diff --git a/evaluation/get_benchmark_fidx.py b/evaluation/get_benchmark_fidx.py
index 4d5e6f7c..97e2c6fd 100644
--- a/evaluation/get_benchmark_fidx.py
+++ b/evaluation/get_benchmark_fidx.py
@@ -26,7 +26,7 @@ def get_subset_fidx(testname: str) -> list:
     subset_of_fidx = []
     try:
         subprocess.run(
-            [f"{WASMR3_PATH}/target/release/slice_dice", replay_wasm_path],
+            [f"{WASMR3_PATH}/crates/target/release/slice_dice", replay_wasm_path],
             check=True,
             capture_output=True,
             text=True,
diff --git a/package-lock.json b/package-lock.json
index 4f0ac70d..c4c629e6 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -1,10 +1,9 @@
 {
-  "name": "tool",
+  "name": "wasm-r3",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
-      "name": "tool",
       "dependencies": {
         "acorn": "^8.11.2",
         "command-line-args": "^5.2.1",
diff --git a/src/test-slice-dice.ts b/src/test-slice-dice.ts
index 88bc9ea6..133368b2 100644
--- a/src/test-slice-dice.ts
+++ b/src/test-slice-dice.ts
@@ -24,7 +24,7 @@ export default async function runSliceDiceTests(names: string[], options) {
 function runSliceDice(replayWasmPath: string, fidx: string) {
   const startTime = Date.now();
   process.stdout.write('    Running slice-dice: ');
-  const command = `./target/release/slice_dice ${replayWasmPath} ${fidx} 1`;
+  const command = `./crates/target/release/slice_dice ${replayWasmPath} ${fidx} 1`;
   execSync(command);
   const endTime = Date.now();
   // We do not actually check what's generated runs to completion. This is misleading.
@@ -32,6 +32,7 @@ function runSliceDice(replayWasmPath: string, fidx: string) {
   console.log(`${endTime - startTime}ms`);
 }
 
+// TODO: this got slower by 2x from 01b34a3952ea29706d29d29d07b6f5148e119065. Investigate why.
 async function runWasmR3(options: any, subsetPath: string, benchmarkPath: string, fidx: string) {
   const startTime = Date.now();
   process.stdout.write('    Running wasm-r3: ');
@@ -78,7 +79,7 @@ function getSubsetFidx(replayWasmPath: string, name: string) {
   try {
   } catch (e) {
   }
-  const command = `./target/release/slice_dice ${replayWasmPath}`;
+  const command = `./crates/target/release/slice_dice ${replayWasmPath}`;
   const stdout = execSync(command, { stdio: ['pipe', 'pipe', 'ignore'] });
   const matches = stdout.toString().match(/\d+/g);
   if (matches) {
diff --git a/src/test.ts b/src/test.ts
index d3a77b12..03448071 100644
--- a/src/test.ts
+++ b/src/test.ts
@@ -104,7 +104,8 @@ async function analyzeAndSaveBenchmark(options: any, testJsPath: string, website
       path.join(process.cwd(), "benchmarks")
     );
     if (options.testcases !== undefined) {
-      testNames = testNames.filter((n) => options.testcases === (n));
+      testNames = testNames.filter((n) => options.testcases.includes(n));
+      testNames = Array.from(new Set([...testNames, ...options.testcases]));
     }
     await runSliceDiceTests(testNames, options);
   }
diff --git a/third_party/wizard-engine b/third_party/wizard-engine
index a0d07f77..b800e5df 160000
--- a/third_party/wizard-engine
+++ b/third_party/wizard-engine
@@ -1 +1 @@
-Subproject commit a0d07f77076c414910cf3594d76752e72bd2fb03
+Subproject commit b800e5dfe5ed957695c4d889ece9711d0a5669fb