How the firm tests a claim
The method is the product. It is the same whether the claim comes from a vendor, a forum, a client or the firm itself, and every published study says how closely it followed it.
Write it down first
Before any result exists, the study fixes the claim in a form that can be coded, the exact rule, the data, the costs, the sample split and the bar a result has to clear. That document is the pre-registration. From October 2026 the firm hashes each pre-registration before the run and publishes the hash with the study, so the order can be proved. Only one earlier study, FPR-26-15, has one.
Test on the tape, net of costs
Studies run on CME futures minute bars from 2010 with contract rolls matched, plus a point-in-time archive of releases, central-bank events and positioning, so no test sees information it would not have had. Every fill pays commission and slippage at the rates each study states; some studies also charge stop entries extra around news. Vendor data stays in the lab; studies publish results, not data files.
Split the sample and use controls
Most studies judge a rule on an in-sample period and confirm it on a later out-of-sample period. Where it fits, the test runs beside controls built to look like the real thing: a coin-flip version of the same trade, placebo dates, shuffled event days. A result has to beat luck, not just zero.
Count every test
Each study counts the arms it ran and sets its bar for that many tries (for example, one-sided p below 0.0167 for three rules, or t of at least 3 for a family). The library has run 950 arms so far; a handful passing by chance is expected, and the firm says so next to anything that passed.
Publish the verdict, with the files
Every study behind a published claim is in the library with its pre-registration status, its timestamps, the written result and the code to rerun it, including the ones that failed and the ones that were never pre-registered. Other studies the firm has run, such as cross-instrument, jobless-claims and order-book tests, back no published claim and are not in the library yet. The verdict words are fixed (below), and the running record of any trading is never published.
Correct in public
If you rerun a study and get a different result, send it. The firm checks it and publishes a dated correction on the study page, whichever way it goes.
The verdict words
Every claim in the library carries one of these, and nothing in between.
| Validated | Passed every frozen test, in sample and out of sample, after costs. No published claim holds this label today. |
| In forward test | Passed in the sample where it was found and is now graded only on new data under a frozen rule. It is not an edge until the forward test passes. |
| Refuted | Failed the frozen pass mark after costs. The study says by how much. |
| Inconclusive | Too few trades, or results that point both ways. Not evidence either way. |
| Weak, post hoc | Found after the fact in a screen of many variants, with a statistic below the firm's bar. Reported so nobody mistakes it for a finding. |
| Described, not tested | A descriptive table with no test attached. |
Limits of the record
No outside replication yet
The studies are designed and run by the firm's founder, Dr. Alex Djahankhah. No study in the library has yet been replicated by anyone outside the firm. The files are published so that someone can.
Not every study met the standard
Two studies were never pre-registered and four only partly; seven pre-registrations carry a written freeze time that disagrees with the file system. Each study page says which, and how.
Not everything is in the library
The library holds the studies behind the firm's published claims. Other internal studies are not published yet, and a study commissioned by a client belongs to the client and is published only with its permission.
What the firm does not do
No trade signals, no managed money, no resale of exchange data and no paid placement. A finding is never sized into a recommendation.