docs(CropLens): publish evidence for nine successful CPU model smoke checks

README.md

@@ -5,6 +5,7 @@
 
 [![Build verification](https://github.com/FuTseYi/CropLens/actions/workflows/build-check.yml/badge.svg)](https://github.com/FuTseYi/CropLens/actions/workflows/build-check.yml)
 [![Inference regression](https://github.com/FuTseYi/CropLens/actions/workflows/inference-tests.yml/badge.svg)](https://github.com/FuTseYi/CropLens/actions/workflows/inference-tests.yml)
+[![Real YOLO CPU smoke](https://github.com/FuTseYi/CropLens/actions/workflows/real-yolo-smoke.yml/badge.svg)](https://github.com/FuTseYi/CropLens/actions/workflows/real-yolo-smoke.yml)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![Python](https://img.shields.io/badge/Python-3.8+-blue.svg)](https://www.python.org/)
 [![Vue.js](https://img.shields.io/badge/Vue.js-3.x-brightgreen.svg)](https://vuejs.org/)
@@ -100,6 +101,8 @@ See [architecture](docs/ARCHITECTURE.md) and [migration notes](docs/MIGRATION.md
 
 ## Testing
 
+**Real checkpoint validation:** All nine bundled crop models successfully processed one corresponding sample each on a GitHub-hosted CPU on October 8, 2026. [Review the recorded test evidence and limitations](docs/VERIFICATION-2026-10-08.md). This is not a measurement of prediction accuracy.
+
 Regular pull requests run the lightweight CI checks. An optional manual workflow can run **one real YOLO checkpoint on a sample image using CPU**; see [testing and limitations](docs/TESTING.md). Its results must be checked separately before claiming any real inference validation.
 
 ## Contributing

README_zh-CN.md

@@ -5,6 +5,7 @@
 
 [![Build verification](https://github.com/FuTseYi/CropLens/actions/workflows/build-check.yml/badge.svg)](https://github.com/FuTseYi/CropLens/actions/workflows/build-check.yml)
 [![Inference regression](https://github.com/FuTseYi/CropLens/actions/workflows/inference-tests.yml/badge.svg)](https://github.com/FuTseYi/CropLens/actions/workflows/inference-tests.yml)
+[![真实 YOLO CPU 推理](https://github.com/FuTseYi/CropLens/actions/workflows/real-yolo-smoke.yml/badge.svg)](https://github.com/FuTseYi/CropLens/actions/workflows/real-yolo-smoke.yml)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![Python](https://img.shields.io/badge/Python-3.8+-blue.svg)](https://www.python.org/)
 [![Vue.js](https://img.shields.io/badge/Vue.js-3.x-brightgreen.svg)](https://vuejs.org/)
@@ -100,6 +101,8 @@ docs/          架构、部署与安全说明
 
 ## 测试与验证
 
+**真实模型验证：** 2026 年 10 月 8 日，仓库内的 9 种作物模型均在 GitHub CPU 环境中成功完成对应样本单图推理。详见 [验证记录及局限性](docs/VERIFICATION-2026-10-08.md)。该结果不代表检测准确率经过验证。
+
 日常 PR 使用轻量级自动化检查；另提供可手动触发的 **CPU 真实 YOLO 模型单图推理测试**。执行方法、测试范围与局限性详见 [测试说明](docs/TESTING.md)，实际运行通过后才可称为真实推理已验证。
 
 ## 贡献方式

docs/TESTING.md

@@ -8,6 +8,8 @@
 | Documentation links | `python scripts/check_docs_links.py` | Relative Markdown links resolve | External link availability |
 | **Real YOLO smoke test** | GitHub Actions: `Real YOLO CPU smoke test` | Loads all nine crop checkpoints and predicts one corresponding bundled JPG per crop on a GitHub-hosted CPU | Diagnostic accuracy, full application workflow, multi-user concurrency, camera/video streaming, disease classification accuracy |
 
+**Verified execution:** [2026-10-08 nine-checkpoint CPU smoke result](VERIFICATION-2026-10-08.md) — all nine model/sample pairs completed basic inference. This is a compatibility smoke result, not an accuracy benchmark.
+
 ## Run the real inference smoke test
 
 The workflow runs automatically on `main` when model checkpoints, its test script, or the smoke workflow are changed. It can also be triggered manually from an authenticated GitHub CLI terminal:

docs/VERIFICATION-2026-10-08.md

@@ -0,0 +1,33 @@
+# CropLens — Real CPU inference verification
+
+**Date:** 2026-10-08  
+**GitHub Actions run:** [Real YOLO CPU smoke test #37754791593](https://github.com/FuTseYi/CropLens/actions/runs/37754791593)  
+**Repository commit:** [ae959b79fd206177525b24c896f7a9d76423d852](https://github.com/FuTseYi/CropLens/commit/ae959b79fd206177525b24c896f7a9d76423d852)  
+**Conclusion:** Success.
+
+## What was actually executed
+
+The workflow ran `inference/tests/smoke_real_model.py` with PyTorch + Ultralytics on a GitHub-hosted CPU runner. It loaded each checked-in `inference/weights/<crop>_best.pt` checkpoint and ran inference on the first matching JPG from `assets/samples/`.
+
+| Crop | Sample category | Checkpoint | Inference |
+| --- | --- | --- | --- |
+| Apple | 苹果 | `apple_best.pt` | Passed |
+| Corn | 玉米 | `corn_best.pt` | Passed |
+| Cotton | 棉花 | `cotton_best.pt` | Passed |
+| Grape | 葡萄 | `grape_best.pt` | Passed |
+| Potato | 马铃薯 | `potato_best.pt` | Passed |
+| Rice | 水稻 | `rice_best.pt` | Passed |
+| Strawberry | 草莓 | `strawberry_best.pt` | Passed |
+| Tomato | 番茄 | `tomato_best.pt` | Passed |
+| Wheat | 小麦 | `wheat_best.pt` | Passed |
+
+All nine invocations returned one decoded image result and a valid detection result structure. The execution logs report one detection per sample; these counts **do not establish whether the labels or bounding boxes were correct**.
+
+## What this does *not* establish
+
+- No mAP, precision, recall, robustness or statistically sound accuracy assessment was performed.
+- Only one selected sample per crop was tested, not all 132 demonstration images or independent holdout datasets.
+- The test does not use Flask, Spring Boot, Vue, MySQL, file uploads or DeepSeek; those full application paths remain unverified.
+- Concurrent video/camera streaming, target-device runtime and third-party image/model licensing require separate review.
+
+The latest state of the smoke workflow may differ from this historical successful run. Check [the workflow page](https://github.com/FuTseYi/CropLens/actions/workflows/real-yolo-smoke.yml) for subsequent results.