Sharpen frontier benchmark positioning

Update the title, hero, and paper framing to match the rebuilt FlavourBench release.

README.md

@@ -8,13 +8,13 @@ sdk_version: 6.9.0
 app_file: app.py
 pinned: false
 license: other
-short_description: Culinary reasoning for frontier LLMs, without a model judge.
+short_description: The executable culinary benchmark for frontier LLMs.
 ---
 
-# FlavourBench: An Executable Benchmark for Culinary Reasoning Without a Model Judge
+# FlavourBench: Executable Culinary Evaluation of Frontier Language Models
 
-An executable benchmark and evidence explorer for 20 current frontier language-model endpoints, scored against
-executable culinary answer keys without a human or model judge.
+The executable culinary benchmark and evidence explorer for 20 frontier language-model endpoints,
+scored against answer keys computed before evaluation.
 
 [Paper](https://github.com/josefchen/flavourbench/blob/main/paper/build/flavourbench.pdf) ·
 [Dataset](https://huggingface.co/datasets/josefchen/flavourbench) ·

app.py

@@ -313,8 +313,8 @@ def _hero_html() -> str:
     return f"""
     <div class="fb-shell fb-hero">
       <section>
-        <div class="fb-kicker">Frontier culinary reasoning benchmark · 20 endpoints</div>
-        <h1>Culinary reasoning without a model judge.</h1>
+        <div class="fb-kicker">The executable culinary benchmark · 20 frontier endpoints</div>
+        <h1>How well do frontier models reason about flavour?</h1>
         <p class="fb-dek">FlavourBench scores 20 current language-model endpoints against answer
         keys compiled by Epicure.
         <strong>Blue is the FlavourBench Score. Gray shows its Wilson 95% interval.</strong>
@@ -510,7 +510,7 @@ theme = gr.themes.Base(
 )
 
 
-with gr.Blocks(title="FlavourBench · Frontier culinary reasoning benchmark") as demo:
+with gr.Blocks(title="FlavourBench · Executable culinary benchmark") as demo:
     gr.HTML(_hero_html())
 
     with gr.Tabs():