Update questions, pages, answers and document files

README.md

@@ -13,8 +13,8 @@ configs:
     path: questions/multi_hop.parquet
   - split: rag_misses
     path: questions/rag_misses.parquet
-  - split: unseen
-    path: questions/unseen.parquet
+  - split: test
+    path: questions/test.parquet
 - config_name: pages
   data_files:
   - split: train
@@ -34,15 +34,15 @@ harnesses, and every answer, score and tool call is here. Variant: **public**.
 | --- | ---: | --- |
 | questions | 50 | Question, gold answer, scoring rule, evidence pages. Splits follow the evaluation stages. |
 | pages | 71 | Page image, converted Markdown, source URL, licence, the questions that selected the page. |
-| answers | 1885 | Model, quantization, provider, seed, configuration, repeat, answer, score, tool calls, seconds, tokens, full tool trace. |
+| answers | 2845 | Model, quantization, provider, seed, configuration, repeat, answer, score, tool calls, seconds, tokens, full tool trace. |
 
 ## Evaluation stages
 
-- `rag_misses`, `multi_hop`: tuning questions. The agent was changed after reading their traces.
-- `holdout`: ten questions held back from the first agent, then used to change the harness. Treat them as tuning data
+- `rag_misses`, `multi_hop`: validation questions. The agent was changed after reading their traces.
+- `holdout`: ten questions held back from the first agent, then used to change the harness. Treat them as validation data
   for any later revision.
-- `unseen`: twenty questions selected before their first run. No harness change came from their results or traces.
-  Fifteen ran first as `holdout2`; five more were added to match the twenty tuning questions.
+- `test`: twenty questions selected before their first run. No harness change came from their results or traces.
+  Fifteen ran first as `holdout2`; five more were added to match the twenty validation questions. Validation and test search the same 531 pages. In `answers`, corpus `pages_validation` holds the validation runs and `pages_unseen` the test runs; `pages_multihop` holds earlier validation runs on 222 of those pages, where the harness was tuned.
 
 ## Retrieval and models
 
@@ -54,7 +54,7 @@ harnesses, and every answer, score and tool call is here. Variant: **public**.
 - Answer model: Qwen3.5-9B at temperature 0. Final runs use the 4-bit (fp4) weights on Darkbloom through OpenRouter,
   `qwen/qwen3.5-9b`, three runs. Earlier runs use the nvfp4 weights in Ollama, `qwen3.5:9b`, one run. Each row of `answers`
   names its `model`, `quantization`, `hosted_provider` and the `seed` sent; at temperature 0 the seed has little effect.
-- Iterations: Improving Correctness is configuration `merged_grep_revision2`, Improving Speed is `merged_grep_revision3`, and
+- Iterations: Trace Fixes is configuration `merged_grep_revision2`, Token Reduction is `merged_grep_revision3`, and
   Thinking is `merged_grep_revision3_thinking`.
 
 ## Files
@@ -68,7 +68,7 @@ lists every document with its source URL, including those not hosted here. `page
 
 ## Variant notes
 
-Every question, gold answer, scoring rule, evidence quote, model answer and score is here. Five XL-DocBench questions, four of them in `unseen`, come from an earlier release of that benchmark. Pages with `redistribution` other than `permitted` are excluded, because their source licences do not allow copies; `source_url` and `open_page_url` lead to them. Tool outputs and prompts that reproduce those documents read `[omitted from the public variant: source licence]`.
+Every question, gold answer, scoring rule, evidence quote, model answer and score is here. Five XL-DocBench questions, four of them in `test`, come from an earlier release of that benchmark. Pages with `redistribution` other than `permitted` are excluded, because their source licences do not allow copies; `source_url` and `open_page_url` lead to them. Tool outputs and prompts that reproduce those documents read `[omitted from the public variant: source licence]`.
 
 ## Sources and attribution
 

answers/train.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5f01739491d4d52fa80853e081c2f1ab8b10ca0a091afef7ba17f09155bc767d
-size 348053
+oid sha256:89e147f8cbd2b020ab8063f696f82a4fc8be3a9729c25eba85de09c6b3520831
+size 548851

questions/{unseen.parquet → test.parquet}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c4c98fd6b2656e3ac0e753c226561a2302f985410c9c66fc12c720c6a94a231
-size 13526
+oid sha256:81d5fee5d6d319ae7af00c9cf893836241a566ddd620f7e34f990c7f61b14219
+size 13516