Progress bar: fresh-this-run counter; fix KeyError 'success'

- Completed X/Y now carries (+N new): samples generated by THIS run,
  excluding the checkpoint-restored head start (previously 98/164 told
  you nothing about how much work this invocation actually did)
- rate/eta computed over fresh samples only (restored ones counted
  toward 141/6s = 23/s when 1 sample had been generated)
- fix KeyError 'success': advance() still read the success/failed task
  fields after they were dropped from the column set -- first advance
  on a resumed bench killed the whole benchmark
- set_overall: give the overall task the fields the shared columns
  actually read (cur/retries/elapsed/eta); the old field set was from
  a previous column layout

Co-Authored-By: Claude <noreply@anthropic.com>

evalharness/progress/rich_terminal.py

@@ -43,7 +43,7 @@ class RichTerminalProgress:
             TextColumn("[progress.description]{task.description}"),
             BarColumn(complete_style="green", finished_style="bold green"),
             TaskProgressColumn(),
-            TextColumn("• Completed {task.completed}/{task.total}"),
+            TextColumn("• Completed {task.completed}/{task.total} [dim]{task.fields[new]}[/dim]"),
             TextColumn("• in-flight [yellow]{task.fields[inflight]}[/yellow] ([yellow]{task.fields[cur]}[/yellow])"),
             TextColumn("• [red]retries {task.fields[retries]}[/red]"),
             TextColumn("• [dim]{task.fields[rate]}/s[/dim]"),
@@ -59,6 +59,7 @@ class RichTerminalProgress:
         self.started = 0.0
         self.current_started = 0.0
         self.inflight = 0
+        self.restored = 0     # checkpoint head start (drives the '+N new' marker)
         self.heartbeat_task = None
 
     def set_overall(self, total: int, done: int, label: str = 'benches'):
@@ -73,8 +74,8 @@ class RichTerminalProgress:
             # overall task the same fields or rendering raises KeyError
             self.overall_id = self.progress.add_task(
                 f'[cyan]{label}[/cyan]', total=total, completed=min(done, total),
-                inflight=0, rate='0.00', success=done, failed=0,
+                new='', inflight=0, cur='0s', retries=0, rate='0.00',
-                waiting='00:00', last_result='')
+                elapsed='0s', eta='-')
         else:
             self.progress.update(self.overall_id, completed=min(done, total))
 
@@ -88,6 +89,7 @@ class RichTerminalProgress:
         if self.disabled:
             return
         self.started = time.monotonic()
+        self.restored = max(completed, 0)
         if self.task_id is not None:
             return  # one live reporter at a time; reuse across benchmarks
         self.progress.start()
@@ -95,18 +97,24 @@ class RichTerminalProgress:
             f"[green]{description}",
             total=total,
             completed=min(completed, total),
-            success=completed,
+            new=self._new_txt(completed),
-            failed=0,
             rate="0.00",
             inflight=0,
             cur="0s",
+            retries=0,
             elapsed="0s",
             eta="-",
-            waiting="00:00",
-            last_result="restored",
         )
         self.heartbeat_task = asyncio.create_task(self._heartbeat())
 
+    def _new_txt(self, absolute_done: int) -> str:
+        """'(+N new)' marker: samples completed by THIS run, i.e. absolute
+        progress minus the checkpoint-restored head start. Empty when the
+        run started fresh (nothing was restored, nothing to distinguish)."""
+        if not getattr(self, 'restored', 0):
+            return ''
+        return f'(+{max(absolute_done - self.restored, 0)} new)'
+
     def reset_samples(self, total: int, description: str, completed: int = 0):
         if self.disabled:
             return
@@ -114,20 +122,20 @@ class RichTerminalProgress:
         self.started = time.monotonic()
         self.inflight = 0
         self.bench_name = description
+        self.restored = max(completed, 0)  # checkpoint head start this bench
         desc = f'[green]{self.bench_tag}{description} · generating[/green]'
         if self.task_id is None:
             self.progress.start()
             self.task_id = self.progress.add_task(
                 desc, total=total, completed=min(completed, total),
-                success=completed, failed=0, rate='0.00', inflight=0,
+                new=self._new_txt(completed), rate='0.00', inflight=0,
-                cur='0s', elapsed='0s', eta='-', retries=0,
+                cur='0s', elapsed='0s', eta='-', retries=0)
-                waiting='00:00', last_result='restored')
         else:
             self.progress.update(self.task_id, description=desc,
                                  total=total, completed=min(completed, total),
-                                 success=completed, failed=0, rate='0.00',
+                                 new=self._new_txt(completed), rate='0.00',
-                                 inflight=0, cur='0s', elapsed='0s', eta='-', retries=0,
+                                 inflight=0, cur='0s', elapsed='0s', eta='-',
-                                 last_result='redone')
+                                 retries=0)
         # ALWAYS recreate the heartbeat: the previous one may have died during
         # pause/resume cycles between benchmarks (stale reference -> silent
         # death -> frozen clock while the spinner still animates)
@@ -161,8 +169,7 @@ class RichTerminalProgress:
             return
         self.inflight += 1
         self.current_started = time.monotonic()
-        self.progress.update(self.task_id, inflight=self.inflight, cur='0s',
+        self.progress.update(self.task_id, inflight=self.inflight, cur='0s')
-                             waiting="00:00", last_result=f"waiting {label}")
 
     def set_retries(self, n: int):
         """Show the retry count on the bar (from the adapter's attempt)."""
@@ -185,22 +192,20 @@ class RichTerminalProgress:
             return
         task = self.progress.tasks[self.task_id]
         completed = task.completed + 1
-        ok = task.fields["success"] + (1 if success else 0)
-        failed = task.fields["failed"] + (0 if success else 1)
         self.inflight = max(0, self.inflight - 1)
         elapsed = max(time.monotonic() - self.started, 1e-6)
+        # rate/eta over THIS RUN's fresh samples only: counting the restored
+        # head start would print 141/6s = 23/s when 1 sample was generated
+        fresh = max(completed - getattr(self, 'restored', 0), 0)
         self.progress.update(
             self.task_id,
             advance=1,
-            success=ok,
+            new=self._new_txt(completed),
-            failed=failed,
+            rate=f"{fresh / elapsed:.2f}",
-            rate=f"{completed / elapsed:.2f}",
             inflight=self.inflight, cur='0s',
             elapsed=_fmt(elapsed),
-            eta=_fmt((task.total - completed) * elapsed / completed)
+            eta=_fmt((task.total - completed) * elapsed / fresh)
-                 if completed and task.total and task.total > completed else '-',
+                 if fresh and task.total and task.total > completed else '-',
-            waiting="00:00",
-            last_result="success" if success else "failed",
         )
 
     async def _heartbeat(self):