Retitle FlavourBench around executable evaluation

README.md

@@ -8,12 +8,13 @@ sdk_version: 6.9.0
 app_file: app.py
 pinned: false
 license: other
-short_description: Inspect how Epicure changes frontier model decisions.
+short_description: Executable culinary reasoning without a model judge.
 ---
 
-# FlavourBench Explorer
+# FlavourBench: An Executable Benchmark for Culinary Reasoning Without a Model Judge
 
-An evidence-first explorer for the FlavourBench Epicure-native automated pilot.
+An evidence explorer and leaderboard for 20 current language-model endpoints, scored against
+executable culinary answer keys without a human or model judge.
 
 [Paper](https://github.com/josefchen/flavourbench/blob/main/paper/build/flavourbench.pdf) ·
 [Dataset](https://huggingface.co/datasets/josefchen/flavourbench) ·

app.py

@@ -277,10 +277,11 @@ def _hero_html() -> str:
     return f"""
     <div class="fb-shell fb-hero">
       <section>
-        <div class="fb-kicker">Epicure-native automated pilot</div>
-        <h1>Measure the<br>tool effect.</h1>
-        <p class="fb-dek">FlavourBench asks every model the same culinary question twice.
-        <strong>Blue is what the model knows alone. Gold is what Epicure adds.</strong>
+        <div class="fb-kicker">Frontier culinary reasoning benchmark · 20 endpoints</div>
+        <h1>Culinary reasoning without a model judge.</h1>
+        <p class="fb-dek">FlavourBench scores 20 current language-model endpoints against answer
+        keys compiled by Epicure.
+        <strong>Blue measures Model only. Gold measures the gain from Epicure.</strong>
         Open any pair to inspect the prompt, answers, tool trace, and hashes.</p>
         <div class="fb-stats">
           <div class="fb-stat"><strong>{counts["models"]}</strong><span>models</span></div>
@@ -289,11 +290,11 @@ def _hero_html() -> str:
           <div class="fb-stat"><strong>{counts["observed_response_arms"]:,}</strong><span>observed arms</span></div>
         </div>
       </section>
-      <section class="fb-frontier" aria-label="Tool-off scores and Epicure uplift">
-        <div class="fb-frontier-head"><strong>Baseline plus Epicure uplift</strong><span>Top 12 · percent correct</span></div>
+      <section class="fb-frontier" aria-label="FlavourBench Score and Epicure Gain">
+        <div class="fb-frontier-head"><strong>FlavourBench Score plus Epicure Gain</strong><span>Top 12 · percent correct</span></div>
         {_frontier_html()}
         <div class="fb-note">One task equals 3.125 percentage points. Read adjacent rows as a
-        pilot score group, then inspect the underlying pairs.</div>
+        close score group, then inspect the underlying pairs.</div>
       </section>
     </div>
     """
@@ -309,9 +310,9 @@ def _leaderboard_frame() -> pd.DataFrame:
             {
                 "Rank": model["rank"],
                 "Model": model["display_name"],
-                "Tool off": f"{off['accuracy_percent']:.3g}%",
-                "Tool on": f"{on['accuracy_percent']:.3g}%",
-                "Uplift": f"+{model['uplift_percentage_points']:.3g} pp",
+                "Model only": f"{off['accuracy_percent']:.3g}%",
+                "Model + Epicure": f"{on['accuracy_percent']:.3g}%",
+                "Epicure Gain": f"+{model['uplift_percentage_points']:.3g} pp",
                 "Observed arms": f"{observed}/64",
                 "Backend": model["execution_backend"],
             }
@@ -326,14 +327,14 @@ def _model_detail(model_name: str) -> tuple[str, pd.DataFrame]:
     summary = f"""
     <div class="fb-pair-status">
       <div class="fb-condition good">
-        <small>Tool-off exact-choice score</small>
+        <small>FlavourBench Score</small>
         <strong>{off["accuracy_percent"]:.3g}%</strong>
         <span>Wilson 95%: {off["wilson_95"][0] * 100:.1f}% to {off["wilson_95"][1] * 100:.1f}%</span>
       </div>
       <div class="fb-condition good">
-        <small>Epicure-assisted score</small>
+        <small>Model + Epicure accuracy</small>
         <strong>{on["accuracy_percent"]:.3g}%</strong>
-        <span>Uplift: +{model["uplift_percentage_points"]:.3g} percentage points</span>
+        <span>Epicure Gain: +{model["uplift_percentage_points"]:.3g} percentage points</span>
       </div>
     </div>
     """
@@ -342,8 +343,8 @@ def _model_detail(model_name: str) -> tuple[str, pd.DataFrame]:
         family_rows.append(
             {
                 "Family": family.title(),
-                "Tool off": f"{off['family_accuracy'][family] * 100:.1f}%",
-                "Tool on": f"{on['family_accuracy'][family] * 100:.1f}%",
+                "Model only": f"{off['family_accuracy'][family] * 100:.1f}%",
+                "Model + Epicure": f"{on['family_accuracy'][family] * 100:.1f}%",
                 "Change": (
                     f"{(on['family_accuracy'][family] - off['family_accuracy'][family]) * 100:+.1f} pp"
                 ),
@@ -400,8 +401,8 @@ def _pair_detail(
     on = _observation(model["model_id"], task_id, "epicure_on")
     status = (
         "<div class='fb-pair-status'>"
-        + _status_card("Epicure off", off)
-        + _status_card("Epicure on", on)
+        + _status_card("Model only", off)
+        + _status_card("Model + Epicure", on)
         + "</div>"
     )
     reference = json.dumps(
@@ -471,7 +472,7 @@ theme = gr.themes.Base(
 )
 
