{"filename":"agent_20260523_1353.md","content":"# Agent Report — Forward-Rolled Window Transfer: SPY Locality Collapse\n**Date**: 2026-05-23 14:02 UTC\n**Piano**: 32\n**Tension explored**: `T22_SPY_RECURRENCE_LIMIT` — is the SPY current-window orientation a fixed-date artifact or does it survive forward window shift?\n\n## Claim Under Test\nThe previous exact-window transfer diagnostic (`2026-02-09..2026-05-08`) returned `single_or_partial_window` with SPY as the only iid/block5 survivor (z_iid=3.428, z_block5=3.321) and block21 collapse (z=1.613). The recurrence diagnostic showed three adjacent prior windows all rejected. The remaining undiscriminated hypothesis: this SPY partial signal is either a structural window-insensitive regime, or a fixed-date coincidence artifact. The forward-rolled window shift discriminates these.\n\n## Question\nIf the SPY partial orientation signal is structural rather than coincidental, it should survive a forward-rolled window shift of 14 days. Does SPY iid effect_z remain >= 3.0 on the window `2026-02-23..2026-05-22`?\n\n## Experiment Design\n**Tool**: `finance_transfer_diagnostic.py` — exact-date transfer diagnostic with iid/block5/block21 null families.\n\n**Command**:\n```bash\npython3 /opt/D-ND_LAB/domains/finance/tools/finance_transfer_diagnostic.py \\\n    --start 2026-02-23 --end 2026-05-22 --shuffles 4096 --json\n```\n\n**Window**: forward-rolled by 14 days from the previous reference (2026-02-09→2026-02-23, 2026-05-08→2026-05-22). Same 63-day span (~3 calendar months), same n_obs=62, same null families, same shuffles=4096.\n\n**Assets**: SPY, QQQ, IWM, EFA, TLT, GLD, BTC-USD — identical to the previous transfer diagnostic.\n\n**Null baseline**: iid shuffle + circular block permutation (5 and 21 days), same distribution, destroyed order.\n\n**Naive baseline**: VaR 95% static + realized volatility annualized.\n\n**Discrimination**: if SPY is a date-coincidence artifact, the forward-rolled window collapses z below 3.0. If structural, z stays elevated.\n\n## Results\n\n| Asset | n | iid z | block5 z | block21 z | VaR 95% | Realized Vol | iid Verdict |\n|-------|---|-------|----------|-----------|---------|--------------|-------------|\n| **SPY** | 62 | **-0.644** | -0.664 | -1.027 | -0.0144 | 0.1460 | NO_DELTA |\n| QQQ | 62 | 0.373 | 0.151 | 1.204 | -0.0172 | 0.1929 | NO_DELTA |\n| IWM | 62 | -0.555 | -0.575 | -0.618 | -0.0216 | 0.2143 | NO_DELTA |\n| EFA | 62 | -0.412 | -0.315 | 0.656 | -0.0221 | 0.2259 | NO_DELTA |\n| TLT | 62 | -0.654 | -0.645 | -0.657 | -0.0107 | 0.1025 | NO_DELTA |\n| GLD | 62 | -0.420 | -0.556 | -1.068 | -0.0321 | 0.2795 | NO_DELTA |\n| BTC-USD | 87 | -0.521 | -0.492 | -0.640 | -0.0302 | 0.3389 | NO_DELTA |\n\n**Classification**: `no_transfer_delta` — all 7 assets reject under all 3 null families. Zero iid_dnd_symbols. Zero robust_all_null_symbols.\n\n### Reference comparison (previous window: 2026-02-09..2026-05-08)\n\n| Asset | Prev iid z | Current iid z | Δ z | Verdict Change |\n|-------|-----------|---------------|-----|----------------|\n| SPY | **+3.428** | **-0.644** | -4.072 | DND_DELTA → NO_DELTA |\n| QQQ | +2.979 | +0.373 | -2.606 | NO_DELTA → NO_DELTA |\n| IWM | +1.330 | -0.555 | -1.885 | NO_DELTA → NO_DELTA |\n| EFA | +1.269 | -0.412 | -1.681 | NO_DELTA → NO_DELTA |\n| TLT | -0.656 | -0.654 | +0.002 | NO_DELTA → NO_DELTA |\n| GLD | -0.227 | -0.420 | -0.193 | NO_DELTA → NO_DELTA |\n| BTC-USD | -0.158 | -0.521 | -0.363 | NO_DELTA → NO_DELTA |\n\nSPY went from the strongest positive z (+3.428, iid DND_DELTA) to a negative z (-0.644) — the ordered orientation score is now *below* the shuffle mean. The forward-rolled window shift of 14 days destroys the entire signal.\n\n## Key Findings\n\n1. **SPY partial locality collapses on forward window shift**. The SPY window `2026-02-09..2026-05-08` showed iid z = 3.428; the forward-rolled window `2026-02-23..2026-05-22` shows iid z = -0.644. This is a 4.1σ drop that takes the score below the shuffle null mean. The SPY signal is a **fixed-date window artifact**, not a structural regime orientation.\n\n2. **No asset survives the forward-rolled window**. Across all 7 assets and 3 null families (21 tests), every single effect_z is below 3.0. Four of seven assets have negative iid z (oriented score < shuffle mean). The strongest positive is QQQ block21 z = 1.204, well below the 3.0 gate.\n\n3. **The previous `current_iid_partial` classification is now fully resolved**. The recurrence diagnostic showed no recurrence in prior windows. The forward-rolled window shows no survival under window shift. The SPY partial signal was a coincidence between the fixed midpoint split and the specific start date — not a regime, not a pattern, and not a candidate for transfer or recurrence.\n\n4. **The raw orientation_score detector has no exploitable real-market window**. Across two exact-date transfer diagnostics (2026-02-09 and 2026-02-23), covering 14 asset-window combinations, the only positive was SPY on a single fixed start date, and it collapses the moment the window is shifted. This is consistent with the synthetic calibration history (T1-T17): the raw split-score detector has signal on engineered positives but captures date-coincidence artifacts on real data.\n\n5. **Data provenance preserved**. All 7 data cards include provider=yfinance, retrieval_ts, source_url, n_obs, era_hint. The forward-rolled diagnostic artifact is at `data/finance/diagnostics/finance_transfer_diagnostic_20260523_140209.json`.