Sealed before the open. Scored in public
Every weekday before the New York open, AIs from the US and China seal calls on 12 markets — the indexes, the US 10-year, the dollar, VIX, gold, oil, copper, bitcoin — for a day, a week, a month and six months. External timestamps fix the moment: Bitcoin via OpenTimestamps since October 16, 2026 (first issue, 22:30 ET), RFC 3161 since October 20, Sigstore Rekor since October 21; the open book lets anyone recount every result. How to check ↓ What “sealed” means ↓
First your call, then the comparison: Seal your call before the AIs →
Loading live data…
What “sealed” means
Forecast → external timestamp → open ledger → scored after close
The live score today
Numbers straight from the open record: they change after every issue and every close.
Loading the live score…
Brier of the Council, the crowd and their mix on the same questions, with the flag «saw the market price» on every sealed answer.
Loading…
check it in 3 steps
1) download council-forecasts.csv or the sealed book council-book.jsonl; 2) compare today's lines with the seal hash and its OpenTimestamps proof; 3) recount Brier and the right side from the outcomes in the same file.
Every account, studied
Four leagues on the same AIs, every account on one chart, and the check that being right more often is not the same as earning more.
Loading…
Calendar of mistakes
Every position an AI closed at a loss in the league, day by day — the darker the day, the bigger the losses. Tap a day to read who lost, on which market, how much, and the AI’s own reason.
Loading…
How good are the forecasts — checks on ourselves
Does a stated 60% come true 60% of the time, what a good Brier looks like, who is right in which market regime, and two tests sealed in advance.
Loading…
Which AI is stronger on which market
More useful than one overall winner: each AI on each of the 12 markets, and accuracy by type of question.
Loading…
Book of misses — explore every case
Where the Council and the crowd were both wrong, where only one of them was, market by market — and what each AI said that morning, from its own sealed line.
Loading…
How to check any call: call → seal before the open (SHA-256 chain; Bitcoin via OpenTimestamps since Oct 16, 2026; RFC 3161 time stamps since Oct 20; the Rekor log since Oct 21) → outcome at the close → score in the open book; a correction is a new line, the old one never changes.
Nothing deleted, every fix on file — check it yourself
The open book of forecasts only grows: lines are numbered, each carries the link of the line before, and a fix is a new line beside the old one. The latest lines are checked as the page opens; the whole book — in your browser, on one tap.
Loading…
The buttons open the Money live record on polyhelper.ai/trends, running since October 16.
Scoreboard by origin: US · China · Europe
Same questions, one seal. The score accumulates; a verdict comes after 350 comparable questions; until then the comparison is provisional. The sealed record runs 10+ AIs from the US and China; Le Chat (Mistral, Europe) answers the Council poll and joins the sealed book next.
| Who | Right side | Brier | Forecasts | Team account, $100,000 |
|---|---|---|---|---|
| US AIs | 54.7% | 0.2478 | 1,287 | $96,929 |
| Chinese AIs | 54.7% | 0.2466 | 1,197 | $95,828 |
| European AIs | next | next | — | Le Chat (Mistral) joins the sealed book next |
| The crowd (Polymarket, Kalshi) | 74.0%* | — | 100* | — |
| S&P 500, bought and held | — | — | — | $98,177 |
* the crowd and the Council on 100 comparable crowd questions; AI rows — all their resolved forecasts. Cut-off: session of Nov 13, 2026, data Nov 16, 2026, 11:38 ET. council-league.json ↗ · council-stats.json ↗
Your $100,000 against the AIs → · Humans vs machines: one table for everyone → · Share the day in one picture → · Get each issue the moment it is sealed →
What the AIs’ work cost — beside what they made
Each account’s return before and after paying for the AI itself, at the provider’s public list price: does the result survive the bill?
Loading…
More: the US–China paired score, East against West, calibration, the crowd by venue, independence, the reader experiment, the schedule
Paired on the same 388 cells: Brier for US AIs 0.2495, Chinese AIs 0.2506; difference −0.001 (95%: −0.0056 to +0.0037, 20 sessions) — not yet distinguishable. East against West, Nov 16: the widest gap between the groups is copper — West 47%, East 43%; distance from the crowd: West 11.0 points, East 8.7; furthest from its own group: Amazon Nova, 6.7 points; the result comes after 30 outcomes (0 so far). East against West ↗
- AIs and the crowd on the same contracts — same side on 92% of 49; the crowd sits 22.7 points from 50, the AIs 8.9. The gap contract by contract is in the Divergence Radar. export ↗
- Best start day — from Nov 13, 2026: the Council $99,856, the S&P 500 $99,637. export ↗
- How independent the AIs are — on 12 markets ≈ 2.26 independent voices on the one-day horizon; the method and our verdict-level measure are in the Council ↑. export ↗
- Against the crowd, venue by venue — Polymarket: 51 events, AIs right on 36, the crowd on 35, Brier 0.2177 vs 0.1758; Kalshi: 9 events, 8 vs 9, Brier 0.1974 vs 0.0711. These are the league’s resolved events — a different, smaller set than the 100 comparable questions above. export ↗
- Do the AIs make readers more accurate — readers are split for good into two groups: one sees the AIs before its first call of the day, the other only after; first calls are compared, target 200 per group; the count is building. export ↗
- Schedule — forecasts are sealed before the 09:30 ET open, the day’s score comes after the close, at 16:30 ET; Saturdays bitcoin, Sundays the weekly radar.
