---
name: quanterlab-paper
description: The structure of a research paper produced from QuanterLab, and the parts specific to it: the plain-language account of what is bought and when before anything else, the declared hypothesis and family size in the paper, limitations stated, and what is published alongside so the result can be re-run. Use when writing, revising or reviewing a study write-up for quanterlab.com, after the study is compiled. Triggers on requests to write the paper, draft the abstract, prepare the write-up, or turn a walk into a report.
---

# QuanterLab paper

This skill lays out the bones of the paper. The content decisions are made before it: an audit decides what the paper says, the house voice strips the machine register, and the author's own voice comes last. This skill does not compete with those; it makes sure the parts a platform paper needs are present and in order.

Every number in the paper comes from the compiled study and its logged windows, read through the platform's tools. Nothing is computed by hand and nothing is remembered from a run.

## The order of the paper

1. **Title.** The answer with a number in it, over how many windows. Not the question. Under 160 characters.
2. **Abstract.** What was tested, how many windows, the headline number against the benchmark, the one caveat that matters, the trial count. Under 900 characters.
3. **What is bought, when, and for how long.** Before methodology, before any number: a plain-language account of the action. Which names enter, at which price and session, how they are weighted, when the book is reselected, how long a position is held, what closes it. One paragraph per action type, and a figure for each action type: an entry rule gets a figure that shows the entry, a hold gets a figure that shows the hold, a switch gets a figure that shows the switch. This is the section most likely to be skipped and the one the reader needs first.
4. **Methodology.** Universe and membership rule, selection, weighting, rebalance, horizon, benchmark and its construction, cost model, the walk (first anchor, step, windows, in-sample years), and every custom primitive by name with what it does in one sentence and the fact that its source is sealed with the registration.
5. **The declared hypothesis and the family.** The claim as registered before the first window, quoted. The number of registrations. The family the paper answers to and its size, so the deflated Sharpe on the page reads against the right count. A study without a declared family says "unrecorded" in this section, in those words.
6. **Findings.** Every window, wins and losses alike, book against benchmark; excluded windows named with their reason; the pooled numbers. Tables for grids; prose for what the table cannot say.
7. **Limitations.** Stated, not buried: no control arm, costs not modelled, current-member universes, few windows, fixed parameters, whichever apply. Two to five sentences, each one a fact about the design.
8. **Discussion.** What the result means and what it does not show, once, briefly, technically. What a reader can check. The next test, named.
9. **Published alongside.** What the reader needs to re-run the result: the registration records with their circuits, the sealed custom primitive sources, the anchors, the plan, the benchmark construction. On quanterlab.com the study page carries these; the paper says where.
10. **References.** Only work cited for what it showed.

## Rules that hold in every section

- The registered claim appears in the paper in its own words, and the result is read against it. A paper whose finding contradicts its registered claim says so.
- A losing window is reported with the same care as a winning one.
- No em dashes. No self-praise. No sentence that carries only rhythm: if it has no number, mechanism or caveat, it is deleted.
- Caveats are stated once, in the limitations, and not repeated as hedges in every paragraph.
- The benchmark is the same in every window and named once.
- A figure shows one thing and its caption says what.

## The handoff

Write the bones with this skill. Then: the audit decides whether the paper has a finding and what it is; the house voice removes what reads as machine-written; the author voice makes it sound like the person whose name is on it. The paper goes to the platform as a draft, and the platform's review gate decides what publishes.