A solemn research paper about something that does not deserve one

The mark is not a disclaimer you can scroll past. It occupies four surfaces the app asserts at load, and losing any one of them fails a check rather than quietly weakening the app: a high-contrast banner above the paper, a repeating watermark behind it, an inline mark inside the document itself, and the browser tab title, so a bookmark or a shared tab carries it too. A fifth mark sits in this page's footer at all times, whether or not a paper has been written.

It survives leaving the page. Anything copied to the clipboard arrives with the parody notice above and below it. The Markdown export has the notice distributed through it rather than parked at the top, so a fragment cut out of the middle still carries one. Every object at every depth of the JSON export gains a parody: true flag. Every BibTeX entry carries it in note and annote, both of which a rendered bibliography prints, so it shows up in the formatted output and not only in the source. Every export path ends in an assertion that throws rather than returning an unmarked artefact. The print stylesheet keeps the watermark on paper.

Fabricated by construction. The reference apparatus is built in your browser before the model is called, from a closed pool of sixteen self-evidently invented journals — titles along the lines of a register of inconclusive work or a journal of results nobody expected — with invented surnames carried as bare initials and years that have not happened yet. There is no path by which a real citation can enter, because no real citation is ever available to draw from.

Citation integrity check. The finished paper is then scanned anyway, on the way out, for anything that could resolve to real scholarship: a DOI, a doi.org link, an arXiv id, a PMID or PMCID, an ISSN or ISBN, any bare URL, a clinical-trial registry id, a volume-and-page citation signature, or the name of a real publication. A run that contains one is blocked rather than shown with a warning.

The app will not acknowledge inside the document that the document is a parody. The paper plays it straight; the page around it does the disclosing. That division is deliberate — a spoof abstract that winks at the reader stops being a parody of the form.

You get a structured spoof abstract: a title, an invented byline with invented affiliations, then background, methods, results and conclusions; a measured quantities table of three to six numbers nobody would think to measure; and a limitations paragraph that concedes the wrong thing, sincerely.

Nine document registers — clinical trial report, bench science, field observation, theoretical paper, systematic review, case report, survey instrument, computational, economics — and ten disciplines, each carrying its own conventions and its own list of jokes not worth making in that field.

A twelve-axis divergence engine. Before the model is called, the browser draws a coordinate in a twelve-axis space — operation, design, sample, finding, concession, entry, hedge, apparatus, close, figure, declaration and what goes unexamined — choosing the point farthest from everything already on your screen. That is why the tenth paper on one subject is not the first one again.

Before. A free client-side subject scan reads what you typed, in your browser, with no account and no charge, and says what it makes of it. It is reconciled against the finished paper afterwards, so you can see whether the paper answered the subject you actually gave it.

After. Register is measured structurally on the way out rather than asserted: sentence length, hedge density, passive rate and citation density, each compared against the target that was actually sent to the model. Where the paper missed its brief, the checks panel says so.

Finished papers are saved to your SkillSafe account and mirrored in this browser, so a history exists on both sides. Any paper can be exported as Markdown, JSON or BibTeX, or printed.