Writer

QED finishes the argument.

QED 1 · qed-2026b

From notes to prose

The author's notes

For an article titled

The passage, as returned

Fig. 1. QED 1 writing a passage from an author's notes. The numbers, citations and formulas it had to carry over are underlined in both. Recorded output, unedited. Write from your own notes →

1What it does

QED turns an author's notes into academic prose, written in that author's voice.

It has three tasks. Compose writes a passage from notes, either on its own or as the next section of a longer document. Polish edits a draft lightly. Revoice rewrites a draft in full while keeping its content.

The voice is set by example: two or three passages of your own writing. Without them, QED writes in a neutral academic register. It does not imitate other authors.

QED reads and writes Markdown, and carries mathematics ( $…$ ) and citations ([@key]) through unchanged. A companion call, compare, verifies that every number, citation and formula in your notes appears in the result.

2Results

We compared QED 1 with the base model it was trained from, on documents held out from training.

Table 1. QED 1 and its base model on 32 held-out documents and 40 passages by unseen authors.
Measure	Base model	QED 1	Better is
Orders ideas as the author did, from shuffled notes Rank correlation with the author's order	0.36	0.53	higher
Wording copied from the notes Excess over human authors	0.39	0.06	lower
Mathematics, citations and emphasis preserved in a rewrite	48%	93%	higher
Documents with unrequested headings, lists or rules	23%	2%	lower
Similarity to an unseen author's style, from three example passages	1.38	1.38	higher
Claims added, dropped or changed per document	7.0	3.8	lower

The full table, with confidence intervals, is in Research, report 2.

3Limits

Check the output: QED writes only from what you give it, but it can still drop a detail or add an aside. Run compare and read the result against your notes.
Editing: Polish makes light edits and revoice a fuller rewrite, most consistently with passages in your own voice. Neither reliably corrects grammar or spelling.
Long drafts: Polish and revoice work best on passages of up to about 400 words. Longer drafts are edited lightly, and less towards the end.
Size and language: Up to 2,000 words per call. English only.

Text written with QED is AI-assisted. Most journals ask authors to say so, and the API can draft the disclosure for you.

4Use it

Listing 1. shell

curl https://api.etal-labs.com/v1/write \
  -H "Authorization: Bearer $ETAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qed",
    "task": "compose",
    "title": "Sleep deprivation and memory consolidation",
    "brief": "- 64 adolescents and 60 adults, crossover design\n- next-morning recall fell by 21% vs 12%\n- no difference at 48 h (p = 0.21)",
    "voice": {"exemplars": ["A paragraph you wrote.", "Another one."]},
    "target_words": 120
  }'

See the API reference, or try it in the playground.

Table 2. The QED family.
Model	API string	Status
QED Minor	n/a	forthcoming
QED 1	`qed-2026b`, alias `qed`	Available. Up to 2,000 words per call, about 3 s for a 200-word passage.
QED Major	n/a	forthcoming

Get an API key →Responsible use