One claim, end to end.
A detail aid for a placeholder GLP-1 brand, ORVEXA. The library holds the real trials and labels. This is what happens to one sentence between the writer typing it and MLR reading it.
The writer writes.
No tagging, no forms. A paragraph in the editor, the way it was always drafted.
In adults with obesity and without diabetes, once-weekly semaglutide 2.4 mg produced a mean weight loss of 14.9% at 68 weeks versus 2.4% with placebo. More than four in five participants lost at least 5% of their body weight, and half lost 15% or more.
Check claims finds the sentence.
Four statements in that paragraph need a reference. This is the second.
“More than four in five participants lost at least 5% of their body weight”
Near an approved claim · AC-017 · “Over 80% of participants lost at least 5% of their body weight” · verify wording before reuse
The span is located in the copy by exact text match, so it stays attached as the paragraph is edited. Because the brand already has an approved claim with the same meaning, the approved references are offered first. The library is still searched, so the writer sees the primary source too.
The library is graded against it.
Fifty passages retrieved, ten reranked, each graded 0 to 100 against the sentence exactly as worded. The two best are kept in view.
- 01Wilding et al., 2021 — STEP 196%N Engl J Med 2021;384:989 · p.989 · ¶3 · s2
“…achieved weight reductions of 5% or more (1047 participants [86.4%] vs. 182 [31.5%]), 10% or more (838 [69.1%] vs. 69 [12.0%]), and 15% or more (612 [50.5%] vs. 28 [4.9%]) at week 68…”
Same population (adults with overweight or obesity, no diabetes), same endpoint (≥5% loss at week 68), and 86.4% is more than four in five. Primary publication of the pivotal trial; a reviewer would accept it as the reference.
- 02WEGOVY prescribing information, 202484%FDA label · p.23 · table
“Percentage of patients achieving ≥5% body weight loss: 83.5% (WEGOVY) vs 31.1% (placebo)”
Substantiates the claim but from the label's analysis, which reports 83.5% rather than the paper's 86.4%. Still more than four in five; verify which figure the brand wants to stand behind before use.
Page, paragraph and sentence are computed from the stored character offsets of the passage, not asked of the model. The excerpt is copied verbatim. The writer ticks the STEP 1 paper as reference 1; the citation number appears in the copy.
View source shows the page.
The reviewer’s question is always the same: where exactly? The answer is the page, with the passage boxed.
Claim, excerpt, location, confidence and the grader’s reasoning sit beside the page image with the passage outlined in red.
The page prints as an exhibit: source, page, paragraph, claim, excerpt, reasoning and date. It is the artefact that goes into the submission when someone asks for the reference.
Export carries it all.
A Word document with numbered citations, a reference list, and an annotated appendix per claim.
| Claim | Reference | Excerpt | Location |
|---|---|---|---|
| More than four in five participants lost at least 5% of their body weight | 1. Wilding JPH, et al. N Engl J Med. 2021;384(11):989-1002. | “…weight reductions of 5% or more (1047 participants [86.4%] vs. 182 [31.5%])…” | p.989 ¶3 s2 |
In the copy the sentence reads “…at least 5% of their body weight1”. AMA superscripts or Vancouver brackets, per brand setting. Approved content blocks in the piece are frozen at the versions that went out.
That is the whole product, one sentence at a time.
The method, thresholds and data flow are on How it works.