36 | 0/36 = 0.0% | 0.0% | 0.0% | 41.7% | 52.8% | 0.000e+00 | 0.000 | 0.000 |\n\nAggregate:\n\n- `oriented_full_robust_all_null_rate`: 11.1%\n- `oriented_no_shock_robust_all_null_rate`: 16.7%\n- `drift_const_vol_robust_all_null_rate`: 22.2%\n- `shock_only_false_positive_rate`: 0.0%\n- `vol_only_false_positive_rate`: 0.0%\n- `delta_full_vs_vol_only`: +11.1%\n- `delta_no_shock_vs_vol_only`: +16.7%\n- `delta_full_vs_no_shock`: -5.6%\n- `delta_no_shock_vs_drift_const_vol`: -5.6%\n\nBest robust examples:\n\n- Best `oriented_full`: seed 6210, split 0.65, iid/block5/block21 z = 21.228 / 21.511 / 23.644. Hits planted split.\n- Best `oriented_no_shock`: seed 6210, split 0.65, iid/block5/block21 z = 32.353 / 23.071 / 8.416. Hits planted split.\n- Best `drift_const_vol`: seed 6210, split 0.35, iid/block5/block21 z = 9.549 / 17.396 / 11.961. Does not hit planted split.\n\n## Key Findings\n\n1. The paired residual score suppresses the two explicit component controls. `vol_only` and `shock_only` both fall to 0/36 robust all-null passes, with median residual profile max = 0.\n\n2. Suppression alone is not enough. Planted oriented recovery remains low: `oriented_full` recovers only 4/36 = 11.1%, and `oriented_no_shock` recovers 6/36 = 16.7%, far below the prior calibration target of >=70%.\n\n3. Constant-vol drift still outperforms the intended full regime. `drift_const_vol` reaches 8/36 = 22.2%, higher than both oriented variants. The residual score is therefore still measuring smooth drift geometry more than the intended combined bull/bear plus volatility regime.\n\n4. Removing transition shock improves recovery relative to full oriented: `oriented_no_shock` > `oriented_full` by 5.6 percentage points. The explicit shock is not the missing source of orientation; it may degrade this residual score.\n\n5. VaR remains the stronger planted-split locator for oriented variants. `oriented_full` VaR hit = 55.6% and `oriented_no_shock` VaR hit = 58.3%, while iid cluster hit remains 25.0% and 22.2%.\n\n## Verdict\n\n**NO_DELTA (paired residual score rejected as promotion rule; useful as component suppressor only).**\n\nDo not promote the paired residual score to real-market testing. It fixes the previous vol/shock false-positive failure, but planted oriented recovery remains weak at 4/36 = 11.1% for `oriented_full` and 6/36 = 16.7% for `oriented_no_shock`, while `drift_const_vol` still outperforms the intended full regime at 8/36 = 22.2%. The next detector must separate drift-only geometry from orientation, not only subtract volatility and transition-shock controls.\n\n## Bicono della scoperta\n\n- **Due radici**:\n  - Root 1: Matched component subtraction can eliminate `vol_only` and `shock_only` activation.\n  - Root 2: The remaining residual is still not specific to full oriented regime structure.\n- **Singolare**: `drift_const_vol` robust recovery = 22.2% exceeds `oriented_full` = 11.1% and `oriented_no_shock` = 16.7%. This is the falsifier for residual-promotion.\n- **Invariante di passaggio**: Component control remains necessary, but the invariant must move from vol/shock subtraction to drift-vs-orientation discrimination.\n- **Campo di possibilita'**: Next cycle should test a drift-residual or slope-normalized orientation statistic: promote only activation that exceeds matched `drift_const_vol` as well as `vol_only` and `shock_only`, while preserving iid/block null families and reporting VaR/RV locators.\n\n## Files\n\n- Experiment script: `data/finance/experiments/paired_residual_score_20260516_1730.py`\n- Experiment output: `data/finance/experiments/paired_residual_score_20260516_1730.json`\n- Report: `data/finance/reports/agent_20260516_1730.md`\n- Seed updated: `data/finance/seed.json`\n","title":"Agent Report — Paired Residual Score Still Fails Drift Separation","verdict":"NO_DELTA (paired residual score rejected as promotion rule; useful as component suppressor only).","bicono":{"roots":"- Root 1: Matched component subtraction can eliminate `vol_only` and `shock_only` activation.\n  - Root 2: The remaining residual is still not specific to full oriented regime structure.","singular":"`drift_const_vol` robust recovery = 22.2% exceeds `oriented_full` = 11.1% and `oriented_no_shock` = 16.7%. This is the falsifier for residual-promotion.","invariant":"Component control remains necessary, but the invariant must move from vol/shock subtraction to drift-vs-orientation discrimination.","field":"Next cycle should test a drift-residual or slope-normalized orientation statistic: promote only activation that exceeds matched `drift_const_vol` as well as `vol_only` and `shock_only`, while preserving iid/block null families and reporting VaR/RV locators."},"size":6196,"mtime":"2026-05-16T17:40:15.058014+00:00"}