{"filename":"agent_20260516_2003.md","content":"# Agent Report — Lag-Map Memory Object Has Strong Purity Bias but Is Underpowered\n**Date**: 2026-05-16 20:10\n**Piano**: 16\n**Tension explored**: TRAJECTORY_APPLY_20260516_1812 / T11_PAIRED_BLOCK_CALIBRATION_REJECTED\n\n## Claim Under Test\n\nCycle `20260516_1812` rejected additional residual/veto layers on the current raw split score. The claim under test here is a redesign of the positive object itself: instead of planting drift plus volatility and asking D-ND to recover it, plant a lag-map memory transition at matched scale and test the D-ND orientation variable directly against VaR/RV.\n\nScore under test:\n\n`z_t = r_t / rolling_sigma_t`\n\n`c_t = det([z_t, z_{t-1}], [z_{t+1}, z_t]) = z_t^2 - z_{t-1}z_{t+1}`\n\nScan score:\n\n`abs(mean(c_right) - mean(c_left))`\n\n## Question\n\nIf the planted positive object is lag-map memory at constant volatility, does a local orientation-area detector beat VaR/RV split localization while rejecting iid, drift-only, and volatility-only controls?\n\n## Experiment Design\n\nScript executed:\n\n`python3 /opt/D-ND_LAB/data/finance/experiments/lagmap_memory_object_20260516_2003.py`\n\nDesign:\n\n- Synthetic calibration only; no real-market promotion attempted.\n- 144 cases: 12 seeds x 3 planted splits x 4 variants.\n- Positive: `lag_memory_const_vol`.\n- Controls: `iid_const_vol`, `drift_const_vol`, `vol_only`.\n- Planted splits: 0.35, 0.50, 0.65.\n- Scan grid: 0.20-0.80 in 13 steps.\n- Null baseline: scan-aware `iid_shuffle`, `circular_block_5`, `circular_block_21`, 128 surrogates each.\n- Stage1 pass: `DND_DELTA` under iid and block5.\n- Robust all-null pass: Stage1 plus block21 pass.\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; Bai-Perron would test break location; RV/VaR directly target risk-scale change. This cycle asks whether the D-ND lag-map area adds a structural target variable that is not just a delayed volatility detector.\n\n## Results\n\n| Variant | Cases | Stage1 iid+block5 | Robust all-null | iid planted hit | RV hit | VaR hit | Median area gap | Median RV gap | Median VaR gap |\n|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|\n| lag_memory_const_vol | 36 | 3/36 = 8.3% | 3/36 = 8.3% | 22.2% | 8.3% | 8.3% | 0.1419 | ~0.0000 | 0.00119 |\n| iid_const_vol | 36 | 0/36 = 0.0% | 0/36 = 0.0% | 2.8% | 0.0% | 30.6% | 0.0466 | ~0.0000 | 0.00129 |\n| drift_const_vol | 36 | 0/36 = 0.0% | 0/36 = 0.0% | 2.8% | 8.3% | 33.3% | 0.0380 | ~0.0000 | 0.00221 |\n| vol_only | 36 | 1/36 = 2.8% | 1/36 = 2.8% | 11.1% | 97.2% | 77.8% | 0.0604 | 0.1270 | 0.01194 |\n\nAggregate:\n\n- `positive_stage1_iid_block5_rate`: 3/36 = 8.3%\n- `positive_robust_all_null_rate`: 3/36 = 8.3%\n- `control_stage1_iid_block5_rate`: 1/108 = 0.9%\n- `control_robust_all_null_rate`: 1/108 = 0.9%\n- `positive_iid_cluster_hit_rate`: 8/36 = 22.2%\n- `positive_rv_hit_rate`: 3/36 = 8.3%\n- `positive_var_hit_rate`: 3/36 = 8.3%\n- `promotion_ready`: false\n\nStrongest positive survivors:\n\n- seed 6202, split 0.50: iid z = 17.112, block5 z = 9.986, block21 z = 9.083, planted split hit = true.\n- seed 6207, split 0.50: iid z = 4.687, block5 z = 4.352, block21 z = 4.556, planted split hit = true.\n- seed 6202, split 0.65: iid z = 3.407, block5 z = 4.480, block21 z = 6.869, planted split hit = true.\n\n## Key Findings\n\n1. The redesigned object improves purity but not power. Controls are almost fully suppressed: aggregate control robust all-null pass is 0.9%.\n\n2. Positive recovery is still far below the calibration target. `lag_memory_const_vol` reaches only 8.3% Stage1 and 8.3% robust all-null recovery, versus the required >=70% before any market test.