The part you can try today今天就能试的部分
swell-ruler: a proposed chicken-muscle ruler, untested on real imagesswell-ruler：一把提议中的"鸡肉标尺"，还没在真实图像上测试过
You don't need the SWELL machine to try its ruler. A lab that already does expansion microscopy could expand a few chicken myofibrils alongside its samples, image them before and after on its own microscope, and get the axial expansion factor of each myofibril long enough to measure (on red-team gel S3, 28% of the true pairs got one), a whole-gel strain map and QC flags (which on S3 caught none of the under- and over-expanded patches). One small Python package, free, no account. It is one more member of a family of in-gel rulers (sarcomeres in Sheard et al. 2019, the GelMap grid of Damstra et al. 2023, nuclear pores, photobleached patterns), not a new kind of QC; no journal-grade test of it exists yet.
不需要 SWELL 这台机器，也能试它的标尺。已经在做膨胀显微的实验室，可以在样品旁边一起膨胀几根鸡肉肌原纤维，用自己的显微镜拍膨胀前后的图，得到每一根够长的肌原纤维的轴向膨胀倍数（在红队凝胶 S3 上，真实配对里有 28% 得到了倍数）、整块凝胶的应变图和质控标记（在 S3 上，它们一处膨胀不足或过度的区域都没标出来）。一个很小的 Python 包，免费，不用注册。它只是"凝胶内标尺"这一家族里的又一员（Sheard 等 2019 用肌节、Damstra 等 2023 的 GelMap 网格、核孔、光漂白图案），不是一种新的质控；目前还没有达到期刊标准的检验。
| Red-team gel S3, as the red team scored it红队凝胶 S3，按红队的评分 | result结果 |
|---|---|
| gel-mean axial error (median)整块凝胶平均轴向误差（中位数） | −0.14% (−0.01%) |
| per-myofibril axial RMS, on the pairs that got a ratio每根肌原纤维轴向误差均方根（只算得到比值的配对） | 1.8% · 28% |
| transverse ruler bias横向标尺偏差 | −17.8% |
| under- and over-expanded patches flagged膨胀不足和过度区域被标出 | 0% · 0% |
Install and try安装和试用
python -m pip install "swell-ruler[images,figures] @ https://x-wow.com/downloads/swell_ruler-0.1.2-py3-none-any.whl"
python -m swell_ruler demo
SHA-256 of swell_ruler-0.1.2-py3-none-any.whl, to check a downloaded copy: f3709576545a05b6dc9dd0ef05ae89bf58f7cc9c9b12fa46f3a9f8eb928e126d
swell_ruler-0.1.2-py3-none-any.whl 的 SHA‑256 校验值，可用来核对下载的文件：f3709576545a05b6dc9dd0ef05ae89bf58f7cc9c9b12fa46f3a9f8eb928e126d
The demo draws a synthetic before/after pair with a planted under-expanded patch, measures it and scores each check PASS or FAIL. Today it passes the axial factor and fails the transverse factor and the patch: the same weak spots S3 found. Then run it on your own pair, with your own measured blur widths (PSFs):
演示会生成一对合成的膨胀前后图像，里面埋了一块膨胀不足的区域，然后测量，并给每一项检查打"通过/不通过"。现在它的轴向倍数通过，横向倍数和那块区域不通过：正是 S3 找到的弱点。接着可以用你自己的图片，并填上你自己测出的模糊宽度（PSF）：
python -m swell_ruler images pre.tif post.tif --px-pre 0.194 --px-post 0.194 --e-macro 4.5 --pre-mode dpc --psf-pre 0.55 --psf-post 0.41
Needs Python 3.9+ (NumPy, SciPy and Pillow come with it). It is not on PyPI, so install it from this link or from the SWELL download, which has the source, tests and data format.
需要 Python 3.9 以上（NumPy、SciPy 和 Pillow 会一起装上）。它不在 PyPI 上，请用这个链接安装，或者用 SWELL 下载包，里面有源码、测试和数据格式。
What it reports它会报告什么
- Axial factor per myofibril: post ÷ pre sarcomere period, fitted on the post image's own Z-disc ladder. On S3 the gel-level average held up; per-myofibril precision did not (1.8% RMS).
- Transverse factor: post ÷ pre width. Biased on S3 by a blur shape it does not model, even with the right blur widths.
- Landmark factor: from neighbours' positions, no calibration needed.
- Maps and flags: local expansion, fibril vs gel, gel stretch, tears and folds, places with no partner. On S3 they caught the bubble and most of the fold, but none of the under- or over-expanded patches.
- Anisotropy is reported only beyond ±1%, and no call is made while the transverse and landmark rulers disagree by more than 3%.
- No expansion yet?
stripeschecks whether your label-free images carry enough striation contrast (go if q ≥ 1.2, rescaled to 0.194 µm pixels).
- 轴向倍数（每根肌原纤维）：膨胀后与膨胀前的肌节周期之比，用膨胀后图像自己的 Z 盘"梯子"拟合。在 S3 上，整块凝胶的平均值扛住了，每根纤维的精度没有（均方根 1.8%）。
- 横向倍数：膨胀前后的宽度之比。在 S3 上，即使给对了模糊宽度，也被一种它没建模的模糊形状带偏了。
- 地标倍数：由周围纤维的位置算出，不需要标定。
- 图和标记：局部膨胀、纤维与凝胶是否一致、凝胶被拉伸、撕裂和折叠、找不到配对的地方。在 S3 上抓到了气泡和大部分折叠，但膨胀不足和过度的区域一个都没抓到。
- 各向异性超过 ±1% 才报告；横向标尺和地标标尺相差超过 3% 时，不下结论。
- 还没做膨胀？
stripes能先检查你的免标记图像条纹反差够不够（q ≥ 1.2 就可以继续，按 0.194 µm 像素换算）。
simulated Tested on synthetic gels only, including a red-team gel built by a separate AI agent (see Tested harder). It has never seen a real micrograph, and its image front-end is deliberately simple, so check what it detects. Before anyone should trust it: real before/after gels in two chemistries, agreement with an independent in-gel reporter such as a GelMap grid in the same gel, and a fresh blind test.
模拟只在合成凝胶上测试过，其中包括一块由另一个 AI 智能体造的"红队"凝胶（见更严的考试）。它还没见过一张真实的显微照片，图像前端也故意做得很简单，所以请检查它找到的东西。要让人信它，还需要：两种化学方法下的真实膨胀前后凝胶、在同一块凝胶里和独立的标尺（比如 GelMap 网格）对得上，以及一次新的盲测。