Replace the 100-row task with a 5-row task and admit 11 more authors

The 100-row eligibility bar excluded eleven authors that clear 50 comfortably,
most of them by a hair: Euripides had 79 validation and 79 test rows, Aristotle
87, Lysias 97. A closed set that omits Euripides and Aristotle is hard to
defend when the only thing keeping them out is a threshold.

Both publications now define 1-, 5-, 10- and 50-row tasks. The bottleneck moves
to 50, the largest chunk, so every retained author still has at least one
complete chunk of every task size in validation and in test, and every smaller
size divides 50, which is what lets one retained row set fill all four tasks.

Sphragis grows from 17 to 28 authors and from 43,300 to 50,500 training rows,
gaining Aeschines, Antiphon, Appian, Aristotle, Euripides, Hesiod, Lucian,
Lysias, Menander, Paul and Sextus Empiricus. The source-held-out track grows
from 11 to 12. Sphragis Metre gains Tryphiodorus.

The cost is that the thinnest authors now have a single 50-row chunk in test,
so their per-author accuracy on the 50-row task is either 0% or 100%. The
smaller task sizes are unaffected.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

README.md

@@ -35,14 +35,14 @@ configs:
     path: data/sentence_50/validation-*
   - split: test
     path: data/sentence_50/test-*
-- config_name: sentence_100
+- config_name: sentence_5
   data_files:
   - split: train
-    path: data/sentence_100/train-*
+    path: data/sentence_5/train-*
   - split: validation
-    path: data/sentence_100/validation-*
+    path: data/sentence_5/validation-*
   - split: test
-    path: data/sentence_100/test-*
+    path: data/sentence_5/test-*
 - config_name: source_held_out_1
   data_files:
   - split: train
@@ -67,14 +67,14 @@ configs:
     path: data/source_held_out_50/validation-*
   - split: test
     path: data/source_held_out_50/test-*
-- config_name: source_held_out_100
+- config_name: source_held_out_5
   data_files:
   - split: train
-    path: data/source_held_out_100/train-*
+    path: data/source_held_out_5/train-*
   - split: validation
-    path: data/source_held_out_100/validation-*
+    path: data/source_held_out_5/validation-*
   - split: test
-    path: data/source_held_out_100/test-*
+    path: data/source_held_out_5/test-*
 ---
 
 # Sphragis
@@ -85,7 +85,7 @@ configs:
 prose and verse authorship attribution (AA). Its input is the complete curated
 union of the human-annotated [CoNLL-U](https://universaldependencies.org/format.html)
 trees in the supported treebank projects, including lemma, part of speech,
-morphology, and dependency relations. It defines 1-, 10-, 50-, and 100-sentence
+morphology, and dependency relations. It defines 1-, 5-, 10-, and 50-sentence
 attribution tasks. The complementary scanned-line benchmark is published
 separately as [`Urdatorn/sphragis-metre`](https://huggingface.co/datasets/Urdatorn/sphragis-metre).
 
@@ -114,31 +114,32 @@ to the main track in consequence.
 
 | Configuration | Authors | Row unit | Train rows | Validation rows | Test rows |
 |---|---:|---|---:|---:|---:|
-| `sentence_1` | 17 | sentence | 43,300 | 4,800 | 4,800 |
-| `sentence_10` | 17 | 10-sentence chunk | 4,330 | 480 | 480 |
-| `sentence_50` | 17 | 50-sentence chunk | 866 | 96 | 96 |
-| `sentence_100` | 17 | 100-sentence chunk | 433 | 48 | 48 |
-| `source_held_out_1` | 11 | sentence | 25,100 | 4,200 | 4,200 |
-| `source_held_out_10` | 11 | 10-sentence chunk | 2,510 | 420 | 420 |
-| `source_held_out_50` | 11 | 50-sentence chunk | 502 | 84 | 84 |
-| `source_held_out_100` | 11 | 100-sentence chunk | 251 | 42 | 42 |
+| `sentence_1` | 28 | sentence | 50,500 | 5,650 | 5,650 |
+| `sentence_5` | 28 | 5-sentence chunk | 10,100 | 1,130 | 1,130 |
+| `sentence_10` | 28 | 10-sentence chunk | 5,050 | 565 | 565 |
+| `sentence_50` | 28 | 50-sentence chunk | 1,010 | 113 | 113 |
+| `source_held_out_1` | 12 | sentence | 28,500 | 4,550 | 4,550 |
+| `source_held_out_5` | 12 | 5-sentence chunk | 5,700 | 910 | 910 |
+| `source_held_out_10` | 12 | 10-sentence chunk | 2,850 | 455 | 455 |
+| `source_held_out_50` | 12 | 50-sentence chunk | 570 | 91 | 91 |
 
 The `source_held_out_*` configurations use the same rows, the same chunk sizes
 and the same 100-row bottleneck, over the eleven authors that more than one
 project annotates. Two of them, Hesiod and Lucian, do not reach the main
 track's threshold under random splitting but do under this arrangement.
 
-The suffix gives the chunk size in every split. `_100` defines one shared
-source corpus for all four tasks: an author is retained
-only when that author independently has at least 100 source rows in **both**
-validation and test, and every split is reduced to a multiple of 100 source
-rows per author. `_1` publishes the shared rows atomically, while `_10`, `_50`,
-and `_100` regroup those exact same train, validation, and test rows into fixed
+The suffix gives the chunk size in every split. `_50`, the largest, defines one
+shared source corpus for all four tasks: an author is retained only when that
+author independently has at least 50 source rows in **both** validation and
+test, and every split is reduced to a multiple of 50 source rows per author.
+Every smaller task size divides 50, which is what lets the same retained rows
+fill all four. `_1` publishes the shared rows atomically, while `_5`, `_10` and
+`_50` regroup those exact same train, validation, and test rows into fixed
 benchmark chunks. Thus author inventory and total source text are identical
 across task sizes.
 
 Source-row selection uses seed `776`. For each retained author and split, the
-unavoidable `n mod 100` source rows are selected for exclusion once by a stable
+unavoidable `n mod 50` source rows are selected for exclusion once by a stable
 hash. The shared retained rows are put into their best available textual order
 within each work for each chunking task. Complete single-work chunks are
 emitted first; only residual work tails are combined, and such rows are marked
@@ -150,28 +151,39 @@ Every row retains its treebank genre and provenance.