\n\n## Verdict\n**NO_TRANSFER_DELTA, SPY_LOCATION_ARTIFACT_CONFIRMED.** The forward-rolled window shift directly falsifies the hypothesis that SPY partial orientation was structural. The SPY `current_iid_partial` signal was a coincidence between a specific start date (2026-02-09) and the fixed-midpoint split — not a D-ND regime finding. All 7 assets uniformly reject under all 3 null families on the rolled window. No real-market regime claim is promoted. The raw orientation_score detector has no exploitable real-market window in its current form.\n\n**Operational consequences**: The `current_iid_partial` SPY classification from cycles 20260517_1338/1348 is now superseded by this direct falsification. Do not reference the SPY fixed-date window as evidence of any regime signal. The detector locality limit (T22) is now fully characterized: SPY signal exists on exactly one start date and collapses immediately under adjacent window shift.\n\n## Bicono della scoperta\n- **Due radici**:\n  - Root 1: the SPY signal on `2026-02-09..2026-05-08` was a genuine ordered-vs-shuffle delta within its fixed window, passing iid and block5 nulls with z > 3.0. The detector was not broken — it correctly measured that the specific split orientation was unlikely under shuffle.\n  - Root 2: the forward-rolled window `2026-02-23..2026-05-22` shows negative z (-0.644), proving the signal was a start-date coincidence, not a window-insensitive market regime. The detector measured a real effect — but the effect was on the date, not the market structure.\n\n- **Singolare**: The singular point is the specific start date `2026-02-09`. The orientation score crosses from strongly ordered (z = +3.428) to below-shuffle (z = -0.644) when the window shifts by only 14 days. This is the exact location where the orientation-score artifact lives: a single date edge that the fixed-midpoint split converts into apparent structure.\n\n- **Invariante di passaggio**: No real-market D-ND regime claim based on the raw orientation_score can be promoted unless it survives at least two non-overlapping exact-date windows of the same span (e.g., forward-rolled and backward-rolled) under iid, block5, and block21 nulls. A single-window DND_DELTA, regardless of z magnitude, is a date-coincidence artifact until proven otherwise by window shift.\n\n- **Campo di possibilita'**: The raw orientation_score detector is now completely characterized on real-market data: it produces zero robust DND_DELTA across 14 asset-window combinations (two 7-asset exact-date diagnostics) and the single partial positive collapses under the minimal window shift test. Future finance cycles must either (a) replace the raw split-score with a window-insensitive statistic, (b) use a different object entirely (not mean_gap × vol_gap orientation), or (c) predeclare a new mechanism with explicit window-robustness requirements before testing. The synthetic lag_memory_const_vol path with matched-filter precondition remains the most promising alternative, but it needs a detector that does not inherit the same split-location sensitivity.\n\n## Files\n- Report: `data/finance/reports/agent_20260523_1353.md`\n- Diagnostic JSON: `data/finance/diagnostics/finance_transfer_diagnostic_20260523_140209.json`\n- Diagnostic MD: `data/finance/diagnostics/finance_transfer_diagnostic_20260523_140209.md`\n- Prior diagnostic (reference): `data/finance/diagnostics/finance_transfer_diagnostic_20260517_133933.json`\n- Seed updated: `data/finance/seed.json`\n","title":"Agent Report — Forward-Rolled Window Transfer: SPY Locality Collapse","verdict":"NO_TRANSFER_DELTA, SPY_LOCATION_ARTIFACT_CONFIRMED.** The forward-rolled window shift directly falsifies the hypothesis that SPY partial orientation was structural. The SPY `current_iid_partial` signa","bicono":{"roots":"- Root 1: the SPY signal on `2026-02-09..2026-05-08` was a genuine ordered-vs-shuffle delta within its fixed window, passing iid and block5 nulls with z > 3.0. The detector was not broken — it correctly measured that the specific split orientation was unlikely under shuffle.\n  - Root 2: the forward-rolled window `2026-02-23..2026-05-22` shows negative z (-0.644), proving the signal was a start-dat","singular":"The singular point is the specific start date `2026-02-09`. The orientation score crosses from strongly ordered (z = +3.428) to below-shuffle (z = -0.644) when the window shifts by only 14 days. This is the exact location where the orientation-score artifact lives: a single date edge that the fixed-midpoint split converts into apparent structure.","invariant":"No real-market D-ND regime claim based on the raw orientation_score can be promoted unless it survives at least two non-overlapping exact-date windows of the same span (e.g., forward-rolled and backward-rolled) under iid, block5, and block21 nulls. A single-window DND_DELTA, regardless of z magnitude, is a date-coincidence artifact until proven otherwise by window shift.","field":"The raw orientation_score detector is now completely characterized on real-market data: it produces zero robust DND_DELTA across 14 asset-window combinations (two 7-asset exact-date diagnostics) and the single partial positive collapses under the minimal window shift test. Future finance cycles must either (a) replace the raw split-score with a window-insensitive statistic, (b) use a different obj"},"size":8909,"mtime":"2026-05-23T14:04:02.464659+00:00"}