History of the models — the Council’s ghost
AIs get new versions. Every version is kept with its own dates and answers, and earlier versions answer the same questions beside the new ones: is the newer one really more accurate?
Loading…
More: all 14 league accounts and 43 more modules of the live record on polyhelper.ai — leagues, market-by-market signals, checks on ourselves, your turn, data · all open
Leagues and virtual accounts
- AI Portfolio League — each AI runs a virtual $100,000: cash plus up to 12 markets, 5–25% a position, filled at the first price after the open, costs inside; the rules are an open file in the method. Right more often ≠ earned more: both are shown side by side. All 14 accounts are in the table above. live on polyhelper.ai · Free open ↗
- Pro league — a second $100,000 per AI with shorts and leverage up to 2×, margin calls and a closed account on a wipeout; its own calendar of mistakes. Nov 13 close: leader Gemini $102,475 (+2.47%), the S&P 500 −1.56% over the same dates from Oct 20. live on polyhelper.ai · Free open ↗
- From the day you pick — $100,000 put into the Council’s portfolio at any start date, against the same sum in the S&P 500. live on polyhelper.ai · Free open ↗
- Seasons — season 1 (Oct 19 – Nov 8) stays frozen as the baseline; from Nov 9 each new league runs from its own start date with its own score. live on polyhelper.ai · Free open ↗
Experiments on the same accounts — the library
Each card: what is tested and the rule, the result right now, the full table one tap further.
Loading…
Do the AIs make readers sharper? → · Learn to read a forecast by playing with one →
Experiments on the same accounts
- Risk league — virtual $10,000 per AI, the whole account on ONE most confident position at 3× every day; a zero is a wipeout. As of Nov 16, 2026: leader ERNIE (Wenxin) $19,942, wipeouts 0 of 14, 5 of 30 resolved bets each; control (random side): a coin $9,498. live on polyhelper.ai · Free open ↗
- Tail league — $100 on the cheap side (under 20¢) of every crowd market, every zero listed. As of Nov 16, 2026: $10,000 → $6,929; 2 of 42 contracts paid. live on polyhelper.ai · Free open ↗
- Fade the Council — virtual $10,000 that bets AGAINST the Council wherever it differs from the crowd by more than 15 points, $100 a bet, with a random-side control. As of Nov 16, 2026: 0 of 30 settled. live on polyhelper.ai · Free open ↗
- Paper alpha fund — rules published before the count: long or short on the Council’s consensus, equal weights, rebalanced on Mondays, against the S&P 500 and cash; started Nov 9. live on polyhelper.ai · Free open ↗
- Confidence poker — each AI spreads a bank across the markets by its own confidence; compared with equal shares. As of Nov 16, 2026: biggest stake Amazon Nova, $1,500 on chips falling. live on polyhelper.ai · Free open ↗
- Whale mirror — the month’s top-5 Polymarket traders by public profit: where an AI sides with them its virtual $10,000 copies, where against — it bets against. live on polyhelper.ai · Free open ↗
- Polymarket–Kalshi spread — the same market on two venues, the cheaper side of each, after fees, against a random-side control. As of Nov 16, 2026: 11 pairs, widest gap 12.2 pp, +$352 vs +$161 for the control, 9 of 30 settled. live on polyhelper.ai · Free open ↗
- Hunt the herd — when the crowd is over 80% sure, an AI may bet against it: +3 for a right hunt, −1 for a wrong one. As of Nov 16, 2026: 81 bets, 0 right. live on polyhelper.ai · Free open ↗
- Shadow modes — the same AIs under other conditions, each with its own accounts and file: the 12 real funds (SOXX, TLT, IBIT…), a closed book without web search, an S&P 100 league, role characters. live on polyhelper.ai · Free open ↗
- Market-aware vs blind Council — the market-aware Council moves halfway to the crowd’s price, the blind one never sees it; lower Brier wins after 30 outcomes. live on polyhelper.ai · Free open ↗
- Council weights v2 — weights set on season 1 and sealed before season 2 (hash published Nov 9): the weighted Council against the simple mean on new questions. live on polyhelper.ai · Free: the result and the hash · the weights: no one open ↗
What the council says, market by market
- Map +100% — sealed before the open, each AI may name one market that could double within a week, with its chance; scored when the week closes. live on polyhelper.ai · Free: the summary line · $2 at launch: in full while fresh, free once resolved open ↗
- Court of the Council — after the seal, up to three AIs that disagreed with the majority name one dated fact it may be missing, and the level at which they would concede. live on polyhelper.ai · Free: the summary line · $2 at launch: in full while fresh, free once resolved open ↗
- Heat map: market × model × horizon — side and confidence of every AI on every market for a day, a week, a month and six months; each cell links to its slice. live on polyhelper.ai · Free: the consensus cell of the day · $2 at launch: the full map while fresh, free once resolved open ↗
- Why they did it — each AI’s own headline, chain of reasoning and why its portfolio looks the way it does, as written. live on polyhelper.ai · Free: each AI’s headline · $2 at launch: the full reasoning while fresh · the reasoning archive: contract open ↗
- Market cards — 12 markets — S&P 500, Nasdaq-100, Dow, chips, the US 10-year, the dollar, VIX, gold, WTI, Brent, copper, bitcoin: each AI’s action, chance of a rise, target, ‘wrong if’ and entry; a price chart with every AI’s entry level and ‘wrong if’ line, the Polymarket and Kalshi odds beside the AIs’; the AIs can stay hidden until you make your own call. live on polyhelper.ai · Free: the consensus line · $2 at launch: every AI while fresh open ↗
- Four roles on one market — bull, bear, historian and crowd advocate on the same market; the historian-vs-advocate gap goes into the debate record. live on polyhelper.ai · Free: the summary line · $2 at launch: in full while fresh, free once resolved open ↗
- Expected moves, 5th / 50th / 95th percentile — the mean of the percentiles the AIs gave, market by market. As of Nov 16, 2026: VIX −13.2% / 3.6% / 29.8%. live on polyhelper.ai · Free open ↗
- Mirror of fear — how far the AIs disagree on tomorrow’s VIX: a disagreement index. As of Nov 16, 2026: spread 5.8 points. live on polyhelper.ai · Free open ↗
- Crowdometer — the crowd’s price of ‘up’ minus the Council’s mean, market by market; right of the line the crowd is pricier. As of Nov 16, 2026: dollar index +15.6 points, gold −5.3, S&P 500 −8.6. live on polyhelper.ai · Free open ↗
- First: the AI or the price — after a news event, did an AI call the move before the price made it. As of Nov 16, 2026: 0 of 9 events. live on polyhelper.ai · Free open ↗
- Council ghost: older vs newer versions — earlier versions of the same AIs on the same questions; is the newer version more accurate (a gap under 5 points counts as the same). live on polyhelper.ai · Free open ↗
- Anti-herd points — a point for changing sides with a stated reason; counts the change, not whether it was right. live on polyhelper.ai · Free open ↗
- Sources the AIs leaned on · links that did not open — every cited link counted by site; a 404 or a non-existent site is counted against the answer. As of Nov 16, 2026: 9 of 189 links (4.8%) did not open. live on polyhelper.ai · Free open ↗
Checks on ourselves
- Museum of the seal — three sealed lists a day — the last before the open, the first after it, the first after the close — with their line counts. live on polyhelper.ai · Free open ↗
- Calibration by AI — ECE and Brier for each AI; who sounds surer than it turns out and who is underconfident. live on polyhelper.ai · Free open ↗
- Who is right in which regime — Council and best-AI Brier when VIX is below 18 or 18–25, when oil moves 2%+, when the S&P 500 is up or down. live on polyhelper.ai · Free open ↗
- Oracle league (Elo) — every AI against every other on the same questions, lower Brier wins; form over 7 days and the AI of the month. live on polyhelper.ai · Free open ↗
- Book of misses — the days when the Council and the crowd were both wrong, with the quote that led the call. As of Nov 16, 2026: 29 of 155 cases; the three latest shown are in the table below. live on polyhelper.ai · Free open ↗
- Places by the 30/10 rule — a place after 30 resolved forecasts over 10 trading days, by horizon, with a 95% interval and tie groups. live on polyhelper.ai · Free open ↗
- Scientific checks — horizon profile, how distinct each AI’s view is, and the test of whether US and Chinese AIs form two blocks. live on polyhelper.ai · Free open ↗
- What speaks against us · what studies showed · nearest projects — outside results set beside ours, each with its source and how our method differs. live on polyhelper.ai · Free open ↗
Your turn
- Make your call first · your own $100,000 — seal your own chance before you see the AIs, or spread virtual $100,000 over the same markets; a shared table of humans against machines and the weekly ‘Council or crowd?’. live on polyhelper.ai · Free open ↗
- Try the rules yourself — runs in your browser, nothing is sent: update a belief with a new fact (odds × strength^weight), check a hash against the seal, test your own rule on resolved questions. live on polyhelper.ai · Free open ↗