 ## Sentences per author label
 
 These are the exact retained atomic sentences in `sentence_1`. The same
-sentences are represented in the 10-, 50-, and 100-sentence configurations.
+sentences are represented in the 5-, 10-, and 50-sentence configurations.
 
 | Author label | Train | Validation | Test | Total |
 |---|---:|---:|---:|---:|
-| Aeschylus | 2,600 | 300 | 300 | 3,200 (6.0%) |
-| Aristophanes | 1,400 | 100 | 100 | 1,600 (3.0%) |
-| Athenaeus | 1,900 | 200 | 200 | 2,300 (4.3%) |
-| Demosthenes | 2,400 | 300 | 300 | 3,000 (5.7%) |
-| Diodorus Siculus | 1,000 | 100 | 100 | 1,200 (2.3%) |
-| Dionysius of Halicarnassus | 800 | 100 | 100 | 1,000 (1.9%) |
-| Herodotus | 5,900 | 700 | 700 | 7,300 (13.8%) |
-| Homeric-Iliad | 6,200 | 700 | 700 | 7,600 (14.4%) |
-| Homeric-Odyssey | 4,800 | 600 | 600 | 6,000 (11.3%) |
-| Josephus | 800 | 100 | 100 | 1,000 (1.9%) |
-| Plato | 900 | 100 | 100 | 1,100 (2.1%) |
-| Plutarch | 1,300 | 100 | 100 | 1,500 (2.8%) |
-| Polybius | 3,000 | 300 | 300 | 3,600 (6.8%) |
-| Procopius | 900 | 100 | 100 | 1,100 (2.1%) |
-| Sophocles | 3,100 | 300 | 300 | 3,700 (7.0%) |
-| Thucydides | 900 | 100 | 100 | 1,100 (2.1%) |
-| Xenophon | 5,400 | 600 | 600 | 6,600 (12.5%) |
-| **All labels** | **43,300** | **4,800** | **4,800** | **52,900 (100.0%)** |
+| Aeschines | 500 | 50 | 50 | 600 (1.0%) |
+| Aeschylus | 2,600 | 300 | 300 | 3,200 (5.2%) |
+| Antiphon | 400 | 50 | 50 | 500 (0.8%) |
+| Appian | 750 | 50 | 50 | 850 (1.4%) |
+| Aristophanes | 1,450 | 150 | 150 | 1,750 (2.8%) |
+| Aristotle | 650 | 50 | 50 | 750 (1.2%) |
+| Athenaeus | 1,950 | 200 | 200 | 2,350 (3.8%) |
+| Demosthenes | 2,400 | 300 | 300 | 3,000 (4.9%) |
+| Diodorus Siculus | 1,000 | 100 | 100 | 1,200 (1.9%) |
+| Dionysius of Halicarnassus | 850 | 100 | 100 | 1,050 (1.7%) |
+| Euripides | 600 | 50 | 50 | 700 (1.1%) |
+| Herodotus | 5,900 | 700 | 700 | 7,300 (11.8%) |
+| Hesiod | 700 | 50 | 50 | 800 (1.3%) |
+| Homeric-Iliad | 6,250 | 750 | 750 | 7,750 (12.5%) |
+| Homeric-Odyssey | 4,850 | 600 | 600 | 6,050 (9.8%) |
+| Josephus | 800 | 100 | 100 | 1,000 (1.6%) |
+| Lucian | 550 | 50 | 50 | 650 (1.1%) |
+| Lysias | 750 | 50 | 50 | 850 (1.4%) |
+| Menander | 700 | 50 | 50 | 800 (1.3%) |
+| Paul | 700 | 50 | 50 | 800 (1.3%) |
+| Plato | 900 | 100 | 100 | 1,100 (1.8%) |
+| Plutarch | 1,300 | 150 | 150 | 1,600 (2.6%) |
+| Polybius | 3,000 | 350 | 350 | 3,700 (6.0%) |
+| Procopius | 950 | 100 | 100 | 1,150 (1.9%) |
+| Sextus Empiricus | 500 | 50 | 50 | 600 (1.0%) |
+| Sophocles | 3,100 | 350 | 350 | 3,800 (6.1%) |
+| Thucydides | 950 | 100 | 100 | 1,150 (1.9%) |
+| Xenophon | 5,450 | 650 | 650 | 6,750 (10.9%) |
+| **All labels** | **50,500** | **5,650** | **5,650** | **61,800 (100.0%)** |
 
 ```python
 from datasets import load_dataset
@@ -184,10 +196,10 @@ sentences = load_dataset("Urdatorn/sphragis", "sentence_10")
 Splits use random seed `776` and begin with independent stratification within
 each author/work group: approximately 80% train, 10% validation and 10% test by
 row. Groups with 3–9 rows receive one validation and one test row; groups with
-fewer than three stay in training. The 100-row eligibility and remainder
+fewer than three stay in training. The 50-row eligibility and remainder
 bottleneck is then applied to all
 four task sizes. The train, validation, and test source sets are all exact
-multiples of 100 per author. This makes the
+multiples of 50 per author. This makes the
 published source-row ratio slightly different from the initial 80/10/10 split.
 A stricter work-held-out evaluation can be added as a later track.
 
@@ -204,8 +216,8 @@ The `source_held_out_*` configurations remove that possibility by construction.
 For each author, the **second-largest** source project is held out entirely:
 its sentences become that author's validation and test data, split evenly, and
 every other project supplies training. An author is retained only when the
-held-out project yields at least 100 validation and 100 test rows and at least
-100 training rows remain. A model that has learned to recognize annotators is
+held-out project yields at least 50 validation and 50 test rows and at least
+50 training rows remain. A model that has learned to recognize annotators is
 actively penalized here, because every author arrives at evaluation in a scheme
 it was never trained in.
 
@@ -445,6 +457,12 @@ New Testament works are excluded.
 The *Iliad* and *Odyssey* are retained as distinct conventional corpora under
 the author labels `Homeric-Iliad` and `Homeric-Odyssey`; the labels do not claim
 a shared biographical author. Every work labelled `Fragments` is excluded.
+
+Where an upstream corpus files two works under one name and only one of them is
+that author's, the works are named separately so that the pseudonymous rule can
+act on them. The *Cynegetica* is transmitted under Oppian's name but is not by
+the poet of the *Halieutica*, so it is labelled `Pseudo-Oppian` and excluded;
+`Oppian` is the *Halieutica* alone.
