Narration lines get stage icons (⬇ download / 📦 ready / ✳ few-shot / ◷ checkpoint / 🤖 generating / ⏭ skipped / ★ scoring / 📝 writing / 🔗 endpoint) and cyan-highlighted paths

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

evalharness/cli.py

@@ -198,6 +198,39 @@ def _print_run_plan(console, args, model_spec):
                   justify='center')
 
 
+
+def _narration(msg: str) -> str:
+    """Icon + path-highlighting for narration lines (visual separation at
+    a glance; paths in cyan)."""
+    icon = ''
+    m = msg.lower()
+    if m.startswith('loading/'):
+        icon = '⬇ '
+    elif 'dataset ready' in m:
+        icon = '📦 '
+    elif 'few-shot' in m:
+        icon = '✳ '
+    elif 'checkpoint' in m or m.endswith('samples') or 'to generate' in m:
+        icon = '◷ '
+    elif 'generation skipped' in m:
+        icon = '⏭ '
+    elif 'generating' in m:
+        icon = '🤖 '
+    elif 'generation complete' in m:
+        icon = '✓ '
+    elif 'scoring' in m:
+        icon = '★ '
+    elif 'writing' in m:
+        icon = '📝 '
+    elif 'endpoint ok' in m:
+        icon = '🔗 '
+    # highlight any trailing path after '->'
+    if '-> ' in msg:
+        head, _, tail = msg.partition('-> ')
+        return f'{icon}{head}-> [cyan]{tail}[/cyan]'
+    return f'{icon}{msg}'
+
+
 def _phase_color(message: str) -> str:
     """Narration line color. Currently PLAIN WHITE for everything (user
     preference); flip the returns to 'yellow'/'green' to restore the
@@ -213,7 +246,7 @@ def _phase_color(message: str) -> str:
 def _print_phase(console, index, total, name, message):
     # single-benchmark runs: the [1/1] tag is noise, drop it
     prefix = f'[{index}/{total}] ' if total > 1 else ''
-    text = f'{prefix}{name}: {message}'
+    text = f'{prefix}{name}: {_narration(message)}'
     color = _phase_color(message)
     if console is not None:
         if color:
@@ -221,7 +254,9 @@ def _print_phase(console, index, total, name, message):
         else:
             console.print(text, highlight=False)
     else:
-        print(text, flush=True)
+        import re as _re0
+
+        print(_re0.sub(r'\[/?[a-z ]+\]', '', text), flush=True)
 
 
 def _print_benchmark_result(console, index, total, name, status, elapsed):
@@ -496,9 +531,8 @@ def _cmd_eval_run(args) -> int:
 
             def _emit(msg, _i=i, _n=name):
                 if _shared_reporter is not None:
-                    _col = _phase_color(msg)
+                    _shared_reporter.log(
-                    _line = f'[{_i + 1}/{total_runs}] {_n}: {msg}'
+                        f'[{_i + 1}/{total_runs}] {_n}: {_narration(msg)}')
-                    _shared_reporter.log(f'[{_col}]{_line}[/{_col}]' if _col else _line)
                 else:
                     _print_phase(console, _i + 1, total_runs, _n, msg)
 
@@ -535,9 +569,7 @@ def _cmd_eval_run(args) -> int:
                                     _console=console):
                     _tag = f'[{_idx}/{total_runs}] ' if total_runs > 1 else ''
                     if _reporter is not None:
-                        _col = _phase_color(msg)
+                        _reporter.log(f'{_tag}{_name}: {_narration(msg)}')
-                        _line = f'{_tag}{_name}: {msg}'
-                        _reporter.log(f'[{_col}]{_line}[/{_col}]' if _col else _line)
                         if 'scoring' in msg:
                             _reporter.set_phase('scoring')
                         elif 'generating model responses' in msg: