fix cost

app.py

@@ -347,6 +347,7 @@ def build_demo(entries=None):
                         "Tasks": f"{entry.passed_tasks}/{entry.total_tasks}",
                         "Avg duration": round2(entry.avg_duration_seconds),
                         "Total duration": round2(entry.total_duration_seconds),
+                        "Total steps": entry.total_agent_steps,
                         "Avg steps": round2(entry.avg_agent_steps),
                         "Tokens/task": round2(entry.avg_tokens_per_task),
                         "Total tokens": entry.total_tokens,

results/deepseek_deepseek-v4-flash__browser-use.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.8235294117647058,
     "agent_dab_agreement_rate": 0.6274509803921569,
     "section_avg_rate": 0.5166143380429095,
-    "total_duration_seconds": 21972.302697375417
+    "total_duration_seconds": 21972.302697375417,
+    "total_agent_steps": 616
   },
   "sections": {
     "shop": {

results/deepseek_deepseek-v4-flash__openhands.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.49019607843137253,
     "agent_dab_agreement_rate": 0.7647058823529411,
     "section_avg_rate": 0.3830455259026687,
-    "total_duration_seconds": 10914.631350658834
+    "total_duration_seconds": 10914.631350658834,
+    "total_agent_steps": 967
   },
   "sections": {
     "shop": {

results/deepseek_deepseek-v4-flash__openmanus.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.82,
     "agent_dab_agreement_rate": 0.74,
     "section_avg_rate": 0.5972004186289901,
-    "total_duration_seconds": 18126.616548523307
+    "total_duration_seconds": 18126.616548523307,
+    "total_agent_steps": 602
   },
   "sections": {
     "shop": {

results/deepseek_deepseek-v4-flash__ouroboros-cut.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.8235294117647058,
     "agent_dab_agreement_rate": 0.6666666666666666,
     "section_avg_rate": 0.5385923600209315,
-    "total_duration_seconds": 24993.06771468371
+    "total_duration_seconds": 24993.06771468371,
+    "total_agent_steps": 592
   },
   "sections": {
     "shop": {

results/deepseek_deepseek-v4-flash__ouroboros-full-evolving.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.7058823529411765,
     "agent_dab_agreement_rate": 0.7843137254901961,
     "section_avg_rate": 0.5300889586603873,
-    "total_duration_seconds": 50013.8336566519
+    "total_duration_seconds": 50013.8336566519,
+    "total_agent_steps": 1159
   },
   "sections": {
     "shop": {

results/deepseek_deepseek-v4-flash__ouroboros-full-isolated.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.6862745098039216,
     "agent_dab_agreement_rate": 0.7450980392156863,
     "section_avg_rate": 0.64141810570382,
-    "total_duration_seconds": 27027.0369386971
+    "total_duration_seconds": 27027.0369386971,
+    "total_agent_steps": 866
   },
   "sections": {
     "shop": {

results/google_gemini-2.5-flash__browser-use.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.8235294117647058,
     "agent_dab_agreement_rate": 0.5882352941176471,
     "section_avg_rate": 0.45866038723181585,
-    "total_duration_seconds": 11037.299745082855
+    "total_duration_seconds": 11037.299745082855,
+    "total_agent_steps": 723
   },
   "sections": {
     "shop": {

results/google_gemini-2.5-flash__openhands.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.9607843137254902,
     "agent_dab_agreement_rate": 0.13725490196078433,
     "section_avg_rate": 0.2099686028257457,
-    "total_duration_seconds": 11133.858564271126
+    "total_duration_seconds": 11133.858564271126,
+    "total_agent_steps": 523
   },
   "sections": {
     "shop": {

results/google_gemini-2.5-flash__openmanus.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.8235294117647058,
     "agent_dab_agreement_rate": 0.6470588235294118,
     "section_avg_rate": 0.5173338566195709,
-    "total_duration_seconds": 10862.327687103301
+    "total_duration_seconds": 10862.327687103301,
+    "total_agent_steps": 659
   },
   "sections": {
     "shop": {

results/google_gemini-2.5-flash__ouroboros-cut.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.8235294117647058,
     "agent_dab_agreement_rate": 0.6078431372549019,
     "section_avg_rate": 0.4696493982208268,
-    "total_duration_seconds": 10506.739309936762
+    "total_duration_seconds": 10506.739309936762,
+    "total_agent_steps": 726
   },
   "sections": {
     "shop": {

results/google_gemini-2.5-flash__ouroboros-full-evolving.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.6862745098039216,
     "agent_dab_agreement_rate": 0.45098039215686275,
     "section_avg_rate": 0.18524332810047098,
-    "total_duration_seconds": 8925.054827867076
+    "total_duration_seconds": 8925.054827867076,
+    "total_agent_steps": 579
   },
   "sections": {
     "shop": {