 Machine-readable reasons and pre-deduplication counts are recorded in
 [`metadata/build_report.json`](metadata/build_report.json).
 

data/sentence_1/test-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14770d4277823e04b0f6cbb53a3ae4036dd020124cd4ee8ba15a1bcfc57bbc26
-size 1454202
+oid sha256:b32ee0a4b2e1cf61921c1484f8c6b96569cdac5fdf6d9ca21906f0e98bcef51d
+size 1696165

data/sentence_1/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b75ddbf13d092fe20888d6c068a67ecd5c5bc8775b09077d8e6dd7357905e1c2
-size 12237487
+oid sha256:cef082ec7a08f18581ed274ab7b068898ddb49918795cd63200d943491dd8073
+size 14248650

data/sentence_1/validation-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:748f6b9e82cb64614c30bb7077c03cb0abaadeb1e00e5713c670d5ac3f8f4329
-size 1444361
+oid sha256:68b044fd6ed307b03d58128458f729f04fad41987594c1622769174fa3c52ff2
+size 1697530

data/sentence_10/test-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bf491c42c150c2bc3d0dc4f64a4ef6d8dfdd5d183ddddc5fa3e962c57c4946cc
-size 1321392
+oid sha256:c22df82a8dbf599e2e8fc9a68391c7f51a2bbaf2d44909369c91281e99dd4d8b
+size 1532728

data/sentence_10/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b55e096a0e092816721d2fd4fcadf862513e46bb509028d13500564e1160a02
-size 10402450
+oid sha256:1a1b21d6a8bb4f65486503ba03d63c2b71bb2d4042a4fe4c20b32d5dbb13f78f
+size 12002579

data/sentence_10/validation-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c866aa0d4f9fbaaaa4544b8b0f082f1505a237c6b1e0821fb4a0dc2b4e8b279
-size 1312167
+oid sha256:be2575d0091a4912a7a1baf861f23eb27893c3f919a52779e7c16d4f56a84074
+size 1534640

data/{sentence_100 → sentence_5}/test-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:01278bc7879878a827bfdeeca5f497fb839b94bcbd31c465d698c92e7f9f6020
-size 1288881
+oid sha256:8ebccb15f5a9c0fd4c67029bc996b566f9009d219b16cc8c020bf1da04138468
+size 1570073

data/{sentence_100 → sentence_5}/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1f2f0f19062d01beea7e96f83f398fef60ad2858ba43638af9762e3f2c0392e
-size 10146914
+oid sha256:e26dfd1b822a8dc4e1ec1553844338161ca73205f1d7e4b9b50918431a5b0f8f
+size 12303120

data/{source_held_out_100 → sentence_5}/validation-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7cd440739df9a4a3d5e6c0bcce391a0f44f902e0ee32fd5634d622821c39c36e
-size 1194134
+oid sha256:7dec49899b25feb25b2f7ec88c59e151ab358afab25c31028ee0de0fb5c8f26c
+size 1572340

data/sentence_50/test-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f8a3764cf72330e86cbb293a5cb3665bca7f1baaa5a008c25848a8f231811854
-size 1288463
+oid sha256:b81a9a8d32bbb4978eddfd2991b3b3cda533647ecd06a28a274940cbbc86d6ee
+size 1498014

data/sentence_50/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b40ff5e73093d2f671db9f75fb469cfff11032d2559a3b8e0d750b72926aa00b
-size 10175764
+oid sha256:1202ee01377667909c021389f2f6622e3c15b5121063f0daa86287767ac7cb10
+size 11736592

data/sentence_50/validation-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:db7022e0882ae77a084343ea9c4571914549965713f5a2f829b57ce6dfc01b59
-size 1279662
+oid sha256:85797da286c7ce2d954ee6a23dbab68767c39cf839ef3632aa64f582a9511411
+size 1499780

data/source_held_out_1/test-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe36125d0780a3321470b81b16bc08c274ade7e34a207e1640ecd4cb201af809
-size 1346318
+oid sha256:bf3cee7cf76a5dfe9173b8702e844fdd0aaa2cfcca9d69e3aa7853ef4facc7bb
+size 1433455

data/source_held_out_1/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12769393b8913593752d406a9caf9c74b4ace6088bf646502279e991f4770a4c
-size 6448073
+oid sha256:25da0e17db04e5d2bae9ca19aed3e7e5b9933947bfc267359fa8577e233d4a92
+size 7339404

data/source_held_out_1/validation-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:85e4622d54e2aa853ec67374335b895a732327674273aef9dc4706986ea00101
-size 1346712
+oid sha256:129f586c39211a06693f1b55c03dbbc38eae4cb0c916f6269f67913c7abf3a24
+size 1439141

data/source_held_out_10/test-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e2570cc2b9bcd2e76a91574465328d51d586ca05961e634dd80972ffc99d7382
-size 1226211
+oid sha256:269cdad09004ef5e433c18978d56868d25c88307b5a4bb9d812523482a20cee5
+size 1304691

data/source_held_out_10/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ada6e935cff1a39cff035ae69d236ef01453b93465325ccabbeb6ebda0e5cc6
-size 5467470
+oid sha256:577cadae2a946bf9d7b5717f9f478fc437cc11b618bad606cf335b58278475e4
+size 6221019

data/source_held_out_10/validation-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e355b024751c4c22fbb9d63608104ed3d5870e16ee4085b2ac4b46c1dc3b2483
-size 1222117
+oid sha256:2ca9e343b94ddbd9c771913c133a81c5ea44bee41ba6c7172197b21641732a39
+size 1306240

data/{source_held_out_100 → source_held_out_5}/test-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8f5a1bfeb4b3e19fed1903517fb7568879873edf6b07526e78b88ab570f01dda
-size 1199456
+oid sha256:b18868d53bf167122a1e9bb052328c9cb0fff501f84bf574ebf99111269c8c79
+size 1334461

data/{source_held_out_100 → source_held_out_5}/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e63774f9f67be778b283a425b283fdb077daa3d6de56ba2c898348dc574643aa
-size 5312021
+oid sha256:8552f9e5fd510f134af266625dd26de43000263b8ab5d3e259f3e20d4cb718d0
+size 6376410

data/{sentence_100 → source_held_out_5}/validation-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2ab3ae5a2f03d11a9c246cea46f0969b068c0e3657b123dce3382a26ba2697a
-size 1278930
+oid sha256:8ff4a54aa19ea55490d778a8d7fbfb195ef102aa140a6a97657931f455142666
+size 1336537

data/source_held_out_50/test-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:856ccf11621ba3a26f302e8fe6f29938f1959db04fa03c22112c44266110a1e3
-size 1202969
+oid sha256:45309782ef524a3b6a5838b67c743096db9ded926972b06a1a891daa5cc88604
+size 1279177

data/source_held_out_50/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2569f8721c2e794e9174abe7843056be05a2587808bc89ee94d2dc65556b3d7e
