Every ORBIT tool called for real on 7 September 2026: how long it took, what it costs, and what the answer was checked against. 3 of the 11 rows carry an accuracy figure and 8 deliberately do not — where there is no correct answer to score against, or nobody has built the corpus yet, the row says so instead of showing a number.
1 credit · 2 for a custom template. 2,893 ms, single run.
Accuracy: Every advertised spec matches the engine. Measured against: The publisher's own template. 151 automated checks tie every margin, font, size and spacing this site advertises to what the formatter actually writes into the .docx, on every build.
1 credit · 2 against a named journal. 2,588 ms, single run.
No accuracy figure is published. Needs a manuscript with a known number of planted violations, scored on how many are found and how many are invented. The corpus does not exist yet.
1 credit · 2 for a full manuscript. 6,720 ms, single run.
Accuracy: 6 of 6 on the hard-case corpus. Measured against: Crossref and Retraction Watch. Three real DOIs resolved, two fabricated ones refused, and a known retraction flagged as retracted. Six cases, chosen one per class — evidence the classes are separated, not a precision rate.
Free. 7,016 ms, single run.
No accuracy figure is published. The reference is excellent — every published paper knows which journal it appeared in. The measurement is not: the tool allows one account twenty suggestions an hour, so a corpus worth publishing takes a day rather than a session. A first attempt scored three papers and was refused the rest; three papers is not a number.
Free. 1,139 ms, median of 5 runs.
No accuracy figure is published. Scoring search means judging which results were relevant, one by one, by hand. That work is real and has not been done.
Free. 807 ms, median of 5 runs.
No accuracy figure is published. Same as Literature Discovery, with less to compare against: there is no equivalent of a citation list to check dataset results for.
1 credit. 2,236 ms, single run.
Accuracy: Nothing invented, checked on every build. Measured against: Your manuscript. Outline mode asks no model anything, so every sentence on a slide must already exist in your paper — and 50 automated checks assert exactly that, including that the order is never changed.
3 credits. Not timed.
No accuracy figure is published. Not yet timed — at three credits a run it was left out of the first pass, and we would rather show nothing than reuse the Outline figure. On quality: this mode writes a spoken script, there is no correct script, and we are not going to invent a scale and score ourselves on it.
1 credit. 3,585 ms, single run.
No accuracy figure is published. There is no correct cover letter. What can be checked is that it leaves blanks rather than inventing a name, and it does.
Free. Not timed.
No accuracy figure is published. This one is a page, not a job you wait for — there is nothing to time. What would be worth measuring is whether the deadlines are right and current, and that has not been checked against the conferences' own sites.
1 credit per table. 5,589 ms, single run.
No accuracy figure is published. A chart choice can be defended for clear cases and argued about at the margins. Writing that rubric honestly is work that has not been done.
This page compares ORBIT to nothing but itself. A fair comparison means running the same papers through someone else's product, and we have not done that — so no competitor is named here and no competitor's numbers are quoted.