clawdotnet · Telli · Sep 27, 2026 · Sep 27, 2026 · Sep 27, 2026 · Sep 27, 2026
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -86,9 +86,7 @@ jobs:
         run: python3 eng/verify-vault-publish-boundary.py
 
       - name: Test local decision service and evaluation tools
-        run: |
-          python3 -B -m unittest discover -s tests/laya-service -v
-          python3 -B -m unittest discover -s tests/routing-eval -p test_jev_report.py
+        run: dotnet test tools/laya_service/tests/LayaService.Tests.csproj -c Release --no-build
 
       - name: Test standard targets
         run: dotnet test --no-build -c Release --verbosity normal --logger "trx;LogFileName=results.trx" src/OpenClaw.Tests

diff --git a/OpenClaw.Net.slnx b/OpenClaw.Net.slnx
@@ -55,6 +55,13 @@
     <Project Path="src/OpenClaw.WhatsApp.BaileysWorker/OpenClaw.WhatsApp.BaileysWorker.csproj" />
     <Project Path="src/OpenClawNet.Sandbox.OpenSandbox/OpenClawNet.Sandbox.OpenSandbox.csproj" />
   </Folder>
+  <Folder Name="/tools/" />
+  <Folder Name="/tools/laya_service/">
+    <Project Path="tools/laya_service/LayaService.csproj" />
+  </Folder>
+  <Folder Name="/tools/laya_service/tests/">
+    <Project Path="tools/laya_service/tests/LayaService.Tests.csproj" />
+  </Folder>
   <Project Path="src/OpenClaw.Agent/OpenClaw.Agent.csproj" />
   <Project Path="src/OpenClaw.Channels/OpenClaw.Channels.csproj" />
   <Project Path="src/OpenClaw.Core/OpenClaw.Core.csproj" />

diff --git a/docs/README.md b/docs/README.md
@@ -43,6 +43,7 @@ Use this page as the map. If you are unsure where to go next, the groups below a
 | [opensquilla-meta-skill-migration.md](opensquilla-meta-skill-migration.md) | OpenClaw.NET meta-skill migration notes for OpenSquilla-style DAG orchestration, current parity, and remaining migration gaps. |
 | [dynamic-turn-routing-model-profiles.md](dynamic-turn-routing-model-profiles.md) | Collaboration playbook for how Dynamic Turn Routing and Model Profiles work together (precedence, field projection, fallback, and restore semantics). |
 | [laya-routing.md](laya-routing.md) | Local Laya decision routing, model preparation, calibration, attribution, and rollback. |
+| [zh-CN/integrations/laya-routing.md](zh-CN/integrations/laya-routing.md) | 本地 Laya 路由服务、.NET CLI、模型校验、校准与回滚（简体中文）。 |
 | [jev-routing.md](jev-routing.md) | Optional Jev shadow evaluation and model routing, configuration, diagnostics, quality reports, and rollback. |
 | [EXTERNAL_CLI_CONNECTORS.md](EXTERNAL_CLI_CONNECTORS.md) | Governed external CLI connectors, optional presets, named command allowlists, approvals, redaction, and audit behavior. |
 | [plugins/payment.md](plugins/payment.md) | Native payment tool, virtual cards, machine payments, providers, and safe agent-facing actions. |

diff --git a/docs/SITE_MAP.md b/docs/SITE_MAP.md
@@ -56,6 +56,9 @@ Use this map when turning the Markdown docs into a documentation website. It kee
 | Integrations | Microsoft Teams | [TEAMS_SETUP.md](TEAMS_SETUP.md) |
 | Integrations | WhatsApp | [WHATSAPP_SETUP.md](WHATSAPP_SETUP.md) |
 | Integrations | A2A | [a2a.md](a2a.md) |
+| Integrations | Local Laya Decision Routing | [laya-routing.md](laya-routing.md) |
+| Integrations | Jev Decision Routing | [jev-routing.md](jev-routing.md) |
+| Integrations | Local Laya Routing (zh-CN) | [zh-CN/integrations/laya-routing.md](zh-CN/integrations/laya-routing.md) |
 | Integrations | External Coding Backends | [external-coding-backends.md](external-coding-backends.md) |
 | Integrations | Nacos MCP Router (PoC) | [nacos-mcp-router.md](nacos-mcp-router.md) |
 | Integrations | Nacos MCP Router (PoC, zh-CN) | [zh-CN/nacos-mcp-router.md](zh-CN/nacos-mcp-router.md) |
@@ -146,6 +149,9 @@ Integrations
   Microsoft Teams
   WhatsApp
   A2A
+  Local Laya Decision Routing
+  Jev Decision Routing
+  Local Laya Routing (zh-CN)
   External Coding Backends
   Nacos MCP Router (PoC)
   Tailscale Deployment