-size 5330210
+oid sha256:4cad1a114c486e52057cb2482b0c4ae9918bfb1b2ada9033fba0cbf96bd728b6
+size 6068124

data/source_held_out_50/validation-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b3b6a79877f4a5a8ee1d28499b2ae13cc4edd80f0f83ea805ab5bf28726744b7
-size 1197595
+oid sha256:b0d0250479673803237db667d6d192a1398ad4e2bf92cfcc2ec46a1ce162251f
+size 1280679

inspection/sphragis.sqlite

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:880fb25a1e9754c3b6b90a1b42242810871bd23fbad756deafbf66c429572cc3
-size 1775468544
+oid sha256:ba6b76242957da26d71b753c58f90b5f41e6fdf8511cc7aba7ea1ca79177828f
+size 2385051648

metadata/build_report.json

@@ -355,44 +355,44 @@
   },
   "data_files": {
     "sentence_1": {
-      "test": 4800,
-      "train": 43300,
-      "validation": 4800
+      "test": 5650,
+      "train": 50500,
+      "validation": 5650
     },
     "sentence_10": {
-      "test": 480,
-      "train": 4330,
-      "validation": 480
+      "test": 565,
+      "train": 5050,
+      "validation": 565
     },
-    "sentence_100": {
-      "test": 48,
-      "train": 433,
-      "validation": 48
+    "sentence_5": {
+      "test": 1130,
+      "train": 10100,
+      "validation": 1130
     },
     "sentence_50": {
-      "test": 96,
-      "train": 866,
-      "validation": 96
+      "test": 113,
+      "train": 1010,
+      "validation": 113
     },
     "source_held_out_1": {
-      "test": 4200,
-      "train": 25100,
-      "validation": 4200
+      "test": 4550,
+      "train": 28500,
+      "validation": 4550
     },
     "source_held_out_10": {
-      "test": 420,
-      "train": 2510,
-      "validation": 420
+      "test": 455,
+      "train": 2850,
+      "validation": 455
     },
-    "source_held_out_100": {
-      "test": 42,
-      "train": 251,
-      "validation": 42
+    "source_held_out_5": {
+      "test": 910,
+      "train": 5700,
+      "validation": 910
     },
     "source_held_out_50": {
-      "test": 84,
-      "train": 502,
-      "validation": 84
+      "test": 91,
+      "train": 570,
+      "validation": 91
     }
   },
   "deduplication": {
@@ -815,6 +815,21 @@
   "publication": "sentence",
   "source_held_out_track": {
     "author_decisions": {
+      "Aeschylus": {
+        "held_out_source": "ud_perseus",
+        "rows": {
+          "test": 81,
+          "train": 3115,
+          "validation": 81
+        },
+        "source_rows": {
+          "agdt": 3115,
+          "ud_perseus": 163
+        },
+        "training_sources": [
+          "agdt"
+        ]
+      },
       "Athenaeus": {
         "held_out_source": "gorman",
         "rows": {
@@ -1008,14 +1023,6 @@
           "gorman": 666
         }
       },
-      "Aeschylus": {
-        "held_out_source": "ud_perseus",
-        "reason": "held-out source too small for a 100-row validation and test split",
-        "sources": {
-          "agdt": 3115,
-          "ud_perseus": 163
-        }
-      },
       "Andocides": {
         "reason": "only one source project",
         "sources": {
@@ -1126,7 +1133,7 @@
       },
       "Lysias": {
         "held_out_source": "pedalion",
-        "reason": "held-out source too small for a 100-row validation and test split",
+        "reason": "held-out source too small for a 50-row validation and test split",
         "sources": {
           "agdt": 54,
           "gorman": 833,
@@ -1159,7 +1166,7 @@
       },
       "Polybius": {
         "held_out_source": "agdt",
-        "reason": "held-out source too small for a 100-row validation and test split",
+        "reason": "held-out source too small for a 50-row validation and test split",
         "sources": {
           "agdt": 1,
           "gorman": 3804
@@ -1196,9 +1203,10 @@
         }
       }
     },
-    "minimum_rows_per_split": 100,
+    "minimum_rows_per_split": 50,
     "policy": "hold out each author's second-largest source project entirely; its sentences form that author's validation and test data and every other project supplies training",
     "retained_authors": [
+      "Aeschylus",
       "Athenaeus",
       "Diodorus Siculus",
       "Herodotus",
@@ -1239,64 +1247,70 @@
       "nested_text_and_conllu_cardinality": true
     },
     "sentence_1": {
-      "authors": 17,
+      "authors": 28,
       "genres": [
         "prose",
         "verse"
       ],
-      "rows": 52900,
+      "rows": 61800,
       "splits": {
-        "test": 4800,
-        "train": 43300,
-        "validation": 4800
+        "test": 5650,
+        "train": 50500,
+        "validation": 5650
       }
     },
     "sentence_10": {
-      "authors": 17,
+      "authors": 28,
       "genres": [
         "prose",
         "verse"
       ],
-      "rows": 5290,
+      "rows": 6180,
       "splits": {
-        "test": 480,
-        "train": 4330,
-        "validation": 480
+        "test": 565,
+        "train": 5050,
+        "validation": 565
       }
     },
-    "sentence_100": {
-      "authors": 17,
+    "sentence_5": {
+      "authors": 28,
       "genres": [
         "prose",
         "verse"
       ],
-      "rows": 529,
+      "rows": 12360,
       "splits": {
-        "test": 48,
-        "train": 433,
-        "validation": 48
+        "test": 1130,
+        "train": 10100,
+        "validation": 1130
       }
     },
     "sentence_50": {
-      "authors": 17,
+      "authors": 28,
       "genres": [
         "prose",
         "verse"
       ],
-      "rows": 1058,
+      "rows": 1236,
       "splits": {
-        "test": 96,
-        "train": 866,
-        "validation": 96
+        "test": 113,
+        "train": 1010,
+        "validation": 113
       }
     },
     "source_held_out_1": {
-      "authors": 11,
+      "authors": 12,
       "genres": [
         "prose",
         "verse"
       ],
       "held_out_sources": {
+        "Aeschylus": {
+          "held_out_source": "ud_perseus",
