The briefs on this site are written by software. This page says what the software is allowed to do, what it is not, and who answers for it.
What a brief is
Each brief covers one scientific paper on quantum computing whose full text is free to read. It says what the paper shows, whom that concerns, and what is still missing before anyone can use it. It carries a verdict, a verbatim quotation from the paper with the place it was taken from, and a link to the paper.
How a paper is judged
A language model reads the full text and records a finding: what was done, on hardware or on paper, with how many qubits, compared against what, and how far the result is from a task someone has. That finding is then held against the paper by code and by a second model, in six checks:
- Every quotation stands in the paper word for word.
- Every number the model wrote stands in the paper with that value. Nothing rounded, nothing computed.
- Every number in the finding stands in one of the quotations given for it.
- Every name of a device, method or organisation stands in the paper.
- A checker that sees only the quotations and the sentences confirms, sentence by sentence, that the quotations carry them.
- A second model reads the paper independently and arrives at the same verdict.
Checks one to four are code. A paper that fails any of the six does not become a brief without a person looking at it.
How a brief is written
The model writes the explanation. It does not write the parts that are facts: the verdict, the quotation, the market estimate with its source, the roadmap entries, the list of affiliations and the note at the end of each brief are inserted by code from the checked finding and from two tables kept by hand. The finished text is then checked again against the finding, because a model that rewrites a sentence can change a number while doing so. A second model reviews the brief; below a set score it is not published.
The verdicts
- Real today. Shown on hardware, compared against the best classical method, usable now.
- Milestone. A hardware result with no application of its own that reaches a quantity on which estimates for useful machines are built.
- Plausible by a year. The paper states in numbers what the task needs, and at least two independent sources have announced a machine of that size. The year is the latest of theirs.
- One vendor’s roadmap, with a year. Only one vendor has announced a machine of that size. The year is that vendor’s announcement, not a judgement by this site.
- Announced in parts. Each resource the task needs has been announced, but by different sources; no one has announced a single machine that does all of it. No year.
- Beyond current roadmaps. At least one resource the task needs exceeds every announcement on file. No year.
A year never comes from the model. It comes from a table of published announcements, each with its address, the date it was retrieved and the passage it was taken from. An announcement is not a delivered machine, and the briefs say so.
What the checks do not do
- They do not check whether the paper itself is right. A brief reports what a paper shows, with its review status.
- They hold numbers, names and quotations. Whether an explanation reads the paper correctly remains a judgement, made by models, and two models can share a misreading.
- A statement that something is absent from a paper has no quotation to stand on.
Market figures and companies
Market figures are quoted with publisher, date and page, and marked as estimates. They are never computed for a single paper. Briefs name the institutions a paper lists and say which technical approach a result strengthens. They do not rank companies and are not investment advice.
Corrections
A brief that turns out to be wrong is corrected in place, with a dated note that states the old and the new verdict. Nothing is changed silently and nothing is deleted. If a figure looks wrong to you, write: name the brief and the figure.
Who answers for it
Georg Singer sets the rules described here and is responsible for what appears on this site. He does not read each brief before it is published; he answers for the rules that decide whether it may be.