Update README.md

README.md

@@ -25,16 +25,19 @@ configs:
      hand-transcribed pages; this open release is the 472 that can be shared. -->
 
 A hand-transcribed benchmark for evaluating Tibetan OCR across writing styles and
-technologies. This is the **open-access subset: 472 page images** with ground-truth
-transcriptions and per-page metadata (script, technology, legibility). The full
-benchmark (1,071 pages) additionally has a **held-out private split** kept secret to
-remain leakage-free — so numbers here are directly comparable across systems without
-anyone being able to train on the test set.
+technologies. This is an **open-access set of 472 page images** with ground-truth
+transcriptions and per-page metadata (script, technology, legibility).
 
 Companion to the model [`BDRC/tibetan-ocr`](https://huggingface.co/BDRC/tibetan-ocr)
 and the **[leaderboard](https://huggingface.co/spaces/BDRC/tibetan-ocr-leaderboard)**
 (dozens of OCR systems scored on this benchmark).
 
+The transcriptions were produced by [Dharmaduta](dharmaduta.in).
+
+The images were selected through detailed research on Tibetan writing styles in collaboration
+with Pentsok W. Rtsang who selected and cataloged the images. Most writing styles are represented
+in the benchmark, except those only present in images that cannot be made fully open access.
+
 ## Contents & schema
 
 One row per page. We deliberately keep **only fully-documentable** metadata; the
@@ -78,17 +81,7 @@ Character Error Rate is computed after normalization, both sides:
 4. Remove placeholder characters (`K O B I S`).
 5. `CER = Levenshtein(hyp, ref) / len(ref)` (rapidfuzz).
 
-We also report **SER** (syllable-level edit rate), a **repetition-loop** metric, and
-a **catastrophic-page rate** (pages with CER > 0.5) so that runaway repetition is
-visible rather than hidden in the mean. Aggregate CER is an **unweighted mean over
-pages**.
-
-## Splits & integrity
-
-- `test` = **472 open images** (released here).
-- A **private holdout** (~382) and further withheld images (third-party / non-open)
-  are **not released**, preserving a leakage-free evaluation set. Contact BDRC to run
-  a system against the full benchmark.
+Aggregate CER is an **unweighted mean over pages**.
 
 ## License
 
@@ -101,12 +94,11 @@ pages**.
 ```bibtex
 @misc{bdrc_tibetan_ocr_benchmark_2026,
   title  = {BDRC Tibetan OCR Benchmark},
-  author = {Roux, Elie and Werner, Eric and Buddhist Digital Resource Center},
+  author = {Buddhist Digital Resource Center},
   year   = {2026},
   howpublished = {Hugging Face},
   note   = {https://huggingface.co/datasets/BDRC/tibetan-ocr-benchmark}
 }
 ```
 
-Funded by the **Khyentse Foundation** ("The BDRC Etext Corpus"). Numbers are
-**preliminary (v1)** and may be updated.
+Funded by the **Khyentse Foundation** ("The BDRC Etext Corpus").