Persistent benchmark counter [i/N] on the single progress bar; phase lines all route through the live channel for strict ordering

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

evalharness/cli.py

@@ -487,8 +487,17 @@ def _cmd_eval_run(args) -> int:
     for i, name in enumerate(args.datasets):
         t0 = _time.time()
         try:
-            _print_phase(console, i + 1, total_runs, name,
-                         'loading/downloading dataset')
+            if _shared_reporter is not None and total_runs > 1:
+                _shared_reporter.set_bench_tag(f'[{i + 1}/{total_runs}]')
+
+            def _emit(msg, _i=i, _n=name):
+                if _shared_reporter is not None:
+                    _col = _phase_color(msg)
+                    _shared_reporter.log(f'[{_col}][{_i + 1}/{total_runs}] {_n}: {msg}[/{_col}]')
+                else:
+                    _print_phase(console, _i + 1, total_runs, _n, msg)
+
+            _emit('loading/downloading dataset')
             ds = get_dataset(name, **overrides)
             if _shared_reporter is not None:
                 _shared_reporter.pause()   # let hub tqdm print cleanly
@@ -496,8 +505,7 @@ def _cmd_eval_run(args) -> int:
             if _shared_reporter is not None:
                 _shared_reporter.resume()
             origin = ds.lineage.get('from', 'unknown')
-            _print_phase(console, i + 1, total_runs, name,
-                         f'dataset ready · samples={sample_count} · source={origin}')
+            _emit(f'dataset ready · samples={sample_count} · source={origin}')
             if model_spec:  # generate + score in one go
                 from evalharness.model import run_eval
 
@@ -556,8 +564,7 @@ def _cmd_eval_run(args) -> int:
             if args.out:
                 report.save(args.out)
             if out_dir:
-                _print_phase(console, i + 1, total_runs, name,
-                             'writing results')
+                _emit('writing results')
                 from pathlib import Path as _P
 
                 bench_dir = _P(out_dir) / name

evalharness/progress/rich_terminal.py

@@ -39,6 +39,8 @@ class RichTerminalProgress:
         )
         self.task_id = None
         self.bench_name = ''
+        self.bench_tag = ''   # e.g. '[1/6]': persistent benchmark counter
+        self._last_phase = ''
         self.started = 0.0
         self.current_started = 0.0
         self.inflight = 0
@@ -86,7 +88,7 @@ class RichTerminalProgress:
         self.started = time.monotonic()
         self.inflight = 0
         self.bench_name = description
-        desc = f'[green]{description} · generating[/green]'
+        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(
@@ -100,13 +102,20 @@ class RichTerminalProgress:
                                  success=completed, failed=0, rate='0.00',
                                  inflight=0, last_result='restored')
 
+    def set_bench_tag(self, tag: str):
+        """Persistent counter shown on the sample bar, e.g. '[1/6]'."""
+        self.bench_tag = tag + ' ' if tag else ''
+        if self._last_phase:
+            self.set_phase(self._last_phase)
+
     def set_phase(self, phase: str):
+        self._last_phase = phase
         """Retag the sample bar with what is happening (generating/scoring/
         writing) -- the bar alone does not say which stage we are in."""
         if self.task_id is not None:
             self.progress.update(
                 self.task_id,
-                description=f'[green]{self.bench_name} · {phase}[/green]')
+                description=f'[green]{self.bench_tag}{self.bench_name} · {phase}[/green]')
 
     def begin_sample(self, label: str):
         if self.task_id is None: