slopbrick 0.34.3 → 0.34.7

package/dist/engine/worker.cjs

@@ -34497,6 +34497,7 @@ var cppMagicNumbersRule = createRule({
     if (!/\.(cpp|cc|cxx|h|hpp|hh|hxx|H)$/i.test(facts.filePath)) return issues;
     const lines = source.split("\n");
     const allowSet = /* @__PURE__ */ new Set([
+      // v0.24 originals
       "1024",
       "65535",
       "65536",
@@ -34515,17 +34516,42 @@ var cppMagicNumbersRule = createRule({
       "2",
       "3",
       "4",
-      "5"
+      "5",
+      // v0.34.4 additions
+      "-1",
+      // sentinel value for "not found" / "all bits set"
+      "100",
+      // percent literal, very common
+      "0.0",
+      "0.5",
+      "1.0",
+      "2.0",
+      // common probability / ratio
+      "4096",
+      // page size, hash bucket count
+      "2048",
+      "512",
+      // power-of-2 sizes
+      "32",
+      "64",
+      "128",
+      // bit widths, byte sizes
+      "16",
+      "8",
+      // common small constants
+      "50"
+      // percentile literal
     ]);
     for (let i = 0; i < lines.length; i++) {
       const line = lines[i] ?? "";
       if (!COMPARE_OR_RETURN_REGEX.test(line)) continue;
+      const codeLine = line.replace(/\/\/.*$/, "").replace(/"(?:[^"\\]|\\.)*"/g, '""').replace(/'(?:[^'\\]|\\.)*'/g, "''");
       let m;
       MAGIC_NUMBER_REGEX.lastIndex = 0;
-      while ((m = MAGIC_NUMBER_REGEX.exec(line)) !== null) {
+      while ((m = MAGIC_NUMBER_REGEX.exec(codeLine)) !== null) {
         const literal = m[0] ?? "";
         if (allowSet.has(literal)) continue;
-        if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(line.slice(0, m.index).trimEnd())) continue;
+        if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(codeLine.slice(0, m.index).trimEnd())) continue;
         const prevLine = i > 0 ? lines[i - 1] ?? "" : "";
         if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(prevLine.trim())) continue;
         issues.push({
@@ -34536,7 +34562,7 @@ var cppMagicNumbersRule = createRule({
           message: `magic number ${literal} at line ${i + 1} \u2014 name it: constexpr int MAX = ${literal};`,
           line: i + 1,
           column: 1,
-          advice: 'Replace the bare literal with a named constant declared as `constexpr int MAX_SIZE = 1024;` (or `static constexpr`). The named constant lives next to its value, can be searched, and forces the reader to mean what they say. Magic numbers in comparisons hide intent ("7" against what?). AI agents produce magic-number-heavy code because their training-data examples rarely bother to name the constant. Reference: cpp/magic-numbers v0.24.'
+          advice: 'Replace the bare literal with a named constant declared as `constexpr int MAX_SIZE = 1024;` (or `static constexpr`). The named constant lives next to its value, can be searched, and forces the reader to mean what they say. Magic numbers in comparisons hide intent ("7" against what?). AI agents produce magic-number-heavy code because their training-data examples rarely bother to name the constant. Reference: cpp/magic-numbers v0.34.4 (expanded allowSet + string/comment exclusion).'
         });
       }
     }
@@ -34549,6 +34575,7 @@ var PRINTF_FAMILY_REGEX = /\b(?:printf|fprintf|sprintf|snprintf)\s*\(/g;
 var COUT_LITERAL_REGEX = /std\s*::\s*(?:cout|cerr|clog)\s*<<\s*"[^"]*"/g;
 var STD_COUT_BARE_REGEX = /std\s*::\s*(?:cout|cerr|clog)\s*<<\s*'[^']*'/g;
 var THRESHOLD_DEFAULT = 1;
+var TEST_FILE_REGEX = /(?:\/tests?\/|_test\.cc|_test\.cpp|Test\.cc|Test\.cpp|Tests\.cc|Tests\.cpp|_unittest\.cc|_unittest\.cpp)/;
 var cppPrintfDebugRule = createRule({
   id: "cpp/printf-debug",
   category: "typo",
@@ -34563,6 +34590,7 @@ var cppPrintfDebugRule = createRule({
     const source = facts.v2?._source;
     if (!source) return issues;
     if (!/\.(cpp|cc|cxx|h|hpp|hh|hxx|H)$/i.test(facts.filePath)) return issues;
+    if (TEST_FILE_REGEX.test(facts.filePath)) return issues;
     const printfCount = (source.match(PRINTF_FAMILY_REGEX) ?? []).length;
     let viaPrintf = false;
     if (printfCount > context.threshold) {
@@ -34579,7 +34607,7 @@ var cppPrintfDebugRule = createRule({
       message: viaPrintf ? `${printfCount} printf-family calls \u2014 use spdlog / glog / AbslLog` : "std::cout/cerr/clog with a string literal \u2014 use spdlog / glog / AbslLog",
       line: 1,
       column: 1,
-      advice: "Replace with `spdlog::info(...)`, `LOG(INFO) << ...` (glog), or `ABSL_LOG(INFO) << ...` (Abseil). All of these are level-aware and have a configurable sink, and they route to stderr by default. `printf` / `std::cout` have no levels, no redaction, no sink routing, and can't be silenced in release builds. AI agents reach for these because their training data has countless C++ textbook examples with them. Reference: cpp/printf-debug v0.24."
+      advice: "Replace with `spdlog::info(...)`, `LOG(INFO) << ...` (glog), or `ABSL_LOG(INFO) << ...` (Abseil). All of these are level-aware and have a configurable sink, and they route to stderr by default. `printf` / `std::cout` have no levels, no redaction, no sink routing, and can't be silenced in release builds. AI agents reach for these because their training data has countless C++ textbook examples with them. Reference: cpp/printf-debug v0.34.7 (refined to skip test files for higher precision)."
     });
     return issues;
   }
@@ -37189,7 +37217,6 @@ var javaSystemOutPrintlnRule = createRule({
 });
 