- Submit your own model — probability forecasts for the same 12 markets and four horizons in the same JSON, sealed like ours, in a separate table of invited models. live on polyhelper.ai · Free open ↗
Data, widget and languages
- Widget: AIs vs the market — free to put in an article, no sign-up; the numbers update with each issue (served by the live record). Code:
<iframe src="https://polyhelper.ai/trends/markets/embed/en/" width="100%" height="230" style="border:0;max-width:520px" loading="lazy" title="AIs vs the market"></iframe>. live on polyhelper.ai · Free open ↗ - How Money grew — one chart of the Council, the S&P 500 and cash, with the dated steps: Oct 16 first issue and the Bitcoin seal · Oct 19 the league starts · Oct 20 RFC 3161 time stamps · Oct 21 the Sigstore Rekor log · Oct 22 AIs set against the crowd · Nov 6 a block for newcomers · Nov 9 season 2. live on polyhelper.ai · Free open ↗
- Daily alerts — a calendar feed (.ics), RSS and web push, one line whenever an issue comes out; the Open API with every file of the record (CC BY 4.0) — index of all files ↗, webhooks
issueanddaysigned with HMAC. live on polyhelper.ai · Free open ↗ - The live record in all site languages — the live record on polyhelper.ai in English and ten translations: Russian · Ukrainian · Spanish · Portuguese · French · German · Japanese · Chinese · Korean · Hindi. live on polyhelper.ai · Free open ↗
Every module above is live and free on polyhelper.ai/trends; “open” takes you straight to it. Figures dated Nov 16, 2026 come from the live record. Method and seal checks — method ↗.
Cut-off: session of Nov 13, 2026, data Nov 16, 2026, 11:38 ET · snapshot 138d253f1c98 · all money is virtual.
See what moves markets
Markets worldwide. 10+ AI models, their independence measured on every poll. A view you can challenge, trace to sources and check against the public record.
The whole of Money on one page: every chapter, dataset, explanation and source. See everything →
Money gives analytics and probabilities, not personal investment advice; figures carry their source and date, illustrations are marked as such, and past outcomes do not guarantee future ones.
All 14 disclosures
- Money provides analytics and probabilities, not individual investment advice; for personalized recommendations in the US and EU the Goal Navigator works through a registered investment adviser (RIA) or a licensed partner — as Betterment and Wealthfront are set up.
- The top strip shows Nov 13, 2026 closing quotes from open sources. Product data updates live.
- The council screen captures the real council answer of Nov 14, 2026; the Pulse ranking refreshes every hour.
- GitHub star counts and private-company revenue are third-party estimates; ranges are shown where available.
- Historical outcomes and scenarios do not guarantee future results. Interface illustrations are not trading signals.
- The “10-year above 5%” episodes and historical regimes use rounded reference data. The product calculates them from dated series.
- The Scenario Lab uses real facts to show the format; probabilities come from the live council.
- Model, data and broker connections require a terms and licensing review at launch; every model and data connection has a fallback.
- Survey figures describe their own samples: Morningstar covers 500+ institutional asset owners in 11 countries, not all investors. State Street flows cover the ETFs in its sample; the World Gold Council figure covers ten months (Jan–Oct), not a full year; the IEA’s $3.4tn is industry spending, not fund flows.
- The 2027 capital map is a working map, not a forecast. Forecasters disagree, and rising investment in a sector does not guarantee returns on its stocks.
- The Thesis Monitor card is an illustration assembled from facts on this page, not a recommendation on grid-equipment stocks.
- Web search is not access to all data: licensed research and expert calls are not on the open web. Open-source code does not grant rights to financial data.
- No public evidence shows a general-purpose LLM consistently out-forecasting markets.
- Word and deed records what public figures said and disclosed, with a short verbatim quote, its source and date, a link and a neutral result — not a judgment of anyone’s honesty or intent. Numbers about named people and funds come from primary filings and their own letters; interpretations are labeled “Money’s assessment”; Money is not affiliated with or endorsed by anyone it names, and none of them takes part in it. Commentators who are not public figures appear under a type label (“video commentator 7”) until they choose to be named; a disclosed position may be partial or out of date.