Add agent rehearsal harness, findings, and per-run evidence for split-loop-with-lambda

Six scripted pi sub-agent runs (3 tool regimes × 2 models) rehearsed the
exercise; deterministic gate = solution test + approved file. Key results:
mandated CLI triples cost on the strong model (4:06 vs 0:39, PASS both);
weak model fails only under the mandate (A+Q), passes when free or optional.
Drift documented with code: scorer.updateSelection left inside the extracted
method (A+Q, C+Q). Harness in tools/rehearse; summary in
rehearsal-findings.md; per-run evidence in rehearsal-runs/; cheat sheet in
rehearse-with-agent.md corrected (lambda-typed parameters unsupported).

tools/rehearse/verify_outcome.sh

@@ -0,0 +1,66 @@
+#!/usr/bin/env bash
+# Deterministic verification of one run's end state:
+#   1. diff vs pristine exercise  ->  diff.patch (the agent's final diff)
+#   2. drop in the solution's test + approved file, run mvn test
+#      PASS = deterministic end state matches the reference solution's behavior
+#   3. structural check: does collectExpectedWeather exist and take the lambda?
+# usage: bash verify_outcome.sh <run_dir>
+set -uo pipefail
+RUN_DIR="$1"
+WS="$RUN_DIR/workspace"
+REPO=/home/cardfanatic/qwan/exercises/learning-hour-maker
+SOLUTION="$REPO/ideas/split-loop-with-lambda/exercise/solution"
+PRISTINE=/tmp/rehearsal/pristine
+
+echo "--- Run $(basename "$RUN_DIR") ---"
+
+# 1. final diff vs pristine
+diff -ru --exclude=target "$PRISTINE" "$WS" > "$RUN_DIR/diff.patch"
+if [ -s "$RUN_DIR/diff.patch" ]; then
+  echo "final diff vs pristine: $(grep -c '^[+-][^+-]' "$RUN_DIR/diff.patch") changed lines -> $RUN_DIR/diff.patch"
+else
+  echo "final diff vs pristine: EMPTY (agent made no changes)"
+fi
+
+# 2. deterministic test: replace test tree with the solution's test + approved file
+if [ ! -d "$SOLUTION" ]; then
+  echo "solution dir missing, skipping mvn verification"
+  exit 0
+fi
+rm -rf "$WS/src/test"
+mkdir -p "$WS/src/test/java/codingdojo"
+cp -r "$SOLUTION/src/test/java/codingdojo/." "$WS/src/test/java/codingdojo/"
+
+echo "mvn test with solution's test + approved file:"
+if ( cd "$WS" && JAVA_HOME=/tmp/jdk21 PATH="/tmp/jdk21/bin:/tmp/apache-maven-3.9.9/bin:$PATH" \
+       mvn -q test ) > "$RUN_DIR/mvn-verify.log" 2>&1; then
+  echo "  RESULT: PASS (deterministic)"
+else
+  echo "  RESULT: FAIL (see $RUN_DIR/mvn-verify.log, last lines:)"
+  tail -15 "$RUN_DIR/mvn-verify.log" | sed 's/^/    /'
+fi
+
+# 3. structural checks on the produced LongRangeForecasts.java
+echo "structural checks:"
+LR="$WS/src/main/java/codingdojo/LongRangeForecasts.java"
+grep -q "collectExpectedWeather" "$LR" \
+  && echo "  [x] method collectExpectedWeather exists" \
+  || echo "  [ ] method collectExpectedWeather MISSING"
+grep -q "LongFunction<Boolean>" "$LR" \
+  && echo "  [x] takes LongFunction<Boolean> parameter" \
+  || echo "  [ ] LongFunction<Boolean> parameter MISSING"
+grep -q "scorer" "$LR" \
+  && echo "  [i] note: 'scorer' still referenced in LongRangeForecasts.java" \
+  || echo "  [x] no direct scorer reference left in LongRangeForecasts.java"
+if grep -qF -- "->" "$LR"; then
+  echo "  [x] lambda syntax present"
+else
+  echo "  [ ] lambda syntax MISSING (the classic drift: extracted method closes over scorer)"
+fi
+
+# comparison against reference solution shape
+if diff -q "$SOLUTION/src/main/java/codingdojo/LongRangeForecasts.java" "$LR" > /dev/null; then
+  echo "  [x] main class byte-identical to reference solution"
+else
+  echo "  [i] main class differs from reference solution (may still be a valid variant)"
+fi