 
-with gr.Blocks(title="FlavourBench · Epicure evidence explorer") as demo:
+with gr.Blocks(title="FlavourBench · Frontier culinary reasoning benchmark") as demo:
     gr.HTML(_hero_html())
 
     with gr.Tabs():
@@ -480,8 +481,8 @@ with gr.Blocks(title="FlavourBench · Epicure evidence explorer") as demo:
                 """
                 <div class="fb-section-title">
                   <div class="fb-kicker">Automated exact-choice track</div>
-                  <h2>The public pilot, without hidden rows</h2>
-                  <p>Rank follows tool-off score. Tool-on and uplift show the paired Epicure effect.</p>
+                  <h2>The complete public benchmark</h2>
+                  <p>Rank follows FlavourBench Score. Model + Epicure and Epicure Gain show the matched intervention.</p>
                 </div>
                 """
             )
@@ -494,8 +495,8 @@ with gr.Blocks(title="FlavourBench · Epicure evidence explorer") as demo:
                 column_widths=[55, 260, 95, 95, 105, 110, 110],
             )
             gr.Markdown(
-                "**Reading the table.** A rank is a deterministic ordering of this 32-task pilot, "
-                "not a claim about general model quality. Equal scores are ordered by the release's "
+                "**Reading the table.** Rank follows Model only accuracy over this 32-task release; "
+                "it is not a claim about general model quality. Equal scores follow the release's "
                 "frozen tie-break rules. Use Pair Lens before interpreting small differences."
             )
 
