Find out if a strategy is real before you trust it.
Describe an idea or bring a paper. We find the research behind it, replicate it on professional-grade historical data and put it through strict method tests. You get a verdict, the reasons, and a fact sheet you can compare.
Earnings-revision drift, large caps
Replicated from a published paper, spec frozen before the run
- Paper replicationready
- Method testscomplete
- Comparable fact sheetavailable
Reason: the result cannot be distinguished from chance.
Every strategy looks good in a backtest. What matters is what survives real costs, real data and honest counting.
A backtest is a claim, not evidence.
Papers report results on cleaned data, without trading costs, after trying more variants than they mention. Re-run them honestly and the effect often shrinks or disappears.
Checking this properly takes weeks per paper: the data, the code, the statistics. Most people skip it, and the gap between the paper and reality ends up in their P&L.
How it works
Scroll to follow one idea through the pipeline.
Idea
Describe a strategy in plain words, or paste a paper.
Evidence
We search the literature and show which papers support or contradict it.
Replication
The paper becomes a frozen spec and code, run on professional-grade historical data.
Verdict
Each strategy goes through our method tests and gets a clear verdict, with the reasons.
Fact sheet
A comparable profile of how the strategy behaves.
What you can do
-
Start from an idea, get the evidence
Describe a strategy in plain words, or bring a paper. The system searches the academic literature, picks the papers that actually test the idea and shows whether the evidence supports or contradicts it.
It checks that the numbers a paper reports appear in the paper itself, turns it into a frozen spec and code, and tests it on professional-grade historical data. You can also ask the paper questions directly.
-
This week in research
A weekly digest of new research papers in finance, each with a short summary. Papers that propose a testable stock strategy are listed first.
One click turns a paper into a test.
-
Know why it passed or failed
Every test returns a verdict with a plain-language reason, plus a fact sheet you can compare across strategies: Sharpe, information ratio, alpha and beta against the market, correlation, maximum drawdown, turnover, breakeven cost, upside and downside capture, behavior in crises (2008, COVID 2020, the 2022 rate shock), concentration in the top names and parameter sensitivity.
You set the thresholds on these characteristics, not us.
Then we make it stronger, without fooling ourselves.
A verdict is not the end of the work. Once the results are in, AI studies the strategy to find its weakest points and searches for ways to improve it, with methods built not to overfit the data.
Earnings-revision drift, large caps
Weakest points found in the results
- Most of the edge goes to trading costshigh
- Depends on a handful of namesmedium
- Sensitive to the rebalancing daymedium
Proposed change: trade less often and cap the weight of each name. It is tested as a new trial, against a higher bar.
Find the weak points
AI goes through the results and the fact sheet and points to where the strategy is weakest: the periods where it loses, the costs that eat its edge, the few names it leans on, the settings it is sensitive to.
Propose fixes you can follow
Each suggestion is a concrete change, explained in plain words and grounded in what the data shows, not in trying settings until one happens to look good.
Test them without overfitting
Every variant counts as a new trial and raises the bar for the next one. Variants are compared with validation that keeps past and future apart, and the final exam runs on a period no earlier step ever used.
Tuning a strategy until it looks good is the fastest way to fool yourself. Here every improvement has to earn its place against a stricter test.
Who it's for
Independent quants
Test the papers you read without building a data and statistics stack first.
Trading firms
Screen ideas fast and drop the ones that fail before they cost desk time.
Funds
Comparable, reproducible evidence for your research and due-diligence process.
Principles
No performance numbers here
This page shows how we judge strategies, not returns.
Public trial counting
Every trial raises the required standard, and the count will be public from launch.
Reproducible
Each result is tied to the code version and the data that produced it.
Research, not advice
We don't manage money and we don't give personalized advice.
Questions
Are you an investment adviser?
No. This is a research tool. We don't manage money, recommend trades or give personalized advice.
When will access be available?
Invitations go out in small waves, and we onboard each one personally.
What data do you use?
Professional-grade historical data. Each result is tied to the exact data and code that produced it.
What does "certified" mean?
That the strategy passed all of our method tests. It says nothing about future returns.
Will you show the strategies that fail?
Yes. A failed test is a result too, and we publish it with the reasons.
Still have questions? Write to us and we will get back to you.
contact@thirdpeg.appBefore you risk capital, check that the strategy exists.
Request early accessThe access form needs JavaScript. Turn it on and reload this page to request access.
Privacy notice
We collect what you enter in the access form: your email and, if you choose, your name, role, capital band, the idea you want tested and whether you're open to a call. We also record the page language, the site that referred you and campaign tags in the link, if any.
We use this only to manage early-access invitations and to contact you about them. We don't sell it or share it for advertising. It is stored with our database provider.
You can ask us to see, correct or delete your data by replying to any email we send you. This page loads no trackers, analytics or external fonts.