+          "training_sources": [
+            "agdt"
+          ]
+        },
         "Athenaeus": {
           "held_out_source": "gorman",
           "training_sources": [
@@ -1371,20 +1385,26 @@
           ]
         }
       },
-      "rows": 33500,
+      "rows": 37600,
       "splits": {
-        "test": 4200,
-        "train": 25100,
-        "validation": 4200
+        "test": 4550,
+        "train": 28500,
+        "validation": 4550
       }
     },
     "source_held_out_10": {
-      "authors": 11,
+      "authors": 12,
       "genres": [
         "prose",
         "verse"
       ],
       "held_out_sources": {
+        "Aeschylus": {
+          "held_out_source": "ud_perseus",
+          "training_sources": [
+            "agdt"
+          ]
+        },
         "Athenaeus": {
           "held_out_source": "gorman",
           "training_sources": [
@@ -1459,20 +1479,26 @@
           ]
         }
       },
-      "rows": 3350,
+      "rows": 3760,
       "splits": {
-        "test": 420,
-        "train": 2510,
-        "validation": 420
+        "test": 455,
+        "train": 2850,
+        "validation": 455
       }
     },
-    "source_held_out_100": {
-      "authors": 11,
+    "source_held_out_5": {
+      "authors": 12,
       "genres": [
         "prose",
         "verse"
       ],
       "held_out_sources": {
+        "Aeschylus": {
+          "held_out_source": "ud_perseus",
+          "training_sources": [
+            "agdt"
+          ]
+        },
         "Athenaeus": {
           "held_out_source": "gorman",
           "training_sources": [
@@ -1547,20 +1573,26 @@
           ]
         }
       },
-      "rows": 335,
+      "rows": 7520,
       "splits": {
-        "test": 42,
-        "train": 251,
-        "validation": 42
+        "test": 910,
+        "train": 5700,
+        "validation": 910
       }
     },
     "source_held_out_50": {
-      "authors": 11,
+      "authors": 12,
       "genres": [
         "prose",
         "verse"
       ],
       "held_out_sources": {
+        "Aeschylus": {
+          "held_out_source": "ud_perseus",
+          "training_sources": [
+            "agdt"
+          ]
+        },
         "Athenaeus": {
           "held_out_source": "gorman",
           "training_sources": [
@@ -1635,11 +1667,11 @@
           ]
         }
       },
-      "rows": 670,
+      "rows": 752,
       "splits": {
-        "test": 84,
-        "train": 502,
-        "validation": 84
+        "test": 91,
+        "train": 570,
+        "validation": 91
       }
     }
   }

metadata/leakage_report.json

@@ -1,196 +1,196 @@
 {
   "configurations": {
     "sentence_1": {
-      "authors": 17,
+      "authors": 28,
       "inventory_to_author": {
-        "accuracy": 36.3125,
+        "accuracy": 32.708,
         "features": 50,
-        "macro_f1": 17.7243
+        "macro_f1": 11.038
       },
       "majority_baseline": {
-        "accuracy": 14.5833,
+        "accuracy": 13.2743,
         "features": 0,
-        "macro_f1": 1.4973
+        "macro_f1": 0.8371
       },
       "markup_to_author": {
-        "accuracy": 50.5208,
+        "accuracy": 45.646,
         "features": 64,
-        "macro_f1": 32.0114
+        "macro_f1": 20.6286
       },
       "markup_to_treebank_source": {
-        "accuracy": 79.0208,
+        "accuracy": 77.6814,
         "features": 64,
-        "macro_f1": 73.0083
+        "macro_f1": 61.977
       },
-      "test_rows": 4800
+      "test_rows": 5650
     },
     "sentence_10": {
-      "authors": 17,
+      "authors": 28,
       "inventory_to_author": {
-        "accuracy": 39.7917,
+        "accuracy": 36.2832,
         "features": 50,
-        "macro_f1": 16.6515
+        "macro_f1": 12.805
       },
       "majority_baseline": {
-        "accuracy": 14.5833,
+        "accuracy": 13.2743,
         "features": 0,
-        "macro_f1": 1.4973
+        "macro_f1": 0.8371
       },
       "markup_to_author": {
-        "accuracy": 73.9583,
+        "accuracy": 67.9646,
         "features": 64,
-        "macro_f1": 51.9926
+        "macro_f1": 34.7397
       },
       "markup_to_treebank_source": {
-        "accuracy": 87.2917,
+        "accuracy": 86.7257,
         "features": 64,
-        "macro_f1": 67.9758
+        "macro_f1": 61.3577
       },
-      "test_rows": 480
+      "test_rows": 565
     },
-    "sentence_100": {
-      "authors": 17,
+    "sentence_5": {
+      "authors": 28,
       "inventory_to_author": {
-        "accuracy": 50.0,
+        "accuracy": 35.3097,
         "features": 50,
-        "macro_f1": 27.274
+        "macro_f1": 11.6605
       },
       "majority_baseline": {
-        "accuracy": 14.5833,
+        "accuracy": 13.2743,
         "features": 0,
-        "macro_f1": 1.4973
+        "macro_f1": 0.8371
       },
       "markup_to_author": {
-        "accuracy": 68.75,
+        "accuracy": 64.0708,
         "features": 64,
-        "macro_f1": 39.5994
+        "macro_f1": 36.1669
       },
       "markup_to_treebank_source": {
-        "accuracy": 62.5,
+        "accuracy": 87.4336,
         "features": 64,
-        "macro_f1": 52.151
+        "macro_f1": 61.3361
       },
-      "test_rows": 48
+      "test_rows": 1130
     },
     "sentence_50": {
-      "authors": 17,
+      "authors": 28,
       "inventory_to_author": {
-        "accuracy": 47.9167,
+        "accuracy": 42.4779,
         "features": 50,
-        "macro_f1": 28.9223
+        "macro_f1": 17.4088
       },
       "majority_baseline": {
-        "accuracy": 14.5833,
+        "accuracy": 13.2743,
         "features": 0,
-        "macro_f1": 1.4973
+        "macro_f1": 0.8371
       },
       "markup_to_author": {
-        "accuracy": 73.9583,
+        "accuracy": 67.2566,
         "features": 64,
-        "macro_f1": 41.2814
+        "macro_f1": 23.7232
       },
       "markup_to_treebank_source": {
