loopx-project · shangzh0 · Sep 18, 2026
@@ -229,12 +229,16 @@ creator dogfooding, reproducible demos, and explicit evidence-strength labels.
 - **[SWE-Marathon](https://huangruiteng.github.io/loopx/benchmarks/swe-marathon/):**
   Five execution modes on 15 matched tasks compare self-verification, scores,
   and cost. More self-verification did not consistently yield higher scores.
+- **[LHTB × LoopX](https://huangruiteng.github.io/loopx/benchmarks/lhtb/):**
+  Five execution mechanisms on 46 long-horizon terminal tasks compare durable
+  state, bounded Todos, replanning, and fresh executor sessions.
 - **[DeepSWE behavior analysis](https://huangruiteng.github.io/loopx/benchmarks/deepswe/behavior-discovery/)** (Chinese):
   Selected cases examine how domain hints relate to requirement retention and
   verification choices, offering mechanism hypotheses for further testing.
 
-SWE-Marathon has one trial per task and mode; DeepSWE uses selected cases and
-post-hoc analysis. Neither establishes a general performance gain.
+SWE-Marathon and LHTB have one effective trial per task and mode; DeepSWE uses
+selected cases and post-hoc analysis. None establishes a general performance
+gain.
 
 More inspectable surfaces:
 

@@ -200,11 +200,14 @@ creator dogfooding、reproducible demo 和证据强度标签见
 - **[SWE-Marathon](https://huangruiteng.github.io/loopx/benchmarks/swe-marathon/?lang=zh)**：
   在相同的 15 个任务上对照 5 种执行模式，比较自验证行为、得分与成本。
   更多自验证并未稳定转化为更高得分。
+- **[LHTB × LoopX](https://huangruiteng.github.io/loopx/benchmarks/lhtb/?lang=zh)**：
+  在 46 个长程终端任务上对比 5 种执行机制，研究持久状态、Todo、replan
+  与 fresh executor session。
 - **[DeepSWE 行为分析](https://huangruiteng.github.io/loopx/benchmarks/deepswe/behavior-discovery/)**：
   通过精选案例观察领域提示、需求保留与验证选择之间的关系，提出有待复验的机制假设。
 
-SWE-Marathon 每个任务、每种模式仅运行一次；DeepSWE 包含精选案例与事后分析。
-目前两者均不足以证明普遍的性能提升。
+SWE-Marathon 与 LHTB 每个任务、每种模式仅保留一条有效轨迹；DeepSWE 包含精选案例与
+事后分析。目前三者均不足以证明普遍的性能提升。
 
 更多可检查入口：
 

@@ -140,6 +140,7 @@ const content = {
       cards: [
         ["Product demo", "Personal Workspace", "A shared view of Goals, Tasks, Chat, and outputs. Watch the demo and start your own local workspace.", "Watch demo & read guide"],
         ["Research & evaluation", "SWE-Marathon", "Compare three retained execution modes across 15 matched tasks, with results, costs, and study limitations.", "Read the study"],
+        ["Research & evaluation", "LHTB × LoopX", "Compare five execution mechanisms across 46 long-horizon terminal tasks and inspect where recoverable state helps.", "Read the study"],
         ["Research · Chinese", "DeepSWE behavior discoveries", "Explore how domain hints affect implementation and validation in individual cases.", "Read the behavior analysis"],
         ["Research · Chinese", "DeepSWE × Sol", "Read the historical 113-task results, continuation mechanisms, and patch-delivery checks.", "Read the research brief"],
         ["Cases", "All LoopX showcases", "Browse public cases, interactive walkthroughs, and their evidence boundaries.", "Browse all cases"],
@@ -254,6 +255,7 @@ const content = {
       cards: [
         ["产品演示", "Personal Workspace", "在一个工作区查看目标、任务、对话与产出。观看演示，开始使用本地工作区。", "观看演示与使用指南"],
         ["研究与评测", "SWE-Marathon", "查看三种保留模式在 15 个匹配任务上的结果、成本与研究局限。", "阅读研究简报"],
+        ["研究与评测", "LHTB × LoopX", "对比五种执行机制在 46 个长程终端任务上的结果，观察可恢复状态在什么场景有效。", "阅读研究简报"],
         ["研究与评测", "DeepSWE 行为分析", "从具体案例观察领域提示如何影响实现选择与验证行为。", "阅读行为分析"],
         ["研究与评测", "DeepSWE × Sol", "解读 113 任务的历史结果、持续执行机制，以及补丁如何进入最终验收。", "阅读研究简报"],
         ["案例", "完整案例目录", "浏览公开案例、交互式讲解及其证据边界。", "浏览全部案例"],
@@ -1011,6 +1013,7 @@ export function App() {
                 const paths = [
                   "docs/guides/personal-workspace-user-guide/",
                   `benchmarks/swe-marathon/${language === "zh" ? "?lang=zh" : ""}`,
+                  `benchmarks/lhtb/${language === "zh" ? "?lang=zh" : ""}`,
                   "benchmarks/deepswe/behavior-discovery/",
                   "benchmarks/deepswe-sol/",
                   `docs/showcases/index${language === "en" ? ".en" : ""}.html`,

@@ -48,6 +48,9 @@ by itself establish a C2 uplift claim.
 
