volt: land the Q144 entry-point record, its CHECK path and the path-claim repair

CHECK-q143

@@ -1,5 +1,6 @@
 #!/bin/bash
-# CHECK-q143 -- three legs over every CHECK entry point in this tree.
+# CHECK-q143 -- every CHECK entry point in this tree, graded against a baseline.
+# Two legs by default (selftest, paths); `--full` adds the sweep leg.
 #
 #   selftest  20 cells on fixture trees: discovery (a near-miss named CHECKER is
 #             not an entry point), the five codes the runner can produce
@@ -11,16 +12,47 @@
 #             anything -- 0.2 s, and it is the leg that notices evidence leaving
 #             the tree.
 #   sweep     every entry point run, its exit code compared with the code the
-#             baseline records. Both directions: a green that went red is a
+#             baseline records, at the SAME --timeout 120 --jobs 4 the baseline
+#             is taken at (a 45 s take compared against a 120 s run calls every
+#             slow artifact MOVED). Both directions: a green that went red is a
 #             reverted repair, a red that went green is a repair nobody signed.
 #
+# WHERE THE TOOL LIVES
+# This wrapper sits at the repo root (the house convention for a top-level
+# gate), and its tool, baseline and report live three levels down in
+# wizard-work/volt/q143-check-sweep/. The first version derived TOOL and BASE
+# from $HERE -- i.e. from the repo root, where neither file exists -- so all
+# three legs printed UNGRAD rc=2 and the entry point graded nothing at all
+# (Q144). Both paths are named explicitly and their absence is rc 2.
+#
+# DEFAULT LEGS vs --full
+# The default run is the two legs that are deterministic and fast (selftest
+# 20 cells, paths 0.2 s): a repo-root gate that re-runs 97 entry points cannot
+# be graded by the sweep that calls it, and a reading that depends on host load
+# (REPORT §5) is not a gate. The full three-leg grade, including the sweep leg,
+# is `bash CHECK-q143 --full` -- explicit, on an idle host.
+#
 # Exit codes: 0 all legs green, 1 a leg faulted, 2 a leg could not grade.
 set -u
 HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-ROOT="$(cd "$HERE/../../.." && pwd)"
-TOOL="$HERE/check_sweep.py"
-BASE="$HERE/check_entry_baseline.tsv"
+ROOT="$HERE"
+ART="$HERE/wizard-work/volt/q143-check-sweep"
+TOOL="$ART/check_sweep.py"
+BASE="$ART/check_entry_baseline.tsv"
 PY="${PYTHON:-python3}"
+FULL=0
+for arg in "$@"; do
+    case "$arg" in
+        --full) FULL=1 ;;
+        *) printf 'usage: %s [--full]\n' "$0" >&2; exit 2 ;;
+    esac
+done
+for f in "$TOOL" "$BASE"; do
+    if [ ! -f "$f" ]; then
+        printf 'CANNOT GRADE: missing %s\n' "$f"
+        exit 2
+    fi
+done
 rc=0
 legs=0
 faults=0
@@ -45,7 +77,12 @@ run_leg() {
 
 run_leg selftest --selftest
 run_leg paths --paths
-run_leg sweep --check
+if [ "$FULL" -eq 1 ]; then
+    # Same timeout and concurrency the baseline was taken at: a code recorded
+    # under a 45 s cap and compared against a 120 s run reports MOVED for every
+    # row that merely needed longer, which is noise rather than a move.
+    run_leg sweep --check --timeout 120 --jobs 4
+fi
 
 if [ "$rc" -eq 0 ]; then
     echo "q143: $legs leg(s), all passed"