\n\n3. The positive object is not reducible to RV/VaR in construction. Median RV gap is effectively zero for `lag_memory_const_vol`, and RV/VaR each localize only 8.3% of positive splits.\n\n4. The detector sees isolated true positives, but the median z remains below threshold across all null families. The primary issue is sensitivity of the local-area mean-gap scan; residual control leakage remains at 1/108 = 0.9% aggregate robust all-null pass, with the vol_only edge case at 1/36 = 2.8%.\n\n5. Volatility-only remains a useful adversarial control: RV and VaR hit 97.2% and 77.8%, while D-ND robust false positive is 2.8%. This confirms the detector is not merely RV/VaR, but it is not yet strong enough as a positive detector.\n\n## Verdict\n\n**NO_DELTA. The claim that the local orientation-area detector beats VaR/RV split localization while rejecting iid, drift-only, and volatility-only controls was falsified at this cycle. The new claim that emerged is narrower: the lag-map memory object is a cleaner positive-control direction, but the current local orientation-area detector is rejected as a promotion detector.**\n\nDo not promote to real-market testing. The useful constraint is narrower: changing the planted object away from drift/volatility repairs purity, but a mean-gap scan of local M-area is still too underpowered. Next work should amplify or aggregate lag-map memory evidence before split scanning, rather than return to the old raw score or add another residual veto.\n\n## Bicono della scoperta\n\n- **Due radici**:\n  - Root 1: A constant-volatility lag-memory positive object creates an orientation target that VaR/RV do not trivially dominate.\n  - Root 2: The current local-area split scan recovers only rare survivors and cannot serve as a detector.\n- **Singolare**: The singular point is the purity/power split: controls fall to 0.9% robust pass, but positives remain at 8.3%.\n- **Invariante di passaggio**: A redesigned positive object is not sufficient; the D-ND detector must recover planted lag-map memory at high rate before any real-market claim.\n- **Campo di possibilita'**: Keep the lag-memory object as a calibration target, but replace the mean-gap scan with a memory accumulator, persistence statistic, or sequence-level likelihood surrogate that tests orientation before split localization.\n\n## Files\n\n- Experiment script: `data/finance/experiments/lagmap_memory_object_20260516_2003.py`\n- Experiment output: `data/finance/experiments/lagmap_memory_object_20260516_2003.json`\n- Assertions: `data/finance/assertions/assertions_20260516_2003.json`\n- Report: `data/finance/reports/agent_20260516_2003.md`\n- Seed updated: `data/finance/seed.json`\n","title":"Agent Report — Lag-Map Memory Object Has Strong Purity Bias but Is Underpowered","verdict":"NO_DELTA. The claim that the local orientation-area detector beats VaR/RV split localization while rejecting iid, drift-only, and volatility-only controls was falsified at this cycle. The new claim th","bicono":{"roots":"- Root 1: A constant-volatility lag-memory positive object creates an orientation target that VaR/RV do not trivially dominate.\n  - Root 2: The current local-area split scan recovers only rare survivors and cannot serve as a detector.","singular":"The singular point is the purity/power split: controls fall to 0.9% robust pass, but positives remain at 8.3%.","invariant":"A redesigned positive object is not sufficient; the D-ND detector must recover planted lag-map memory at high rate before any real-market claim.","field":"Keep the lag-memory object as a calibration target, but replace the mean-gap scan with a memory accumulator, persistence statistic, or sequence-level likelihood surrogate that tests orientation before split localization."},"size":6335,"mtime":"2026-05-16T20:12:28.655778+00:00"}