From 6ff055b9cefc120860c863efd04aad488987f5d6 Mon Sep 17 00:00:00 2001 From: Daniel Maddern Date: Mon, 17 Aug 2026 13:14:13 +0700 Subject: [PATCH] Resume Guard overnight from base evaluation --- scripts/run_guard3b_overnight_spark.sh | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/scripts/run_guard3b_overnight_spark.sh b/scripts/run_guard3b_overnight_spark.sh index 883e073..f72b895 100755 --- a/scripts/run_guard3b_overnight_spark.sh +++ b/scripts/run_guard3b_overnight_spark.sh @@ -57,7 +57,8 @@ fi cp "$PROJECT/artifex_overnight/guard3b_v03_canary/pilot_metrics.json" "$RUN_ROOT/canary_metrics.json" # Base and the initial Challenger use the identical fixed harness/version. -evaluate base +# Reuse a completed base result when resuming after an infrastructure repair. +if [[ ! -s "$RUN_ROOT/base_evmbench.json" ]]; then evaluate base; else docker_run python /workspace/scripts/score_evmbench_strict.py "/workspace/artifex_overnight/guard3b_v03_overnight/base_evmbench.json" > "$RUN_ROOT/base_strict.json" 2>&1; fi cp "$RUN_ROOT/base_strict.json" "$RUN_ROOT/current_strict.json" CURRENT=guard3b_v03_canary evaluate canary "$CURRENT"