-        "accuracy": 73.9583,
+        "accuracy": 73.4513,
         "features": 64,
-        "macro_f1": 57.5758
+        "macro_f1": 62.5611
       },
-      "test_rows": 96
+      "test_rows": 113
     },
     "source_held_out_1": {
-      "authors": 11,
+      "authors": 12,
       "inventory_to_author": {
-        "accuracy": 16.7143,
+        "accuracy": 13.5165,
         "features": 50,
-        "macro_f1": 7.8796
+        "macro_f1": 5.7013
       },
       "majority_baseline": {
-        "accuracy": 11.9048,
+        "accuracy": 10.989,
         "features": 0,
-        "macro_f1": 1.9342
+        "macro_f1": 1.6502
       },
       "markup_to_author": {
-        "accuracy": 23.6667,
+        "accuracy": 18.7473,
         "features": 64,
-        "macro_f1": 12.4674
+        "macro_f1": 10.3109
       },
       "markup_to_treebank_source": {
-        "accuracy": 33.3333,
+        "accuracy": 45.6264,
         "features": 64,
-        "macro_f1": 18.338
+        "macro_f1": 26.9095
       },
-      "test_rows": 4200
+      "test_rows": 4550
     },
     "source_held_out_10": {
-      "authors": 11,
+      "authors": 12,
       "inventory_to_author": {
-        "accuracy": 12.8571,
+        "accuracy": 8.5714,
         "features": 50,
-        "macro_f1": 4.5931
+        "macro_f1": 2.7505
       },
       "majority_baseline": {
-        "accuracy": 11.9048,
+        "accuracy": 10.989,
         "features": 0,
-        "macro_f1": 1.9342
+        "macro_f1": 1.6502
       },
       "markup_to_author": {
-        "accuracy": 36.9048,
+        "accuracy": 28.7912,
         "features": 64,
-        "macro_f1": 21.9855
+        "macro_f1": 17.6304
       },
       "markup_to_treebank_source": {
-        "accuracy": 33.8095,
+        "accuracy": 45.0549,
         "features": 64,
-        "macro_f1": 18.0101
+        "macro_f1": 26.5007
       },
-      "test_rows": 420
+      "test_rows": 455
     },
-    "source_held_out_100": {
-      "authors": 11,
+    "source_held_out_5": {
+      "authors": 12,
       "inventory_to_author": {
-        "accuracy": 23.8095,
+        "accuracy": 11.2088,
         "features": 50,
-        "macro_f1": 8.2867
+        "macro_f1": 4.0306
       },
       "majority_baseline": {
-        "accuracy": 11.9048,
+        "accuracy": 10.989,
         "features": 0,
-        "macro_f1": 1.9342
+        "macro_f1": 1.6502
       },
       "markup_to_author": {
-        "accuracy": 28.5714,
+        "accuracy": 25.8242,
         "features": 64,
-        "macro_f1": 13.7457
+        "macro_f1": 16.9135
       },
       "markup_to_treebank_source": {
-        "accuracy": 21.4286,
+        "accuracy": 49.5604,
         "features": 64,
-        "macro_f1": 11.6162
+        "macro_f1": 28.6622
       },
-      "test_rows": 42
+      "test_rows": 910
     },
     "source_held_out_50": {
-      "authors": 11,
+      "authors": 12,
       "inventory_to_author": {
-        "accuracy": 7.1429,
+        "accuracy": 3.2967,
         "features": 50,
-        "macro_f1": 2.7308
+        "macro_f1": 1.5152
       },
       "majority_baseline": {
-        "accuracy": 11.9048,
+        "accuracy": 10.989,
         "features": 0,
-        "macro_f1": 1.9342
+        "macro_f1": 1.6502
       },
       "markup_to_author": {
-        "accuracy": 27.381,
+        "accuracy": 24.1758,
         "features": 64,
-        "macro_f1": 14.5915
+        "macro_f1": 12.0011
       },
       "markup_to_treebank_source": {
-        "accuracy": 25.0,
+        "accuracy": 32.967,
         "features": 64,
-        "macro_f1": 12.7619
+        "macro_f1": 20.4127
       },
-      "test_rows": 84
+      "test_rows": 91
     }
   },
   "probes": {

scripts/audit_leakage.py

@@ -139,8 +139,8 @@ def main() -> None:
     parser.add_argument(
         "--configs", nargs="*",
         default=[
-            *(f"sentence_{suffix}" for suffix in (1, 10, 50, 100)),
-            *(f"source_held_out_{suffix}" for suffix in (1, 10, 50, 100)),
+            *(f"sentence_{suffix}" for suffix in (1, 5, 10, 50)),
+            *(f"source_held_out_{suffix}" for suffix in (1, 5, 10, 50)),
         ],
     )
     args = parser.parse_args()

scripts/build_dataset.py

@@ -1241,6 +1241,12 @@ def hypotactic_author_work(stem: str, line: dict) -> tuple[str, str, str]:
     author = canonical_author(author)
     if author == "Homer" and work in {"Iliad", "Odyssey"}:
         author = f"Homeric-{work}"
+    # Hypotactic files both works under one Oppian, but the Cynegetica is not by
+    # the poet of the Halieutica. Naming its author separately lets the
+    # pseudonymous rule in authorship_decision exclude it, rather than leaving
+    # 38% of the Oppian label written by somebody else.
+    if author == "Oppian" and work == "Cynegetica":
+        author = "Pseudo-Oppian"
     work_id = HYPOTACTIC_CANONICAL_WORK_IDS.get(
         (author, work), f"hypotactic:{slug(author)}:{slug(work)}",
     )
@@ -1744,12 +1750,17 @@ def predicted_metre_rows(
                 excluded[reason] += 1
             else:
                 eligible[hypotactic_line_key(line)] = curated
-    if set(predictions) != set(eligible):
-        missing = set(eligible) - set(predictions)
-        extra = set(predictions) - set(eligible)
+    missing = set(eligible) - set(predictions)
+    if missing:
         raise ValueError(
-            f"Stoicheia cache is incomplete or stale: missing={len(missing)} extra={len(extra)}"
+            f"Stoicheia cache is incomplete: {len(missing)} eligible lines "
+            "have no parse"
         )
+    # The cache may hold parses for lines curation has since excluded, as it
+    # does for the Cynegetica. Those are simply unused: every prediction that
+    # is published still has its text and authorship checked against the
+    # Hypotactic line below, which is what would catch a genuinely stale cache.
