{"filename":"agent_20260517_1325.md","content":"# Agent Report - Equity-Cluster Transfer Diagnostic\n**Date**: 2026-05-17 13:25 UTC\n**Piano**: 27\n**Tension explored**: `REAL_MARKET_TRANSFER_DIAGNOSTIC / T4_NULL_FAMILY_ROBUSTNESS`\n\n## Claim Under Test\nThe latest real-market diagnostic found a `local_robust` SPY window: current 3-month SPY passed the ordered-vs-shuffle D-ND gate, while three adjacent SPY windows rejected. The unresolved claim is whether this is transferable evidence across assets or only a local equity-window observation.\n\nThis is not a trading signal, forecast, alpha, buy/sell indication, or public finance claim.\n\n## Question\nDoes the current SPY `DND_DELTA` transfer to another equity ETF and to a non-equity crypto asset under the same ordered-vs-shuffle detector, or does it remain local/non-operational?\n\n## Experiment Design\nTool family used: `exp_regime_shift` with real-market `market_data` payloads, `seed=42`, `shuffles=2048`.\n\nExecuted windows:\n\n- SPY current window: `2026-02-09..2026-05-09`\n- QQQ cached adjacent current window: `2026-02-14..2026-05-15`\n- Bitcoin CoinGecko 365d window: latest cached `2025-05-18..2026-05-17`\n\nAttempted but not usable:\n\n- QQQ exact `2026-02-09..2026-05-09` failed because DNS resolution for Yahoo crumb host `guce.yahoo.com` was unavailable in this sandboxed run. The experiment therefore used the available audited cache window `2026-02-14..2026-05-15`, and this date mismatch is part of the limitation.\n\nNull baseline: shuffle returns with the same marginal distribution and destroyed temporal order.\n\nNaive baseline: static VaR 95% and realized volatility reported for each window.\n\nPrior-art boundary: Hamilton/HMM would model latent regimes, Bai-Perron would test structural breaks, and RV-based regimes would rank volatility. This experiment only asks whether D-ND orientation survives the sample-conditional shuffle null better than naive VaR/RV framing. It does not establish recurrence or operational utility.\n\n## Results\n| Asset | Window | n returns | Verdict | effect_z | ordered | shuffle mean | shuffle std | realized vol | VaR 95 | Cassini residue |\n|---|---|---:|---|---:|---:|---:|---:|---:|---:|---:|\n| SPY | 2026-02-09..2026-05-08 | 62 | `DND_DELTA` | 3.385 | 0.00003940 | 0.00000673 | 0.00000965 | 0.1474 | -0.01526 | 0.02910 |\n| QQQ | 2026-02-17..2026-05-14 | 61 | `DND_DELTA` | 3.502 | 0.00006431 | 0.00001082 | 0.00001528 | 0.1910 | -0.01731 | 0.03306 |\n| BTC | 2025-05-18..2026-05-17 | 365 | `NO_DELTA` | -0.270 | 0.00000725 | 0.00001124 | 0.00001477 | 0.3538 | -0.03644 | 0.00326 |\n\nProvenance:\n\n- SPY: `yfinance://Ticker(SPY).history(start=2026-02-09, end=2026-05-09, interval=1d)`; retrieved `2026-05-17T13:22:01.127319+00:00`; era `2026Q1-Q2`; n_obs 63.\n- QQQ: `yfinance://Ticker(QQQ).history(start=2026-02-14, end=2026-05-15, interval=1d)`; retrieved `2026-05-17T13:20:50.448358+00:00`; era `2026Q1-Q2`; n_obs 62.\n- BTC: `https://api.coingecko.com/api/v3/coins/bitcoin/market_chart?vs_currency=usd&days=365`; retrieved `2026-05-17T13:19:45.750183+00:00`; era `2025-2026`; n_obs 366. CoinGecko free tier provides close prices; OHL are close proxies in the lab schema.\n\nAssertions:\n\n| ID | Status | Detail |\n|---|---|---|\n| `XFER_01` | PASS | SPY current window reproduces the prior local result: effect_z 3.385, `DND_DELTA`. |\n| `XFER_02` | PASS_WITH_LIMIT | QQQ also passes in an adjacent cached current window: effect_z 3.502, `DND_DELTA`; exact same-date QQQ was not fetched because DNS was unavailable. |\n| `XFER_03` | FAIL_FOR_CROSS_DOMAIN | BTC 365d rejects: effect_z -0.270, `NO_DELTA`, despite much higher realized vol 0.3538. |\n| `XFER_04` | FAIL_FOR_OPERATIONAL | The pass cluster is equity-correlated and local. Prior SPY adjacent windows remain 3/3 `NO_DELTA`, so recurrence is absent. |\n\n## Key Findings\n1. The prior SPY `local_robust` observation is not strictly SPY-only: QQQ also passes on a close current window, with effect_z 3.502.