From 27b2daf7ce42f2efcebbb118a4a957804c949499 Mon Sep 17 00:00:00 2001 From: Gahow Wang Date: Sun, 19 Jul 2026 19:12:51 +0800 Subject: [PATCH] Distinguish proxy and exact state verdicts --- .../analyze_qwen235_fixed_pd_state.py | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) diff --git a/runs/frontier-fidelity-envelope-v1/analyze_qwen235_fixed_pd_state.py b/runs/frontier-fidelity-envelope-v1/analyze_qwen235_fixed_pd_state.py index 934605b..103d176 100644 --- a/runs/frontier-fidelity-envelope-v1/analyze_qwen235_fixed_pd_state.py +++ b/runs/frontier-fidelity-envelope-v1/analyze_qwen235_fixed_pd_state.py @@ -695,10 +695,16 @@ def analyze( if decision_contrast is None: verdict = "STOP: real Running proxy has insufficient exact simulator support" elif decision_contrast["total"] < 0: - verdict = ( - "State-composition mismatch is insufficient: after reweighting Frontier " - "to measured real decode composition, it still predicts TP8 faster." - ) + if exact_contrast is not None: + verdict = ( + "State-composition mismatch is insufficient: after reweighting Frontier " + "to measured real decode composition, it still predicts TP8 faster." + ) + else: + verdict = ( + "Active-batch-count mismatch is insufficient: after reweighting Frontier " + "to the frozen-real Running proxy, it still predicts TP8 faster." + ) else: verdict = ( "Coarse active-batch state can flip Frontier's ordering, but iteration-level "