Review Abstract

/review-abstract

Diagnose an abstract and hand back specific, actionable rewrites. This is a reviewer's pass over the writing, not over the science: you are not judging whether the results are good, only whether the text delivers them.

Before anything else

1. Load the style rules. Invoke paper-writing-style and read references/human-voice.md. If it is not installed, work from the condensed rules in create-abstract.
2. Lock the language. Use the language of the user's first message for the whole run and do not switch. Any rewritten text you produce is in English.
3. Get the abstract. Read it from the path given, from the manuscript in the working directory, or from what the user pasted. If there is a full manuscript available, read it too: it is the only way to tell whether the abstract's claims match the paper.

Step 1: categorization

This is the exercise the course builds the whole method on. Go sentence by sentence and label each with the part it implements:

One sentence may carry two parts; label both. Present the categorization as a table with each sentence and its label, so the researcher sees the structure of their own text.

Step 2: structural verdict

Report, in this order:

1. Missing mandatory parts. Purpose, results and conclusions are mandatory. A missing one is the most serious defect available and goes first.
2. Missing recommended parts. Context and gap. Judge against the target journal: a specialist venue may legitimately drop them, a broad one may not. Ask which journal if it is not obvious and it changes the verdict.
3. Order. The parts should run context, gap, purpose, methods, results, conclusions. Note any part that appears out of sequence and whether it actually hurts.
4. Unlabelled sentences. Anything labelled NONE is a candidate for deletion. Say what, if anything, would be lost.
5. Proportion. Flag an abstract that spends half its words on context and two sentences on results, which is the most common imbalance.

Do not report a review article or short communication as defective for missing parts. Those legitimately follow a different structure; check what kind of document this is first.

Step 3: the gap and the results

These two carry the abstract, so give each its own judgement.

The gap. Is it specific enough that the purpose answers it? A gap phrased as "more research is needed" is not a gap. Look for what is unexplained, contested, unmeasured, or too costly in the current state of the art, and check that the purpose lands on exactly that. A purpose that does not answer the stated gap is a structural fault, not a wording one.

The results. Are there real numbers? An abstract that says performance "was improved" or that a material "showed good stability" without values wastes its most important sentences. Quote what is there and name what is absent.

Step 4: style pass

Check against the house style and report each hit with the sentence and a rewrite:

• Em dashes. Any ---, or -- outside a numeric range, must go.
• Specificity. Any sentence that could appear unchanged in a different paper in the field.
• Sentence length. Any sentence long enough that its opening is forgotten by its end.
• Nominalization. an improvement was observed for improved, made the measurement of for measured.
• Redundancy. completely eliminated, introduce a new, and anything the editor already assumes: that results were analysed, that samples were prepared carefully.
• Words to use sparingly. preliminary, careful, novel, successfully.
• Plain English. utilization, in order to, the majority of, due to the fact that.
• Empty openers. it is important to note that, it is relevant to mention here that.
• AI tells. delve, harness, leverage, underscore, pave the way, plays a crucial role, it is not just X, it is Y.
• Stacked passive voice.
• Ambiguity. as or since where because is meant; any it, this, or they with more than one possible referent.
• Abbreviations. Defined at first use, and not several crowded into one short sentence.
• Tense. Past for what was done, present for what the results indicate.

Do not flag however, moreover, in contrast, in this paper, here we report, or single deliberate hedges such as may and suggests. Those are correct scientific register. Flagging them is a false positive and makes the review untrustworthy.

Step 5: consistency with the paper

Only if the full manuscript is available. Check that:

• Every number in the abstract appears identically in the results.
• The conclusion claimed is the conclusion the paper actually supports.
• The gap promised in the abstract is the gap the introduction develops.
• Nothing important in the paper is missing from the abstract.

Report any divergence as a discrepancy, naming both values or claims. Do not silently prefer one; the researcher decides which is right.

Step 6: deliver

Give the researcher, in this order:

1. The categorization table.
2. The structural verdict, most serious first.
3. The style hits, each with sentence and rewrite.
4. Any discrepancies against the manuscript.
5. One rewritten abstract as a LaTeX fragment with structure labels, applying everything above, with % TODO: where a missing value must come from them.

Then ask whether to apply it. Never edit their files unasked.

Hard rules

• Never invent a number or citation to fill a gap you found. Name what is missing.
• Never emit an em dash.
• Separate the writing verdict from the science. If a result looks scientifically weak, that is outside this skill's remit; say so briefly and move on.
• Report honestly. If the abstract is already good, say that plainly and keep the review short rather than manufacturing findings.