\n\n2. The transfer does not cross asset class. BTC rejects under the same detector even with higher realized volatility, which blocks a generic \"risk-on/risk-off market regime\" interpretation.\n\n3. The result is better classified as `equity_cluster_local`: current large-cap US equity ETFs show ordered-vs-shuffle separation in this narrow era, but recurrence and cross-domain transfer are absent.\n\n4. Because QQQ is highly correlated with SPY and its exact same-date fetch failed, this is not two independent assets in the promotion sense. It is a transfer diagnostic, not an operational finding.\n\n## Verdict\n**EQUITY_CLUSTER_LOCAL, NON_OPERATIONAL.** The SPY-only local interpretation was falsified at this cycle: QQQ also passed in the adjacent cached current window with effect_z 3.502. The new claim that emerged is a correlated-equity local cluster, but it does not become an operational finance claim. Promotion remains blocked until the detector shows recurrence across independent windows and materially independent assets, with data_card provenance preserved and no trading-signal language.\n\n## Bicono della scoperta\n- **Due radici**:\n  - Root 1: SPY and QQQ preserve ordered-vs-shuffle orientation in the current equity window.\n  - Root 2: BTC rejects under the same null baseline, and previous adjacent SPY windows reject, so the field does not become a market-wide regime claim.\n- **Singolare**: The singular point is the current US equity era: two correlated equity ETFs cross the 3 sigma gate while adjacent SPY eras and BTC do not.\n- **Invariante di passaggio**: A real-market `DND_DELTA` must be labeled by transfer class: single-window, correlated-cluster, cross-asset, recurring, operational. This cycle reaches only correlated-cluster local.\n- **Campo di possibilita'**: Future real transfer tests should either use same-date cached/fetched windows across multiple independent assets or stop at dashboard-level diagnostic copy. No exposure decision or public claim follows from this result.\n\n## Files\n- Report: `data/finance/reports/agent_20260517_1325.md`\n- Prior diagnostic: `data/finance/diagnostics/finance_diagnostic_20260517_132200.json`\n- Seed updated: `data/finance/seed.json`\n","title":"Agent Report - Equity-Cluster Transfer Diagnostic","verdict":"EQUITY_CLUSTER_LOCAL, NON_OPERATIONAL.** The SPY-only local interpretation was falsified at this cycle: QQQ also passed in the adjacent cached current window with effect_z 3.502. The new claim that em","bicono":{"roots":"- Root 1: SPY and QQQ preserve ordered-vs-shuffle orientation in the current equity window.\n  - Root 2: BTC rejects under the same null baseline, and previous adjacent SPY windows reject, so the field does not become a market-wide regime claim.","singular":"The singular point is the current US equity era: two correlated equity ETFs cross the 3 sigma gate while adjacent SPY eras and BTC do not.","invariant":"A real-market `DND_DELTA` must be labeled by transfer class: single-window, correlated-cluster, cross-asset, recurring, operational. This cycle reaches only correlated-cluster local.","field":"Future real transfer tests should either use same-date cached/fetched windows across multiple independent assets or stop at dashboard-level diagnostic copy. No exposure decision or public claim follows from this result."},"size":6101,"mtime":"2026-05-17T13:29:08.796369+00:00"}