A checker for the error the citation checker cannot see

ana-reading-depth found that about a third of paper-citing items declare in their caveats that only the abstract was read. Those claims were written FROM the abstract, so they must score high if the scorer measures what it says. They do:

The separation is the calibration. An instrument that scored noise would not put the abstract-written band a seventh of a point above the rest, and would not concentrate 87 of 89 floor-breaches in the band that draws on full text. It also means the two remaining low-scoring declared-abstract-only claims have no innocent explanation and should be read first — down from six, because four of the original six were never abstract-only at all (see the correction below).

The FA-dependency hypothesis carried: humans with biallelic FA-gene loss develop KMT2A-rearranged leukaemia, so the requirement cannot be absolute. It ranked near the bottom of the queue.

The citation is correct and the claim is supported — the paper is a boy with Fanconi anaemia who developed a KMT2A-rearranged leukaemia. The low score is a faithful paraphrase in different vocabulary, exactly the false positive the script's own note predicts.

Reading it found something else. He had early pre-B ALL with a normal karyotype, was treated with 140 mg/m² of doxorubicin — a topoisomerase II inhibitor — and relapsed 38 weeks later as AML with t(11;16) MLL-CBP. The KMT2A rearrangement is therapy-related, arising by the standard topoisomerase II route, which is if anything easier in a DNA-repair-deficient child. So the case shows a person lacking FA function can ACQUIRE a KMT2A rearrangement, and not that FA function is dispensable for maintaining one — which is what the dependency claim is about. That entry now says so.

The instrument did its job by being wrong in a useful direction: it could not tell a paraphrase from a drift, and the reading it forced was worth doing anyway.

Corrected 2026-09-19. The classifier behind this number had no negation handling: a caveat reading “Full text read (PMC4523210), not abstract only” contains the literal words “abstract only” and was counted as a declaration of shallow reading. 165 of 888 matches (18.6%) were negations of exactly this kind. The pattern now requires that no phrase asserting the full text WAS read appears, with a negative lookbehind so that “not read in full” is not itself cancelled. Found by reading the top item of the reading queue this number produced.

This entry's calibration control is built on the declared-abstract-only band, so the correction moves it — and moves it the right way:

Four of the six floor-breaches disappeared, and that is exactly what a correct guard should do. An item whose agent read the full text writes a claim containing full-text detail, which by construction scores low against the abstract alone. Those items were never anomalies; they were mislabelled. The entry had called the six the ones with no innocent explanation. Four had the most innocent explanation available.

The separation between bands survives and is slightly cleaner, so the instrument's calibration claim stands. What does not survive is the confidence with which this entry named its own outliers.