 // src/rules/java/thread-sleep-in-loop.ts
-var THREAD_SLEEP_REGEX = /\bThread\.sleep\s*\(/g;
 var javaThreadSleepInLoopRule = createRule({
   id: "java/thread-sleep-in-loop",
   category: "perf",
@@ -37205,11 +37232,104 @@ var javaThreadSleepInLoopRule = createRule({
     if (!source) return issues;
     if (!/\.java$/i.test(facts.filePath)) return issues;
     if (!/\bThread\.sleep\s*\(/.test(source)) return issues;
-    if (!/\b(?:for|while|do)\b/.test(source)) return issues;
-    let m;
-    THREAD_SLEEP_REGEX.lastIndex = 0;
-    while ((m = THREAD_SLEEP_REGEX.exec(source)) !== null) {
-      const line = source.slice(0, m.index).split("\n").length;
+    const braceStack = [];
+    const loopSet = /* @__PURE__ */ new Set();
+    let pendingLoopKeyword = null;
+    let parenDepth = 0;
+    const sleepEvents = [];
+    let inString = false;
+    let inLineComment = false;
+    let inBlockComment = false;
+    for (let i = 0; i < source.length; i++) {
+      const c = source[i] ?? "";
+      const next = source[i + 1] ?? "";
+      const prev = source[i - 1] ?? "";
+      if (inLineComment) {
+        if (c === "\n") inLineComment = false;
+        continue;
+      }
+      if (inBlockComment) {
+        if (c === "*" && next === "/") {
+          inBlockComment = false;
+          i++;
+        }
+        continue;
+      }
+      if (inString) {
+        if (c === "\\") {
+          i++;
+          continue;
+        }
+        if (c === inString) inString = false;
+        continue;
+      }
+      if (c === "/" && next === "/") {
+        inLineComment = true;
+        i++;
+        continue;
+      }
+      if (c === "/" && next === "*") {
+        inBlockComment = true;
+        i++;
+        continue;
+      }
+      if (c === '"' || c === "'") {
+        inString = c;
+        continue;
+      }
+      if (c === "(") {
+        parenDepth++;
+        continue;
+      }
+      if (c === ")") {
+        parenDepth = Math.max(0, parenDepth - 1);
+        continue;
+      }
+      if (/[A-Za-z_]/.test(c) && !/[A-Za-z0-9_]/.test(prev)) {
+        const next3 = source.slice(i, i + 3);
+        const next5 = source.slice(i, i + 5);
+        const next2 = source.slice(i, i + 2);
+        const after3 = source[i + 3] ?? "";
+        const after5 = source[i + 5] ?? "";
+        const after2 = source[i + 2] ?? "";
+        if (next3 === "for" && !/[A-Za-z0-9_]/.test(after3)) {
+          pendingLoopKeyword = { kind: "for", idx: i };
+          i += 2;
+          continue;
+        }
+        if (next5 === "while" && !/[A-Za-z0-9_]/.test(after5)) {
+          pendingLoopKeyword = { kind: "while", idx: i };
+          i += 4;
+          continue;
+        }
+        if (next2 === "do" && !/[A-Za-z0-9_]/.test(after2)) {
+          pendingLoopKeyword = { kind: "do", idx: i };
+          i += 1;
+          continue;
+        }
+      }
+      if (c === "{") {
+        braceStack.push(i);
+        if (pendingLoopKeyword && parenDepth === 0) {
+          loopSet.add(i);
+          pendingLoopKeyword = null;
+        }
+      } else if (c === "}") {
+        const popped = braceStack.pop();
+        if (popped !== void 0 && loopSet.has(popped)) {
+          loopSet.delete(popped);
+        }
+        if (pendingLoopKeyword) {
+          pendingLoopKeyword = null;
+        }
+      } else if (c === "T" && source.slice(i, i + 13) === "Thread.sleep(") {
+        sleepEvents.push({ idx: i, loopDepth: loopSet.size });
+        i += 12;
+      }
+    }
+    for (const ev of sleepEvents) {
+      if (ev.loopDepth === 0) continue;
+      const line = source.slice(0, ev.idx).split("\n").length;
       issues.push({
         ruleId: "java/thread-sleep-in-loop",
         category: "perf",
@@ -37218,7 +37338,7 @@ var javaThreadSleepInLoopRule = createRule({
         message: `Thread.sleep() at line ${line}`,
         line,
         column: 1,
-        advice: 'Use ScheduledExecutorService for periodic work, or BlockingQueue.take() for event-driven work. Thread.sleep in a loop is the classic "polling with sleep" anti-pattern \u2014 the thread blocks for the sleep duration each iteration. In server contexts this ties up Tomcat/Jetty/Netty threads. Reference: java/thread-sleep-in-loop v0.30.'
+        advice: 'Use ScheduledExecutorService for periodic work, or BlockingQueue.take() for event-driven work. Thread.sleep in a loop is the classic "polling with sleep" anti-pattern \u2014 the thread blocks for the sleep duration each iteration. In server contexts this ties up Tomcat/Jetty/Netty threads. Reference: java/thread-sleep-in-loop v0.34.6 (refined to require Thread.sleep inside the loop block, not just in the file).'
       });
     }
     return issues;
@@ -37465,6 +37585,7 @@ var kotlinObjectSingletonMisuseRule = createRule({
 // src/rules/kotlin/println-as-log.ts
 var PRINTLN_REGEX = /\bprintln\s*\(/g;
 var REAL_LOGGING_IMPORT_REGEX2 = /\bimport\s+(?:android\.util\.Log|org\.slf4j\.|io\.github\.oshai\.kotlinlogging|kotlin\.logging|co\.touchlab\.kermit|com\.github\.ajalt\.timber|org\.apache\.logging\.log4j)/;
+var TEST_FILE_REGEX2 = /(?:\/src\/test\/|\/test\/|\/Tests\/|\/Test\.kt|\/Tests\.kt|Tests\.kt$|Test\.kt$)/;
 var kotlinPrintlnAsLogRule = createRule({
   id: "kotlin/println-as-log",
   category: "logic",
@@ -37479,7 +37600,7 @@ var kotlinPrintlnAsLogRule = createRule({
     const source = facts.v2?._source;
     if (!source) return issues;
     if (!/\.kts?$/i.test(facts.filePath)) return issues;
-    if (/\/test\//i.test(facts.filePath) || /\.test\.kts?$/i.test(facts.filePath)) return issues;
+    if (TEST_FILE_REGEX2.test(facts.filePath)) return issues;
     if (REAL_LOGGING_IMPORT_REGEX2.test(source)) return issues;
     let m;
     PRINTLN_REGEX.lastIndex = 0;
@@ -37493,7 +37614,7 @@ var kotlinPrintlnAsLogRule = createRule({
         message: `println() as logger at line ${line}`,
         line,
         column: 1,
-        advice: "Use a real logging library: slf4j (JVM), android.util.Log (Android), Timber (Android), kermit (multiplatform), or kotlin-logging. println() has no log level, no timestamp, no correlation ID, and cannot be filtered. Reference: kotlin/println-as-log v0.29."
+        advice: "Use a real logging library: slf4j (JVM), android.util.Log (Android), Timber (Android), kermit (multiplatform), or kotlin-logging. println() has no log level, no timestamp, no correlation ID, and cannot be filtered. Reference: kotlin/println-as-log v0.34.5 (refined to skip test files for higher precision)."
       });
     }
     return issues;
@@ -40516,7 +40637,7 @@ var swiftImplicitlyUnwrappedOptionalRule = createRule({
 // src/rules/swift/print-debug.ts
 var PRINT_REGEX = /\bprint\s*\(/g;
 var DEFAULT_THRESHOLD2 = 1;
-var TEST_FILE_REGEX = /(?:\/Tests\/|\/Tests\.swift|\/Test\.swift|Tests\.swift$|Test\.swift$)/;
+var TEST_FILE_REGEX3 = /(?:\/Tests\/|\/Tests\.swift|\/Test\.swift|Tests\.swift$|Test\.swift$)/;
 var swiftPrintDebugRule = createRule({
   id: "swift/print-debug",
   category: "typo",
@@ -40531,7 +40652,7 @@ var swiftPrintDebugRule = createRule({
     const source = facts.v2?._source;
     if (!source) return issues;
     if (!/\.swift$/i.test(facts.filePath)) return issues;
-    if (TEST_FILE_REGEX.test(facts.filePath)) return issues;
+    if (TEST_FILE_REGEX3.test(facts.filePath)) return issues;
     const matches = [];
     let m;
     PRINT_REGEX.lastIndex = 0;
@@ -45467,7 +45588,7 @@ var signal_strength_default = {
     precision: 0.1268,
     lastCalibratedAt: "2026-07-03T00:00:00Z",
     verdict: "OK",
-    _calibrationNote: "v0.29: v9 Kotlin calibration (2698 neg, 213 pos). ratio=1.84 (\u22651.5) \u2014 first rule in this entire session with a positive direction! Fires 1.84x more on post-2024 AI/demos than pre-2022 production. Precision=12.7% (below 50% USEFUL threshold); verdict=OK. The signal is real: post-2024 Kotlin code (especially AI-generated examples) uses println for output; pre-2022 production code uses slf4j/kermit. INSUFFICIENT_DATA: pos arm only 213 files. defaultOff: still set to true (verdict is OK but precision is below 50% \u2014 the guardrail expects OK/USEFUL rules to be defaultOff:false only when calibrated with a meaningful pos arm).",
+    _calibrationNote: "v0.34.5: REFINED \u2014 rule now skips test files (JUnit4/5 + Android: *Tests.kt, *Test.kt, src/test/, test/ dirs). Per-file unique v9 Kotlin calibration (2698 neg, 213 pos): ratio=1.84 (\u22651.5, OK). The refinement mirrors v0.34.2's swift/print-debug fix: the previous exclusion only matched `\\/test\\/` and `.test.kts?$`, missing JUnit5's `FooTests.kt` naming. Expected post-refinement precision: 12.7% \u2192 25%+ by removing the JUnit assertion / debug-output fires that were ~50% of the FPs. Same direction as java/system-out-println (1.73 refined) and the v0.34.2 swift/print-debug pipeline. Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.29: v9 Kotlin calibration (2698 neg, 213 pos). ratio=1.84 (\u22651.5) \u2014 first rule in this entire session with a positive direction! Fires 1.84x more on post-2024 AI/demos than pre-2022 production. Precision=12.7% (below 50% USEFUL threshold); verdict=OK. The signal is real: post-2024 Kotlin code (especially AI-generated examples) uses println for output; pre-2022 production code uses slf4j/kermit. INSUFFICIENT_DATA: pos arm only 213 files. defaultOff: still set to true (verdict is OK but precision is below 50% \u2014 the guardrail expects OK/USEFUL rules to be defaultOff:false only when calibrated with a meaningful pos arm).",
     aiSpecific: false,
     _v9Verdict: "OK",
     _v9Lift: 1.84,
@@ -45531,7 +45652,7 @@ var signal_strength_default = {
     precision: 0.1091,
     lastCalibratedAt: "2026-07-03T00:00:00Z",
     verdict: "DORMANT",
-    _calibrationNote: "v0.30: v9 Java calibration (81891 neg, 10305 pos). ratio=0.97 (just below 1.0) \u2014 fires equally on both arms. Borderline era-confound: pre-2022 Java had more Thread.sleep in loops (less mature concurrency APIs); modern Java uses ScheduledExecutorService. Same direction as kotlin/runblocking-misuse (0.50). 2154 total fires \u2014 high absolute count, meaningful measurement. defaultOff.",
+    _calibrationNote: "v0.34.6: REFINED \u2014 rule now requires Thread.sleep to be INSIDE the loop block (verified via brace-counting), not just in a file that happens to contain a for/while/do keyword. v0.30 baseline: 2154 total fires, ratio 0.97 (DORMANT). The refinement targets: (1) Thread.sleep in `main()` no longer fires if a different method has a `for` loop (the v0.30 heuristic fired on every Thread.sleep in the file); (2) Thread.sleep before/after a loop block (not inside it) no longer fires; (3) string literals containing `Thread.sleep(...)` are skipped via the string-state-machine. Expected post-refinement ratio: 1.5+ (positive-signal OK verdict) by removing the cross-method over-fires. Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.30: v9 Java calibration (81891 neg, 10305 pos). ratio=0.97 (just below 1.0) \u2014 fires equally on both arms. Borderline era-confound: pre-2022 Java had more Thread.sleep in loops (less mature concurrency APIs); modern Java uses ScheduledExecutorService. Same direction as kotlin/runblocking-misuse (0.50). 2154 total fires \u2014 high absolute count, meaningful measurement. defaultOff.",
     aiSpecific: false,
     _v9Verdict: "DORMANT",
     _v9Lift: 0.97,
@@ -45763,7 +45884,7 @@ var signal_strength_default = {
     precision: 0.4407,
     lastCalibratedAt: "2026-07-03T00:00:00Z",
     verdict: "OK",
-    _calibrationNote: "v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 156 TP files, 198 FP files, ratio=2.43 (\u22651.5) \u2014 fourth positive-signal rule in v9 history (and the strongest ratio so far). precision=44.07% (just below 50% USEFUL threshold); verdict=OK. Total fires: 182 TP, 198 FP. The signal is real: post-2024 C++ (mostly AI-generated demos in llama.cpp, whisper.cpp, openai-cpp) uses printf/cout for output; pre-2022 production C++ (folly, protobuf, abseil) uses spdlog / glog / AbslLog. Same direction as kotlin/println-as-log (1.84), java/system-out-println (1.73 refined), swift/print-debug (1.13). INSUFFICIENT_DATA: pos arm 1655 files (below 10k floor).",
+    _calibrationNote: "v0.34.7: REFINED \u2014 rule now skips test files (gtest, catch2, doctest conventions: *_test.cpp, *_test.cc, /tests/ dir, *Test.cpp, *Test.cc, *_unittest.cpp). Per-file unique v9 C++ calibration (5107 neg, 1655 pos): 156 TP files, 198 FP files, ratio=2.43 (OK, fourth positive-signal in v9 history). precision=44.07% (below 50% USEFUL threshold); verdict=OK. The refinement is expected to push precision from 44% to 50%+ by removing gtest test output fires (which were a significant portion of FPs in the v0.33.0 measurement). The full v9 re-calibration is part of the broader v0.34.5/v0.34.7 pipeline; see v0.34.1's Four-language validation section. Same direction as kotlin/println-as-log (1.84), java/system-out-println (1.73 refined, 3.29 unrefined), swift/print-debug (1.13). INSUFFICIENT_DATA: pos arm 1655 files (below 10k floor).",
     aiSpecific: true,
     _v7Verdict: "DORMANT",
     _v7Lift: 1,
@@ -45786,7 +45907,7 @@ var signal_strength_default = {
     precision: 0.2187,
     lastCalibratedAt: "2026-07-03T00:00:00Z",
     verdict: "DORMANT",
-    _calibrationNote: "v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 220 TP, 786 FP, ratio=0.86. Both arms have similar proportions of files with magic numbers \u2014 this is a general C++ anti-pattern that hasn't been eradicated. INSUFFICIENT_DATA: pos arm 1655 files.",
+    _calibrationNote: "v0.34.4: REFINED \u2014 expanded allowSet with common constants (-1 sentinel, 100 percent literal, 0.5/1.0/2.0 ratios, 4096/2048/512/128/64/32/16/8 power-of-2 sizes, 50 percentile). Added string-literal and `//` comment stripping so substrings like `\"got 42 errors\"` and `// ticket #4242` no longer fire. v0.33 baseline: 220 TP / 786 FP per-file (ratio 0.86, DORMANT). The refinement targets: (1) sentinel `-1` literals are skipped because `1` is allowlisted and the `-` doesn't match MAGIC_NUMBER_REGEX anyway; (2) hex literals (`0xFF`) are skipped because the digit-only regex requires a word-boundary after digits, and `0x...` has none; (3) literals inside string/comment contexts no longer fire. Expected post-refinement ratio: 1.0-1.2 (still DORMANT but better-targeted). Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 220 TP, 786 FP, ratio=0.86. Both arms have similar proportions of files with magic numbers \u2014 this is a general C++ anti-pattern that hasn't been eradicated. INSUFFICIENT_DATA: pos arm 1655 files.",
     aiSpecific: true,
     _v7Verdict: "DORMANT",
     _v7Lift: 1,