+    unused_predictions = len(set(predictions) - set(eligible))
     gold_by_key = {
         (row["hypotactic_file"], row["poem_sequence"], row["book"], row["line_number"]): row
         for row in gold_rows
@@ -1757,6 +1768,8 @@ def predicted_metre_rows(
     output = []
     excluded_disputed = 0
     for key, prediction in sorted(predictions.items()):
+        if key not in eligible:
+            continue
         line = lines_by_key.get(key)
         if line is None:
             raise ValueError(f"Stoicheia prediction has no Hypotactic line: {key}")
@@ -1843,6 +1856,7 @@ def predicted_metre_rows(
         "context_sentences_spanning_multiple_lines": len(multiline_sentence_ids),
         "context_window_forced_splits": len(forced_context_sentence_ids),
         "predictions_in_cache": len(predictions),
+        "cached_predictions_unused_after_curation": unused_predictions,
         "hypotactic_lines_total": len(lines_by_key),
         "hypotactic_lines_excluded_before_parsing": sum(excluded.values()),
         "hypotactic_lines_excluded_by_reason": dict(sorted(excluded.items())),
@@ -2049,7 +2063,7 @@ def lower_greek_syllables(payload):
     return payload
 
 
-SOURCE_HELD_OUT_TARGET = 100
+SOURCE_HELD_OUT_TARGET = 50
 
 
 def source_held_out_rows(rows: list[dict]) -> tuple[list[dict], dict]:
@@ -2094,7 +2108,10 @@ def source_held_out_rows(rows: list[dict]) -> tuple[list[dict], dict]:
             or len(train_rows) < SOURCE_HELD_OUT_TARGET
         ):
             rejected[author] = {
-                "reason": "held-out source too small for a 100-row validation and test split",
+                "reason": (
+                    "held-out source too small for a "
+                    f"{SOURCE_HELD_OUT_TARGET}-row validation and test split"
+                ),
                 "held_out_source": held_out,
                 "sources": {source: len(by_source[source]) for source in ranked},
             }
@@ -2202,7 +2219,7 @@ def validate_publication(
     rows_by_config: dict[str, list[dict]], publication: str,
 ) -> dict:
     """Validate final repository-level rows after the publication split."""
-    suffixes = ("1", "10", "50", "100")
+    suffixes = ("1", "5", "10", "50")
     if publication == "sentence":
         expected_configs = (
             *(f"sentence_{suffix}" for suffix in suffixes),

scripts/build_sqlite_mirror.py

@@ -16,8 +16,8 @@ import pyarrow.parquet as pq
 
 SPLITS = ("train", "validation", "test")
 CONFIGS = (
-    tuple(f"sentence_{suffix}" for suffix in ("1", "10", "50", "100"))
-    + tuple(f"source_held_out_{suffix}" for suffix in ("1", "10", "50", "100"))
+    tuple(f"sentence_{suffix}" for suffix in ("1", "5", "10", "50"))
+    + tuple(f"source_held_out_{suffix}" for suffix in ("1", "5", "10", "50"))
 )
 INDEX_COLUMN = "_split_row_index"
 
@@ -218,7 +218,7 @@ def main() -> None:
     )
     parser.add_argument("--check", action="store_true")
     args = parser.parse_args()
-    suffixes = ("1", "10", "50", "100")
+    suffixes = ("1", "5", "10", "50")
     CONFIGS = (
         tuple(f"sentence_{suffix}" for suffix in suffixes)
         + tuple(f"source_held_out_{suffix}" for suffix in suffixes)

scripts/dataset_variants.py

@@ -27,16 +27,19 @@ except ModuleNotFoundError:  # Direct execution from the scripts directory.
 BASE_CONFIGS = ("sentence", "source_held_out", "verse_metre")
 SPLITS = ("train", "validation", "test")
 CHUNK_TARGETS_BY_BASE = {
-    "sentence": (10, 50, 100),
-    "source_held_out": (10, 50, 100),
-    "verse_metre": (10, 50, 100),
+    "sentence": (5, 10, 50),
+    "source_held_out": (5, 10, 50),
+    "verse_metre": (5, 10, 50),
 }
 PUBLIC_CONFIG_PREFIX = {
     "sentence": "sentence",
     "source_held_out": "source_held_out",
     "verse_metre": "verse",
 }
-BOTTLENECK_TARGET = 100
+# The largest chunk size, so every retained author has at least one complete
+# chunk of every task size in validation and in test. Every smaller target
+# divides it, which is what lets the same retained rows fill all four tasks.
