Ask Marty — Tetlock-method analysis for prediction markets

THE DIFFERENCE

Coin-flip calls, or a tracked methodology.

WITHOUT ASK MARTY

Refresh Kalshi every 10 minutes
Doomscroll Twitter for context
Ask ChatGPT and hope for the best
No record or analysis of past calls
Coin-flip calls

WITH ASK MARTY

Live odds with the analysis built in
A six-step playbook on every market
Past patterns + the traps your brain falls for
A public report card on every pick
Honest probability ranges — and where the edge is

HOW IT WORKS

Six steps. Applied to every market. Logged for every call.

The framework is the same one Tetlock's superforecasters used to beat intelligence analysts with classified data by 30%. Marty applies it to live markets in seconds.

DEEP DIVE

Six ways your brain lies to you about prediction markets.

CAUGHT THIS YEAR

2,026

Step two of every analysis runs the same six cognitive checks. Each catch is logged with the market and the override reasoning.

BIAS · RECENCY

Recency bias

Weighting the last news cycle as if it predicts the next ten.

Caught312

BIAS · BASE RATE

Base-rate neglect

Ignoring how often things like this actually happen.

Caught487

BIAS · AVAILABILITY

Availability heuristic

What's vivid feels likely. It usually isn't.

Caught256

BIAS · ANCHORING

Price anchoring

Treating the current market price as ground truth.

Caught401

BIAS · MOTIVATED

Motivated reasoning

Wanting an outcome and finding evidence after the fact.

Caught198

BIAS · OVERCONFIDENCE

Overconfidence

Tight ranges that imply more knowledge than we actually have.

Caught372

THE SCORECARD

Public, ongoing, undeniable.

Every market Marty has analyzed, with the estimate, the market price at the time, the resolution, and the Brier score. Misses stay published. That's the point.

RESOLVED

1

CORRECT

1

AVG BRIER

0.023

OPEN

8

RESOLVED PREDICTIONS

Date	Market	Platform	Our Call	Mkt Price	Result	Brier
2026-03-04	Fed March Rate Decision	Kalshi	No cut (85–90%)	~64% Yes cut	✓ NO CUT	0.023

OPEN PREDICTIONS · AS OF MARCH 2026

Date	Market	Platform	Our Call	Mkt Price	Resolves
2026-03-04	U.S. Recession 2026	Kalshi	No (60–70%)	~25–30% Yes	Year-end
2026-03-04	S&P 500 > 7000 EOY	Kalshi	~50/50	TBD	Year-end
2026-03-04	Gov't Shutdown < 42 days	Kalshi	Lean under (~55%)	42.7d avg	When resolved
2026-03-04	Iran Ceasefire by Mar 31	Polymarket	~40–45%	61% Yes	Mar 31
2026-03-04	Iran Regime Fall by Jun 30	Polymarket	~15–20%	36% Yes	Jun 30
2026-03-04	Best AI Model March	Polymarket	Anthropic ~55–60%	66% Anthropic	Mar 31
2026-03-04	BTC below $65K in March	Polymarket	~70–75%	76% Yes	~Apr 1
2026-03-04	U.S. Ground Invasion Iran	Polymarket	~8–12%	19% Yes	End 2026

Brier score = mean squared error between predicted probability and resolved outcome (0/1). Lower is better. How we compute it →

METHODOLOGY · LINEAGE

The same forecasting method that beat the CIA. Now pointed at Kalshi.

The Good Judgment Project's superforecasters beat intelligence analysts with classified data by 30%. Marty applies their framework — reference classes, bias scans, calibrated ranges, pre-committed update triggers — to live prediction markets.

“The strongest predictor of rising into the ranks of superforecasters is perpetual beta — the degree to which one is committed to belief updating.”
— PHILIP E. TETLOCK · SUPERFORECASTING (2015)

PRICING

One plan that pays for itself, or a free seat at the table.

Pro covers active traders end-to-end. Tester is genuinely free — read every analysis, see every Brier, no card.

BETA Free during public beta. All Pro features unlocked while we build the track record. Pricing locks in when we exit beta — early users keep their tier at the launch rate.

TESTER

Free

Read every analysis. No card required.

Start testing

Read all published analyses
Public scorecard + Brier history
Daily top-3 markets digest
1 saved watchlist

Direct answers to the questions a skeptical Kalshi trader actually asks.

Is this financial advice?

No. Ask Marty produces calibrated probability estimates and a structured rationale for prediction-market questions. Nothing on this site or in the product is a recommendation to buy or sell any contract. You are responsible for your own positions.

How do you compute Brier scores?

Brier = (probability − outcome)², where outcome is 1 if the market resolved YES, 0 if NO. We use the midpoint of Marty's range. Lower is better. Marty's running score is 0.023 — a perfectly calibrated coin flip scores 0.25, and perfect foresight scores 0.000. That puts Marty roughly 11× more accurate than random guessing. Every component is shown on the scorecard page.

What happens when you get a call wrong?

It stays on the public scorecard with the full original analysis, the resolved outcome, and a post-mortem note. We never delete or rewrite past calls. The track record is the moat — sanitizing it would defeat the entire premise.

How is this different from ChatGPT?

Three things. Structured methodology — every analysis runs the same six-step framework, no prompt drift. Live market data — Marty pulls Kalshi/Polymarket prices, not a stale training corpus. Tracked accuracy — every output is logged with a Brier score, so you can see when it works and when it doesn't.

Which markets do you cover?

Kalshi and Polymarket today, with Manifold support coming during beta. We support any binary or multi-outcome market with a public resolution criterion. Continuous markets (e.g. "S&P closing price on Dec 31") are handled as a series of binary thresholds.

Can I cancel?

Any time. Your scorecard history stays accessible to you in read-only mode.

Tetlock-method analysis for prediction markets. Every call tracked and graded.