Export a research paper
A typeset body, its figures, and a disclosure box built on the Arnott–Harvey–Markowitz protocol. It closes on the hypothetical-performance disclaimer that 17 CFR 4.41 calls for. The Paper Editor is available with Atelier and Portfolio. Starter has no access and no limited edition.
Edit the rendered document
Open a page in text-edit mode to inspect and revise the research document before saving or exporting it.
Open full sizeTrial counts, split design, costs, and what was not corrected
The box states how many configurations were searched in this run, the cross-validation design, the costs that were applied, and the fact that the universe came from you with no survivorship correction. A reader can dispute each of those, which is the point of printing them.
When the run was produced as a batch, an extra item carries the size of that batch. A single paper out of a batch, presented alone, is the shape a data-snooping problem takes.
The page above comes from a paper exported after a real run on 28 August 2026; the dialog shows how its sections were chosen. The layout is the same for a NO-GO and for a pass: a negative outcome is typeset like a positive one.
Open full sizeThe box is re-inserted, so a layout cannot drop it
You compose the paper by choosing its sections. A layout that omits the disclosure box gets it back, placed after the conclusion, and the header section is put back at the top the same way. There is no code path that produces a paper without them.
The call returns the layout that was actually rendered, so what came out is readable from the result and does not have to be inferred from what was asked for.
The verdict is computed by the run, and the paper reports it
The paper does not judge the strategy. It prints the verdict the run computed and puts it in a sentence. Where no gate was wired, the default rule is an out-of-sample result that holds together with a selection score under its threshold, that score being a Romano-Wolf stepwise p-value adjusted over the unfiltered population.
The set of outcomes is wider than a pass and a fail. A run that produced no trade, a preview, and a run whose verdict could not be established each have their own value, and each prints under the same layout as a pass. A negative outcome is typeset like a positive one, at the same length.
A forbidden lexicon, and prose that is checked against the numbers
A lexicon of promotional words is refused in the generated text. Generated prose is opt-in, and a sentence that states a figure the run did not produce is dropped — with the reason for dropping it printed in the paper, where a reader sees it.
The same input produces the same document, byte for byte. A section id that does not exist raises, so a silent drop cannot happen.
The printed list is a bibliography of the protocol
The references in the PDF cite the literature the protocol comes from, and they are printed when the references section is part of your layout. The methods the software implements are listed on the research page.
Two entries in the File menu
Compose the paper section by section, or export it directly. Both are in the File menu, and both are also reachable from the MCP server.