diff --git a/evalharness/cli.py b/evalharness/cli.py
index 606f654..9d2cfe8 100644
--- a/evalharness/cli.py
+++ b/evalharness/cli.py
@@ -198,13 +198,26 @@ def _print_run_plan(console, args, model_spec):
                   justify='center')
 
 
+def _phase_color(message: str) -> str:
+    """Start/finish color code: events that START or are IN progress are
+    yellow; events that COMPLETE are green. One glance = what stage,
+    did it finish."""
+    m = message.lower()
+    if any(k in m for k in ('complete', 'ready', 'downloaded', 'parsed',
+                            'restored', 'ok (')):
+        return 'green'
+    return 'yellow'
+
+
 def _print_phase(console, index, total, name, message):
-    # single-benchmark runs: the [1/1] tag is noise, drop it.
-    # narration lines stay UNCOLORED everywhere (matches reporter.log and
-    # the runner's plain prints): color is reserved for ✓/✗/errors only
+    # single-benchmark runs: the [1/1] tag is noise, drop it
     prefix = f'[{index}/{total}] ' if total > 1 else ''
     text = f'{prefix}{name}: {message}'
-    print(text, flush=True)
+    color = _phase_color(message)
+    if console is not None:
+        console.print(f'[{color}]{text}[/{color}]', highlight=False)
+    else:
+        print(text, flush=True)
 
 
 def _print_benchmark_result(console, index, total, name, status, elapsed):
@@ -508,7 +521,8 @@ def _cmd_eval_run(args) -> int:
                                     _reporter=progress_reporter,
                                     _console=console):
                     if _reporter is not None:
-                        _reporter.log(f'{_name}: {msg}')
+                        _col = _phase_color(msg)
+                        _reporter.log(f'[{_col}]{_name}: {msg}[/{_col}]')
                         if 'scoring' in msg:
                             _reporter.set_phase('scoring')
                         elif 'generating model responses' in msg: