results.parquet: every parser's verdict on every test; card wording

README.md

@@ -16,6 +16,10 @@ configs:
   data_files:
   - split: test
     path: manifest.parquet
+- config_name: results
+  data_files:
+  - split: test
+    path: results.parquet
 ---
 
 # Omni Parse Bench
@@ -24,13 +28,14 @@ Choosing a document parser usually starts with a leaderboard: one number per par
 average document, and the parser that wins on average can be the one that fails on yours. Omni Parse Bench
 is built to answer the question you actually have: which parser is best on documents like mine? It is:
 
-- simple to reason about: every test is a yes/no question about one page (is this content in the output, in
-  this order, in a table cell under these headings, in a layout block of this kind?), and every score is the
-  share of tests passed;
+- clear: every test is a yes/no question about one page (is this content in the output, in this order, in a
+  table cell under these headings, in a layout block of this kind?), every score is the share of tests
+  passed, and every score comes with the number of tests and documents behind it, so a thin slice says it
+  is thin;
+- auditable: every test records how its answer was established, and the scorer is open code with no model
+  judging the outputs; and
 - explorable: every test is tagged by what its content is (a table cell, handwriting, a rotated page, Thai),
-  so any slice of the benchmark gets its own score; and
-- clear about its gaps: every score comes with the number of tests and documents behind it, so a thin slice
-  says it is thin.
+  so any slice of the benchmark gets its own score.
 
 It has 16,288 tests on 2,937 pages from 2,343 documents over 22 suites, in
 92 languages. Omni Parse Bench is developed by Datalab.
@@ -96,11 +101,23 @@ tagged pieces of content and found no wrong tag.
 | Azure | 83.0 | 74.6 | 81.7 | 92.8 |
 | Tesseract | none | 17.6 | 0.0 | unsupported |
 
-\* Used in creating and verifying at least some tests, so not ranked.
+\* Used in creating and verifying at least some tests.
 
 The headline is the mean of three family scores: text (`present`, `order` and `repeat` tests), tables
 (`table_cell`) and layout (`layout_kind`). Tesseract returns no layout, so it has no headline.
 
+Every verdict behind these numbers is in `results.parquet`: one row per parser and test, with the test's
+page, document, suite, type, derivation and tags, its `result` (`pass`, `fail`, or `unsupported` when the
+parser can't produce that output) and `why` it failed. Browse it as the `results` subset in the viewer
+above, or query it; for instance, the share of handwriting tests each parser passes:
+
+```sql
+select system, round(100 * avg((result = 'pass')::int), 1) as score, count(*) as tests
+from 'hf://datasets/datalab-to/omni_parse_bench/results.parquet'
+where result in ('pass', 'fail') and list_contains(tags, 'handwriting')
+group by system order by score desc;
+```
+
 ![Headline against median seconds per page](./assets/latency.svg)
 
 ![Headline against price per 1,000 pages](./assets/cost.svg)
@@ -113,6 +130,7 @@ The headline is the mean of three family scores: text (`present`, `order` and `r
 manifest.parquet        one row per page
 pages/<page_id>.pdf     the page that gets sent: a one-page PDF, or a .png for an image page
 gold/<page_id>.json     its tests: {"tests": [...]}
+results.parquet         every parser's verdict on every test (Results above)
 licenses/               the license, and the credit it requires, of each source
 assets/                 the charts on this card
 ```

results.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d66a39b952cfd97f89b60fe03515d033fe0045794e953c031f6d5ae97e49c41a
+size 1679444