@@ -565,8 +566,8 @@ with gr.Blocks(title="FlavourBench · Epicure evidence explorer") as demo:
             )
             choices = gr.JSON(value=initial_pair[2], label="Choices")
             with gr.Row():
-                off_answer = gr.Markdown(value=initial_pair[3], label="Epicure off answer")
-                on_answer = gr.Markdown(value=initial_pair[4], label="Epicure on answer")
+                off_answer = gr.Markdown(value=initial_pair[3], label="Model only answer")
+                on_answer = gr.Markdown(value=initial_pair[4], label="Model + Epicure answer")
             tool_trace = gr.JSON(value=initial_pair[5], label="Observed Epicure trace")
             reference = gr.Textbox(
                 value=initial_pair[6],
@@ -599,9 +600,10 @@ with gr.Blocks(title="FlavourBench · Epicure evidence explorer") as demo:
                 <div class="fb-method-grid">
                   <div>
                     <h3>Scoring contract</h3>
-                    <p><strong>Epicure-native score</strong> is exact-choice accuracy with Epicure
-                    disabled. <strong>Tool-assisted score</strong> uses the same model-task cells with
-                    Epicure available. <strong>Uplift</strong> is the paired percentage-point change.</p>
+                    <p><strong>FlavourBench Score</strong> is Model only exact-choice accuracy over
+                    all 32 tasks. <strong>Model + Epicure</strong> uses the same endpoint-task cells
+                    with one named Epicure operation. <strong>Epicure Gain</strong> is the matched
+                    percentage-point change and does not affect rank.</p>
                     <p>Tasks cover substitution, composition, cookability, and evidence. Every expected
                     answer is derived from a fixed read-only Epicure operation.</p>
                     <h3>Public records</h3>
@@ -642,7 +644,7 @@ public research preview.
     gr.HTML(
         """
         <div class="fb-shell fb-footer">
-          FlavourBench · Evidence-first culinary model evaluation · Public automated pilot
+          FlavourBench · Executable culinary evaluation · Public automated benchmark
         </div>
         """
     )

data/epicure-native-release.json

@@ -1,5 +1,5 @@
 {
-  "artifact_sha256": "603e4aa4db86a6e28900736322943dde2593c9e1ffd473f26b9a469bcd5c912a",
+  "artifact_sha256": "197b84d75ccaa549650ad08cc88f934a4ac1173c8e51c9fb149a703ce2602ce8",
   "benchmark": "FlavourBench",
   "counts": {
     "assigned_arms": 1280,
@@ -20,7 +20,7 @@
       "reference_tool_match_pairs": 606,
       "uplift_percentage_points": 50.0
     },
-    "artifact_sha256": "6153fc38fb27107386b127562809762facb9729807c8a9f8a5e91ed499cfad4e",
+    "artifact_sha256": "6076a904af22c6d58b11cb6a0b267431dbfeff639ab511bf6c936d9cf55eebf4",
     "benchmark": "FlavourBench",
     "design": {
       "conditions": [
@@ -1483,16 +1483,16 @@
     "official_track": true,
     "primary_metric": {
       "chance_level": 25,
-      "definition": "100 times equal-family exact-choice accuracy against Epicure answers, without Epicure tool access",
-      "name": "Epicure Benchmark Score",
+      "definition": "100 times Model only correct answers divided by 32, against the published Epicure answer keys",
+      "name": "FlavourBench Score",
       "range": [
         0,
         100
       ],
       "rank_order": [
-        "Epicure Benchmark Score descending",
-        "Epicure-on accuracy descending",
-        "Epicure-on reliability descending",
+        "FlavourBench Score descending",
+        "Model + Epicure accuracy descending",
+        "Model + Epicure completion descending",
         "model ID ascending"
       ]
     },
@@ -34493,7 +34493,7 @@
       "release_id": "exploratory-unmatched-1790-runtime",
       "schema_version": "epicure-runtime-provenance-v1"
     },
-    "leaderboard_artifact_sha256": "6153fc38fb27107386b127562809762facb9729807c8a9f8a5e91ed499cfad4e",
+    "leaderboard_artifact_sha256": "6076a904af22c6d58b11cb6a0b267431dbfeff639ab511bf6c936d9cf55eebf4",
     "manifest_sha256": "44220f0a6c26798871f830f6ddd62ed99d8872ad5dcb7516491b90bfc71887fc",
     "task_set_sha256": "c033bf8e349350b186a47b28c94d4b0f356cbb519d0cffb3d5a7389ce7d7ca2f",
     "taskset_artifact_sha256": "615e6164a29ea57c507d979934f3ce484d6b71f4d6da55e9e8d338147c68e536"