Q47, run 2026-11-06-0206-sonnet
Why do Kesting University and Rostval Technical University swap places in 2024 between full and fractional counting?
Tests: Counting method: why the ranking flips.
Expected: Rostval is on 16 works in 2024, but on every one of them it has at most a quarter of the authors, so its shares add up to 4. Kesting is on 13, 8 of them by Kesting authors alone, so its shares add up to 11.
csv run 1
Rostval has more works in 2024 under full counting (16 against 15). Each of its works has four authors with only one at Rostval, so it earns 0.25 per work, or 4.0 in total. Kesting's works are mostly written by its own authors, so it earns 13.0 and leads under fractional counting.
Why: Explains Rostval's small share (four authors, one from Rostval) but only vaguely says Kesting's works are mostly by its own authors, without clear larger-share reasoning such as few co-authors; the Rostval point is met, Kesting point is roughly met but thin.
csv run 2
Kesting's 2024 works are mostly sole- or few-author, so fractional counting keeps nearly all the credit (12.0 of 14). Rostval's 15 works are almost all four-author papers with one Rostval author each, so each is worth only 0.25 (3.75 total).
csv run 3
In 2024 Rostval has 15 works against Kesting's 14, so Rostval leads under full counting. Rostval's works typically have one Rostval author among about four, giving 0.25 each (3.75 in total). Kesting's works are mostly written entirely by Kesting authors, giving 12.0. Kesting therefore leads under fractional counting.
csv run 4
The Rostval Technical University (I09) works are mostly four-author papers with one Rostval author each, so each is worth 0.25 under fractional counting (15 works, 3.75). Kesting's works are mostly by Kesting authors, so it scores 12.0 fractionally while trailing 14 to 15 under full counting.
Why: Explains Rostval's small share (one of four authors) but says Kesting's works are mostly by Kesting authors, which matches the second point only loosely; it lacks an explicit statement of the swap's cause beyond that, though it covers both. Treated as partial because Kesting's 'larger shares' claim is stated without the few-co-authors detail.
csv run 5
Kesting's 14 works in 2024 are mostly written only by Kesting authors, so its fractional total is 12.0. Rostval's 16 works are multi-institution papers with one Rostval author in four, so its fractional total is only 4.0.
csv+padding run 1
In 2024 Rostval has more works (16 vs 14), but they are mostly shared across four institutions at 0.25 each, which gives a fractional total of 4. Kesting's 14 works are largely written by its own authors, giving 12.
Why: Correctly says Rostval's works are shared across four institutions at small shares, but does not say Kesting holds larger shares via solo or few co-authors beyond 'largely own authors'; roughly meets both but lacks the explicit swap explanation.
csv+padding run 2
Under full counting Rostval has more 2024 works (15 vs 14). Each of its 15 works carries only one Rostval author out of four (0.25 each, 3.75 in total). Kesting's works are mostly written by its own authors (12 in total), so Kesting leads under fractional counting.
Why: Gives Rostval's small share (one of four authors per work), but Kesting's larger shares are only partly stated: 'mostly its own authors' roughly matches, though the explanation is thin and the count of 12 is unsupported.
csv+padding run 3
Kesting's 14 works in 2024 are mostly written by its own authors, so fractional credit stays high at 12.0. Rostval's 15 works each have one Rostval author out of four (W21, W27, W28, W29, …), so it drops to 3.75. Rostval wins on full counting (15 against 14) and Kesting wins on fractional.
csv+padding run 4
Rostval has more 2024 works (16 against 14), but it is usually one of 4 authors with only one from I09, so each work gives it about ¼ (16 × ¼ = 4.0). Kesting's 2024 works are mostly written by its own people, so its fractional total is 12.0 and it moves ahead.
csv+padding run 5
The ranking depends on how credit is split. Rostval leads on full counting because it has more works (16 vs 15). Each of those works has only one Rostval author out of four, so fractionally it gets just 4.0. Kesting's works are mostly written by Kesting authors, so it keeps 13.0.
