Democrats retook the Senate on Nov. 3, the side all twelve BULX agents had backed327.
The Associated Press reported the result at 11:40 p.m. Eastern. "Democrats have retaken control of the U.S. Senate," it wrote27.
On Nov. 2 at 1:51 p.m. Eastern, the agents posted their final forecast3. Their estimates ran from 61% to 68%, and the average was 64%. Kalshi traders paid 75 cents on the dollar that day4, and Decision Desk HQ's election model said 65%23.
Nathan Scherner, BULX's chief executive, spoke with AI Lately on Thursday, before a board meeting on the live test1. Four days before the vote, he wrote to AI Lately after reading Your Chatbot Now Quotes the Betting Odds. He offered his committee for public grading20. AI Lately published the call the day it was made26.
Disclosure: AI Lately has offered BULX a paid, labeled advertising placement. Coverage stays independent of advertising, and this report follows AI Lately's editorial standards.
Breaking the Artificial Hive Mind
Scherner read the result as a test of independence. "I would say that it proves that BULX doesn't interact like a hive mind," he said. "It has the ability and capabilities to go against the trend in its entirety whilst also keeping logistics and analytics sound."1
That goal took 45 days. Early versions kept producing the same forecast from every agent, and the team rebuilt the system to stop it1.
Today the seats run on three model families: Anthropic's Claude, OpenAI's GPT and Nvidia's Nemotron22. Beneath them sits OASIS, an open-source simulator from a 23-author paper that scales to one million agents7. A memory layer built on Zep's Graphiti stores facts with timestamps, so agents can follow how news changes over days or weeks8.
Five roles stay fixed on every run: superforecaster, resolution analyst, prediction market trader, Bayesian statistician and contrarian20. Each seat reads the same case file alone and writes a probability before it sees anyone else's. Prompts tell each one to start from a base rate, the long-run average for similar past events.
The research behind the problem is clear. Stanford researchers led by Shibani Santurkar described the challenge in 2023. "We find substantial misalignment between the views reflected by current LMs and those of US demographic groups."9
A 2024 Stanford study led by Joon Sung Park found that agents built from two-hour interviews with 1,052 Americans did better10. BULX takes a third route. It gives each seat an analytical job instead of a voter identity20.
An internal 16-agent test showed the effect, Scherner said. "It showed that they'll talk against each other to come to a solution that isn't joint," Scherner said1.
Two Ways to Grade an Election Call
Statisticians score a probability forecast with the Brier score. It squares the gap between the forecast and the outcome, so lower is better25. Democratic control counts as 1.
Kalshi's 75% scored 0.06254. Decision Desk HQ's 65% scored 0.122523. BULX's 64% scored 0.12963.
Kalshi's 75% scored 0.0625 on the Senate result, ahead of Decision Desk HQ's 0.1225 and BULX's 0.1296
Brier score of each Nov. 2 forecast after Democrats retook the Senate; lower is better
Each bar is one forecast's Brier score: the shortest bar was closest to the result, and all three forecasts backed the winning side.
Sources: BULX community [3]; Kalshi [4]; Chris Cillizza (Substack) [23]; Wikipedia [25]; PBS NewsHour [27]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value |
|---|---|
| Kalshi traders, 75% | 0.0625 |
| Decision Desk HQ, 65% | 0.1225 |
| BULX committee, 64% | 0.1296 |
All three backed the winner. Kalshi's higher confidence scored best, so the market takes this round. One event settles one round, and a fair comparison needs many forecasts.
The committee's path was its own. On Sept. 26 it priced Democratic control at 69% while Kalshi sat at 62%2. By Nov. 2 Kalshi had risen 13 points to 75%. The committee moved the other way, five points lower, to 64%3.
Seat notes show the same independence. One seat on the BULX Investment Committee, a prediction market trader, wrote on Nov. 2: "The 75% quote looks rich by 10 points; with this volume, I'd still avoid a blind short until fees, depth and slippage confirm there's a clean entry."3
Scherner offered a second, informal yardstick on Thursday: the share of races Democrats won across the map. "It wasn't obviously grounded in science," he said of it1. AI Lately has yet to check that comparison. The Brier score stays the standard test.
Tuesday's vote carried other AI stories. Voters in ten states banned or restricted data centers, as Voters Punish Data Centers as AI Backlash Spreads reported. The side AI industry money backed lost seven of twelve close races, per AI Super PACs Back Seven Losers.
Robots, Base Rates and Five Thousand Tests
Scherner was quick to name what the result left open. "One thing that it did not prove is complete certainty," he said1. The agents are models, and "we can try to get them as general intelligence as humanly possible, but they're still robots."
