Simplify prerelease dataset card

README.md

@@ -12,38 +12,35 @@ size_categories:
 - n<1K
 ---
 
-# 2D WebMCP Browser Focus — Prerelease
+# 2D WebMCP Browser Focus (Prerelease)
 
-This prerelease benchmark evaluates whether agent-visible tool results support correct, browser-verified focus outcomes in a workflow canvas.
+## What this is
 
-Each trial asks an agent to create a Retry node, atomically replace the affected workflow connections, reveal the new node, and verify focus. Success requires all of the following:
+This is an early test of whether agents need useful tool results to complete an accessible browser task.
 
-- the Retry item exists;
-- its three expected relationships exist;
-- the browser's active element is the Retry item;
-- its role is `group` and its accessible name is `Retry node: Retry`;
-- the app receipt reports that the requested node was focused and visible.
+The agent must add a Retry step to a workflow, connect it correctly, and move keyboard focus to that new step. The test checks the real browser, not just the agent's final answer.
 
-## Results
+## What happened
 
-The release contains 20 matched trials per arm using `gpt-5-mini` with low reasoning effort.
+We ran each version 20 times with `gpt-5-mini` using low reasoning effort.
 
-| Tool-result arm | Verified task success |
+| Tool result | Verified success |
 | --- | ---: |
-| Success/failure only | 0% (0/20) |
-| Generic structured result | 95% (19/20) |
-| Compact Surface contract | 95% (19/20) |
+| Success or failure only | 0% (0/20) |
+| Simple structured result | 95% (19/20) |
+| Compact Surface result | 95% (19/20) |
 
-The success/failure-only arm produced plausible final answers in every trial despite no verified canvas or focus success. This supports the narrower claim that meaningful tool results and independent browser verification are necessary for this task. In this one task/model configuration, the compact Surface contract did not improve verified success over the concise generic result and used more input tokens.
+The success-or-failure version often gave a confident answer, but never completed and verified the real task. A useful result made a big difference.
 
-## Data
+For this task, the compact Surface result did not beat the simpler structured result. It also used more input tokens.
 
-`data/test.jsonl` contains 60 scored trials. Each row includes the benchmark prompt, variant, model configuration, tool-call transcript, final browser state, and deterministic outcome metrics. Private LangSmith identifiers have been removed.
+## Files
 
-`report.md` contains aggregate results and matched comparisons.
+- `data/test.jsonl` has all 60 scored runs. It includes tool calls, final browser state, and scores. Private LangSmith IDs were removed.
+- `report.md` has the full tables and comparisons.
 
-## Limitations
+## Limits
 
-- This is a prerelease, based on one application fixture, one task, and one model setting.
-- It is browser-automation evidence, not a screen-reader usability study; it does not establish user benefit without research involving screen-reader users.
-- The underlying benchmark worktree was dirty at the time of the run. A future version will be tied to a committed source revision.
+- This is a prerelease. It covers one app, one task, and one model setting.
+- It is an automated browser test, not a study with screen-reader users.
+- The code was not committed when this run was made. A later release will point to a committed version.