Active paper experiment · AI and markets
What happens when AI meets the market?
Two experiments, one public record. AI helped research a fixed rulebook and propose high-upside hypotheses. The rules make the calls; no AI or agent places trades.
Paper = simulated. No real money is invested or moves; all values are hypothetical.
Loading the latest daily snapshot…
Checking source
Experiment 01 · AI proposes, arithmetic judges
Moonshot Scout
A deliberately small record of long-shot, big-if-right ideas. Every loss stays visible beside A fund that trades like a stock but tracks the S&P 500 — the “what if I’d just bought the index” comparison..
Two honest views
— recordsSame hypotheses. Different questions.
How did the whole batch do, sized by how confident we were?
Confidence (1–5) rates how good the payoff looks if the bet is right — not how likely it is to happen. sets $25–$100 in the model. SPY gets the same dollars on the same days.
- Model dollars
- —
- Paper value
- — —
- SPY, same $ / days
- — —
- Ahead / behind SPY
- — —
Is the AI’s picking any good?
Here every pick counts the same: $100 flat. It answers a different question — is the picking any good, apart from how much we bet on each one?
- Model dollars
- —
- Paper value
- — —
- SPY, same $ / days
- — —
- Ahead / behind SPY
- — —
Paper batchSPY comparison
Exit discipline · public and identity-masked
Exit record.
Every public paper opening, plan revision and rule-fired close is dated here, newest first. Based on verified daily closes — not real-time or trade alerts.
Dated public record
Loading
Signal log
- Revised
- —
- Automatic exits
- —
- Closed
- —
Show older paper events
Open results · private pick details
The Moonshot record
Private details
Ticker, company, The idea and Why now are hidden.
Sign in with Google to reveal them. Paper results and exit plans stay public.
Your Google email is used only to check access.
Experiment 02 · AI researches, rules decide
The Rules Machine
Prices frozen as of 25 Aug 2026. A disciplined strategy built and tested with AI’s help; this panel does not update every trading day yet.
We test ideas hard and drop what does not hold up.
We tried making the strategy more aggressive and adding more moving parts; the tests said the extra risk was not worth it.
How to read this
Method, assumptions & limits
The arithmetic is deliberately simple. Every result here comes with its caveats attached.
Sized by confidence
A confidence score of 1–2 gets $25 in the model, 3 gets $50, 4 gets $75, 5 gets $100 — set on the day we made the pick. Confidence is about how good the payoff looks if we’re right, not how likely we are to be right.
Two SPY comparisons
The first view puts the same confidence-sized dollars into SPY on the same days, so you can see if the strategy beats just buying the index. The second gives every pick and SPY a flat $100 each, so you can judge the picks themselves, without the sizing.
Plans change only through a dated event
The original plan is never edited or deleted. The public signal log adds a structured entry for each revision or close without exposing private reasoning. Once a pick closes, its paper cash sits still while the SPY comparison keeps running.
Closing prices, never real-time
Right now this page runs on one saved set of prices as a stand-in. Once it’s fully live, it’ll pull fresh closing prices every trading day. New picks join the public record only after their first verified close; approved friends may see an awaiting-first-close record earlier.
What is hidden or excluded
Ticker, company, The idea, Why now and free-text exit reasoning require access. Paper results, confidence, sizing, mechanical exit levels and the identity-masked signal log remain public. Broker details, real balances and execution never appear here.
What this cannot prove
A short, selected paper sample cannot establish that AI investing works. It can show what these experiments did and whether the record remained honest.