Before the vote, BULX had posted 22 times on X and scored eight stock simulations22. A call earns 50 points for the right direction and up to 50 for landing near the close. Seven of the eight got the direction right. The best score was 98.9, on SPY13.
On Kalshi, the committee called a 15-minute Bitcoin contract18 and a Trump approval contract19 correctly. A third call leaned 68% toward the Packers, and the Falcons won 35 to 1417. With the Senate call, that makes 10 right out of 12 resolved calls.
BULX got 10 of its 12 resolved public calls right; the misses were Wendy's and the Packers game
BULX's public stock, Kalshi and Senate calls resolved by Nov. 5, 2026, one cube per call
Each cube is one resolved call: magenta cubes are calls BULX got right, and the two gray cubes are the Wendy's stock call and the Packers game.
A small sample of calls BULX chose to post; it shows a record and leaves calibration open.
Sources: X (@bulxai) [11]; X (@bulxai) [12]; X (@bulxai) [13]; X (@bulxai) [14]; X (@bulxai) [15]; X (@bulxai) [16]; X (@bulxai) [17]; X (@bulxai) [18]; X (@bulxai) [19]; X (@bulxai) [21]; AI Lately [22]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value | Note |
|---|---|---|
| Called right | 10 | Seven stock calls, two Kalshi calls and the Senate call |
| Called wrong | 2 | Wendy's, scored 0, and the Packers game |
| Whole | 12 |
Twelve calls is a small sample, and BULX chose which ones to post. Scherner said so himself, and he has a plan. "I mean, a clean record is great until it fails, right?" Scherner said1.
He wants thousands of public, timestamped tests, on a Google Sheet for example. His working figure is 5,000. "Then when we get one wrong, they can be like, oh, well, they weren't hiding anything," Scherner said1.
Learning From Misses and Injecting Shockwaves
The clearest miss was Wendy's. On Oct. 27, a 500-agent BULX simulation projected a 3% dip to $6.00 for the next session21. Before the Oct. 28 open, Nelson Peltz's Trian Fund Management offered $7.35 a share in cash, about $1.4 billion. The stock rose 14% in morning trading24.
bulx@bulxaiWe used our 500 agent symphony to simulate another stock for next trading day.
This request was $WEN and the agents simulated a -3% drop, leaving it at $6 even for tomorrow!
98 agents sold
82 agents bought
320 agents held https://t.co/GuCUDwVslJ pic.twitter.com/kcYIR2xtiZ
The simulator reads only public news from prior days, so a private boardroom bid stayed hidden. Scherner put it plainly: "we can't predict that. We would need inside information."1
He welcomes misses like it. "The more fails we can hit in the beginning, the sooner the better, because we want to be that reliable, dependent service," Scherner said1.
BULX has also changed its engine so users can alter a simulation while it runs. "We just came out with a scenario update where you can now inject changes into the simulation while it's going on," Scherner said1. His team calls them shockwaves.
Here is a hypothetical, written by AI Lately. An analyst could start with the 500 retail trader agents near $6.00. A buyout rumor dropped in mid-run would show how the agents reprice the stock during the session.
Two Hundred Users and the One-Tap Future
The platform crossed 200 active users this past week1. Scherner said he expects to sign closing pre-seed documents with his board over the weekend.
Two upgrades top the list. Scenario reports will get infographics that update after each run. Presets will let a newcomer start a tailored simulation in two clicks. The aim, Scherner said, is "just a couple buttons and go."1
Education is the larger task. MiroFish, an open-source engine built on the same OASIS simulator, has 83,500 GitHub stars6. Scherner estimates open source covered the first 10% of the public learning curve1.
He expects wider use to move markets. "If more people are using BULX and more people are using these systems and they're trusting them, the odds are going to shift in that way," Scherner said1.
Tuesday gave BULX its first graded newsroom benchmark26. Now the team turns to the next 5,000 tests.
What to do
- Judge any forecast by its Brier score, and wait for volume25. One election is one round. Ask to see the 5,000-test ledger when BULX publishes it1.
- Try a shockwave on a case you already know, such as a takeover bid or a rate decision. Note which agents move first1.
- Treat a simulated crowd as a second opinion beside the market price. Scherner himself says "they're still robots"1.
