Research | Forecast Alpha

The engine computes every number; the AI layer only explains them. Select a market and generate a structured brief grounded in the engine's persisted inputs snapshot. Without an LLM key, a deterministic template renderer produces the brief from the same snapshot. Briefs are persisted and timestamped.

BTC closes above $100k this week?

Template renderer (no LLM key)

Generated Fri, 26 Jun 2026 12:01:19 GMTData quality

65Moderate

simModel confidence 0.74sim

Summary

Market prices BTC closes above $100k this week at 61.0%; the model estimates 56.6% (confidence 0.74). Best-side EV after modeled costs is 1.2% on NO. Engine signal: NO TRADE.

What the market implies

The market-implied probability is 61.0%. At a liquidity score of 72/100 the modeled cost of taking a position is 3.2% (2.0% fees + 1.2% slippage), which any edge must clear before it is tradeable.

What the model estimates

Starting from a 52.0% base rate (0.080 log-odds), the factor stack moves the raw model estimate to 55.1%: crypto.btc_eth_momentum_7d +0.67 (+0.135 log-odds), pm.momentum_7d -0.03 (-0.011 log-odds), pm.cross_market_divergence +0.00 (+0.000 log-odds), macro.rate_surprise +0.00 (+0.000 log-odds), macro.yield_curve_shift +0.00 (+0.000 log-odds), news.source_weighted_signal +0.00 (+0.000 log-odds), crowd.calibrated_aggregate +0.00 (+0.000 log-odds). Confidence of 0.74 then shrinks the final estimate toward the market price, landing at 56.6%.

Why they differ

The model sits 4.4% below the market. The largest single driver is the crypto.btc_eth_momentum_7d factor at +0.67 (+0.135 log-odds): 7-day BTC/ETH return z-score, tanh-squashed. Note that the sentiment and source factors are simulated inputs in this MVP — the disagreement structure is real, the inputs are not.

Base-rate analysis

The engine's prior for this market is 52.0%, intended to reflect how often events of this class resolve YES historically. The prior contributes 0.080 log-odds before any market-specific evidence; with factor adjustments totalling 0.124 log-odds, the raw model lands at 55.1%.

Comparable events

Event	Outcome	Relevance
BTC all-time-high retests after >20% drawdowns (2017–2025)	Reclaimed within 6 months in most bull regimes	Price-threshold markets depend heavily on prevailing regime.
Round-number threshold markets on Polymarket 2024–25	Markets systematically overpriced near-miss thresholds	Anchoring bias inflates YES prices near salient levels.

Recent information

Trend and momentum factors are computed from the stored 30-day price path (trend undefined, momentum undefined). Sentiment and source-quality inputs are simulated in this MVP and are labeled as such wherever displayed.

Bull case

For NO: the model's 43.4% estimate against an entry near 39.0% leaves 1.2% of EV after costs. If the factor evidence is genuine information the market has not priced, expected value accrues as the market converges toward the model.

Bear case

Against the position: the market aggregates more information than any four-factor model. If the crypto.btc_eth_momentum_7d signal is stale or spurious, the true probability is closer to the market's 61.0% and the position's EV is roughly the negative of its costs (3.2%).

Key uncertainty

Whether the crypto.btc_eth_momentum_7d factor reflects real, unpriced information — and whether resolution follows the thesis or turns on a definitional edge in the criteria.

What could make this wrong?

The model's edge depends on its inputs being right. Concretely: the base rate of 52.0% may not apply if this event differs structurally from its reference class; the crypto.btc_eth_momentum_7d factor could be noise rather than information at this horizon; and with confidence at 0.74, the model itself concedes meaningful estimation error. The risk engine also flags: Inside the 48h resolution-risk window — late-breaking information dominates any model edge.

Turn model inputs into evidence-backed market briefs.

Brief context loaded

Research Engine