diff --git a/docs/cli/routing.md b/docs/cli/routing.md
@@ -27,6 +27,14 @@ Router mode behavior:
 
 `routing status` also reports `jevMode`, `jevModel`, `layaMode`, `layaModel`, and `layaCalibrationId`. See [local Laya routing](../laya-routing.md) for offline setup and calibration. Jev is disabled by default; its `shadow` and `active` modes are configured through `DynamicTurnRouting.Jev`. See [Jev routing](../jev-routing.md) for the evaluation workflow. The existing `diagnostics on|off` command controls the ONNX policy flag; Jev journaling is controlled separately by `Jev.DiagnosticsPath`.
 
+Jev and Laya journal summaries share the standalone .NET report tool. It can write JSON and optional ScottPlot PNG reliability/risk-coverage charts; see the provider guides for label schema and examples:
+
+```bash
+dotnet run --project tools/laya_service -c Release -- report \
+	/path/to/decisions.snapshot.jsonl --labels /path/to/labels.jsonl \
+	--output /path/to/report.json --plot /path/to/reliability.png
+```
+
 Examples:
 
 ```bash

diff --git a/docs/jev-routing.md b/docs/jev-routing.md
@@ -62,7 +62,8 @@ The tier/profile fields describe routing preferences before the existing model s
 Copy a completed journal snapshot and summarize it:
 
 ```bash
-python3 scripts/evaluate-jev-routing.py /path/to/jev-decisions.snapshot.jsonl --output /tmp/jev-report.json
+dotnet run --project tools/laya_service -c Release -- report \
+  /path/to/jev-decisions.snapshot.jsonl --output /tmp/jev-report.json
 ```
 
 For representative decisions, use existing session records to assign human labels in a separate JSONL file:
@@ -72,7 +73,9 @@ For representative decisions, use existing session records to assign human label
 ```
 
 ```bash
-python3 scripts/evaluate-jev-routing.py /path/to/jev-decisions.snapshot.jsonl --labels /path/to/labels.jsonl --output /tmp/jev-quality.json
+dotnet run --project tools/laya_service -c Release -- report \
+  /path/to/jev-decisions.snapshot.jsonl --labels /path/to/labels.jsonl \
+  --output /tmp/jev-quality.json
 ```
 
 Without labels the report includes coverage, failures, disagreement, latency and estimated decision spend; it makes no accuracy claim. With labels it compares the baseline, always-T2, and Jev with safeguards/fallback using accuracy, confusion matrices, per-tier F1, under-routing, and high-risk capability retention. Label coverage and per-tier sample counts matter. Separate model/rubric cohorts before calibrating thresholds.
@@ -104,7 +107,7 @@ Restart afterward and remove any environment override that would re-enable Jev.
 
 ```bash
 dotnet test src/OpenClaw.Tests/OpenClaw.Tests.csproj -p:OpenClawSkipDashboardBuild=true --filter 'FullyQualifiedName~JevRoutingTests|FullyQualifiedName~TurnRoutingPolicyTests|FullyQualifiedName~RoutingCommandsTests'
-python3 -m unittest discover -s tests/routing-eval -p test_jev_report.py
+dotnet test tools/laya_service/tests/LayaService.Tests.csproj -c Release --filter 'FullyQualifiedName~RoutingJournalReportTests|FullyQualifiedName~ReliabilityPlotTests'
 ```
 
 The transport follows the [TypeSafe API](https://docs.typesafe.ai/api). The model pin and initial estimated price come from its [model reference](https://docs.typesafe.ai/models), checked September 20, 2026. See [confidence semantics](https://docs.typesafe.ai/confidence) and [known model limitations](https://docs.typesafe.ai/model-jaggedness/jev-1.13) before tuning policy.