results/google_gemini-2.5-flash__ouroboros-full-isolated.json

@@ -21,7 +21,8 @@
     "agent_completion_rate": 0.5882352941176471,
     "agent_dab_agreement_rate": 0.47058823529411764,
     "section_avg_rate": 0.17857142857142858,
-    "total_duration_seconds": 30995.941995203495
+    "total_duration_seconds": 30995.941995203495,
+    "total_agent_steps": 679
   },
   "sections": {
     "shop": {

src/export_results.py

@@ -189,6 +189,18 @@ def backfill_summary_costs(results_dir: Path | None = None) -> int:
     return updated
 
 
+def _total_agent_steps_from_payload(payload: dict[str, Any]) -> int | None:
+    run = payload.get("run") or {}
+    total = run.get("total_agent_steps")
+    if isinstance(total, int):
+        return total
+    if isinstance(total, float):
+        return int(total)
+    tests = payload.get("tests") or []
+    steps = [row["agent_steps"] for row in tests if isinstance(row.get("agent_steps"), int)]
+    return sum(steps) if steps else None
+
+
 def export_entry(payload: dict[str, Any], *, source_path: str | None = None) -> LeaderboardEntry:
     config = load_bench_config()
     run = payload.get("run") or {}
@@ -241,6 +253,7 @@ def export_entry(payload: dict[str, Any], *, source_path: str | None = None) ->
         total_tasks=int(run.get("total_tasks") or 0),
         avg_duration_seconds=run.get("avg_duration_seconds"),
         total_duration_seconds=run.get("total_duration_seconds"),
+        total_agent_steps=_total_agent_steps_from_payload(payload),
         avg_agent_steps=run.get("avg_agent_steps"),
         avg_tokens_per_task=run.get("avg_tokens_per_task"),
         total_tokens=run.get("total_tokens"),
@@ -323,6 +336,36 @@ def export_latest_runs(
     return written
 
 
+def backfill_summary_total_agent_steps(results_dir: Path | None = None) -> int:
+    """Add total_agent_steps to summary JSON from copied raw results when missing."""
+    root = results_dir or (_space_root() / "results")
+    if not root.is_dir():
+        return 0
+    updated = 0
+    for path in sorted(root.glob("*.json")):
+        payload = json.loads(path.read_text(encoding="utf-8"))
+        if "metrics" not in payload:
+            continue
+        metrics = dict(payload.get("metrics") or {})
+        if metrics.get("total_agent_steps") is not None:
+            continue
+        entry = LeaderboardEntry.from_json(payload)
+        raw_path = raw_results_path_for(entry, root.parent)
+        if not raw_path.is_file() and entry.raw_results_path:
+            raw_path = root.parent / entry.raw_results_path
+        if not raw_path.is_file():
+            continue
+        raw = json.loads(raw_path.read_text(encoding="utf-8"))
+        total_steps = _total_agent_steps_from_payload(raw)
+        if total_steps is None:
+            continue
+        metrics["total_agent_steps"] = total_steps
+        payload["metrics"] = metrics
+        path.write_text(json.dumps(payload, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
+        updated += 1
+    return updated
+
+
 def backfill_summary_total_duration(results_dir: Path | None = None) -> int:
     """Add total_duration_seconds to summary JSON from copied raw results when missing."""
     root = results_dir or (_space_root() / "results")

src/formatting.py

@@ -41,6 +41,12 @@ def fmt_pct_value(value: float | None) -> str:
     return fmt_number(value, suffix="%")
 
 
+def fmt_steps(value: float | int | None) -> str:
+    if value is None or isinstance(value, bool):
+        return "—"
+    return str(int(round(float(value))))
+
+
 def fmt_duration(seconds: float | None) -> str:
     if seconds is None:
         return "—"

src/load_entries.py

@@ -10,6 +10,7 @@ from src.bench_config import load_bench_config
 from src.export_results import (
     _cost_from_payload,
     _recorded_cost_from_payload,
+    _total_agent_steps_from_payload,
     raw_results_path_for,
     ui_classes_from_payload,
 )
@@ -77,6 +78,10 @@ def _enrich_entry_from_raw(entry: LeaderboardEntry, raw: dict[str, Any]) -> None
         entry.avg_tokens_per_task = float(run["avg_tokens_per_task"])
     if entry.total_duration_seconds is None and isinstance(run.get("total_duration_seconds"), (int, float)):
         entry.total_duration_seconds = float(run["total_duration_seconds"])
+    if entry.total_agent_steps is None:
+        total_steps = _total_agent_steps_from_payload(raw)
+        if total_steps is not None:
+            entry.total_agent_steps = total_steps
     slots = run.get("ouroboros_model_slots")
     if isinstance(slots, dict) and slots and not entry.ouroboros_model_slots:
         entry.ouroboros_model_slots = {
@@ -104,6 +109,7 @@ def load_entry_file(path: Path) -> LeaderboardEntry | None:
         or entry.total_tokens is None
         or entry.avg_tokens_per_task is None
         or entry.total_duration_seconds is None
+        or entry.total_agent_steps is None
         or (entry.harness.startswith("ouroboros-full") and not entry.ouroboros_model_slots)
     )
     if needs_raw:

