Cross-Asset Lead-Lag Intelligence — March 2, 2026 10:15 ET — 0DTE Bear Spread Session
| Symbol | Last | Change | Signal |
|---|---|---|---|
| ES | 6,849.50 | -0.57% | Failed at 6850 resistance |
| NQ | 24,855.25 | -0.60% | Tech weakness |
| SPY | 682.55 | -0.50% | |
| QQQ | 604.05 | -0.53% | |
| PLTR | 143.36 | +4.50% | Leading everything — canary |
| Symbol | Last | Change | Signal |
|---|---|---|---|
| GC | 5,343.50 | +2.16% | Gold ripping — fear |
| ZN | 113.34 | -0.45% | Bonds weak |
| ZB | 118.06 | -0.68% | Long bonds weaker |
| BTC | 67,026.50 | +2.64% | Crypto diverging up |
| ETH | 1,977.35 | +3.26% | Crypto diverging up |
| Pair | Lag | r | Stability | Bits | SNR (dB) | Class | Physics |
|---|---|---|---|---|---|---|---|
| ZN → ZB | sync | +0.675 | 75% | 0.382 | -0.8 | REAL | Same curve, different tenors — calibration check |
| NQ → ZB | 1 bar (30s) | +0.360 | 100% | 0.073 | -7.6 | REAL | Equity attention leads bond repricing. Genuine info asymmetry. |
| ZN → ES | 18 bar (9 min) | +0.321 | 75% | 0.057 | -9.4 | ACTIONABLE | Bond "smart money" prices macro before equity algos re-price components. 9 min early warning. |
| PLTR → ZN | 8 bar (4 min) | +0.325 | 100% | 0.059 | -8.9 | SUSPECT | Not causal. PLTR is high-beta proxy reacting faster to risk-on/off. |
| PLTR → NQ | sync | +0.513 | 50% | 0.166 | -4.5 | SUSPECT | Regime-dependent. Both respond to AI/tech sentiment. |
| ES → GC | 9 bar (4.5 min) | +0.307 | 67% | 0.052 | -9.8 | NOISE | Positive ES/gold correlation is regime anomaly. 67% stability = unreliable. |
Most correlations (r ≈ 0.32) are within one standard error of zero given the window sizes (13 points per stability window → SE = 1/√10 = 0.316). You cannot statistically distinguish them from noise.
Real signals: ZN→ZB (trivially correlated, calibration check) and NQ→ZB at 100% stability (equity attention genuinely leads bond repricing). Suspicious: PLTR "leading" bonds is not causal — it's a high-beta proxy reacting faster, not causing. Need 4x more bars before other signals become trustworthy.
ZN→ES carries 0.057 bits per bar — roughly 1-2% of ES return entropy. SNR = -9.4 dB. Tradeable signals all operate below 0 dB.
Reliable communication at sub-zero dB is possible (I proved it in 1948) but requires coding gain: aggregate many bets with small edge. Weight pairs by I(X;Y) × stability² not |r| × stability.
Bug found: correlation uses global mean/std but computes on lag-shifted subwindows — biases toward zero for large lags. Need windowed normalization.
Watch ZN like your life depends on it. If ZN breaks 113.25, you have 9 minutes before ES follows. If ZN rallies past 113.50, close everything.
ES at 6850 is the trap line. If ES reclaims 6855 and holds 15 min, cover the call spread immediately. Third rejection at round-number resistance resolves down ~62% of the time.
Deadline: If the break hasn't materialized by 2 PM ET, the trade is dead. Close. The house wins overtime.
The bulls failed at the 6850 redoubt. Gold surging = rear echelons already withdrawing. PLTR leading alone = the Old Guard carrying the entire line — when it exhausts, nothing is behind it.
Hold. Do not adjust a winning position from impatience. The decisive hour is 14:00-15:00 ET. If bulls can't take the heights by morning, their morale breaks. Time and theta fire on the enemy.
Retreat only if ES breaks 6855 with conviction and volume before noon.
Not architecture astronautics. ~500 lines across 5 files, clean separation (market-pylon → signal-engine), proper OHLCV with log returns, honest Pearson implementation. Written by someone who programs, not architects.
But not yet useful. You built a telescope, not a trading system. The gap between "NQ leads ES by 2 bars" and "place this order at this price" is where all the money lives — and that gap is currently empty. Missing: (1) action layer (condor-pilot consuming /leaders), (2) backtesting on days of data not hours, (3) rolling stability windows, (4) persistence (SQLite, not just ring buffers).