package/dist/engine/worker.js

@@ -34468,6 +34468,7 @@ var cppMagicNumbersRule = createRule({
     if (!/\.(cpp|cc|cxx|h|hpp|hh|hxx|H)$/i.test(facts.filePath)) return issues;
     const lines = source.split("\n");
     const allowSet = /* @__PURE__ */ new Set([
+      // v0.24 originals
       "1024",
       "65535",
       "65536",
@@ -34486,17 +34487,42 @@ var cppMagicNumbersRule = createRule({
       "2",
       "3",
       "4",
-      "5"
+      "5",
+      // v0.34.4 additions
+      "-1",
+      // sentinel value for "not found" / "all bits set"
+      "100",
+      // percent literal, very common
+      "0.0",
+      "0.5",
+      "1.0",
+      "2.0",
+      // common probability / ratio
+      "4096",
+      // page size, hash bucket count
+      "2048",
+      "512",
+      // power-of-2 sizes
+      "32",
+      "64",
+      "128",
+      // bit widths, byte sizes
+      "16",
+      "8",
+      // common small constants
+      "50"
+      // percentile literal
     ]);
     for (let i = 0; i < lines.length; i++) {
       const line = lines[i] ?? "";
       if (!COMPARE_OR_RETURN_REGEX.test(line)) continue;
+      const codeLine = line.replace(/\/\/.*$/, "").replace(/"(?:[^"\\]|\\.)*"/g, '""').replace(/'(?:[^'\\]|\\.)*'/g, "''");
       let m;
       MAGIC_NUMBER_REGEX.lastIndex = 0;
-      while ((m = MAGIC_NUMBER_REGEX.exec(line)) !== null) {
+      while ((m = MAGIC_NUMBER_REGEX.exec(codeLine)) !== null) {
         const literal = m[0] ?? "";
         if (allowSet.has(literal)) continue;
-        if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(line.slice(0, m.index).trimEnd())) continue;
+        if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(codeLine.slice(0, m.index).trimEnd())) continue;
         const prevLine = i > 0 ? lines[i - 1] ?? "" : "";
         if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(prevLine.trim())) continue;
         issues.push({
@@ -34507,7 +34533,7 @@ var cppMagicNumbersRule = createRule({
           message: `magic number ${literal} at line ${i + 1} \u2014 name it: constexpr int MAX = ${literal};`,
           line: i + 1,
           column: 1,
-          advice: 'Replace the bare literal with a named constant declared as `constexpr int MAX_SIZE = 1024;` (or `static constexpr`). The named constant lives next to its value, can be searched, and forces the reader to mean what they say. Magic numbers in comparisons hide intent ("7" against what?). AI agents produce magic-number-heavy code because their training-data examples rarely bother to name the constant. Reference: cpp/magic-numbers v0.24.'
+          advice: 'Replace the bare literal with a named constant declared as `constexpr int MAX_SIZE = 1024;` (or `static constexpr`). The named constant lives next to its value, can be searched, and forces the reader to mean what they say. Magic numbers in comparisons hide intent ("7" against what?). AI agents produce magic-number-heavy code because their training-data examples rarely bother to name the constant. Reference: cpp/magic-numbers v0.34.4 (expanded allowSet + string/comment exclusion).'
         });
       }
     }
@@ -34520,6 +34546,7 @@ var PRINTF_FAMILY_REGEX = /\b(?:printf|fprintf|sprintf|snprintf)\s*\(/g;
 var COUT_LITERAL_REGEX = /std\s*::\s*(?:cout|cerr|clog)\s*<<\s*"[^"]*"/g;
 var STD_COUT_BARE_REGEX = /std\s*::\s*(?:cout|cerr|clog)\s*<<\s*'[^']*'/g;
 var THRESHOLD_DEFAULT = 1;
+var TEST_FILE_REGEX = /(?:\/tests?\/|_test\.cc|_test\.cpp|Test\.cc|Test\.cpp|Tests\.cc|Tests\.cpp|_unittest\.cc|_unittest\.cpp)/;
 var cppPrintfDebugRule = createRule({
   id: "cpp/printf-debug",
   category: "typo",
@@ -34534,6 +34561,7 @@ var cppPrintfDebugRule = createRule({
     const source = facts.v2?._source;
     if (!source) return issues;
     if (!/\.(cpp|cc|cxx|h|hpp|hh|hxx|H)$/i.test(facts.filePath)) return issues;
+    if (TEST_FILE_REGEX.test(facts.filePath)) return issues;
     const printfCount = (source.match(PRINTF_FAMILY_REGEX) ?? []).length;
     let viaPrintf = false;
     if (printfCount > context.threshold) {
@@ -34550,7 +34578,7 @@ var cppPrintfDebugRule = createRule({
       message: viaPrintf ? `${printfCount} printf-family calls \u2014 use spdlog / glog / AbslLog` : "std::cout/cerr/clog with a string literal \u2014 use spdlog / glog / AbslLog",
       line: 1,
       column: 1,
-      advice: "Replace with `spdlog::info(...)`, `LOG(INFO) << ...` (glog), or `ABSL_LOG(INFO) << ...` (Abseil). All of these are level-aware and have a configurable sink, and they route to stderr by default. `printf` / `std::cout` have no levels, no redaction, no sink routing, and can't be silenced in release builds. AI agents reach for these because their training data has countless C++ textbook examples with them. Reference: cpp/printf-debug v0.24."
+      advice: "Replace with `spdlog::info(...)`, `LOG(INFO) << ...` (glog), or `ABSL_LOG(INFO) << ...` (Abseil). All of these are level-aware and have a configurable sink, and they route to stderr by default. `printf` / `std::cout` have no levels, no redaction, no sink routing, and can't be silenced in release builds. AI agents reach for these because their training data has countless C++ textbook examples with them. Reference: cpp/printf-debug v0.34.7 (refined to skip test files for higher precision)."
     });
     return issues;
   }
@@ -37160,7 +37188,6 @@ var javaSystemOutPrintlnRule = createRule({
 });
 
