Describe evaluation content and distribution

README.md

@@ -23,27 +23,57 @@ configs:
 
 # Fluid 2 — dictation cleanup eval
 
-This repository is the compact, text-only copy of the exact **7,161-row**
-development partition used by the Fluid 2 Qwen3.5 0.8B training run. It is
-intended for reproducible comparison of Fluid 2 checkpoints without fetching
-or decoding the parent corpus's large audio payloads.
+This repository contains **7,161 text-only voice-dictation cleanup evaluation
+rows** for like-for-like model comparison. No audio is included or fetched.
 
 | Split | Rows | Documents | Audio |
 | --- | ---: | ---: | --- |
 | `eval` | 7,161 | 2,718 | Not included |
 
-The rows are the original deterministic document-grouped 2% assignment from
-[`johnbean393/fluid-2-sft-asr`](https://huggingface.co/datasets/johnbean393/fluid-2-sft-asr).
-The split algorithm is SHA-256 over `document_id` with seed `20260813`; no
-document overlaps the parent dataset's train or test partitions. Exact file,
-row-order, and order-independent membership hashes are recorded in
-[`split_manifest.json`](split_manifest.json).
+## What this benchmark tests
 
-> This is a development/evaluation set, not the final blind test set. It is
-> evaluated 20 times during each Fluid 2 training run, so its metrics are useful
-> for model selection and like-for-like comparisons but should not be presented
-> as an unbiased final test result. The separate 7,152-row `test` split in the
-> parent dataset remains untouched.
+The benchmark measures whether a model can turn noisy voice dictation into the
+intended written text without answering it or adding content. The rows cover:
+
+- local ASR, spelling, capitalization, punctuation, and grammar repair;
+- filler removal, abandoned starts, cancellations, and correction chains;
+- spoken punctuation plus line, paragraph, and list formatting;
+- short notes and questions through long, multi-paragraph passages; and
+- dictation with no surrounding context, one-sided context, or both previous
+  and following context fields.
+
+The 121 empty-target rows represent fully cancelled dictation and test whether
+the model stops without emitting cleaned text.
+
+## Measured content distribution
+
+The following counts are computed over all 7,161 rows. Target-length counts
+include the 121 empty targets.
+
+| Clean-target length | Rows | Share |
+| --- | ---: | ---: |
+| Empty | 121 | 1.69% |
+| 1–50 characters | 3,198 | 44.66% |
+| 51–150 characters | 2,601 | 36.32% |
+| 151–500 characters | 1,123 | 15.68% |
+| 501+ characters | 118 | 1.65% |
+
+| Available context | Rows | Share |
+| --- | ---: | ---: |
+| Neither side | 2,069 | 28.89% |
+| Previous only | 2,665 | 37.22% |
+| Following only | 655 | 9.15% |
+| Previous and following | 1,772 | 24.75% |
+
+The median ASR input is 14 words (95th percentile: 53; maximum: 596). Among
+non-empty targets, the median is 11 words (95th percentile: 44; maximum: 427).
+Of the 7,040 non-empty targets, 6,764 (96.08%) differ exactly from their ASR
+input; 5,068 are shorter, 1,389 are longer, and 583 have the same character
+length while still potentially differing in content.
+
+Formatting behaviors are materially represented and the categories overlap:
+1,090 targets contain a line break, 562 contain a blank-line paragraph break,
+203 contain a Markdown-style list item, and 423 end in a question mark.
 
 ## Schema and loading
 
@@ -104,10 +134,6 @@ completion-token limit are likewise reported separately and excluded from those
 text-quality scores. Throughput and inference-failure counts still cover the full
 7,161-request census.
 
-The corpus combines sources with different upstream terms. The parent dataset
-card documents provenance and license labels; this repository does not replace
-those terms with a blanket license.
-
 <!-- FLUID2_EVAL_RESULTS_START -->
 ## Published model results