Fix no_changes_guard: reward.txt now written by Python (38 verifiers)

The no_changes_guard in write_scored_result correctly set reward=0.0 in
Python when git confirmed zero agent changes, but reward.txt was still
written by a bash `echo "$score"` that used the original (non-zeroed)
shell variable. Harbor reads reward.txt for the final score, so the
guard was ineffective — agents that crashed before running could still
get non-zero scores from heuristic checks matching pre-existing repo state.

Fix: Python block now writes reward.txt directly (respecting the guard),
bash echo removed. Affects 38 test.sh files across 9 SDLC suites.

Also adds OH rerun configs for 12 tainted tasks.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: sjarmak

benchmarks/csb_sdlc_debug/tidb-query-plan-regression-debug-001/tests/test.sh

@@ -52,6 +52,8 @@ payload = {
 }
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
     echo "0.0" > /logs/verifier/reward.txt
 }
@@ -134,8 +136,10 @@ if details:
     payload["details"] = details
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
-    echo "$score" > /logs/verifier/reward.txt
+    # reward.txt now written by Python (respects no_changes_guard)
 }
 
 

benchmarks/csb_sdlc_design/elasticsearch-shard-alloc-design-001/tests/test.sh

@@ -52,6 +52,8 @@ payload = {
 }
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
     echo "0.0" > /logs/verifier/reward.txt
 }
@@ -134,8 +136,10 @@ if details:
     payload["details"] = details
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
-    echo "$score" > /logs/verifier/reward.txt
+    # reward.txt now written by Python (respects no_changes_guard)
 }
 
 

benchmarks/csb_sdlc_document/godot-gdscript-api-docgen-001/tests/test.sh

@@ -52,6 +52,8 @@ payload = {
 }
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
     echo "0.0" > /logs/verifier/reward.txt
 }
@@ -134,8 +136,10 @@ if details:
     payload["details"] = details
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
-    echo "$score" > /logs/verifier/reward.txt
+    # reward.txt now written by Python (respects no_changes_guard)
 }
 
 

benchmarks/csb_sdlc_document/grpc-channel-api-docgen-001/tests/test.sh

@@ -52,6 +52,8 @@ payload = {
 }
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
     echo "0.0" > /logs/verifier/reward.txt
 }
@@ -134,8 +136,10 @@ if details:
     payload["details"] = details
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
-    echo "$score" > /logs/verifier/reward.txt
+    # reward.txt now written by Python (respects no_changes_guard)
 }
 
 

benchmarks/csb_sdlc_feature/cilium-policy-audit-logger-feat-001/tests/test.sh

@@ -52,6 +52,8 @@ payload = {
 }
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
     echo "0.0" > /logs/verifier/reward.txt
 }
@@ -134,8 +136,10 @@ if details:
     payload["details"] = details
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
-    echo "$score" > /logs/verifier/reward.txt
+    # reward.txt now written by Python (respects no_changes_guard)
 }
 
 

benchmarks/csb_sdlc_feature/cilium-policy-quota-feat-001/tests/test.sh

@@ -52,6 +52,8 @@ payload = {
 }
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
     echo "0.0" > /logs/verifier/reward.txt
 }
@@ -134,8 +136,10 @@ if details:
     payload["details"] = details
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
-    echo "$score" > /logs/verifier/reward.txt
+    # reward.txt now written by Python (respects no_changes_guard)
 }
 
 

benchmarks/csb_sdlc_feature/curl-http3-priority-feat-001/tests/test.sh

@@ -52,6 +52,8 @@ payload = {
 }
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
     echo "0.0" > /logs/verifier/reward.txt
 }
@@ -134,8 +136,10 @@ if details:
     payload["details"] = details
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
-    echo "$score" > /logs/verifier/reward.txt
+    # reward.txt now written by Python (respects no_changes_guard)
 }
 
 

benchmarks/csb_sdlc_feature/django-rate-limit-middleware-feat-001/tests/test.sh

@@ -52,6 +52,8 @@ payload = {
 }
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
     echo "0.0" > /logs/verifier/reward.txt
 }
@@ -134,8 +136,10 @@ if details:
     payload["details"] = details
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
-    echo "$score" > /logs/verifier/reward.txt
+    # reward.txt now written by Python (respects no_changes_guard)
 }
 
 

benchmarks/csb_sdlc_feature/envoy-custom-header-filter-feat-001/tests/test.sh

@@ -52,6 +52,8 @@ payload = {
 }
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
     echo "0.0" > /logs/verifier/reward.txt
 }
@@ -134,8 +136,10 @@ if details:
     payload["details"] = details
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
-    echo "$score" > /logs/verifier/reward.txt
+    # reward.txt now written by Python (respects no_changes_guard)
 }
 
 

benchmarks/csb_sdlc_feature/numpy-rolling-median-feat-001/tests/test.sh

@@ -52,6 +52,8 @@ payload = {
 }
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
     echo "0.0" > /logs/verifier/reward.txt
 }
@@ -134,8 +136,10 @@ if details:
     payload["details"] = details
 with open("/logs/verifier/validation_result.json", "w") as f:
     json.dump(payload, f, indent=2)
+with open("/logs/verifier/reward.txt", "w") as f:
+    f.write(f"{reward:.4f}\n")
 PYEOF
-    echo "$score" > /logs/verifier/reward.txt
+    # reward.txt now written by Python (respects no_changes_guard)
 }