Prediction markets, trader behavior & AI agents. Building HowSure. Market breakdowns, experiments and things that changed my mind.
future
Joined July 2014
- Tweets6.8K
- Following404
- Followers51.5K
- Likes4.6K
Pinned Tweet
I’m building HowSure for prediction-market traders and agent builders.
Check the contract, inspect its sources and estimate your size.
Public reads are free. Saved team workflows: $100 / 30-day pilot.
howsure.com/pilot
568
I want a trading tool with good taste.
This is an extremely annoying requirement to put in a product spec.
353
Yesterday I was talking about whale tracking.
Metaculus’s $50k tournament for forecasting bots has asked where Percy the shark will be on Dec 15.
The shark is unlikely to provide forward guidance.
1
322
These are real Fall FutureEval bot questions, already closed for forecasts:
Percy: metaculus.com/questions/4634…
Pantone: metaculus.com/questions/4634…
GTA VI: metaculus.com/questions/4627…
$50k is the tournament’s total prize pool.
63
bez | Howsure retweeted
Some career news: I’ve joined @TrustWallet to lead marketing and growth.
Trust Wallet was my first wallet, where my crypto journey began. Now I’m working closely with @felix_fan and the team to bring the freedom of onchain finance to the next hundred million users.
We have one of the largest user bases in Web3, a product that keeps getting better, and a team that’s passionate about the future of self-custody.
The goal now is clear: make Trust Wallet the default for everyone.
I’ll be leading growth, brand, and community, and building a stronger bridge between our users and the team.
Excited for what’s ahead. Let’s build. 💙
I want whale tracking in HowSure.
Prediction Frontier’s Oct 11 Polymarket midterms snapshot shows a $3.43M position with matched YES and NO.
Apparently the little whale icon is going to be the easy part.
1
916
Prediction Frontier’s Nov 6 report; Oct 11 snapshot.
rdba held matched YES + NO on a Democratic sweep, then closed the position on Oct 25.
This describes one position, not the wallet’s whole book.
predictionfrontier.com/resea…
100
“Yes, but not this month” is a perfectly good opinion.
I like this because it gives “eventually” an actual price tag. Crypto price targets have enjoyed that free extension for long enough.
we have yield curves for interest rates. why don't we have something similar for probabilities?
the chance of btc hitting 150k in 30 days, 90 days and 180 days should form a tradable curve.
you could trade the steepness, hedge timing risk or position for an event happening later than the market expects.
there's probably an entire derivatives market hiding inside the timing of outcomes.
513
Sunday beat Election Day on Polymarket in contracts traded: 959M vs 767M.
Election Day won in dollars traded: $250M vs $232M.
1
348
Source: Polymarket’s official data. US + International, taker volume, Nov 1 vs Nov 3.
US Eastern days: Sunday had 25 hours (DST), Tuesday 24. Both rankings survive a per-hour comparison.
data.polymarket.com/
110
Imagine your agent warns of a GPU shortage. The provider moves capacity. No shortage.
Great warning. Losing YES bet if held to resolution.
So how should that agent get paid?
one of the more overlooked applications of prediction markets is resource allocation.
i) a cloud provider using markets to estimate GPU shortages across regions over the next 24 hours.
ii) agents could trade those forecasts while the provider uses the prices to decide where to move capacity or adjust pricing.
eventually the market isn't just predicting the shortage but also helping prevent it.
418
My dream trading agent has an embarrassingly specific first job: make five datasets agree on what a timestamp is.
Anthropic’s new study: 48.1% of Claude Code scientific research sessions were data analysis. 2.1% were idea generation.
5
1
422
Source: Anthropic, Nov 5.
78,135 interactive Claude Code research sessions, Aug 2–Sep 30. Weighted shares; consumer plans only. Ideas woven into other tasks may be undercounted.
anthropic.com/research/how-r…
90
Yesterday I asked an AI to study a Polymarket wallet for trade ideas.
Today I found a Hyperliquid paper where wallet history improved 1-second return forecasts.
Apparently this is a literature review now.
1
374
Ding & Zhai, Nov 3. BTC/ETH/SOL, 1-second return forecasts.
Ridge R²: 10.88% → 12.31%
Boosted trees: 19.48% → 20.65%
Retrospective test, July 21–27. R² measures forecast fit, not trading returns.
arxiv.org/abs/2611.05218
1
95
He Has an Excuse for Every Job | DEEP MUD youtube.com/shorts/49yRMYQSH…
387
Kalshi’s 7-day midterm map, by account location:
New York’s most traded market: Michigan Senate.
D.C.’s: Los Angeles mayor.
I would love to hear the D.C. → L.A. origin story.
1
432
Source: Kalshi Trade Watch, Nov 4, 14:50 UTC. Previous 7 days, national markets included.
Ranked by contracts traded by individual accounts; institutions and block trades excluded.
kalshi.com/research/trade-wa…
69
A monthly bill for being stubborn would do incredible numbers on this website.
I’d still want to know who’s better informed and who just has a bigger budget.
a primitive nobody's built: a funding rate on beliefs.
hold yes on an open-ended event & pay or earn funding vs the crowd instead of waiting for a date.
> fed cuts in 2027 trades for 14 months & bleeds or pays you each hour
> wrong & stubborn gets expensive, right & early gets paid
conviction with a carry cost is the most honest signal there is
377
FABench describes agents applying a gold coin’s timeline to ANY Trump coin release. A different $1 coin made the deadline.
I’m building HowSure, and this is the failure that worries me: the research looks solid until you notice what it’s actually about.
1
383
FABench, Appendix G. Forecast date: July 1; deadline: Sept 15. Agents applied a gold coin’s timeline too broadly. A different $1 coin went on sale Sept 2.
arxiv.org/abs/2610.40940
66