 - [`swe-marathon/README.md`](swe-marathon/README.md) links the published
   [SWE-Marathon research brief](https://huangruiteng.github.io/loopx/benchmarks/swe-marathon/).
+- [`LHTB/studies/five-arm-gpt56sol-max/README.md`](LHTB/studies/five-arm-gpt56sol-max/README.md)
+  documents the public-safe aggregate behind the bilingual
+  [LHTB research brief](https://huangruiteng.github.io/loopx/benchmarks/lhtb/).
 - [`deepswe/behavior-discovery/README.md`](deepswe/behavior-discovery/README.md)
   links the standalone
   [DeepSWE behavior-discovery article](https://huangruiteng.github.io/loopx/benchmarks/deepswe/behavior-discovery/).

@@ -80,3 +80,31 @@ def test_published_data_and_bilingual_tables_share_scope(exporters):
     for localized in copy.values():
         assert {row[0] for row in localized["armRows"]} == allowed
         assert set(localized["executiveReads"]) == allowed
+
+
+def test_lhtb_published_data_and_bilingual_copy_share_scope():
+    study = STUDY.parent / "LHTB" / "studies" / "five-arm-gpt56sol-max"
+    data = json.loads((study / "data.json").read_text())
+    arms = {
+        "plain",
+        "native_goal",
+        "ssh_goal",
+        "legacy_heartbeat",
+        "new_heartbeat",
+    }
+    assert set(data["arms"]) == arms
+    assert len(data["tasks"]) == len({row["task"] for row in data["tasks"]}) == 46
+    assert all(set(row) == arms | {"task"} for row in data["tasks"])
+
+    for arm, summary in data["arms"].items():
+        rewards = [row[arm] for row in data["tasks"]]
+        assert summary["mean_reward"] == pytest.approx(sum(rewards) / 46, rel=0, abs=1e-12)
+        assert summary["pass_095"] == sum(reward >= 0.95 for reward in rewards)
+
+    site = STUDY.parents[1] / "apps/presentation/site/src"
+    localized_copy = json.loads((site / "lhtb-copy.json").read_text())
+    assert set(localized_copy) == {"en", "zh"}
+    for localized in localized_copy.values():
+        assert set(localized["armLabels"]) == arms
+        assert set(localized["armKinds"]) == arms
+        assert {row[0] for row in localized["mechanismRows"]} == arms
@@ -6,7 +6,7 @@ const read = (path) => readFileSync(fileURLToPath(new URL(`../${path}`, import.m
 const home = read("apps/presentation/site/src/App.tsx");
 const styles = read("apps/presentation/site/src/styles.css");
 assert.doesNotMatch(home, /href=.[^\n]*(?:deprecated|frontstage\/)/, "homepage must not promote retired surfaces");
-for (const destination of ["docs/guides/personal-workspace-user-guide/", "benchmarks/swe-marathon/", "benchmarks/deepswe/behavior-discovery/", "benchmarks/deepswe-sol/", "docs/showcases/index", "developers/projections/"]) {
+for (const destination of ["docs/guides/personal-workspace-user-guide/", "benchmarks/swe-marathon/", "benchmarks/lhtb/", "benchmarks/deepswe/behavior-discovery/", "benchmarks/deepswe-sol/", "docs/showcases/index", "developers/projections/"]) {
   assert.ok(home.includes(destination), `missing public destination: ${destination}`);
 }
 assert.ok(styles.includes("prefers-reduced-motion"));