Sources
- Ryan Elliott Dennis, "Follow-up interview with Nathan Scherner, chief executive of BULX," AI Lately, Nov. 5, 2026
- BULX, "Got a bet on 'Which party will win the U.S. Senate?' Take a look what our predictive agents say about the betting field," X (@bulxai), Sept. 26, 2026, https://x.com/bulxai/status/2103980833967329623
- BULX, "Investment Committee #committee_759623fd31c7: Who will control the Senate?," BULX community, Nov. 2, 2026, https://bulx.ai/community/post/541b384f-1dbd-48b7-a23a-4d83c479d81e
- Kalshi, "Who will control the Senate?," Kalshi, Nov. 2, 2026, https://kalshi.com/markets/controls/senate-winner/controls-2026
- BULX, "BULX: Behavioral AI Prediction Network," bulx.ai, Nov. 5, 2026, https://bulx.ai
- 666ghj, "MiroFish: A Simple and Universal Swarm Intelligence Engine, Predicting Anything," GitHub, Nov. 2, 2026, https://github.com/666ghj/MiroFish
- Ziyi Yang, Zaibin Zhang, Zirui Zheng and others, "OASIS: Open Agent Social Interaction Simulations with One Million Agents," arXiv, Nov. 18, 2024, https://arxiv.org/abs/2411.11581
- Zep, "Graphiti: Build Real-Time Knowledge Graphs for AI Agents," GitHub, Nov. 2, 2026, https://github.com/getzep/graphiti
- Shibani Santurkar, Esin Durmus, Faisal Ladhak, Cinoo Lee, Percy Liang, Tatsunori Hashimoto, "Whose Opinions Do Language Models Reflect?," arXiv, March 30, 2023, https://arxiv.org/abs/2303.17548
- Joon Sung Park, Carolyn Q. Zou, Jonne Kamphorst and others, "LLM Agents Grounded in Self-Reports Enable General-Purpose Simulation of Individuals," arXiv, Nov. 15, 2024, https://arxiv.org/abs/2411.10109
- BULX, "FINAL TEASER: Here is the final prediction of BULX's 1024 retail trader's simulation on $AMD for the next 3 days," X (@bulxai), Sept. 8, 2026, https://x.com/bulxai/status/2097327360844632536
- BULX, "BULX call on NVDA, one-day simulation," X (@bulxai), Sept. 17, 2026, https://x.com/bulxai/status/2100625458522128522
- BULX, "BULX call on SPY, next session," X (@bulxai), Oct. 18, 2026, https://x.com/bulxai/status/2111971825215496282
- BULX, "BULX call on TSLA, next session," X (@bulxai), Oct. 20, 2026, https://x.com/bulxai/status/2112664539330453760
- BULX, "BULX call on AAPL, next session," X (@bulxai), Oct. 21, 2026, https://x.com/bulxai/status/2112918233514930636
- BULX, "BULX call on NVDA, next session," X (@bulxai), Oct. 22, 2026, https://x.com/bulxai/status/2113381329832485117
- BULX, "Atlanta Falcons are playing the Green Bay Packers at Lambeau Field tonight in the NFL!," X (@bulxai), Sept. 24, 2026, https://x.com/bulxai/status/2103120197993574445
- BULX, "BULX committee call on Kalshi's 15-minute Bitcoin contract," X (@bulxai), Sept. 20, 2026, https://x.com/bulxai/status/2101811225646731625
- BULX, "BULX committee call on Kalshi's Trump RealClearPolitics approval contract," X (@bulxai), Oct. 10, 2026, https://x.com/bulxai/status/2108945623970116086
- Nathan Scherner, "Email from Nathan Scherner to AI Lately on the Senate forecast, the committee and the public record," AI Lately, Oct. 29, 2026
- BULX, "We used our 500 agent symphony to simulate another stock for next trading day. This request was $WEN," X (@bulxai), Oct. 27, 2026, https://x.com/bulxai/status/2115254394438684735
- Nathan Scherner, "Email from Nathan Scherner to AI Lately correcting the AMD close and listing every scored call," AI Lately, Nov. 2, 2026
- Chris Cillizza, Geoffrey Skelley, Maddisen Murrell, "Democrats now have a 2-in-3 chance of winning BOTH the House and Senate," Chris Cillizza (Substack), Nov. 2, 2026, https://chriscillizza.substack.com/p/democrats-now-have-a-2-in-3-chance
- David Moadel, "Wendy's Surges 14% on Trian-Led $1.4B Take-Private Offer; McDonald's Edges Higher, Yum Brands Stays Put," 24/7 Wall St., Oct. 28, 2026, https://247wallst.com/investing/2026/10/28/wendys-surges-14-on-trian-led-1-4b-take-private-offer-mcdonalds-edges-higher-yum-brands-stays-put
- Wikipedia contributors, "Brier score," Wikipedia, Nov. 3, 2026, https://en.wikipedia.org/wiki/Brier_score
- Ryan Elliott Dennis, "Can an AI Committee Tell the Future?," AI Lately, Nov. 2, 2026, https://ailately.com/articles/bulx-crowd-inside-the-machine
- Associated Press, "Democrats take control of the Senate after winning key midterm races," PBS NewsHour, Nov. 3, 2026, https://www.pbs.org/newshour/politics/democrats-take-control-of-the-senate-after-winning-key-midterm-races
