foxalgoBias
We do not sell a directional prediction. Below is what the measured data actually says about "up" vs. "down" — and, separately, three specific directional edges we went looking for and did not find. As of 01:50 ET.
Measured odds. No trade calls. what this means →
SESSION BIAS (today)
Per-market lean for the 6 futures we track: today's active conditions — gap vs. yesterday's close, and whether the overnight Asia session already swept yesterday's high/low — checked against measured findings from verified_findings. A market with no active condition backed by a measured edge shows COIN FLIP, on purpose — that is the honest answer, not a missing feature.
Grade = how far the 95% CI's near edge clears the base rate, in the direction of the observed delta: S ≥ 10pp clearance, A ≥ 5pp, B > 0pp. Below that (CI overlaps the base rate) the card shows COIN FLIP instead — we do not round a non-result up to an edge.
E-mini S&P 500 · COIN FLIP
Asia session still running — asia session still running — not decided yet.
E-mini Nasdaq 100 · COIN FLIP
Asia session still running — asia session still running — not decided yet.
E-mini Dow · COIN FLIP
Asia session still running — asia session still running — not decided yet.
E-mini Russell 2000 · COIN FLIP
Asia session still running — asia session still running — not decided yet.
Gold · LEANS UP GRADE S (strong)
IF the day opens with a gap up THEN price breaks yesterday's high sometime after 9:30am ET.
conditional 68.9% vs. base 41.3% · +31.2 pp · 95% CI [65.9, 71.7] · n = 987 · SKEPTIC-CONFIRMED
WTI Crude · COIN FLIP
Asia session still running — asia session still running — not decided yet.
HTF CONTEXT
Higher-timeframe (daily/weekly) direction: we do not have a measured HTF directional edge to show here. The one HTF-shaped hypothesis we tested — whether agreement across timeframes predicts which way price goes next — came back at 49% across 242,000 board-states, a coin flip; it predicts move size, not direction. See the falsification below and in the graveyard.
HTF direction: measured as a coin flip in our data. Until a genuine higher-timeframe directional finding survives adversarial review, this section stays a coin flip on purpose, same convention as SESSION BIAS above.
Baseline asymmetry
For each market: P(touch 0.5 ATR up) vs. P(touch 0.5 ATR down) from right now, the difference in percentage points, and the sample size behind it. This is a raw geometric view, not the measured-condition leans above.
The asymmetry you see below comes from position in today's range and time remaining in the session — not from a directional forecast. When price sits near today's low, P(up) exceeds P(dn) simply because there is more room to travel upward than downward before the level is reached; that is geometry, the same geometry a coin flip would show if you nailed it to a bounded range. It is not an opinion about where price is headed next.
Futures
| market | P(touch UP 0.5 ATR) | P(touch DOWN 0.5 ATR) | diff (pp) | n | position in range |
|---|---|---|---|---|---|
| E-mini S&P 500 | 48.8% | 47.8% | +1.0 pp | 11,490 | 54% of today's range |
| E-mini Nasdaq 100 | 48.8% | 47.8% | +1.0 pp | 11,490 | 53% of today's range |
| E-mini Dow | 48.8% | 47.8% | +1.0 pp | 11,490 | 88% of today's range |
| E-mini Russell 2000 | 48.8% | 47.8% | +1.0 pp | 11,490 | 80% of today's range |
| Gold | 52.1% | 57.2% | -5.0 pp | 9,580 | 91% of today's range |
| WTI Crude | 50.2% | 52.7% | -2.5 pp | 9,084 | 74% of today's range |
P(touch by close) — horizon convention follows each symbol's detail page.
FX
| market | P(touch UP 0.5 ATR) | P(touch DOWN 0.5 ATR) | diff (pp) | n | position in range |
|---|---|---|---|---|---|
| EUR/USD | 35.0% | 35.1% | -0.1 pp | 8,522 | 13% of today's range |
| GBP/USD | 35.0% | 35.1% | -0.1 pp | 8,522 | 51% of today's range |
| USD/JPY | 35.0% | 35.1% | -0.1 pp | 8,522 | 75% of today's range |
| AUD/USD | 35.0% | 35.1% | -0.1 pp | 8,522 | 88% of today's range |
| USD/CAD | 35.0% | 35.1% | -0.1 pp | 8,522 | 64% of today's range |
| USD/CHF | 35.0% | 35.1% | -0.1 pp | 8,522 | 67% of today's range |
| NZD/USD | 35.0% | 35.1% | -0.1 pp | 8,522 | 71% of today's range |
| Gold spot (XAU/USD) | 35.0% | 35.1% | -0.1 pp | 8,522 | not cached yet |
P(touch within next 915m) — horizon convention follows each symbol's detail page.
Stocks
| market | P(touch UP 0.5 ATR) | P(touch DOWN 0.5 ATR) | diff (pp) | n | position in range |
|---|---|---|---|---|---|
| S&P 500 ETF (SPY) | 40.8% | 42.0% | -1.2 pp | 1,861 | 42% of today's range |
| Nasdaq 100 ETF (QQQ) | 40.8% | 42.0% | -1.2 pp | 1,861 | 65% of today's range |
| Russell 2000 ETF (IWM) | 40.8% | 42.0% | -1.2 pp | 1,861 | 83% of today's range |
| Apple | 41.8% | 39.7% | +2.1 pp | 4,354 | 44% of today's range |
| Microsoft | 41.8% | 39.7% | +2.1 pp | 4,354 | 60% of today's range |
| Nvidia | 41.8% | 39.7% | +2.1 pp | 4,354 | 5% of today's range |
| Amazon | 41.8% | 39.7% | +2.1 pp | 4,354 | 39% of today's range |
| Alphabet (GOOGL) | 41.8% | 39.7% | +2.1 pp | 4,354 | 76% of today's range |
| Meta Platforms | 41.8% | 39.7% | +2.1 pp | 4,354 | 63% of today's range |
| Tesla | 39.9% | 41.1% | -1.2 pp | 1,152 | 81% of today's range |
| AMD | 39.9% | 41.1% | -1.2 pp | 1,152 | 73% of today's range |
| Netflix | 39.9% | 41.1% | -1.2 pp | 1,152 | 32% of today's range |
P(touch by close 16:00 ET) — horizon convention follows each symbol's detail page.
Historical period 2020-2026, these 12 names selected ex-post (survivorship) — calibration applies to these names, not a random stock. Live frozen scorecard accumulates from launch.
Directional edge we looked for and did not find
Three separate research passes tried to turn something in the data into a genuine directional signal. Here is what happened to each, with the number attached.
Timeframe agreement predicts size, not direction
Tested whether multiple timeframes pointing the same way predicts which way price goes next. Across 242,000 board-states, direction accuracy was 49% — a coin flip. Agreement tracks how big the next move is likely to be, not which way it goes.
Vendor selection predicts resolution, not direction
Tested whether the subset of ranges a market-data vendor chooses to publish carries directional information. Selected and unselected ranges have an identical gambler's-ruin baseline (0.813 vs. 0.807) — yet selected ranges resolve 15 percentage points better. The vendor is picking ranges that resolve cleanly, not ranges that go a particular direction.
Expansion as a directional signal — the effect was a bug
Raw backtest showed range expansion predicting direction with a +18.4 pp edge. After removing an intrabar lookahead bug, the edge fell to +0.2 pp — the entire effect was the bug, not the signal.
If we find a directional edge that survives our own adversarial review, it will appear here with its sample size. Until then, this section stays empty on purpose. See the graveyard for findings we tested and killed, and the methodology for how review works.