2026-04-23 - Flow BEAR / Price BULL divergence at open
The observation
From 8:30 to 8:45 CT on 2026-04-23, SPY printed a clean 5/5 green trend-day open, rallying from 710.7. The Cortana scoring engine read the move as BEARISH with composite score 47 (NEUTRAL). No trades were taken.
Engine snapshot at 8:46 CT (SPY 710.69, +$0.35)
| Engine | Signal | Score | Drivers |
|---|---|---|---|
| Premium Flow | BEAR | -7 | Tide ▼20.4%, Net Prem ▼70%, 0DTE ▼0% |
| Smart Money | BEAR | -8 | NPP/NCP 11.2:1 (strong bearish), 10.7M puts |
| Volume Sentiment | BEAR | -9 | P/C 1.27, recent P/C 1.32, daily P/C 1.27 |
| Dealer Position | NEUT | -3 | short γ, 0DTE expiry map |
| Volatility | NEUT | +2 | IV 16.2%, term backwardation 0.98 |
| Price Trend | BULL | +2 | at VWAP 0.02 (5/5 green) |
Composite: 47 NEUTRAL. Signal: BEARISH, trigger=strike_stack, conv=0.60.
The question
Why did the engine go aggressively BEAR while price printed a 5/5 green trend-day open?
Working hypotheses to test:
-
Hedging flow, not directional. Institutions buying puts against existing long cash exposure. The puts register as “bearish” to flow engines (Premium Flow -7, Smart Money -8, Volume Sentiment -9) but are defensive hedges, not bets against SPY. On a trend-day open, hedging demand should be elevated precisely because the cash move is strong.
-
Price Trend under-weighting a trend-day signal. 5/5 green from open with “at VWAP” scored only +2. A VWAP reclaim earlier in the session (8:30:40) scored +9 with 2/5 green. The “at VWAP” path appears heavily muted vs. the “VWAP reclaim” path. A 5-consecutive- green pattern off the open should arguably be a first-class trend-day tell, not a +2 whisper.
-
Three correlated bear engines, one decorrelated bull. Premium Flow, Smart Money, and Volume Sentiment all derive from options flow data. When options flow tilts bearish (hedging), all three move together - triple-counting one signal. Price Trend is the lone non-options engine and gets outvoted 3-to-1 by construction.
-
Composite aggregation is additive, not multiplicative. A strong BULL Price Trend +10 plus BULL Premium Flow +5 plus BULL Volatility +4 = +19 total, but Volume Sentiment -9 alone cuts nearly half of that. One highly-confident BEAR engine can veto three mildly- confident BULL engines even when price action says otherwise.
Why this matters
The North Star mandate is “early and right.” This is the inverse - late and wrong. The engine watched a $1.30 rally (5 candles, 15 minutes) and produced a BEAR signal the entire time. If the fast-path had fired BULL and been gated by a 47-NEUTRAL composite, at least the failure mode would be “too cautious.” This failure mode is actively opposite - the engine was advising against the move that was happening.
Related
concepts/cortana-is-the-asset.md- individual trades are fuel, the engine is the asset. The engine got this one wrong; the data is the fuel for fixing it.- Prior April 16 empirical finding: V1 composite fires AT tops/
bottoms. See
project_score_instability_april16andproject_v1_latency_april16in the legacy memory index.
What to do with this
Not a code change today. Data point to replay when we next tune the composite aggregation or the hedging filter. Specifically:
- Replay 8:30-8:45 CT with engine logs + tape + chart overlay
- Quantify: how often does flow-BEAR + price-BULL correlate with a continuation vs. a reversal? Need N >= 20 trend-day opens.
- Test a “flow-vs-price divergence” circuit breaker: when price prints 3+ consecutive directional bars opposite to flow composite, downweight the flow signal by 0.5x.
Status
Open question. Do not close until replay + quantification are complete.