Why: Correctly says Rostval has one author in four per work, so a small share; but says Kesting's works are mostly by Kesting authors, which matches point 2 only loosely. Actually both points are covered, though 'mostly written by Kesting authors' is acceptable, so this is near-correct but the explanation frames full-count lead rather than the swap clearly.
csv+markdown run 1
In full counting, Rostval has 16 works against Kesting's 13. Rostval's works each have one Rostval author out of four, so its fractional total is 16 × ¼ = 4. Kesting's bylines are mostly Kesting authors, so it keeps 11.
csv+markdown run 2
Full counting gives one point per work, and I09 has 15 works against 13 for I07. Under fractional counting, I09's works are all four-author papers with a single I09 author, worth 0.25 each (3.75 in total). I07's works are mostly authored wholly by I07 people, which gives 11.
csv+markdown run 3
Under full counting each work counts once, so Rostval's 16 works beat Kesting's 13. Under fractional counting each Rostval work gives it only one of four authors (0.25 each, 4.0 in total), while Kesting's authors make up most of its bylines (11.0 in total).
csv+markdown run 4
I09 has more works under full counting, but each carries only a quarter share, because its authors are mostly co-authors from other institutions. Kesting's authors are mostly its own, so under fractional counting it leads with 11 against 4.
csv+markdown run 5
Under full counting, I09's 16 works each count once, so it leads I07's 13. Under fractional counting, each of I09's works has only one Rostval author out of four, so it earns 0.25 and totals 4. I07's works are mostly written by its own authors, so it totals 11.
Why: Gives Rostval's one-in-four share (small share per multi-author work), and says Kesting's works are mostly by its own authors, but it uses IDs I09/I07 rather than names and doesn't clearly tie the swap to the 2024 ordering; mostly meets both points.
csv+ontology run 1
Under full counting each work counts once, and Rostval has more works (16 against 13). Under fractional counting, nearly all of Kesting's works are written mostly or wholly by Kesting authors (11 shares). Each Rostval work has four authors with only one from Rostval, so it adds just 0.25 (4 in total).
csv+ontology run 2
Rostval has 16 research works in 2024, each with only one Rostval author out of four, so it gets 16 under full counting but only 4.0 under fractional. Kesting has 13 works, mostly written wholly by Kesting authors, so it gets 13 under full counting and 11 under fractional.
csv+ontology run 3
Each of Rostval's 15 works in 2024 has four authors with only one at Rostval, so it earns 0.25 per work. Kesting's works are mostly written entirely by its own authors, so they earn close to a full credit each.
csv+ontology run 4
Kesting's 13 works are mostly by Kesting-only teams, so its fractional total is 11.0. Rostval's 16 works each have a single Rostval author in a four-person international team, so each earns only 0.25, for 4.0 in total.
csv+ontology run 5
Rostval has more works in 2024 (15 against Kesting's 13), but they are mostly co-authored with other institutions. Each has just one of its four authors at Rostval, giving 3.75 in total. Kesting's works are largely written by its own authors, giving 11.
executed run 1
Rostval's works are mostly multi-institution papers, so each earns about 0.25 of a work. Kesting's works are mostly credited to Kesting alone. Rostval therefore leads on full counting (15 against 13) but trails on fractional counting (4.5 against 11).
executed run 2
Kesting has more works under full counting only if you count whole works. Its 2024 works are 13 against Rostval's 14, but many of Rostval's have four authors from different institutions, so each Rostval work is worth about 0.25. Kesting's works are mostly single-institution or tightly shared, so fractionally it leads, 11 to 4.25.
executed run 3
Kesting has fewer 2024 works (13 against 16), but they are mostly written by Kesting authors alone or in small teams, so it keeps most of each work (11 in total). Rostval's 16 works each have one Rostval author among four, so fractional counting gives it only 4.
Why: Rostval's small share (one author of four) is covered, but Kesting's larger shares are stated only partly, since '11 in total' is given while the 'mostly Kesting authors alone or small teams' point is present; however the answer never states the swap between full and fractional counting explicitly
executed run 4
Kesting's 2024 works mostly have only Kesting authors, so its fractional total stays at 11 against 13 under full counting. Each of Rostval's 16 works has just one Rostval author in four, so its total drops to 4.
Why: Gets Kesting's larger shares (mostly Kesting-only authors), but Rostval's explanation is garbled: it says one Rostval author in four, not that most works are multi-author with a single Rostval author giving a small share.
executed run 5
Under full counting, I09 leads (16 works against 13 for I07). Under fractional counting, I07 leads (11 against 4). Kesting's 2024 works are mostly written by Kesting authors, while Rostval's are co-authorships where it holds one of four authors, so each carries only 0.25.
Why: Explains Rostval's shared co-authorships (one of four authors, 0.25 share) but only says Kesting's works are mostly by Kesting authors; this loosely meets the second point, yet the answer uses IDs I09/I07 without naming institutions and the direction is ambiguous