Paper money · Real prices · Published live
7 AIs trading in public
Paper money, real prices. Every trade is published the moment it is decided — the losing ones too.
AVGO
CME
CRH
FERG
MCO
SLM
SPY
STX
TDG
DAL
DVA
GOOGL
KHC
KR
OXY
SIRIBrowse
How they are doing
BTCCRV
ETHETHFI
LINK+4BTCETHIBITLINK
NVDA+1
AAPL
AXP
BAC
CB
CVX+10
AMZN
GOOG
META
MSFT
SGOV+1
COST
JNJMSFT
UNP
VSince each portfolio started, after modelled trading costs. Retired portfolios are not ranked here — they no longer trade, and they are published in full in the Graveyard.
Showing every layer. Pick Simple or Advanced in the header to choose one — nothing is removed either way, and the full record is always in this page.
Can AI beat the market?
Sixteen AI-managed portfolios trade $100,000 each, paper-money, under published rules — against the market, in public, permanently. Every decision timestamped before its outcome. Free to read.
+3 more
+9 moreMarket benchmark over the same window: -1.36% (S&P 500 (SPY, total price return))
Picks & Shovels bought ENTG · requested 5% of NAV · conviction 3/5
Decided 2026-08-26 20:05Z — committed to the audit chain before the outcome was knowable.
Leaderboard
| Portfolio | Mandate | Operational | Total return | Benchmark, same window | Max drawdown | Live since | Sharpe |
|---|---|---|---|---|---|---|---|
| P04 Digital Assets & Equities United States · + crypto | mandate active | Catching upCATCHING_UP | +5.84% | — | -1.31% 2026-09-03 → 2026-09-10 | 2026-08-13 21 daily obs | withheld |
| P01 Long-Term Compounders United States | mandate active | Catching upCATCHING_UP | -0.10% | -1.36% SPY | -1.28% 2026-08-25 → 2026-09-09 | 2026-08-13 21 daily obs | withheld |
| P03 Billionaire Consensus United States | mandate active | Frozen until its next trading windowOUTSIDE_TRADING_WINDOW | -3.65% | -1.36% SPY | -4.87% 2026-08-14 → 2026-09-10 | 2026-08-13 21 daily obs | withheld |
| Portfolio | Mandate | Operational | Total return | Benchmark, same window | Max drawdown | Live since | Sharpe |
|---|---|---|---|---|---|---|---|
| P09 Crypto 24/7 (EXPERIMENTAL) experimental Crypto · 24/7 | mandate active | Waiting for its next scheduled runWAITING_FOR_SCHEDULE | +29.39% | — | -7.42% 2026-09-06 → 2026-09-15 | 2026-08-14 44 daily obs | withheld |
| P12 Berkshire 13F Tracker experimental United States | mandate active | Frozen until its next trading windowOUTSIDE_TRADING_WINDOW | +0.64% | -1.36% SPY | -1.43% 2026-09-03 → 2026-09-09 | 2026-08-13 21 daily obs | withheld |
| P13 Pershing Square 13F Tracker experimental United States | mandate active | Frozen until its next trading windowOUTSIDE_TRADING_WINDOW | +0.55% | -1.36% SPY | -1.94% 2026-08-13 → 2026-08-18 | 2026-08-13 21 daily obs | withheld |
| P20 Operator Discretionary experimental United States | mandate active | Waiting for its next decisionWAITING_FOR_SCHEDULE | -0.17% | +0.22% SPY | -0.50% 2026-08-25 → 2026-09-11 | 2026-08-21 15 daily obs | withheld |
| P08 The Learning Book experimental United States | mandate active | Catching upCATCHING_UP | -2.00% | -1.36% SPY | -2.06% 2026-08-17 → 2026-09-14 | 2026-08-13 21 daily obs | withheld |
| P16 Picks & Shovels experimental United States | mandate active | Missed a scheduled runSCHEDULE_MISSED | -3.15% | -1.36% SPY | -3.38% 2026-08-17 → 2026-09-14 | 2026-08-13 21 daily obs | withheld |
| P07 Claude Discretionary experimental United States | mandate active | Catching upCATCHING_UP | -4.25% | -1.36% SPY | -4.98% 2026-08-13 → 2026-09-02 | 2026-08-13 21 daily obs | withheld |
Published experiments, expected to be the most likely to reach the Graveyard — and published for exactly that reason. Same rails, same audit chain, same $100,000, no special treatment in any aggregate.
Measured from the earliest portfolio inception to the latest close available for it — the identical window as the records it sits next to. Index values are derived percentage returns over IEX last-trade closes, which are licensed for publication; no raw vendor price is rendered.
| Portfolio | Mandate | Operational | Total return | Max drawdown | Live since | Sharpe |
|---|---|---|---|---|---|---|
| P06 Intraday AI Lab (EXPERIMENTAL) experimental United States | retired | RetiredRETIRED | 0.00% | 0.00% | 2026-08-14 2 daily obs | withheld |
| P02 AI Swing Trader United States | retired | RetiredRETIRED | no record | — | never traded | withheld |
| P05 Political Trades Tracker United States | retired | RetiredRETIRED | no record | — | never traded | withheld |
| P14 Bottleneck Hunters experimental United States | retired | RetiredRETIRED | no record | — | never traded | withheld |
| P15 Second-Order Winners experimental United States | retired | RetiredRETIRED | no record | — | never traded | withheld |
| P17 Hidden Monopolies experimental United States | retired | RetiredRETIRED | no record | — | never traded | withheld |
A portfolio can be retired but never removed. Why the Graveyard exists →
- Sorted by total return, with maximum drawdown shown in the same row at equal visual weight. A leaderboard that ranks on return alone teaches the wrong lesson.
- Every entry shows when its record began and how many daily observations it holds. A portfolio two days old sits next to one five hundred days old and the difference must be visible instantly.
- No annualised figure is shown below 60 observations. Annualising a short sample produces a number that looks precise and is not.
- Retired portfolios appear below the live ones, present and included in every aggregate.
The experiment to watch: can an AI learn from this platform's mistakes?
Every other portfolio here reads market data. P08 is also handed this platform's own realised record — the closed trades of the books its rulebook names (7 of 16, listed on its own page), what each cost, how long it was held, why it was exited and what it returned — as facts in its prompt, filtered so that no outcome unknown at decision time can leak in. Its own trades are excluded. No training, no fine-tuning: a portfolio whose prompt contains an honest scoreboard.
The control is P07 Claude Discretionary: the same model, the same universe shape, the same deterministic rails, without the record. The difference between the two is the closest this platform can get to measuring what access to its own history is worth. Beating the index would be pleasant; beating P07 is the actual hypothesis — and if the record shows the opposite, that stays up too.
The mandates
Core
Compound capital over a 5-10+ year horizon by owning a concentrated set of high-quality businesses that earn durable returns on invested capital and reinvest at attractive rates. Minimise tu…
Hold the US-listed equities most widely and heavily owned across a fixed, publicly declared list of large institutional investment managers, as disclosed in their SEC Form 13F-HR filings.
+14 moreExpress diversified exposure to the digital-asset economy across two sleeves: spot crypto majors, and US-listed equities and ETPs whose economics are driven by digital assets. Survive a full…
Labs — labelled experiments
Test, in public and with realistic execution costs, whether a large language model given a broad liquid universe and no style constraint can select equities that outperform a passive benchma…
Test whether a language model that can see this platform's own realised trading record — every closed trade, its cost, its holding period, its exit reason and its outcome — selects better th…
Test, in public and with realistic entry-tier retail costs, whether an LLM given delayed venue data can produce positive expectancy trading spot crypto on a 24/7 cadence, across weekends and…
+3 moreMechanically copy the disclosed US-listed equity book of Berkshire Hathaway Inc (SEC CIK 0001067983) from its most recent Form 13F-HR filing: value-proportional weights across the top disclo…
+9 moreMechanically copy the disclosed US-listed equity book of Pershing Square Capital Management, L.P. (SEC CIK 0001336528) from its most recent Form 13F-HR filing: value-proportional weights acr…
Identify major growing industries and invest only in the infrastructure and equipment providers around them — semiconductor equipment, testing and metrology, cooling, power infrastructure, l…
Test, in public and with realistic execution costs, whether the operator selecting equities by hand — free to act on news between the AI books' decision windows — outperforms the same platfo…
Graveyard — retired, never deleted
Capture directional moves lasting roughly 3 to 30 trading days in liquid US equities and ETFs, using a defined-risk framework where every position has a pre-committed stop and a maximum hold…
Publish, and mechanically mirror, the stock purchases that US federal legislators disclose under the STOCK Act. The point is transparency: to show what a portfolio built only from public dis…
Test, in public and with realistic costs, whether an LLM given only delayed intraday data can produce positive expectancy over a session horizon, flat by the close every single day.
Own companies that control something an industry cannot easily function without: sole or limited-source suppliers, extreme switching costs, tiny components inside huge end-markets, specialis…
Instead of buying the obvious beneficiary of a major trend, own whoever benefits two or three steps downstream. Example logic: AI boom -> data centres -> electricity demand -> transformers -…
Own companies with obscure monopolistic or oligopolistic positions: dominant share of a small niche, regulatory or certification moats, extremely high customer retention, proprietary standar…
Why this might be worth reading
The AI ranks and explains. It never prices or executes.
The model's output schema has no field in which a price, a share count, a notional or a risk limit could be expressed. It is not instructed to avoid them — there is nowhere to put them. The one sizing input it may offer is a requested weight, on the books whose rulebooks accept one: a bounded preference the engine clamps to the mandate's own caps. Everything else that touches money is computed by deterministic code the model cannot see or influence, and a test fails the build if anyone widens that schema.
A limit that is published but not enforced is worse than no limit.
A reader cannot tell the difference between a rule that binds and a rule that is decorative. Every limit on this site names the code path that enforces it, and a test walks every mandate, drives the engine into the state each declared control claims to protect against, and fails the build if nothing stops it.
The mistakes stay up.
Every AI proposal the rule engine refused is published with its reason code. Passages the model could not fully ground in source data are published flagged, not deleted. Retired portfolios stay in every aggregate — the Graveyard already has an entry, withdrawn before it ever traded, and it is published rather than dropped.
You can check it rather than trust it.
Every event is committed to a hash chain, and today's head hash is published. Save it, and you can prove later whether history was rewritten. The complete payload every page was rendered from is downloadable.
Real findings, not claims
Two things this project measured against live SEC data, published because both are the kind of error that would otherwise be invisible. See the evidence →
Payload generated 2026-09-26 · audit chain at record 6136 · verified — verified