feat: add Cost column to leaderboard (aggregate per-run cost)

scripts/generate_leaderboard.py

@@ -167,6 +167,11 @@ def compute_stats(records: list[dict]) -> dict:
     failures = sum(1 for r in records if r.get("evaluation_status") in (
         "failure_api_or_parse", "failure_unexpected_type", "error_bad_ground_truth"))
 
+    # Cost is None (not 0) when no record logged it, so the leaderboard can
+    # distinguish "free" from "predates cost logging" and render "?".
+    cost_vals = [r.get("cost") for r in records if r.get("cost") is not None]
+    cost = round(sum(cost_vals), 4) if cost_vals else None
+
     return {
         "score": total_score,
         "correct": correct,
@@ -174,6 +179,7 @@ def compute_stats(records: list[dict]) -> dict:
         "incorrect": incorrect,
         "skipped": skipped,
         "failures": failures,
+        "cost": cost,
         "num_questions": len(records),
     }
 
@@ -190,6 +196,7 @@ def load_summary_json(filepath: str) -> dict | None:
             "incorrect": data.get("overall_incorrect", data.get("overall_incorrect_choice", 0)),
             "skipped": data.get("overall_skipped", 0),
             "failures": data.get("overall_api_parse_failures", 0),
+            "cost": data.get("total_cost_usd"),
             "num_questions": data.get("total_questions_processed", 0),
         }
     except (OSError, json.JSONDecodeError, KeyError):

scripts/leaderboard_template.html

@@ -49,6 +49,7 @@ a{color:var(--accent);}
 <th data-k="model">Model</th>
 <th data-k="score" class="num">Score</th>
 <th data-k="pct" class="num">%</th>
+<th data-k="cost" class="num">Cost</th>
 <th data-k="correct" class="num">Correct</th>
 <th data-k="partial" class="num">Partial</th>
 <th data-k="incorrect" class="num">Incorrect</th>
@@ -90,11 +91,12 @@ function fillYears(){
 
 function render(){
   let rows = entries.filter(e => e.exam === state.exam && String(e.year) === String(state.year));
+  const sortVal = (e) => state.sort === 'pct' ? (pct(e) ?? -1)
+                       : state.sort === 'cost' ? (e.cost ?? -1)
+                       : (e[state.sort] ?? 0);
   rows.sort((a,b) => {
     if(state.sort === 'model') return state.dir * a.model.localeCompare(b.model);
-    let va = state.sort === 'pct' ? (pct(a) ?? -1) : (a[state.sort] ?? 0);
-    let vb = state.sort === 'pct' ? (pct(b) ?? -1) : (b[state.sort] ?? 0);
-    return state.dir * (va - vb);
+    return state.dir * (sortVal(a) - sortVal(b));
   });
 
   const fn = []; const fnIndex = {};
@@ -124,6 +126,7 @@ function render(){
       '<td title="' + title.replace(/"/g,'&quot;') + '">' + e.model + badge + lab + sup + '</td>' +
       '<td class="num">' + e.score + (e.max_score ? (' / ' + e.max_score) : '') + '</td>' +
       '<td class="num">' + (p !== null ? p.toFixed(1) : '?') + '</td>' +
+      '<td class="num">' + (e.cost != null ? '$' + e.cost.toFixed(2) : '?') + '</td>' +
       '<td class="num">' + e.correct + '</td>' +
       '<td class="num">' + e.partial + '</td>' +
       '<td class="num">' + e.incorrect + '</td>' +

tests/test_leaderboard_html.py

@@ -48,6 +48,19 @@ def test_render_html_missing_placeholder_raises(tmp_path):
         gl.render_html([], str(template), "2026-06-20")
 
 
+def test_compute_stats_sums_cost():
+    records = [
+        {"cost": 0.01, "marks_awarded": 4, "evaluation_status": "correct"},
+        {"cost": 0.02, "marks_awarded": 0, "evaluation_status": "incorrect"},
+    ]
+    assert gl.compute_stats(records)["cost"] == 0.03
+
+
+def test_compute_stats_cost_none_when_absent():
+    records = [{"marks_awarded": 4, "evaluation_status": "correct"}]
+    assert gl.compute_stats(records)["cost"] is None
+
+
 def test_render_html_serializes_yaml_date_objects(tmp_path):
     # PyYAML parses unquoted ISO dates into datetime.date, which json can't
     # serialize natively. render_html must handle it.