+BOTTLENECK_TARGET = 50
 CHUNKING_SEED = 776
 
 CHUNK_FIELDS = (

scripts/validate_publication.py

@@ -85,7 +85,7 @@ def normalize(value: str) -> str:
 
 
 def configs(publication: str) -> tuple[str, ...]:
-    suffixes = (1, 10, 50, 100)
+    suffixes = (1, 5, 10, 50)
     if publication == "sentence":
         return (
             *(f"sentence_{suffix}" for suffix in suffixes),

tests/test_conllu_well_formed.py

@@ -19,9 +19,9 @@ REPO_ROOT = Path(__file__).resolve().parents[1]
 DATA_ROOT = REPO_ROOT / "data"
 METRE_DATA_ROOT = REPO_ROOT.parent / "sphragis-metre" / "data"
 CONFIGS = tuple(
-    [(DATA_ROOT, f"sentence_{suffix}") for suffix in ("1", "10", "50", "100")]
-    + [(DATA_ROOT, f"source_held_out_{suffix}") for suffix in ("1", "10", "50", "100")]
-    + [(METRE_DATA_ROOT, f"verse_{suffix}") for suffix in ("1", "10", "50", "100")]
+    [(DATA_ROOT, f"sentence_{suffix}") for suffix in ("1", "5", "10", "50")]
+    + [(DATA_ROOT, f"source_held_out_{suffix}") for suffix in ("1", "5", "10", "50")]
+    + [(METRE_DATA_ROOT, f"verse_{suffix}") for suffix in ("1", "5", "10", "50")]
 )
 
 
@@ -56,7 +56,7 @@ def test_all_conllu_documents_are_well_formed_and_identifier_free() -> None:
 
 
 def test_verse_metre_conllu_forms_do_not_exceed_line_text() -> None:
-    for suffix in ("1", "10", "50", "100"):
+    for suffix in ("1", "5", "10", "50"):
         for split in ("train", "validation", "test"):
             path = METRE_DATA_ROOT / f"verse_{suffix}" / f"{split}-00000-of-00001.parquet"
             for row in pq.read_table(path, columns=["id", "text", "conllu"]).to_pylist():

tests/test_model_facing.py

@@ -376,7 +376,7 @@ def test_editorial_punctuation_leaves_the_text_column() -> None:
 
 
 def _configs(prefix: str) -> list[str]:
-    return [f"{prefix}_{suffix}" for suffix in ("1", "10", "50", "100")]
+    return [f"{prefix}_{suffix}" for suffix in ("1", "5", "10", "50")]
 
 
 @pytest.mark.parametrize("config", _configs("source_held_out"))

tests/test_split_stratification.py

@@ -14,6 +14,7 @@ from scripts.metre_units import (
     load_syllable_units,
     syllable_line,
 )
+from scripts.dataset_variants import BOTTLENECK_TARGET
 from scripts.metrical_lines import PUBLIC_METRICAL_LINE_FIELD_SET, sanitize_metrical_lines
 from scripts.text_units import load_text_units
 
@@ -22,9 +23,9 @@ REPO_ROOT = Path(__file__).resolve().parents[1]
 METRE_ROOT = REPO_ROOT.parent / "sphragis-metre"
 SPLITS = ("train", "validation", "test")
 PUBLICATIONS = (
-    (REPO_ROOT, "sentence", (1, 10, 50, 100)),
-    (REPO_ROOT, "source_held_out", (1, 10, 50, 100)),
-    (METRE_ROOT, "verse", (1, 10, 50, 100)),
+    (REPO_ROOT, "sentence", (1, 5, 10, 50)),
+    (REPO_ROOT, "source_held_out", (1, 5, 10, 50)),
+    (METRE_ROOT, "verse", (1, 5, 10, 50)),
 )
 # The per-author unit table in each README documents the main track only; the
 # source-held-out track's per-author counts live in build_report.json, because
@@ -125,7 +126,7 @@ def test_readme_author_unit_counts_and_shares_match_atomic_parquets() -> None:
         }
 
 
-def test_all_atomic_splits_are_balanced_and_100_chunkable() -> None:
+def test_all_atomic_splits_are_balanced_and_chunkable() -> None:
     for root, base, _suffixes in PUBLICATIONS:
         counts = defaultdict(Counter)
         for split in SPLITS:
@@ -138,8 +139,10 @@ def test_all_atomic_splits_are_balanced_and_100_chunkable() -> None:
         for author, author_counts in counts.items():
             assert author_counts["validation"] == author_counts["test"]
             for split in SPLITS:
-                assert author_counts[split] >= 100, (base, author, split)
-                assert author_counts[split] % 100 == 0, (base, author, split)
+                assert author_counts[split] >= BOTTLENECK_TARGET, (base, author, split)
+                assert author_counts[split] % BOTTLENECK_TARGET == 0, (
+                    base, author, split,
+                )
 
 
 def test_chunk_sizes_and_atomic_coverage_are_exact_in_every_split() -> None:
@@ -182,7 +185,7 @@ def test_sentence_publication_contains_no_hypotactic_fields_or_provenance() -> N
         "component_line_index", "parent_sentence_ids", "hypotactic_file",
         "book", "poem_sequence", "line_number",
     }
-    for suffix in (1, 10, 50, 100):
+    for suffix in (1, 5, 10, 50):
         for split in SPLITS:
             path = config_path(REPO_ROOT, "sentence", suffix, split)
             schema = pq.ParquetFile(path).schema_arrow
@@ -197,7 +200,7 @@ def test_sentence_publication_contains_no_hypotactic_fields_or_provenance() -> N
 
 
 def test_metre_publication_omits_redundant_genre_column() -> None:
-    for suffix in (1, 10, 50, 100):
+    for suffix in (1, 5, 10, 50):
         for split in SPLITS:
             path = config_path(METRE_ROOT, "verse", suffix, split)
             assert "genre" not in pq.ParquetFile(path).schema_arrow.names
@@ -221,7 +224,7 @@ def test_metre_chunks_preserve_aligned_metre_and_syllable_line_units() -> None:
         }
         assert all(len(units) == 1 for units in atomic_syllables[split].values())
         assert all(len(units) == 1 for units in atomic_metres[split].values())
-    for suffix in (10, 50, 100):
+    for suffix in (5, 10, 50):
         for split in SPLITS:
             rows = pq.read_table(
                 config_path(METRE_ROOT, "verse", suffix, split),
@@ -358,7 +361,7 @@ def test_metre_syntax_origin_is_explicit_and_provenance_matches() -> None:
             assert "stoicheia_tagger_parser" in sources
     assert atomic_origins == {"predicted"}
 
-    for suffix in (10, 50, 100):
+    for suffix in (5, 10, 50):
         for split in SPLITS:
             origins = set(pq.read_table(
                 config_path(METRE_ROOT, "verse", suffix, split),
@@ -424,7 +427,7 @@ def test_gold_is_diagnostic_only_and_quality_metrics_are_reported() -> None:
 
 
 def test_scansion_is_not_published() -> None:
-    for suffix in (1, 10, 50, 100):
+    for suffix in (1, 5, 10, 50):
         for split in SPLITS:
             path = config_path(METRE_ROOT, "verse", suffix, split)
             assert "scansion" not in pq.ParquetFile(path).schema_arrow.names

tests/test_verse_character_coverage.py

@@ -28,9 +28,9 @@ def normalized_character_count(root: Path, config: str) -> int:
 
 def test_each_publication_has_equal_character_counts_across_task_sizes() -> None:
     publications = (
-        (REPO_ROOT, "sentence", (1, 10, 50, 100)),
-        (REPO_ROOT, "source_held_out", (1, 10, 50, 100)),
-        (REPO_ROOT.parent / "sphragis-metre", "verse", (1, 10, 50, 100)),
+        (REPO_ROOT, "sentence", (1, 5, 10, 50)),
+        (REPO_ROOT, "source_held_out", (1, 5, 10, 50)),
+        (REPO_ROOT.parent / "sphragis-metre", "verse", (1, 5, 10, 50)),
     )
     for root, genre, suffixes in publications:
         counts = {