src/models.py

@@ -52,6 +52,7 @@ class LeaderboardEntry:
 
     avg_duration_seconds: float | None = None
     total_duration_seconds: float | None = None
+    total_agent_steps: int | None = None
     avg_agent_steps: float | None = None
     avg_tokens_per_task: float | None = None
     total_tokens: int | None = None
@@ -112,6 +113,8 @@ class LeaderboardEntry:
             return self.avg_duration_seconds
         if name == "total_duration_seconds":
             return self.total_duration_seconds
+        if name == "total_agent_steps":
+            return float(self.total_agent_steps) if self.total_agent_steps is not None else None
         if name == "total_cost_usd":
             return self.total_cost_usd
         if name == "value_score":
@@ -134,6 +137,7 @@ class LeaderboardEntry:
                 "total_tasks": self.total_tasks,
                 "avg_duration_seconds": self.avg_duration_seconds,
                 "total_duration_seconds": self.total_duration_seconds,
+                "total_agent_steps": self.total_agent_steps,
                 "avg_agent_steps": self.avg_agent_steps,
                 "avg_tokens_per_task": self.avg_tokens_per_task,
                 "total_tokens": self.total_tokens,
@@ -193,6 +197,7 @@ class LeaderboardEntry:
                 total_tasks=int(metrics.get("total_tasks") or 0),
                 avg_duration_seconds=metrics.get("avg_duration_seconds"),
                 total_duration_seconds=metrics.get("total_duration_seconds"),
+                total_agent_steps=metrics.get("total_agent_steps"),
                 avg_agent_steps=metrics.get("avg_agent_steps"),
                 avg_tokens_per_task=metrics.get("avg_tokens_per_task"),
                 total_tokens=metrics.get("total_tokens"),

src/ui.py

@@ -7,7 +7,7 @@ from pathlib import Path
 from typing import Any
 
 from src.bench_config import load_bench_config, task_count
-from src.formatting import fmt_cost, fmt_duration, fmt_number, fmt_pct_value, round2
+from src.formatting import fmt_cost, fmt_duration, fmt_number, fmt_pct_value, fmt_steps, round2
 from src.load_entries import sort_entries
 from src.models import LeaderboardEntry
 from src.token_display import fmt_tokens, format_model_usage_breakdown_html
@@ -282,7 +282,7 @@ def _speed_row(idx: int, entry: LeaderboardEntry) -> str:
         <div class="wab-badges"><span class="wab-badge">{html.escape(entry.harness)}</span></div></td>
       <td>{html.escape(fmt_duration(entry.total_duration_seconds))}</td>
       <td>{html.escape(fmt_duration(entry.avg_duration_seconds))}</td>
-      <td>{html.escape(fmt_number(entry.avg_agent_steps))}</td>
+      <td>{html.escape(fmt_steps(entry.total_agent_steps))}</td>
       <td class="wab-score-cell">
         <div class="wab-score-value wab-score-{tier}">{html.escape(fmt_pct_display(entry))}</div>
         <div class="wab-bar"><span class="wab-bar-{tier}" style="width:{max(pct, 0):.1f}%"></span></div>
@@ -320,11 +320,11 @@ def build_table(entries: list[LeaderboardEntry], view_id: str) -> str:
     sorted_entries = sort_entries(entries, view_id)
 
     if view_id == "speed":
-        head = "<tr><th>#</th><th>Model</th><th>Total time</th><th>Avg time</th><th>Steps</th><th>Success %</th></tr>"
+        head = "<tr><th>#</th><th>Model</th><th>Total time</th><th>Avg time</th><th>Total steps</th><th>Success %</th></tr>"
         rows = [_speed_row(i, entry) for i, entry in enumerate(sorted_entries)]
         subtitle = (
             "Sorted by total run duration (run.total_duration_seconds). "
-            "Avg time is mean per-task duration (run.avg_duration_seconds)."
+            "Avg time is mean per-task duration. Total steps is the sum of agent_steps across tasks."
         )
     elif view_id == "cost":
         head = (