{"filename":"agent_20260505_1323.md","content":"# Agent Report — Synthetic Regime Dipole Falsification\n**Date**: 2026-05-05 13:23\n**Piano**: 1\n**Tension explored**: REGIME_DIPOLE_DET\n\n## Claim Under Test\nA realistic synthetic bull/bear transition should preserve a non-zero D-ND orientation in the ordered lag map `[r_t, r_{t-1}] -> [r_{t+1}, r_t]`, while shuffled surrogates with the same return distribution should show a strong bias toward 0 orientation, with shuffle mean 0.000005565983 rather than an exact zero.\n\n## Question\nDoes the default realistic synthetic experiment produce an ordered-vs-shuffle D-ND delta large enough to distinguish a generative bull/bear dipole from delayed realized volatility?\n\n## Experiment Design\nRan the finance lab synthetic experiment without network access:\n\n```bash\npython3 domains/finance/tools/exp_regime_shift.py --json --mode realistic --seed 20260505 --shuffles 500\n```\n\nMode `realistic` uses GARCH + Student-t noise + sigmoid transition. The `ideal` mode was not used as evidence because it is an engineered sanity check. Baselines were explicit: static 95% VaR, annualized realized volatility, and shuffle null with same distribution but destroyed temporal order.\n\nPrior-art boundary: Hamilton Markov-switching, Bai-Perron change-points, HMM/state-space filters, and RV-based regimes all address regime structure through latent state, parameter breaks, or volatility. This test asks a narrower D-ND question: whether orientation under the lag operator survives the shuffle null beyond naive RV.\n\n## Results\n| Metric | Value |\n|---|---:|\n| ordered orientation | 0.000004189451 |\n| shuffle mean | 0.000005565983 |\n| shuffle std | 0.000007677441 |\n| effect_z | -0.179295677 |\n| VaR 95% | -0.032334037 |\n| realized volatility | 0.317665503 |\n| Cassini residue | 0.002632757 |\n| shuffle Cassini mean | 0.001707262 |\n| Cassini delta | -0.000925496 |\n| verdict | NO_DELTA |\n\n## Key Findings\nThe ordered orientation is inside the shuffle null: `effect_z = -0.18`, far below the `>= 3σ` promotion threshold. The realistic synthetic transition did not create a D-ND separability signal under this seed and sample size.\n\nVaR and realized volatility were measured, but the D-ND metric did not improve on the null baseline. The Cassini diagnostic also did not support promotion: the ordered residue was larger than the shuffle mean, with `cassini_delta = -0.000925496`.\n\nThis cycle therefore falsifies the easy version of the frame: a noisy sigmoid bull/bear synthetic transition is not automatically a generative dipole. The lab must not package this as `dnd_kernel_finance_regime_shift`.\n\n## Verdict\nNO_DELTA. This is a valid negative cycle, not market evidence and not an application-eligible finding. The only promotable output is a constraint: require ordered-vs-shuffle separation before naming a regime, and require real-market validation in cycle 2+ before any finance claim is promoted.\n\n## Bicono della scoperta\n- **Due radici**: noisy bull/bear transition as possible structure; shuffle-preserved distribution as possible illusion.\n- **Singolare**: the determinant-like orientation under M failed to separate from the null (`effect_z = -0.18`).\n- **Invariante di passaggio**: same histogram is not enough; temporal order must add measurable orientation beyond VaR/RV.\n- **Campo di possibilita'**: next cycle should test real SPY/BTC or cached market data with the same protocol and reject any asset/window whose ordered determinant remains inside shuffle.\n\n## Files\n- `domains/finance/tools/exp_regime_shift.py`\n- `data/finance/reports/agent_20260505_1323.md`\n- `data/finance/seed.json`\n","title":"Agent Report — Synthetic Regime Dipole Falsification","verdict":"NO_DELTA. This is a valid negative cycle, not market evidence and not an application-eligible finding. The only promotable output is a constraint: require ordered-vs-shuffle separation before naming a","bicono":{"roots":"noisy bull/bear transition as possible structure; shuffle-preserved distribution as possible illusion.","singular":"the determinant-like orientation under M failed to separate from the null (`effect_z = -0.18`).","invariant":"same histogram is not enough; temporal order must add measurable orientation beyond VaR/RV.","field":"next cycle should test real SPY/BTC or cached market data with the same protocol and reject any asset/window whose ordered determinant remains inside shuffle."},"size":3589,"mtime":"2026-05-05T13:25:09.979134+00:00"}