ren worker folder adding orig, mv new one into root

worker-toolkit-potion-polyglot/harbor-tasks/_task-scaffold/.toolkit-managed.json

@@ -1,8 +1,8 @@
 {
   "version": 1,
-  "stampedAt": "2026-09-07T17:37:17.141Z",
+  "stampedAt": "2026-09-15T00:07:53.720Z",
   "files": {
-    "tests/test.sh": "34ea5925a7ded396d2d811041236cb9ad655dde08775d0062ba9e8f9ab553600",
+    "tests/test.sh": "9dba71df2a7e3e1c19f928835f5f963d24f46fc638efa0db6d4e77d6c748c8cc",
     "tests/grader-system-prompt-consolidated.md": "032ce032728a8c0b2717478b929dbd7535e07c96ffe2e991097dd2c233543275"
   }
 }

worker-toolkit-potion-polyglot/harbor-tasks/_task-scaffold/task.toml

@@ -3,11 +3,11 @@ version = "1.0"
 [metadata]
 author = "worker"
 repo = "potion-app"
-commit = "6b4fee0c"
+commit = "f89abccf"
 # The toolkit release this task was created with. Written by the toolkit —
 # leave it in place: task tooling reads it to know which toolkit's assets
 # this task grades with.
-toolkit_version = "2f696c53b4"
+toolkit_version = "037cfcf94b"
 # Set true for a task about a UI: the trial gets Playwright + Chromium (`pw <script.js>`),
 # and on claude the `Read` tool so the agent can view a screenshot it takes. Leave false
 # when the point of the task is that something cannot be verified.
@@ -33,6 +33,8 @@ cpus = 2
 memory_mb = 4096
 storage_mb = 10240
 gpus = 0
+# Leave this true, and don't add network_mode or allowed_hosts: the grader reaches the
+# model API over the network, so restricting egress here makes every trial ungradable.
 allow_internet = true
 
 [verifier.env]

worker-toolkit-potion-polyglot/harbor-tasks/_task-scaffold/tests/test.sh

@@ -54,7 +54,7 @@ RENDER_GRADE="$TESTS_DIR/render-grade-consolidated.py"
 # Grader model + number of samples (graded GRADER_SAMPLES times and averaged to
 # reduce noise). Override with GRADER_MODEL=... / GRADER_SAMPLES=...
 GRADER_MODEL="${GRADER_MODEL:-claude-fable-5-1}"
-GRADER_SAMPLES="${GRADER_SAMPLES:-3}"
+GRADER_SAMPLES="${GRADER_SAMPLES:-1}"
 
 # The grader model's Claude Code floor. A task image installs Claude Code when it is first built
 # and Docker reuses that layer on every rebuild, so an image can carry a CLI the API refuses for
@@ -332,6 +332,23 @@ if [ -f "$TESTS_DIR/test-commands.sh" ] && [ "$AGENT_CHANGED" = 1 ]; then
             echo "signal setup ok" >&2
         else
             echo "signal setup FAILED (rc=${rc}, non-fatal) — see /logs/verifier/signals/_setup.log" >&2
+            # Surface it where the grader reads. Guidance routinely says "if the codegen
+            # step failed, treat the downstream failures as environmental" — a judgement
+            # the grader could not make, because this outcome reached only stderr.
+            {
+                echo "===== SETUP STEP FAILED ====="
+                echo "command: \`$1\`"
+                echo "exit: ${rc} (non-fatal; the checks below still ran)"
+                echo "This is a build/codegen step, not a scored check. Failures in the checks"
+                echo "below that follow from it are not the response's doing — but say which,"
+                echo "and do not discount a failure this cannot explain."
+                echo "full output file (readable with your tools): /logs/verifier/signals/_setup.log"
+                echo '```'
+                tail -c 4000 /logs/verifier/signals/_setup.log 2>/dev/null
+                echo '```'
+                echo "===== END SETUP STEP ====="
+                echo
+            } >> "$_SIG"
         fi
     }
     # shellcheck source=/dev/null
@@ -646,7 +663,7 @@ echo "Launching Claude Code grader (requested model: $GRADER_MODEL, samples: $GR
 # the canonical grade.md etc. are copied from the sample closest to the mean. A
 # sample with no valid reward in [0,1] is skipped; need min(2, GRADER_SAMPLES) valid.
 GRADER_MODEL="${GRADER_MODEL:-claude-fable-5-1}"
-GRADER_SAMPLES="${GRADER_SAMPLES:-3}"
+GRADER_SAMPLES="${GRADER_SAMPLES:-1}"
 mkdir -p /tmp/outputs /logs/verifier
 [ -e /tmp/files ] || ln -sfn /workspace /tmp/files