Subscribe to the daily brief

You'll get a personal-style daily brief on Monday, Wednesday, and Friday at 11:30 AM PT — featuring the top scenario, recent prediction grades, and the major events shifting probability that day.

Free during alpha. No credit card. Reply to any email to unsubscribe.

Signed in

Email

—

Status

—

How GeoStrat models work

GeoStrat blends structured signals (market data, GDELT tone, OSINT chatter, vessel tracking, prediction-market flow) with category-specific historical anchors to produce per-scenario probabilities.

• Bootstrap probability — anchored on category × severity base rate, then adjusted by current signals.
• Calibration — every closed prediction is graded YES/NO and rolled into Brier scoring per category, which feeds back into the next forecast.
• Anchor windows — each major scenario maps to a curated set of historical analogs; the model uses their resolved market windows as a prior.
• Shadow models — newer architectures (e.g. gradient-boosted) run in parallel but only get promoted once they clear the holdout-Brier gate.

Full methodology page coming soon. In the meantime, the Track Record tab shows every graded call.

Methodology

How a probability is built. Every scenario starts from a category × severity base rate (the historical frequency of similar events). Live signals — market data, GDELT tone, OSINT chatter, Hormuz vessel traffic, prediction-market flow, statement impact — push the probability up or down according to per-event weights. The number you see on each scenario card is the result.

Anchors. For each major scenario the model loads a curated set of historical analogs (e.g. "Iran shoots down US drone, 2019" for Hormuz escalation). Their resolved market windows act as a prior — if every analog showed Brent +6% in 5 days, the modeled-impact panel reflects that.

Calibration + grading. Every closed prediction is graded YES or NO against a documented resolution rule. Brier scores roll up by category and feed back into the next forecast — a category that's been overshooting gets dampened, one that's been undershooting gets boosted. Track Record exposes the full grading log.

Shadow models. Newer architectures (currently a gradient-boosted shadow) run in parallel with the bootstrap model but only get promoted once they clear a holdout-Brier gate against the same set of resolved events. This prevents in-sample-fit shadows from displacing a calibrated production model.

What GeoStrat is

GeoStrat is a daily geopolitical forecasting platform organized around major events and their downstream impact. Each scenario shows two things you actually care about: how likely it is, and what happens to markets / supply chains / geopolitics if it hits.

We grade every prediction publicly — short-term, long-term, and major. Our track record is the only thing that should make you trust the forecasts.

Built by Trevor Crowe. Currently in alpha — free for everyone.

Get in touch

Reply to any of the daily-brief emails — they come from a real inbox and Trevor reads every reply.

Or email crowetrevor@gmail.com directly.

Bug reports, model corrections, scenarios you wish we tracked — all welcome.