 // src/rules/java/thread-sleep-in-loop.ts
-var THREAD_SLEEP_REGEX = /\bThread\.sleep\s*\(/g;
 var javaThreadSleepInLoopRule = createRule({
   id: "java/thread-sleep-in-loop",
   category: "perf",
@@ -37176,11 +37203,104 @@ var javaThreadSleepInLoopRule = createRule({
     if (!source) return issues;
     if (!/\.java$/i.test(facts.filePath)) return issues;
     if (!/\bThread\.sleep\s*\(/.test(source)) return issues;
-    if (!/\b(?:for|while|do)\b/.test(source)) return issues;
-    let m;
-    THREAD_SLEEP_REGEX.lastIndex = 0;
-    while ((m = THREAD_SLEEP_REGEX.exec(source)) !== null) {
-      const line = source.slice(0, m.index).split("\n").length;
+    const braceStack = [];
+    const loopSet = /* @__PURE__ */ new Set();
+    let pendingLoopKeyword = null;
+    let parenDepth = 0;
+    const sleepEvents = [];
+    let inString = false;
+    let inLineComment = false;
+    let inBlockComment = false;
+    for (let i = 0; i < source.length; i++) {
+      const c = source[i] ?? "";
+      const next = source[i + 1] ?? "";
+      const prev = source[i - 1] ?? "";
+      if (inLineComment) {
+        if (c === "\n") inLineComment = false;
+        continue;
+      }
+      if (inBlockComment) {
+        if (c === "*" && next === "/") {
+          inBlockComment = false;
+          i++;
+        }
+        continue;
+      }
+      if (inString) {
+        if (c === "\\") {
+          i++;
+          continue;
+        }
+        if (c === inString) inString = false;
+        continue;
+      }
+      if (c === "/" && next === "/") {
+        inLineComment = true;
+        i++;
+        continue;
+      }
+      if (c === "/" && next === "*") {
+        inBlockComment = true;
+        i++;
+        continue;
+      }
+      if (c === '"' || c === "'") {
+        inString = c;
+        continue;
+      }
+      if (c === "(") {
+        parenDepth++;
+        continue;
+      }
+      if (c === ")") {
+        parenDepth = Math.max(0, parenDepth - 1);
+        continue;
+      }
+      if (/[A-Za-z_]/.test(c) && !/[A-Za-z0-9_]/.test(prev)) {
+        const next3 = source.slice(i, i + 3);
+        const next5 = source.slice(i, i + 5);
+        const next2 = source.slice(i, i + 2);
+        const after3 = source[i + 3] ?? "";
+        const after5 = source[i + 5] ?? "";
+        const after2 = source[i + 2] ?? "";
+        if (next3 === "for" && !/[A-Za-z0-9_]/.test(after3)) {
+          pendingLoopKeyword = { kind: "for", idx: i };
+          i += 2;
+          continue;
+        }
+        if (next5 === "while" && !/[A-Za-z0-9_]/.test(after5)) {
+          pendingLoopKeyword = { kind: "while", idx: i };
+          i += 4;
+          continue;
+        }
+        if (next2 === "do" && !/[A-Za-z0-9_]/.test(after2)) {
+          pendingLoopKeyword = { kind: "do", idx: i };
+          i += 1;
+          continue;
+        }
+      }
+      if (c === "{") {
+        braceStack.push(i);
+        if (pendingLoopKeyword && parenDepth === 0) {
+          loopSet.add(i);
+          pendingLoopKeyword = null;
+        }
+      } else if (c === "}") {
+        const popped = braceStack.pop();
+        if (popped !== void 0 && loopSet.has(popped)) {
+          loopSet.delete(popped);
+        }
+        if (pendingLoopKeyword) {
+          pendingLoopKeyword = null;
+        }
+      } else if (c === "T" && source.slice(i, i + 13) === "Thread.sleep(") {
+        sleepEvents.push({ idx: i, loopDepth: loopSet.size });
+        i += 12;
+      }
+    }
+    for (const ev of sleepEvents) {
+      if (ev.loopDepth === 0) continue;
+      const line = source.slice(0, ev.idx).split("\n").length;
       issues.push({
         ruleId: "java/thread-sleep-in-loop",
         category: "perf",
@@ -37189,7 +37309,7 @@ var javaThreadSleepInLoopRule = createRule({
         message: `Thread.sleep() at line ${line}`,
         line,
         column: 1,
-        advice: 'Use ScheduledExecutorService for periodic work, or BlockingQueue.take() for event-driven work. Thread.sleep in a loop is the classic "polling with sleep" anti-pattern \u2014 the thread blocks for the sleep duration each iteration. In server contexts this ties up Tomcat/Jetty/Netty threads. Reference: java/thread-sleep-in-loop v0.30.'
+        advice: 'Use ScheduledExecutorService for periodic work, or BlockingQueue.take() for event-driven work. Thread.sleep in a loop is the classic "polling with sleep" anti-pattern \u2014 the thread blocks for the sleep duration each iteration. In server contexts this ties up Tomcat/Jetty/Netty threads. Reference: java/thread-sleep-in-loop v0.34.6 (refined to require Thread.sleep inside the loop block, not just in the file).'
       });
     }
     return issues;
@@ -37436,6 +37556,7 @@ var kotlinObjectSingletonMisuseRule = createRule({
 // src/rules/kotlin/println-as-log.ts
 var PRINTLN_REGEX = /\bprintln\s*\(/g;
 var REAL_LOGGING_IMPORT_REGEX2 = /\bimport\s+(?:android\.util\.Log|org\.slf4j\.|io\.github\.oshai\.kotlinlogging|kotlin\.logging|co\.touchlab\.kermit|com\.github\.ajalt\.timber|org\.apache\.logging\.log4j)/;
+var TEST_FILE_REGEX2 = /(?:\/src\/test\/|\/test\/|\/Tests\/|\/Test\.kt|\/Tests\.kt|Tests\.kt$|Test\.kt$)/;
 var kotlinPrintlnAsLogRule = createRule({
   id: "kotlin/println-as-log",
   category: "logic",
@@ -37450,7 +37571,7 @@ var kotlinPrintlnAsLogRule = createRule({
     const source = facts.v2?._source;
     if (!source) return issues;
     if (!/\.kts?$/i.test(facts.filePath)) return issues;
-    if (/\/test\//i.test(facts.filePath) || /\.test\.kts?$/i.test(facts.filePath)) return issues;
+    if (TEST_FILE_REGEX2.test(facts.filePath)) return issues;
     if (REAL_LOGGING_IMPORT_REGEX2.test(source)) return issues;
     let m;
     PRINTLN_REGEX.lastIndex = 0;
@@ -37464,7 +37585,7 @@ var kotlinPrintlnAsLogRule = createRule({
         message: `println() as logger at line ${line}`,
         line,
         column: 1,
-        advice: "Use a real logging library: slf4j (JVM), android.util.Log (Android), Timber (Android), kermit (multiplatform), or kotlin-logging. println() has no log level, no timestamp, no correlation ID, and cannot be filtered. Reference: kotlin/println-as-log v0.29."
+        advice: "Use a real logging library: slf4j (JVM), android.util.Log (Android), Timber (Android), kermit (multiplatform), or kotlin-logging. println() has no log level, no timestamp, no correlation ID, and cannot be filtered. Reference: kotlin/println-as-log v0.34.5 (refined to skip test files for higher precision)."
       });
     }
     return issues;
@@ -40487,7 +40608,7 @@ var swiftImplicitlyUnwrappedOptionalRule = createRule({
 // src/rules/swift/print-debug.ts
 var PRINT_REGEX = /\bprint\s*\(/g;
 var DEFAULT_THRESHOLD2 = 1;
-var TEST_FILE_REGEX = /(?:\/Tests\/|\/Tests\.swift|\/Test\.swift|Tests\.swift$|Test\.swift$)/;
+var TEST_FILE_REGEX3 = /(?:\/Tests\/|\/Tests\.swift|\/Test\.swift|Tests\.swift$|Test\.swift$)/;
 var swiftPrintDebugRule = createRule({
   id: "swift/print-debug",
   category: "typo",
@@ -40502,7 +40623,7 @@ var swiftPrintDebugRule = createRule({
     const source = facts.v2?._source;
     if (!source) return issues;
     if (!/\.swift$/i.test(facts.filePath)) return issues;
-    if (TEST_FILE_REGEX.test(facts.filePath)) return issues;
+    if (TEST_FILE_REGEX3.test(facts.filePath)) return issues;
     const matches = [];
     let m;
     PRINT_REGEX.lastIndex = 0;
@@ -45438,7 +45559,7 @@ var signal_strength_default = {
     precision: 0.1268,
     lastCalibratedAt: "2026-07-03T00:00:00Z",
     verdict: "OK",
-    _calibrationNote: "v0.29: v9 Kotlin calibration (2698 neg, 213 pos). ratio=1.84 (\u22651.5) \u2014 first rule in this entire session with a positive direction! Fires 1.84x more on post-2024 AI/demos than pre-2022 production. Precision=12.7% (below 50% USEFUL threshold); verdict=OK. The signal is real: post-2024 Kotlin code (especially AI-generated examples) uses println for output; pre-2022 production code uses slf4j/kermit. INSUFFICIENT_DATA: pos arm only 213 files. defaultOff: still set to true (verdict is OK but precision is below 50% \u2014 the guardrail expects OK/USEFUL rules to be defaultOff:false only when calibrated with a meaningful pos arm).",
+    _calibrationNote: "v0.34.5: REFINED \u2014 rule now skips test files (JUnit4/5 + Android: *Tests.kt, *Test.kt, src/test/, test/ dirs). Per-file unique v9 Kotlin calibration (2698 neg, 213 pos): ratio=1.84 (\u22651.5, OK). The refinement mirrors v0.34.2's swift/print-debug fix: the previous exclusion only matched `\\/test\\/` and `.test.kts?$`, missing JUnit5's `FooTests.kt` naming. Expected post-refinement precision: 12.7% \u2192 25%+ by removing the JUnit assertion / debug-output fires that were ~50% of the FPs. Same direction as java/system-out-println (1.73 refined) and the v0.34.2 swift/print-debug pipeline. Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.29: v9 Kotlin calibration (2698 neg, 213 pos). ratio=1.84 (\u22651.5) \u2014 first rule in this entire session with a positive direction! Fires 1.84x more on post-2024 AI/demos than pre-2022 production. Precision=12.7% (below 50% USEFUL threshold); verdict=OK. The signal is real: post-2024 Kotlin code (especially AI-generated examples) uses println for output; pre-2022 production code uses slf4j/kermit. INSUFFICIENT_DATA: pos arm only 213 files. defaultOff: still set to true (verdict is OK but precision is below 50% \u2014 the guardrail expects OK/USEFUL rules to be defaultOff:false only when calibrated with a meaningful pos arm).",
     aiSpecific: false,
     _v9Verdict: "OK",
     _v9Lift: 1.84,
@@ -45502,7 +45623,7 @@ var signal_strength_default = {
     precision: 0.1091,
     lastCalibratedAt: "2026-07-03T00:00:00Z",
     verdict: "DORMANT",
-    _calibrationNote: "v0.30: v9 Java calibration (81891 neg, 10305 pos). ratio=0.97 (just below 1.0) \u2014 fires equally on both arms. Borderline era-confound: pre-2022 Java had more Thread.sleep in loops (less mature concurrency APIs); modern Java uses ScheduledExecutorService. Same direction as kotlin/runblocking-misuse (0.50). 2154 total fires \u2014 high absolute count, meaningful measurement. defaultOff.",
+    _calibrationNote: "v0.34.6: REFINED \u2014 rule now requires Thread.sleep to be INSIDE the loop block (verified via brace-counting), not just in a file that happens to contain a for/while/do keyword. v0.30 baseline: 2154 total fires, ratio 0.97 (DORMANT). The refinement targets: (1) Thread.sleep in `main()` no longer fires if a different method has a `for` loop (the v0.30 heuristic fired on every Thread.sleep in the file); (2) Thread.sleep before/after a loop block (not inside it) no longer fires; (3) string literals containing `Thread.sleep(...)` are skipped via the string-state-machine. Expected post-refinement ratio: 1.5+ (positive-signal OK verdict) by removing the cross-method over-fires. Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.30: v9 Java calibration (81891 neg, 10305 pos). ratio=0.97 (just below 1.0) \u2014 fires equally on both arms. Borderline era-confound: pre-2022 Java had more Thread.sleep in loops (less mature concurrency APIs); modern Java uses ScheduledExecutorService. Same direction as kotlin/runblocking-misuse (0.50). 2154 total fires \u2014 high absolute count, meaningful measurement. defaultOff.",
     aiSpecific: false,
     _v9Verdict: "DORMANT",
     _v9Lift: 0.97,
@@ -45734,7 +45855,7 @@ var signal_strength_default = {
     precision: 0.4407,
     lastCalibratedAt: "2026-07-03T00:00:00Z",
     verdict: "OK",
-    _calibrationNote: "v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 156 TP files, 198 FP files, ratio=2.43 (\u22651.5) \u2014 fourth positive-signal rule in v9 history (and the strongest ratio so far). precision=44.07% (just below 50% USEFUL threshold); verdict=OK. Total fires: 182 TP, 198 FP. The signal is real: post-2024 C++ (mostly AI-generated demos in llama.cpp, whisper.cpp, openai-cpp) uses printf/cout for output; pre-2022 production C++ (folly, protobuf, abseil) uses spdlog / glog / AbslLog. Same direction as kotlin/println-as-log (1.84), java/system-out-println (1.73 refined), swift/print-debug (1.13). INSUFFICIENT_DATA: pos arm 1655 files (below 10k floor).",
+    _calibrationNote: "v0.34.7: REFINED \u2014 rule now skips test files (gtest, catch2, doctest conventions: *_test.cpp, *_test.cc, /tests/ dir, *Test.cpp, *Test.cc, *_unittest.cpp). Per-file unique v9 C++ calibration (5107 neg, 1655 pos): 156 TP files, 198 FP files, ratio=2.43 (OK, fourth positive-signal in v9 history). precision=44.07% (below 50% USEFUL threshold); verdict=OK. The refinement is expected to push precision from 44% to 50%+ by removing gtest test output fires (which were a significant portion of FPs in the v0.33.0 measurement). The full v9 re-calibration is part of the broader v0.34.5/v0.34.7 pipeline; see v0.34.1's Four-language validation section. Same direction as kotlin/println-as-log (1.84), java/system-out-println (1.73 refined, 3.29 unrefined), swift/print-debug (1.13). INSUFFICIENT_DATA: pos arm 1655 files (below 10k floor).",
     aiSpecific: true,
     _v7Verdict: "DORMANT",
     _v7Lift: 1,
@@ -45757,7 +45878,7 @@ var signal_strength_default = {
     precision: 0.2187,
     lastCalibratedAt: "2026-07-03T00:00:00Z",
     verdict: "DORMANT",
-    _calibrationNote: "v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 220 TP, 786 FP, ratio=0.86. Both arms have similar proportions of files with magic numbers \u2014 this is a general C++ anti-pattern that hasn't been eradicated. INSUFFICIENT_DATA: pos arm 1655 files.",
+    _calibrationNote: "v0.34.4: REFINED \u2014 expanded allowSet with common constants (-1 sentinel, 100 percent literal, 0.5/1.0/2.0 ratios, 4096/2048/512/128/64/32/16/8 power-of-2 sizes, 50 percentile). Added string-literal and `//` comment stripping so substrings like `\"got 42 errors\"` and `// ticket #4242` no longer fire. v0.33 baseline: 220 TP / 786 FP per-file (ratio 0.86, DORMANT). The refinement targets: (1) sentinel `-1` literals are skipped because `1` is allowlisted and the `-` doesn't match MAGIC_NUMBER_REGEX anyway; (2) hex literals (`0xFF`) are skipped because the digit-only regex requires a word-boundary after digits, and `0x...` has none; (3) literals inside string/comment contexts no longer fire. Expected post-refinement ratio: 1.0-1.2 (still DORMANT but better-targeted). Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 220 TP, 786 FP, ratio=0.86. Both arms have similar proportions of files with magic numbers \u2014 this is a general C++ anti-pattern that hasn't been eradicated. INSUFFICIENT_DATA: pos arm 1655 files.",
     aiSpecific: true,
     _v7Verdict: "DORMANT",
     _v7Lift: 1,

package/dist/index.cjs

@@ -36,7 +36,7 @@ var VERSION;
 var init_header = __esm({
   "src/types/_header.ts"() {
     "use strict";
-    VERSION = "0.34.3";
+    VERSION = "0.34.7";
   }
 });
 
@@ -27734,6 +27734,7 @@ var init_magic_numbers = __esm({
         if (!/\.(cpp|cc|cxx|h|hpp|hh|hxx|H)$/i.test(facts.filePath)) return issues;
         const lines = source.split("\n");
         const allowSet = /* @__PURE__ */ new Set([
+          // v0.24 originals
           "1024",
           "65535",
           "65536",
@@ -27752,17 +27753,42 @@ var init_magic_numbers = __esm({
           "2",
           "3",
           "4",
-          "5"
+          "5",
+          // v0.34.4 additions
+          "-1",
+          // sentinel value for "not found" / "all bits set"
+          "100",
+          // percent literal, very common
+          "0.0",
+          "0.5",
+          "1.0",
+          "2.0",
+          // common probability / ratio
+          "4096",
+          // page size, hash bucket count
+          "2048",
+          "512",
+          // power-of-2 sizes
+          "32",
+          "64",
+          "128",
+          // bit widths, byte sizes
+          "16",
+          "8",
+          // common small constants
+          "50"
+          // percentile literal
         ]);
         for (let i = 0; i < lines.length; i++) {
           const line = lines[i] ?? "";
           if (!COMPARE_OR_RETURN_REGEX.test(line)) continue;
+          const codeLine = line.replace(/\/\/.*$/, "").replace(/"(?:[^"\\]|\\.)*"/g, '""').replace(/'(?:[^'\\]|\\.)*'/g, "''");
           let m;
           MAGIC_NUMBER_REGEX.lastIndex = 0;
-          while ((m = MAGIC_NUMBER_REGEX.exec(line)) !== null) {
+          while ((m = MAGIC_NUMBER_REGEX.exec(codeLine)) !== null) {
             const literal = m[0] ?? "";
             if (allowSet.has(literal)) continue;
-            if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(line.slice(0, m.index).trimEnd())) continue;
+            if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(codeLine.slice(0, m.index).trimEnd())) continue;
             const prevLine = i > 0 ? lines[i - 1] ?? "" : "";
             if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(prevLine.trim())) continue;
             issues.push({
@@ -27773,7 +27799,7 @@ var init_magic_numbers = __esm({
               message: `magic number ${literal} at line ${i + 1} \u2014 name it: constexpr int MAX = ${literal};`,
               line: i + 1,
               column: 1,
-              advice: 'Replace the bare literal with a named constant declared as `constexpr int MAX_SIZE = 1024;` (or `static constexpr`). The named constant lives next to its value, can be searched, and forces the reader to mean what they say. Magic numbers in comparisons hide intent ("7" against what?). AI agents produce magic-number-heavy code because their training-data examples rarely bother to name the constant. Reference: cpp/magic-numbers v0.24.'
+              advice: 'Replace the bare literal with a named constant declared as `constexpr int MAX_SIZE = 1024;` (or `static constexpr`). The named constant lives next to its value, can be searched, and forces the reader to mean what they say. Magic numbers in comparisons hide intent ("7" against what?). AI agents produce magic-number-heavy code because their training-data examples rarely bother to name the constant. Reference: cpp/magic-numbers v0.34.4 (expanded allowSet + string/comment exclusion).'
             });
           }
         }
@@ -27784,7 +27810,7 @@ var init_magic_numbers = __esm({
 });
 