@@ -268,6 +268,7 @@ async function writeShareReadme(outDir, base, interactivePages) {
   const homepageUrl = base;
   const frontstageUrl = `${base}frontstage/`;
   const sweMarathonBriefUrl = `${base}benchmarks/swe-marathon/`;
+  const lhtbBriefUrl = `${base}benchmarks/lhtb/`;
   const deepSweBehaviorArticleUrl = `${base}benchmarks/deepswe/behavior-discovery/`;
   const previewBlock = base === "/"
     ? `## Try It Locally
@@ -283,6 +284,7 @@ Then open the homepage or showcase:
 http://127.0.0.1:8080${homepageUrl}
 http://127.0.0.1:8080${frontstageUrl}
 http://127.0.0.1:8080${sweMarathonBriefUrl}
+http://127.0.0.1:8080${lhtbBriefUrl}
 http://127.0.0.1:8080${deepSweBehaviorArticleUrl}
 \`\`\`
 `
@@ -298,6 +300,7 @@ Hosted entries:
 ${homepageUrl}
 ${frontstageUrl}
 ${sweMarathonBriefUrl}
+${lhtbBriefUrl}
 ${deepSweBehaviorArticleUrl}
 \`\`\`
 `;
@@ -322,6 +325,7 @@ ${previewBlock}
   catalog-declared interactive case pages.
 - Homepage source: \`apps/presentation/site\`.
 - SWE-Marathon research brief: \`${sweMarathonBriefUrl}\`, built from the pinned public-safe aggregate and case-insight projection under \`benchmark/swe-marathon/\`.
+- LHTB research brief: \`${lhtbBriefUrl}\`, built from the public-safe five-arm aggregate under \`benchmark/LHTB/studies/five-arm-gpt56sol-max/\`.
 - DeepSWE behavior discoveries: \`${deepSweBehaviorArticleUrl}\`, copied byte-for-byte from the reviewed standalone article at \`${deepSweBehaviorArticlePath}\`.
 - DeepSWE × Sol research brief: \`${base}benchmarks/deepswe-sol/\`, a static historical-study interpretation from \`${deepSweSolArticlePath}\`.
 - Homepage evidence assets: ${homepageEvidenceAssets.map((path) => `\`${path}\``).join(", ")}.
@@ -345,6 +349,7 @@ async function writeManifest(outDir, base, interactivePages) {
     status_fixture: `site/${statusFileName}`,
     homepage_entry: "site/index.html",
     swe_marathon_brief_entry: "site/benchmarks/swe-marathon/index.html",
+    lhtb_brief_entry: "site/benchmarks/lhtb/index.html",
     deepswe_behavior_article_entry: "site/benchmarks/deepswe/behavior-discovery/index.html",
     deepswe_sol_article_entry: "site/benchmarks/deepswe-sol/index.html",
     installer_entry: "site/install.sh",
@@ -354,6 +359,7 @@ async function writeManifest(outDir, base, interactivePages) {
     content_sources: {
       public_homepage: "apps/presentation/site",
       swe_marathon_brief: "benchmark/swe-marathon",
+      lhtb_brief: "benchmark/LHTB/studies/five-arm-gpt56sol-max",
       deepswe_behavior_article: deepSweBehaviorArticlePath,
       deepswe_sol_article: deepSweSolArticlePath,
       installer_script: installerScriptPath,
@@ -474,6 +480,7 @@ async function main() {
     site_dir: siteDir,
     homepage_url: args.base,
     swe_marathon_brief_url: `${args.base}benchmarks/swe-marathon/`,
+    lhtb_brief_url: `${args.base}benchmarks/lhtb/`,
     deepswe_behavior_article_url: `${args.base}benchmarks/deepswe/behavior-discovery/`,
     frontstage_url: `${args.base}frontstage/`,
     status_fixture: `site/${statusFileName}`,

@@ -234,6 +234,10 @@ const benchmarkHtml = await readFile(resolve(siteDir, "benchmarks/swe-marathon/i
 if (benchmarkHtml !== homepageHtml) {
   throw new Error("SWE-Marathon static route must reuse the compiled public-site entry");
 }
+const lhtbHtml = await readFile(resolve(siteDir, "benchmarks/lhtb/index.html"), "utf8");
+if (lhtbHtml !== homepageHtml) {
+  throw new Error("LHTB static route must reuse the compiled public-site entry");
+}
 const deepSweBehaviorHtml = await readFile(
   resolve(siteDir, "benchmarks/deepswe/behavior-discovery/index.html"),
   "utf8",
@@ -367,6 +371,7 @@ if (manifest.base !== "/loopx/") {
 if (
   manifest.homepage_entry !== "site/index.html" ||
   manifest.swe_marathon_brief_entry !== "site/benchmarks/swe-marathon/index.html" ||
+  manifest.lhtb_brief_entry !== "site/benchmarks/lhtb/index.html" ||
   manifest.deepswe_behavior_article_entry !== "site/benchmarks/deepswe/behavior-discovery/index.html" ||
   manifest.frontstage_entry !== "site/frontstage/index.html" ||
   manifest.installer_entry !== "site/install.sh"
@@ -379,6 +384,9 @@ if (manifest.content_sources?.public_homepage !== "apps/presentation/site") {
 if (manifest.content_sources?.swe_marathon_brief !== "benchmark/swe-marathon") {
   throw new Error(`manifest benchmark brief source mismatch: ${JSON.stringify(manifest.content_sources)}`);
 }
+if (manifest.content_sources?.lhtb_brief !== "benchmark/LHTB/studies/five-arm-gpt56sol-max") {
+  throw new Error(`manifest LHTB brief source mismatch: ${JSON.stringify(manifest.content_sources)}`);
+}
 if (
   manifest.content_sources?.deepswe_behavior_article !==
   "benchmark/deepswe/behavior-discovery/index.html"
@@ -458,6 +466,9 @@ if (!readmeText.includes("frontstage/")) {
 if (!readmeText.includes("benchmarks/swe-marathon/")) {
   throw new Error("share bundle README must publish the SWE-Marathon research brief entry");
 }
+if (!readmeText.includes("benchmarks/lhtb/")) {
+  throw new Error("share bundle README must publish the LHTB research brief entry");
+}
 if (!readmeText.includes("benchmarks/deepswe/behavior-discovery/")) {
   throw new Error("share bundle README must publish the DeepSWE behavior article entry");
 }