INTERACTIVE COMPANION · EXPLAINER · SCIENTIFIC PRACTICE

The Paper Anatomy Bench

A live model accompanying “How to Read a Scientific Paper Without Pretending You Understood It”

Two benches. The first is a mock paper you can walk through section by section, with the two reading orders laid over it. The second takes one finding and slides it from the way a Results section would phrase it to the way an abstract would, scoring both ends on the same word lists the club's analysis script uses.

Every corpus number quoted here comes from that script: 67 open-access papers from Europe PMC and 298 sections of our own 27 articles. Nothing on this page is a guess.

Bench 1. The mock paper

Block heights are proportional to the median section length we measured. Click a block, or drag the step slider, and the panel on the right tells you what that room is for, what to look for, and what should make you suspicious. Switch the reading order and watch the numbers on the blocks rearrange themselves.

Step 5 · Results Starts on Results, because that is where the evidence is and it is the block that dwarfs everything else. Walk it down to 6 and then back to 1 and notice how little of the paper the abstract stands for. Reading order A · front to back Switch to B. The badges renumber and the abstract moves from first to last. Try stepping through B from 1 to 6 and stopping at step 3, which is where most real readers abandon a paper. Block heights proportional to length Set it to equal and the paper looks like the thing most people picture, six tidy parts of similar weight. Set it back and the Results block takes over. That difference is the whole reason the abstract is misleading.

median length: – hedges per 1,000: – report : interpret: – compression: –

–

Bench 2. The same finding, phrased five ways

One result, five drafts. Draft 0 is how a Results section would put it, with the sample size, the comparison and the numbers in place. Draft 4 is how the abstract of the same paper would put it. The drafts in between are the steps the sentence takes on the way forward through the manuscript.

Each draft is scored live by the same three instruments the analysis script uses: a graded claim lexicon of 81 forms weighted from −2 to +2, a 41-form reporting list, and a 40-form interpretation list. Watch which gauge moves.

Finding A · compound X and cognitive decline All three are constructed, and all three are modelled on claims the club has actually had to check. Run each one from 0 to 4 and the same shape appears every time. Draft 0 · Results phrasing The interesting move is from 1 to 2, where the last number leaves the sentence. After that nothing the reader can check remains. Also watch draft 3, where the hedges spike and the claim score goes negative before rebounding at 4. Abstract word budget 206 words 206 is the median abstract in our sample. Drag it down to 120 and the compression against a 3,148-word Results section goes past 26 to one. The budget is why the register has to change.

claim strength: – report : interpret: – hedges per 1,000: – checkable numbers: –

–

Corpus abstract mean: claim strength 10.33 per 1,000 words, report:interpret 0.75, hedges 5.16
Corpus Results mean: claim strength 10.00 per 1,000 words, report:interpret 7.58, hedges 4.16
Paired difference: +0.33 ± 2.12 per 1,000 words, t = 0.16 on 66 df, n = 67 papers
Checked by hand: 3,148 ÷ 206 = 15.28, shown as 15.3× · 10.33 − 10.00 = +0.33 · 0.33 ÷ 2.12 = 0.156, shown as 0.16

The claim gauge barely moves between the two ends. The ratio gauge falls off a cliff. That is the article's finding in one picture: an abstract does not shout at you, it changes what kind of sentence it is writing and takes the evidence with it.