 // src/rules/cpp/printf-debug.ts
-var PRINTF_FAMILY_REGEX, COUT_LITERAL_REGEX, STD_COUT_BARE_REGEX, THRESHOLD_DEFAULT, cppPrintfDebugRule;
+var PRINTF_FAMILY_REGEX, COUT_LITERAL_REGEX, STD_COUT_BARE_REGEX, THRESHOLD_DEFAULT, TEST_FILE_REGEX, cppPrintfDebugRule;
 var init_printf_debug = __esm({
   "src/rules/cpp/printf-debug.ts"() {
     "use strict";
@@ -27793,6 +27819,7 @@ var init_printf_debug = __esm({
     COUT_LITERAL_REGEX = /std\s*::\s*(?:cout|cerr|clog)\s*<<\s*"[^"]*"/g;
     STD_COUT_BARE_REGEX = /std\s*::\s*(?:cout|cerr|clog)\s*<<\s*'[^']*'/g;
     THRESHOLD_DEFAULT = 1;
+    TEST_FILE_REGEX = /(?:\/tests?\/|_test\.cc|_test\.cpp|Test\.cc|Test\.cpp|Tests\.cc|Tests\.cpp|_unittest\.cc|_unittest\.cpp)/;
     cppPrintfDebugRule = createRule({
       id: "cpp/printf-debug",
       category: "typo",
@@ -27807,6 +27834,7 @@ var init_printf_debug = __esm({
         const source = facts.v2?._source;
         if (!source) return issues;
         if (!/\.(cpp|cc|cxx|h|hpp|hh|hxx|H)$/i.test(facts.filePath)) return issues;
+        if (TEST_FILE_REGEX.test(facts.filePath)) return issues;
         const printfCount = (source.match(PRINTF_FAMILY_REGEX) ?? []).length;
         let viaPrintf = false;
         if (printfCount > context.threshold) {
@@ -27823,7 +27851,7 @@ var init_printf_debug = __esm({
           message: viaPrintf ? `${printfCount} printf-family calls \u2014 use spdlog / glog / AbslLog` : "std::cout/cerr/clog with a string literal \u2014 use spdlog / glog / AbslLog",
           line: 1,
           column: 1,
-          advice: "Replace with `spdlog::info(...)`, `LOG(INFO) << ...` (glog), or `ABSL_LOG(INFO) << ...` (Abseil). All of these are level-aware and have a configurable sink, and they route to stderr by default. `printf` / `std::cout` have no levels, no redaction, no sink routing, and can't be silenced in release builds. AI agents reach for these because their training data has countless C++ textbook examples with them. Reference: cpp/printf-debug v0.24."
+          advice: "Replace with `spdlog::info(...)`, `LOG(INFO) << ...` (glog), or `ABSL_LOG(INFO) << ...` (Abseil). All of these are level-aware and have a configurable sink, and they route to stderr by default. `printf` / `std::cout` have no levels, no redaction, no sink routing, and can't be silenced in release builds. AI agents reach for these because their training data has countless C++ textbook examples with them. Reference: cpp/printf-debug v0.34.7 (refined to skip test files for higher precision)."
         });
         return issues;
       }
@@ -30861,12 +30889,11 @@ var init_system_out_println = __esm({
 });
 
 // src/rules/java/thread-sleep-in-loop.ts
-var THREAD_SLEEP_REGEX, javaThreadSleepInLoopRule;
+var javaThreadSleepInLoopRule;
 var init_thread_sleep_in_loop = __esm({
   "src/rules/java/thread-sleep-in-loop.ts"() {
     "use strict";
     init_rule();
-    THREAD_SLEEP_REGEX = /\bThread\.sleep\s*\(/g;
     javaThreadSleepInLoopRule = createRule({
       id: "java/thread-sleep-in-loop",
       category: "perf",
@@ -30882,11 +30909,104 @@ var init_thread_sleep_in_loop = __esm({
         if (!source) return issues;
         if (!/\.java$/i.test(facts.filePath)) return issues;
         if (!/\bThread\.sleep\s*\(/.test(source)) return issues;
-        if (!/\b(?:for|while|do)\b/.test(source)) return issues;
-        let m;
-        THREAD_SLEEP_REGEX.lastIndex = 0;
-        while ((m = THREAD_SLEEP_REGEX.exec(source)) !== null) {
-          const line = source.slice(0, m.index).split("\n").length;
+        const braceStack = [];
+        const loopSet = /* @__PURE__ */ new Set();
+        let pendingLoopKeyword = null;
+        let parenDepth = 0;
+        const sleepEvents = [];
+        let inString = false;
+        let inLineComment = false;
+        let inBlockComment = false;
+        for (let i = 0; i < source.length; i++) {
+          const c = source[i] ?? "";
+          const next = source[i + 1] ?? "";
+          const prev = source[i - 1] ?? "";
+          if (inLineComment) {
+            if (c === "\n") inLineComment = false;
+            continue;
+          }
+          if (inBlockComment) {
+            if (c === "*" && next === "/") {
+              inBlockComment = false;
+              i++;
+            }
+            continue;
+          }
+          if (inString) {
+            if (c === "\\") {
+              i++;
+              continue;
+            }
+            if (c === inString) inString = false;
+            continue;
+          }
+          if (c === "/" && next === "/") {
+            inLineComment = true;
+            i++;
+            continue;
+          }
+          if (c === "/" && next === "*") {
+            inBlockComment = true;
+            i++;
+            continue;
+          }
+          if (c === '"' || c === "'") {
+            inString = c;
+            continue;
+          }
+          if (c === "(") {
+            parenDepth++;
+            continue;
+          }
+          if (c === ")") {
+            parenDepth = Math.max(0, parenDepth - 1);
+            continue;
+          }
+          if (/[A-Za-z_]/.test(c) && !/[A-Za-z0-9_]/.test(prev)) {
+            const next3 = source.slice(i, i + 3);
+            const next5 = source.slice(i, i + 5);
+            const next2 = source.slice(i, i + 2);
+            const after3 = source[i + 3] ?? "";
+            const after5 = source[i + 5] ?? "";
+            const after2 = source[i + 2] ?? "";
+            if (next3 === "for" && !/[A-Za-z0-9_]/.test(after3)) {
+              pendingLoopKeyword = { kind: "for", idx: i };
+              i += 2;
+              continue;
+            }
+            if (next5 === "while" && !/[A-Za-z0-9_]/.test(after5)) {
+              pendingLoopKeyword = { kind: "while", idx: i };
+              i += 4;
+              continue;
+            }
+            if (next2 === "do" && !/[A-Za-z0-9_]/.test(after2)) {
+              pendingLoopKeyword = { kind: "do", idx: i };
+              i += 1;
+              continue;
+            }
+          }
+          if (c === "{") {
+            braceStack.push(i);
+            if (pendingLoopKeyword && parenDepth === 0) {
+              loopSet.add(i);
+              pendingLoopKeyword = null;
+            }
+          } else if (c === "}") {
+            const popped = braceStack.pop();
+            if (popped !== void 0 && loopSet.has(popped)) {
+              loopSet.delete(popped);
+            }
+            if (pendingLoopKeyword) {
+              pendingLoopKeyword = null;
+            }
+          } else if (c === "T" && source.slice(i, i + 13) === "Thread.sleep(") {
+            sleepEvents.push({ idx: i, loopDepth: loopSet.size });
+            i += 12;
+          }
+        }
+        for (const ev of sleepEvents) {
+          if (ev.loopDepth === 0) continue;
+          const line = source.slice(0, ev.idx).split("\n").length;
           issues.push({
             ruleId: "java/thread-sleep-in-loop",
             category: "perf",
@@ -30895,7 +31015,7 @@ var init_thread_sleep_in_loop = __esm({
             message: `Thread.sleep() at line ${line}`,
             line,
             column: 1,
-            advice: 'Use ScheduledExecutorService for periodic work, or BlockingQueue.take() for event-driven work. Thread.sleep in a loop is the classic "polling with sleep" anti-pattern \u2014 the thread blocks for the sleep duration each iteration. In server contexts this ties up Tomcat/Jetty/Netty threads. Reference: java/thread-sleep-in-loop v0.30.'
+            advice: 'Use ScheduledExecutorService for periodic work, or BlockingQueue.take() for event-driven work. Thread.sleep in a loop is the classic "polling with sleep" anti-pattern \u2014 the thread blocks for the sleep duration each iteration. In server contexts this ties up Tomcat/Jetty/Netty threads. Reference: java/thread-sleep-in-loop v0.34.6 (refined to require Thread.sleep inside the loop block, not just in the file).'
           });
         }
         return issues;
@@ -31177,13 +31297,14 @@ var init_object_singleton_misuse = __esm({
 });
 
 // src/rules/kotlin/println-as-log.ts
-var PRINTLN_REGEX, REAL_LOGGING_IMPORT_REGEX2, kotlinPrintlnAsLogRule;
+var PRINTLN_REGEX, REAL_LOGGING_IMPORT_REGEX2, TEST_FILE_REGEX2, kotlinPrintlnAsLogRule;
 var init_println_as_log = __esm({
   "src/rules/kotlin/println-as-log.ts"() {
     "use strict";
     init_rule();
     PRINTLN_REGEX = /\bprintln\s*\(/g;
     REAL_LOGGING_IMPORT_REGEX2 = /\bimport\s+(?:android\.util\.Log|org\.slf4j\.|io\.github\.oshai\.kotlinlogging|kotlin\.logging|co\.touchlab\.kermit|com\.github\.ajalt\.timber|org\.apache\.logging\.log4j)/;
+    TEST_FILE_REGEX2 = /(?:\/src\/test\/|\/test\/|\/Tests\/|\/Test\.kt|\/Tests\.kt|Tests\.kt$|Test\.kt$)/;
     kotlinPrintlnAsLogRule = createRule({
       id: "kotlin/println-as-log",
       category: "logic",
@@ -31198,7 +31319,7 @@ var init_println_as_log = __esm({
         const source = facts.v2?._source;
         if (!source) return issues;
         if (!/\.kts?$/i.test(facts.filePath)) return issues;
-        if (/\/test\//i.test(facts.filePath) || /\.test\.kts?$/i.test(facts.filePath)) return issues;
+        if (TEST_FILE_REGEX2.test(facts.filePath)) return issues;
         if (REAL_LOGGING_IMPORT_REGEX2.test(source)) return issues;
         let m;
         PRINTLN_REGEX.lastIndex = 0;
@@ -31212,7 +31333,7 @@ var init_println_as_log = __esm({
             message: `println() as logger at line ${line}`,
             line,
             column: 1,
-            advice: "Use a real logging library: slf4j (JVM), android.util.Log (Android), Timber (Android), kermit (multiplatform), or kotlin-logging. println() has no log level, no timestamp, no correlation ID, and cannot be filtered. Reference: kotlin/println-as-log v0.29."
+            advice: "Use a real logging library: slf4j (JVM), android.util.Log (Android), Timber (Android), kermit (multiplatform), or kotlin-logging. println() has no log level, no timestamp, no correlation ID, and cannot be filtered. Reference: kotlin/println-as-log v0.34.5 (refined to skip test files for higher precision)."
           });
         }
         return issues;
@@ -41057,14 +41178,14 @@ var init_implicitly_unwrapped_optional = __esm({
 });
 
