From d7f06395a0458b824314689ef26f4f9e0b43b057 Mon Sep 17 00:00:00 2001
From: sora <2075279110@qq.com>
Date: Fri, 11 Sep 2026 03:09:15 +0000
Subject: [PATCH] Highlight whole noun phrases (1319 samples from cache / 4/4
 predictions / 0 samples left to run), not bare numbers; score rule runs first
 so scores keep solid green

Co-Authored-By: Claude <noreply@anthropic.com>
---
 evalharness/cli.py | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/evalharness/cli.py b/evalharness/cli.py
index c2f6a87..e948708 100644
--- a/evalharness/cli.py
+++ b/evalharness/cli.py
@@ -202,13 +202,6 @@ def _print_run_plan(console, args, model_spec):
 def _narration(msg: str) -> str:
     """Icon + path-highlighting for narration lines (visual separation at
     a glance; paths in cyan)."""
-    # semantic numbers get bold: samples counts, checkpoint fractions,
-    # generation counts, few-shot counts (curated, not rich's blanket highlight)
-    import re as _re0
-
-    msg = _re0.sub(r'(?<![\w/%.])(\d+(?:/\d+)?)(?=[\s]|$)',
-                  r'[bold yellow]\1[/bold yellow]', msg)
-
     icon = ''
     m = msg.lower()
     if m.startswith('loading/'):
@@ -239,6 +232,14 @@ def _narration(msg: str) -> str:
     if m:
         return (f'{icon}{msg[:m.start()]}'
                 f'[bold green]{m.group(1)}[/bold green]')
+
+    # semantic PHRASES get color, not bare numbers: '164 samples from cache',
+    # '4/4 predictions' -- a lone number is easy to lose. Runs AFTER the
+    # score rule so scores keep their solid green.
+    import re as _re0
+
+    msg = _re0.sub(r'(?<![\w/%.])(\d+(?:/\d+)?(?:\s+[a-z-]+){0,4})(?=[\s,.]|$)',
+                  r'[bold yellow]\1[/bold yellow]', msg)
     # highlight any trailing path after '->'
     if '-> ' in msg:
         head, _, tail = msg.partition('-> ')