 // src/rules/swift/print-debug.ts
-var PRINT_REGEX, DEFAULT_THRESHOLD2, TEST_FILE_REGEX, swiftPrintDebugRule;
+var PRINT_REGEX, DEFAULT_THRESHOLD2, TEST_FILE_REGEX3, swiftPrintDebugRule;
 var init_print_debug = __esm({
   "src/rules/swift/print-debug.ts"() {
     "use strict";
     init_rule();
     PRINT_REGEX = /\bprint\s*\(/g;
     DEFAULT_THRESHOLD2 = 1;
-    TEST_FILE_REGEX = /(?:\/Tests\/|\/Tests\.swift|\/Test\.swift|Tests\.swift$|Test\.swift$)/;
+    TEST_FILE_REGEX3 = /(?:\/Tests\/|\/Tests\.swift|\/Test\.swift|Tests\.swift$|Test\.swift$)/;
     swiftPrintDebugRule = createRule({
       id: "swift/print-debug",
       category: "typo",
@@ -41079,7 +41200,7 @@ var init_print_debug = __esm({
         const source = facts.v2?._source;
         if (!source) return issues;
         if (!/\.swift$/i.test(facts.filePath)) return issues;
-        if (TEST_FILE_REGEX.test(facts.filePath)) return issues;
+        if (TEST_FILE_REGEX3.test(facts.filePath)) return issues;
         const matches = [];
         let m;
         PRINT_REGEX.lastIndex = 0;
@@ -51554,7 +51675,7 @@ var init_signal_strength = __esm({
         precision: 0.1268,
         lastCalibratedAt: "2026-07-03T00:00:00Z",
         verdict: "OK",
-        _calibrationNote: "v0.29: v9 Kotlin calibration (2698 neg, 213 pos). ratio=1.84 (\u22651.5) \u2014 first rule in this entire session with a positive direction! Fires 1.84x more on post-2024 AI/demos than pre-2022 production. Precision=12.7% (below 50% USEFUL threshold); verdict=OK. The signal is real: post-2024 Kotlin code (especially AI-generated examples) uses println for output; pre-2022 production code uses slf4j/kermit. INSUFFICIENT_DATA: pos arm only 213 files. defaultOff: still set to true (verdict is OK but precision is below 50% \u2014 the guardrail expects OK/USEFUL rules to be defaultOff:false only when calibrated with a meaningful pos arm).",
+        _calibrationNote: "v0.34.5: REFINED \u2014 rule now skips test files (JUnit4/5 + Android: *Tests.kt, *Test.kt, src/test/, test/ dirs). Per-file unique v9 Kotlin calibration (2698 neg, 213 pos): ratio=1.84 (\u22651.5, OK). The refinement mirrors v0.34.2's swift/print-debug fix: the previous exclusion only matched `\\/test\\/` and `.test.kts?$`, missing JUnit5's `FooTests.kt` naming. Expected post-refinement precision: 12.7% \u2192 25%+ by removing the JUnit assertion / debug-output fires that were ~50% of the FPs. Same direction as java/system-out-println (1.73 refined) and the v0.34.2 swift/print-debug pipeline. Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.29: v9 Kotlin calibration (2698 neg, 213 pos). ratio=1.84 (\u22651.5) \u2014 first rule in this entire session with a positive direction! Fires 1.84x more on post-2024 AI/demos than pre-2022 production. Precision=12.7% (below 50% USEFUL threshold); verdict=OK. The signal is real: post-2024 Kotlin code (especially AI-generated examples) uses println for output; pre-2022 production code uses slf4j/kermit. INSUFFICIENT_DATA: pos arm only 213 files. defaultOff: still set to true (verdict is OK but precision is below 50% \u2014 the guardrail expects OK/USEFUL rules to be defaultOff:false only when calibrated with a meaningful pos arm).",
         aiSpecific: false,
         _v9Verdict: "OK",
         _v9Lift: 1.84,
@@ -51618,7 +51739,7 @@ var init_signal_strength = __esm({
         precision: 0.1091,
         lastCalibratedAt: "2026-07-03T00:00:00Z",
         verdict: "DORMANT",
-        _calibrationNote: "v0.30: v9 Java calibration (81891 neg, 10305 pos). ratio=0.97 (just below 1.0) \u2014 fires equally on both arms. Borderline era-confound: pre-2022 Java had more Thread.sleep in loops (less mature concurrency APIs); modern Java uses ScheduledExecutorService. Same direction as kotlin/runblocking-misuse (0.50). 2154 total fires \u2014 high absolute count, meaningful measurement. defaultOff.",
+        _calibrationNote: "v0.34.6: REFINED \u2014 rule now requires Thread.sleep to be INSIDE the loop block (verified via brace-counting), not just in a file that happens to contain a for/while/do keyword. v0.30 baseline: 2154 total fires, ratio 0.97 (DORMANT). The refinement targets: (1) Thread.sleep in `main()` no longer fires if a different method has a `for` loop (the v0.30 heuristic fired on every Thread.sleep in the file); (2) Thread.sleep before/after a loop block (not inside it) no longer fires; (3) string literals containing `Thread.sleep(...)` are skipped via the string-state-machine. Expected post-refinement ratio: 1.5+ (positive-signal OK verdict) by removing the cross-method over-fires. Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.30: v9 Java calibration (81891 neg, 10305 pos). ratio=0.97 (just below 1.0) \u2014 fires equally on both arms. Borderline era-confound: pre-2022 Java had more Thread.sleep in loops (less mature concurrency APIs); modern Java uses ScheduledExecutorService. Same direction as kotlin/runblocking-misuse (0.50). 2154 total fires \u2014 high absolute count, meaningful measurement. defaultOff.",
         aiSpecific: false,
         _v9Verdict: "DORMANT",
         _v9Lift: 0.97,
@@ -51850,7 +51971,7 @@ var init_signal_strength = __esm({
         precision: 0.4407,
         lastCalibratedAt: "2026-07-03T00:00:00Z",
         verdict: "OK",
-        _calibrationNote: "v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 156 TP files, 198 FP files, ratio=2.43 (\u22651.5) \u2014 fourth positive-signal rule in v9 history (and the strongest ratio so far). precision=44.07% (just below 50% USEFUL threshold); verdict=OK. Total fires: 182 TP, 198 FP. The signal is real: post-2024 C++ (mostly AI-generated demos in llama.cpp, whisper.cpp, openai-cpp) uses printf/cout for output; pre-2022 production C++ (folly, protobuf, abseil) uses spdlog / glog / AbslLog. Same direction as kotlin/println-as-log (1.84), java/system-out-println (1.73 refined), swift/print-debug (1.13). INSUFFICIENT_DATA: pos arm 1655 files (below 10k floor).",
+        _calibrationNote: "v0.34.7: REFINED \u2014 rule now skips test files (gtest, catch2, doctest conventions: *_test.cpp, *_test.cc, /tests/ dir, *Test.cpp, *Test.cc, *_unittest.cpp). Per-file unique v9 C++ calibration (5107 neg, 1655 pos): 156 TP files, 198 FP files, ratio=2.43 (OK, fourth positive-signal in v9 history). precision=44.07% (below 50% USEFUL threshold); verdict=OK. The refinement is expected to push precision from 44% to 50%+ by removing gtest test output fires (which were a significant portion of FPs in the v0.33.0 measurement). The full v9 re-calibration is part of the broader v0.34.5/v0.34.7 pipeline; see v0.34.1's Four-language validation section. Same direction as kotlin/println-as-log (1.84), java/system-out-println (1.73 refined, 3.29 unrefined), swift/print-debug (1.13). INSUFFICIENT_DATA: pos arm 1655 files (below 10k floor).",
         aiSpecific: true,
         _v7Verdict: "DORMANT",
         _v7Lift: 1,
@@ -51873,7 +51994,7 @@ var init_signal_strength = __esm({
         precision: 0.2187,
         lastCalibratedAt: "2026-07-03T00:00:00Z",
         verdict: "DORMANT",
-        _calibrationNote: "v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 220 TP, 786 FP, ratio=0.86. Both arms have similar proportions of files with magic numbers \u2014 this is a general C++ anti-pattern that hasn't been eradicated. INSUFFICIENT_DATA: pos arm 1655 files.",
+        _calibrationNote: "v0.34.4: REFINED \u2014 expanded allowSet with common constants (-1 sentinel, 100 percent literal, 0.5/1.0/2.0 ratios, 4096/2048/512/128/64/32/16/8 power-of-2 sizes, 50 percentile). Added string-literal and `//` comment stripping so substrings like `\"got 42 errors\"` and `// ticket #4242` no longer fire. v0.33 baseline: 220 TP / 786 FP per-file (ratio 0.86, DORMANT). The refinement targets: (1) sentinel `-1` literals are skipped because `1` is allowlisted and the `-` doesn't match MAGIC_NUMBER_REGEX anyway; (2) hex literals (`0xFF`) are skipped because the digit-only regex requires a word-boundary after digits, and `0x...` has none; (3) literals inside string/comment contexts no longer fire. Expected post-refinement ratio: 1.0-1.2 (still DORMANT but better-targeted). Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 220 TP, 786 FP, ratio=0.86. Both arms have similar proportions of files with magic numbers \u2014 this is a general C++ anti-pattern that hasn't been eradicated. INSUFFICIENT_DATA: pos arm 1655 files.",
         aiSpecific: true,
         _v7Verdict: "DORMANT",
         _v7Lift: 1,

package/dist/index.js

@@ -19,7 +19,7 @@ var VERSION;
 var init_header = __esm({
   "src/types/_header.ts"() {
     "use strict";
-    VERSION = "0.34.3";
+    VERSION = "0.34.7";
   }
 });
 
@@ -27716,6 +27716,7 @@ var init_magic_numbers = __esm({
         if (!/\.(cpp|cc|cxx|h|hpp|hh|hxx|H)$/i.test(facts.filePath)) return issues;
         const lines = source.split("\n");
         const allowSet = /* @__PURE__ */ new Set([
+          // v0.24 originals
           "1024",
           "65535",
           "65536",
@@ -27734,17 +27735,42 @@ var init_magic_numbers = __esm({
           "2",
           "3",
           "4",
-          "5"
+          "5",
+          // v0.34.4 additions
+          "-1",
+          // sentinel value for "not found" / "all bits set"
+          "100",
+          // percent literal, very common
+          "0.0",
+          "0.5",
+          "1.0",
+          "2.0",
+          // common probability / ratio
+          "4096",
+          // page size, hash bucket count
+          "2048",
+          "512",
+          // power-of-2 sizes
+          "32",
+          "64",
+          "128",
+          // bit widths, byte sizes
+          "16",
+          "8",
+          // common small constants
+          "50"
+          // percentile literal
         ]);
         for (let i = 0; i < lines.length; i++) {
           const line = lines[i] ?? "";
           if (!COMPARE_OR_RETURN_REGEX.test(line)) continue;
+          const codeLine = line.replace(/\/\/.*$/, "").replace(/"(?:[^"\\]|\\.)*"/g, '""').replace(/'(?:[^'\\]|\\.)*'/g, "''");
           let m;
           MAGIC_NUMBER_REGEX.lastIndex = 0;
-          while ((m = MAGIC_NUMBER_REGEX.exec(line)) !== null) {
+          while ((m = MAGIC_NUMBER_REGEX.exec(codeLine)) !== null) {
             const literal = m[0] ?? "";
             if (allowSet.has(literal)) continue;
-            if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(line.slice(0, m.index).trimEnd())) continue;
+            if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(codeLine.slice(0, m.index).trimEnd())) continue;
             const prevLine = i > 0 ? lines[i - 1] ?? "" : "";
             if (/\b(?:constexpr|const)\s+\w+\s*=\s*$/.test(prevLine.trim())) continue;
             issues.push({
@@ -27755,7 +27781,7 @@ var init_magic_numbers = __esm({
               message: `magic number ${literal} at line ${i + 1} \u2014 name it: constexpr int MAX = ${literal};`,
               line: i + 1,
               column: 1,
-              advice: 'Replace the bare literal with a named constant declared as `constexpr int MAX_SIZE = 1024;` (or `static constexpr`). The named constant lives next to its value, can be searched, and forces the reader to mean what they say. Magic numbers in comparisons hide intent ("7" against what?). AI agents produce magic-number-heavy code because their training-data examples rarely bother to name the constant. Reference: cpp/magic-numbers v0.24.'
+              advice: 'Replace the bare literal with a named constant declared as `constexpr int MAX_SIZE = 1024;` (or `static constexpr`). The named constant lives next to its value, can be searched, and forces the reader to mean what they say. Magic numbers in comparisons hide intent ("7" against what?). AI agents produce magic-number-heavy code because their training-data examples rarely bother to name the constant. Reference: cpp/magic-numbers v0.34.4 (expanded allowSet + string/comment exclusion).'
             });
           }
         }
@@ -27766,7 +27792,7 @@ var init_magic_numbers = __esm({
 });
 
 // src/rules/cpp/printf-debug.ts
-var PRINTF_FAMILY_REGEX, COUT_LITERAL_REGEX, STD_COUT_BARE_REGEX, THRESHOLD_DEFAULT, cppPrintfDebugRule;
+var PRINTF_FAMILY_REGEX, COUT_LITERAL_REGEX, STD_COUT_BARE_REGEX, THRESHOLD_DEFAULT, TEST_FILE_REGEX, cppPrintfDebugRule;
 var init_printf_debug = __esm({
   "src/rules/cpp/printf-debug.ts"() {
     "use strict";
@@ -27775,6 +27801,7 @@ var init_printf_debug = __esm({
     COUT_LITERAL_REGEX = /std\s*::\s*(?:cout|cerr|clog)\s*<<\s*"[^"]*"/g;
     STD_COUT_BARE_REGEX = /std\s*::\s*(?:cout|cerr|clog)\s*<<\s*'[^']*'/g;
     THRESHOLD_DEFAULT = 1;
+    TEST_FILE_REGEX = /(?:\/tests?\/|_test\.cc|_test\.cpp|Test\.cc|Test\.cpp|Tests\.cc|Tests\.cpp|_unittest\.cc|_unittest\.cpp)/;
     cppPrintfDebugRule = createRule({
       id: "cpp/printf-debug",
       category: "typo",
@@ -27789,6 +27816,7 @@ var init_printf_debug = __esm({
         const source = facts.v2?._source;
         if (!source) return issues;
         if (!/\.(cpp|cc|cxx|h|hpp|hh|hxx|H)$/i.test(facts.filePath)) return issues;
+        if (TEST_FILE_REGEX.test(facts.filePath)) return issues;
         const printfCount = (source.match(PRINTF_FAMILY_REGEX) ?? []).length;
         let viaPrintf = false;
         if (printfCount > context.threshold) {
@@ -27805,7 +27833,7 @@ var init_printf_debug = __esm({
           message: viaPrintf ? `${printfCount} printf-family calls \u2014 use spdlog / glog / AbslLog` : "std::cout/cerr/clog with a string literal \u2014 use spdlog / glog / AbslLog",
           line: 1,
           column: 1,
-          advice: "Replace with `spdlog::info(...)`, `LOG(INFO) << ...` (glog), or `ABSL_LOG(INFO) << ...` (Abseil). All of these are level-aware and have a configurable sink, and they route to stderr by default. `printf` / `std::cout` have no levels, no redaction, no sink routing, and can't be silenced in release builds. AI agents reach for these because their training data has countless C++ textbook examples with them. Reference: cpp/printf-debug v0.24."
+          advice: "Replace with `spdlog::info(...)`, `LOG(INFO) << ...` (glog), or `ABSL_LOG(INFO) << ...` (Abseil). All of these are level-aware and have a configurable sink, and they route to stderr by default. `printf` / `std::cout` have no levels, no redaction, no sink routing, and can't be silenced in release builds. AI agents reach for these because their training data has countless C++ textbook examples with them. Reference: cpp/printf-debug v0.34.7 (refined to skip test files for higher precision)."
         });
         return issues;
       }
@@ -30842,12 +30870,11 @@ var init_system_out_println = __esm({
 });
 
 // src/rules/java/thread-sleep-in-loop.ts
-var THREAD_SLEEP_REGEX, javaThreadSleepInLoopRule;
+var javaThreadSleepInLoopRule;
 var init_thread_sleep_in_loop = __esm({
   "src/rules/java/thread-sleep-in-loop.ts"() {
     "use strict";
     init_rule();
-    THREAD_SLEEP_REGEX = /\bThread\.sleep\s*\(/g;
     javaThreadSleepInLoopRule = createRule({
       id: "java/thread-sleep-in-loop",
       category: "perf",
@@ -30863,11 +30890,104 @@ var init_thread_sleep_in_loop = __esm({
         if (!source) return issues;
         if (!/\.java$/i.test(facts.filePath)) return issues;
         if (!/\bThread\.sleep\s*\(/.test(source)) return issues;
-        if (!/\b(?:for|while|do)\b/.test(source)) return issues;
-        let m;
-        THREAD_SLEEP_REGEX.lastIndex = 0;
-        while ((m = THREAD_SLEEP_REGEX.exec(source)) !== null) {
-          const line = source.slice(0, m.index).split("\n").length;
+        const braceStack = [];
+        const loopSet = /* @__PURE__ */ new Set();
+        let pendingLoopKeyword = null;
+        let parenDepth = 0;
+        const sleepEvents = [];
+        let inString = false;
+        let inLineComment = false;
+        let inBlockComment = false;
+        for (let i = 0; i < source.length; i++) {
+          const c = source[i] ?? "";
+          const next = source[i + 1] ?? "";
+          const prev = source[i - 1] ?? "";
+          if (inLineComment) {
+            if (c === "\n") inLineComment = false;
+            continue;
+          }
+          if (inBlockComment) {
+            if (c === "*" && next === "/") {
+              inBlockComment = false;
+              i++;
+            }
+            continue;
+          }
+          if (inString) {
+            if (c === "\\") {
+              i++;
+              continue;
+            }
+            if (c === inString) inString = false;
+            continue;
+          }
+          if (c === "/" && next === "/") {
+            inLineComment = true;
+            i++;
+            continue;
+          }
+          if (c === "/" && next === "*") {
+            inBlockComment = true;
+            i++;
+            continue;
+          }
+          if (c === '"' || c === "'") {
+            inString = c;
+            continue;
+          }
+          if (c === "(") {
+            parenDepth++;
+            continue;
+          }
+          if (c === ")") {
+            parenDepth = Math.max(0, parenDepth - 1);
+            continue;
+          }
+          if (/[A-Za-z_]/.test(c) && !/[A-Za-z0-9_]/.test(prev)) {
+            const next3 = source.slice(i, i + 3);
+            const next5 = source.slice(i, i + 5);
+            const next2 = source.slice(i, i + 2);
+            const after3 = source[i + 3] ?? "";
+            const after5 = source[i + 5] ?? "";
+            const after2 = source[i + 2] ?? "";
+            if (next3 === "for" && !/[A-Za-z0-9_]/.test(after3)) {
+              pendingLoopKeyword = { kind: "for", idx: i };
+              i += 2;
+              continue;
+            }
+            if (next5 === "while" && !/[A-Za-z0-9_]/.test(after5)) {
+              pendingLoopKeyword = { kind: "while", idx: i };
+              i += 4;
+              continue;
+            }
+            if (next2 === "do" && !/[A-Za-z0-9_]/.test(after2)) {
+              pendingLoopKeyword = { kind: "do", idx: i };
+              i += 1;
+              continue;
+            }
+          }
+          if (c === "{") {
+            braceStack.push(i);
+            if (pendingLoopKeyword && parenDepth === 0) {
+              loopSet.add(i);
+              pendingLoopKeyword = null;
+            }
+          } else if (c === "}") {
+            const popped = braceStack.pop();
+            if (popped !== void 0 && loopSet.has(popped)) {
+              loopSet.delete(popped);
+            }
+            if (pendingLoopKeyword) {
+              pendingLoopKeyword = null;
+            }
+          } else if (c === "T" && source.slice(i, i + 13) === "Thread.sleep(") {
+            sleepEvents.push({ idx: i, loopDepth: loopSet.size });
+            i += 12;
+          }
+        }
+        for (const ev of sleepEvents) {
+          if (ev.loopDepth === 0) continue;
+          const line = source.slice(0, ev.idx).split("\n").length;
           issues.push({
             ruleId: "java/thread-sleep-in-loop",
             category: "perf",
@@ -30876,7 +30996,7 @@ var init_thread_sleep_in_loop = __esm({
             message: `Thread.sleep() at line ${line}`,
             line,
             column: 1,
-            advice: 'Use ScheduledExecutorService for periodic work, or BlockingQueue.take() for event-driven work. Thread.sleep in a loop is the classic "polling with sleep" anti-pattern \u2014 the thread blocks for the sleep duration each iteration. In server contexts this ties up Tomcat/Jetty/Netty threads. Reference: java/thread-sleep-in-loop v0.30.'
+            advice: 'Use ScheduledExecutorService for periodic work, or BlockingQueue.take() for event-driven work. Thread.sleep in a loop is the classic "polling with sleep" anti-pattern \u2014 the thread blocks for the sleep duration each iteration. In server contexts this ties up Tomcat/Jetty/Netty threads. Reference: java/thread-sleep-in-loop v0.34.6 (refined to require Thread.sleep inside the loop block, not just in the file).'
           });
         }
         return issues;
@@ -31158,13 +31278,14 @@ var init_object_singleton_misuse = __esm({
 });
 
 // src/rules/kotlin/println-as-log.ts
-var PRINTLN_REGEX, REAL_LOGGING_IMPORT_REGEX2, kotlinPrintlnAsLogRule;
+var PRINTLN_REGEX, REAL_LOGGING_IMPORT_REGEX2, TEST_FILE_REGEX2, kotlinPrintlnAsLogRule;
 var init_println_as_log = __esm({
   "src/rules/kotlin/println-as-log.ts"() {
     "use strict";
     init_rule();
     PRINTLN_REGEX = /\bprintln\s*\(/g;
     REAL_LOGGING_IMPORT_REGEX2 = /\bimport\s+(?:android\.util\.Log|org\.slf4j\.|io\.github\.oshai\.kotlinlogging|kotlin\.logging|co\.touchlab\.kermit|com\.github\.ajalt\.timber|org\.apache\.logging\.log4j)/;
+    TEST_FILE_REGEX2 = /(?:\/src\/test\/|\/test\/|\/Tests\/|\/Test\.kt|\/Tests\.kt|Tests\.kt$|Test\.kt$)/;
     kotlinPrintlnAsLogRule = createRule({
       id: "kotlin/println-as-log",
       category: "logic",
@@ -31179,7 +31300,7 @@ var init_println_as_log = __esm({
         const source = facts.v2?._source;
         if (!source) return issues;
         if (!/\.kts?$/i.test(facts.filePath)) return issues;
-        if (/\/test\//i.test(facts.filePath) || /\.test\.kts?$/i.test(facts.filePath)) return issues;
+        if (TEST_FILE_REGEX2.test(facts.filePath)) return issues;
         if (REAL_LOGGING_IMPORT_REGEX2.test(source)) return issues;
         let m;
         PRINTLN_REGEX.lastIndex = 0;
@@ -31193,7 +31314,7 @@ var init_println_as_log = __esm({
             message: `println() as logger at line ${line}`,
             line,
             column: 1,
-            advice: "Use a real logging library: slf4j (JVM), android.util.Log (Android), Timber (Android), kermit (multiplatform), or kotlin-logging. println() has no log level, no timestamp, no correlation ID, and cannot be filtered. Reference: kotlin/println-as-log v0.29."
+            advice: "Use a real logging library: slf4j (JVM), android.util.Log (Android), Timber (Android), kermit (multiplatform), or kotlin-logging. println() has no log level, no timestamp, no correlation ID, and cannot be filtered. Reference: kotlin/println-as-log v0.34.5 (refined to skip test files for higher precision)."
           });
         }
         return issues;
@@ -41038,14 +41159,14 @@ var init_implicitly_unwrapped_optional = __esm({
 });
 
 // src/rules/swift/print-debug.ts
-var PRINT_REGEX, DEFAULT_THRESHOLD2, TEST_FILE_REGEX, swiftPrintDebugRule;
+var PRINT_REGEX, DEFAULT_THRESHOLD2, TEST_FILE_REGEX3, swiftPrintDebugRule;
 var init_print_debug = __esm({
   "src/rules/swift/print-debug.ts"() {
     "use strict";
     init_rule();
     PRINT_REGEX = /\bprint\s*\(/g;
     DEFAULT_THRESHOLD2 = 1;
-    TEST_FILE_REGEX = /(?:\/Tests\/|\/Tests\.swift|\/Test\.swift|Tests\.swift$|Test\.swift$)/;
+    TEST_FILE_REGEX3 = /(?:\/Tests\/|\/Tests\.swift|\/Test\.swift|Tests\.swift$|Test\.swift$)/;
     swiftPrintDebugRule = createRule({
       id: "swift/print-debug",
       category: "typo",
@@ -41060,7 +41181,7 @@ var init_print_debug = __esm({
         const source = facts.v2?._source;
         if (!source) return issues;
         if (!/\.swift$/i.test(facts.filePath)) return issues;
-        if (TEST_FILE_REGEX.test(facts.filePath)) return issues;
+        if (TEST_FILE_REGEX3.test(facts.filePath)) return issues;
         const matches = [];
         let m;
         PRINT_REGEX.lastIndex = 0;
@@ -51532,7 +51653,7 @@ var init_signal_strength = __esm({
         precision: 0.1268,
         lastCalibratedAt: "2026-07-03T00:00:00Z",
         verdict: "OK",
-        _calibrationNote: "v0.29: v9 Kotlin calibration (2698 neg, 213 pos). ratio=1.84 (\u22651.5) \u2014 first rule in this entire session with a positive direction! Fires 1.84x more on post-2024 AI/demos than pre-2022 production. Precision=12.7% (below 50% USEFUL threshold); verdict=OK. The signal is real: post-2024 Kotlin code (especially AI-generated examples) uses println for output; pre-2022 production code uses slf4j/kermit. INSUFFICIENT_DATA: pos arm only 213 files. defaultOff: still set to true (verdict is OK but precision is below 50% \u2014 the guardrail expects OK/USEFUL rules to be defaultOff:false only when calibrated with a meaningful pos arm).",
+        _calibrationNote: "v0.34.5: REFINED \u2014 rule now skips test files (JUnit4/5 + Android: *Tests.kt, *Test.kt, src/test/, test/ dirs). Per-file unique v9 Kotlin calibration (2698 neg, 213 pos): ratio=1.84 (\u22651.5, OK). The refinement mirrors v0.34.2's swift/print-debug fix: the previous exclusion only matched `\\/test\\/` and `.test.kts?$`, missing JUnit5's `FooTests.kt` naming. Expected post-refinement precision: 12.7% \u2192 25%+ by removing the JUnit assertion / debug-output fires that were ~50% of the FPs. Same direction as java/system-out-println (1.73 refined) and the v0.34.2 swift/print-debug pipeline. Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.29: v9 Kotlin calibration (2698 neg, 213 pos). ratio=1.84 (\u22651.5) \u2014 first rule in this entire session with a positive direction! Fires 1.84x more on post-2024 AI/demos than pre-2022 production. Precision=12.7% (below 50% USEFUL threshold); verdict=OK. The signal is real: post-2024 Kotlin code (especially AI-generated examples) uses println for output; pre-2022 production code uses slf4j/kermit. INSUFFICIENT_DATA: pos arm only 213 files. defaultOff: still set to true (verdict is OK but precision is below 50% \u2014 the guardrail expects OK/USEFUL rules to be defaultOff:false only when calibrated with a meaningful pos arm).",
         aiSpecific: false,
         _v9Verdict: "OK",
         _v9Lift: 1.84,
@@ -51596,7 +51717,7 @@ var init_signal_strength = __esm({
         precision: 0.1091,
         lastCalibratedAt: "2026-07-03T00:00:00Z",
         verdict: "DORMANT",
-        _calibrationNote: "v0.30: v9 Java calibration (81891 neg, 10305 pos). ratio=0.97 (just below 1.0) \u2014 fires equally on both arms. Borderline era-confound: pre-2022 Java had more Thread.sleep in loops (less mature concurrency APIs); modern Java uses ScheduledExecutorService. Same direction as kotlin/runblocking-misuse (0.50). 2154 total fires \u2014 high absolute count, meaningful measurement. defaultOff.",
+        _calibrationNote: "v0.34.6: REFINED \u2014 rule now requires Thread.sleep to be INSIDE the loop block (verified via brace-counting), not just in a file that happens to contain a for/while/do keyword. v0.30 baseline: 2154 total fires, ratio 0.97 (DORMANT). The refinement targets: (1) Thread.sleep in `main()` no longer fires if a different method has a `for` loop (the v0.30 heuristic fired on every Thread.sleep in the file); (2) Thread.sleep before/after a loop block (not inside it) no longer fires; (3) string literals containing `Thread.sleep(...)` are skipped via the string-state-machine. Expected post-refinement ratio: 1.5+ (positive-signal OK verdict) by removing the cross-method over-fires. Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.30: v9 Java calibration (81891 neg, 10305 pos). ratio=0.97 (just below 1.0) \u2014 fires equally on both arms. Borderline era-confound: pre-2022 Java had more Thread.sleep in loops (less mature concurrency APIs); modern Java uses ScheduledExecutorService. Same direction as kotlin/runblocking-misuse (0.50). 2154 total fires \u2014 high absolute count, meaningful measurement. defaultOff.",
         aiSpecific: false,
         _v9Verdict: "DORMANT",
         _v9Lift: 0.97,
@@ -51828,7 +51949,7 @@ var init_signal_strength = __esm({
         precision: 0.4407,
         lastCalibratedAt: "2026-07-03T00:00:00Z",
         verdict: "OK",
-        _calibrationNote: "v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 156 TP files, 198 FP files, ratio=2.43 (\u22651.5) \u2014 fourth positive-signal rule in v9 history (and the strongest ratio so far). precision=44.07% (just below 50% USEFUL threshold); verdict=OK. Total fires: 182 TP, 198 FP. The signal is real: post-2024 C++ (mostly AI-generated demos in llama.cpp, whisper.cpp, openai-cpp) uses printf/cout for output; pre-2022 production C++ (folly, protobuf, abseil) uses spdlog / glog / AbslLog. Same direction as kotlin/println-as-log (1.84), java/system-out-println (1.73 refined), swift/print-debug (1.13). INSUFFICIENT_DATA: pos arm 1655 files (below 10k floor).",
+        _calibrationNote: "v0.34.7: REFINED \u2014 rule now skips test files (gtest, catch2, doctest conventions: *_test.cpp, *_test.cc, /tests/ dir, *Test.cpp, *Test.cc, *_unittest.cpp). Per-file unique v9 C++ calibration (5107 neg, 1655 pos): 156 TP files, 198 FP files, ratio=2.43 (OK, fourth positive-signal in v9 history). precision=44.07% (below 50% USEFUL threshold); verdict=OK. The refinement is expected to push precision from 44% to 50%+ by removing gtest test output fires (which were a significant portion of FPs in the v0.33.0 measurement). The full v9 re-calibration is part of the broader v0.34.5/v0.34.7 pipeline; see v0.34.1's Four-language validation section. Same direction as kotlin/println-as-log (1.84), java/system-out-println (1.73 refined, 3.29 unrefined), swift/print-debug (1.13). INSUFFICIENT_DATA: pos arm 1655 files (below 10k floor).",
         aiSpecific: true,
         _v7Verdict: "DORMANT",
         _v7Lift: 1,
@@ -51851,7 +51972,7 @@ var init_signal_strength = __esm({
         precision: 0.2187,
         lastCalibratedAt: "2026-07-03T00:00:00Z",
         verdict: "DORMANT",
-        _calibrationNote: "v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 220 TP, 786 FP, ratio=0.86. Both arms have similar proportions of files with magic numbers \u2014 this is a general C++ anti-pattern that hasn't been eradicated. INSUFFICIENT_DATA: pos arm 1655 files.",
+        _calibrationNote: "v0.34.4: REFINED \u2014 expanded allowSet with common constants (-1 sentinel, 100 percent literal, 0.5/1.0/2.0 ratios, 4096/2048/512/128/64/32/16/8 power-of-2 sizes, 50 percentile). Added string-literal and `//` comment stripping so substrings like `\"got 42 errors\"` and `// ticket #4242` no longer fire. v0.33 baseline: 220 TP / 786 FP per-file (ratio 0.86, DORMANT). The refinement targets: (1) sentinel `-1` literals are skipped because `1` is allowlisted and the `-` doesn't match MAGIC_NUMBER_REGEX anyway; (2) hex literals (`0xFF`) are skipped because the digit-only regex requires a word-boundary after digits, and `0x...` has none; (3) literals inside string/comment contexts no longer fire. Expected post-refinement ratio: 1.0-1.2 (still DORMANT but better-targeted). Full v9 re-calibration is part of the v0.35.0 re-measurement. v0.33: v9 C++ calibration (5107 neg, 1655 pos). Per-file unique: 220 TP, 786 FP, ratio=0.86. Both arms have similar proportions of files with magic numbers \u2014 this is a general C++ anti-pattern that hasn't been eradicated. INSUFFICIENT_DATA: pos arm 1655 files.",
         aiSpecific: true,
         _v7Verdict: "DORMANT",
         _v7Lift: 1,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "slopbrick",
-  "version": "0.34.3",
+  "version": "0.34.7",
   "description": "Discovered, modeled, and governed repository structure. SlopBrick scans source code, classifies it against 95+ rules in 15 categories, computes 4 scores (aiSlopScore: lower=cleaner, engineeringHygiene, security, repositoryHealth composite), and persists the structure for AI agents and CI.",
   "type": "module",
   "bin": {