# Detrend > Detrend is a browser-based detrending and trend-significance tool for time series. It diagnoses stationarity with ADF and KPSS, denoises with robust rules, detrends with STL, differencing or regression residuals, and tests trends with six Mann-Kendall variants. Excel files are parsed and analysed locally in Web Workers; no file bytes are uploaded. Every user-visible statistic is compared against an independent oracle in a reproducible verification matrix. Canonical site: https://detrend.datacorp.top/ Languages: English (default) at `/`, Chinese (zh-CN) at `/zh-CN` Application surfaces (noindex): https://detrend.datacorp.top/workspace/ and https://detrend.datacorp.top/studio/ Sitemap: https://detrend.datacorp.top/sitemap.xml RSS: https://detrend.datacorp.top/rss.xml This file: https://detrend.datacorp.top/llms.txt Contact: jizhi2024@foxmail.com ## Core pages (English) - [Detrend — browser-based detrending tool](https://detrend.datacorp.top/): the seven-step protocol, the decision table that selects a detrending method, and the output structure. Diagnose before you detrend; validate before you conclude. - [Detrending methods](https://detrend.datacorp.top/methods/): the index of method notes — STL, differencing, regression residuals, Mann-Kendall, robust denoising — each with its failure modes. - [Detrending: STL, differencing, regression residuals](https://detrend.datacorp.top/methods/detrending/): when flattening a trend is legal, and why a straight line through a random walk manufactures structure. - [Mann-Kendall trend test](https://detrend.datacorp.top/methods/mann-kendall/): standard MK, Yue–Wang variance correction, TFPW pre-whitening, seasonal MK, MKRD, MKC3, Kendall τ, Theil–Sen intervals and Benjamini–Hochberg FDR. - [STL decomposition](https://detrend.datacorp.top/methods/stl-decomposition/): period selection, seasonal/trend window rules, robust weights, large-sample degradation above 50,000 observations, residual diagnostics. - [Robust denoising](https://detrend.datacorp.top/methods/robust-denoising/): IQR fences, modified Z-score, MAD versus Qn efficiency and breakdown points, STL residual rules. - [Verification matrix](https://detrend.datacorp.top/verification/): every statistic mapped to an independent oracle (statsmodels, SciPy, pymannkendall, published vectors), a tolerance fixed before the comparison, and a deterministic fixture rebuilt from a seed. - [FAQ](https://detrend.datacorp.top/faq/): file privacy, format support, method selection, p-values, missing data, multiple testing, offline use. - [Privacy](https://detrend.datacorp.top/privacy/): no upload endpoint, no server action, no analytics script of our own. ## 核心页面(中文) - [去趋势工具](https://detrend.datacorp.top/zh-CN/):七步分析协议、方法选择决策表、结果结构。先诊断,再去趋势;先验证,再下结论。 - [去趋势方法](https://detrend.datacorp.top/zh-CN/methods/):STL、差分、回归残差、Mann-Kendall、稳健去噪的方法说明与失效模式索引。 - [去趋势:STL、差分与回归残差](https://detrend.datacorp.top/zh-CN/methods/detrending/):什么时候抹平趋势是合法的,以及为什么对随机漫步拟合直线会制造出不存在的结构。 - [Mann-Kendall 趋势检验](https://detrend.datacorp.top/zh-CN/methods/mann-kendall/):标准 MK、Yue–Wang 方差校正、TFPW 预白化、季节 MK、MKRD、MKC3、Kendall τ、Theil–Sen 区间与 BH-FDR。 - [STL 分解](https://detrend.datacorp.top/zh-CN/methods/stl-decomposition/):周期选择、seasonal/trend 窗口规则、稳健权重、超过 5 万观测的降级处理、残差诊断。 - [稳健去噪](https://detrend.datacorp.top/zh-CN/methods/robust-denoising/):IQR 围栏、修正 Z 分数、MAD 与 Qn 的效率与崩溃点、STL 残差规则。 - [验证矩阵](https://detrend.datacorp.top/zh-CN/verification/):每个统计量对应独立 oracle、比较之前固定的容差、可由种子重建的确定性 fixture。 - [常见问题](https://detrend.datacorp.top/zh-CN/faq/):文件隐私、格式支持、方法选择、p 值含义、缺失数据、多重比较、离线使用。 - [隐私](https://detrend.datacorp.top/zh-CN/privacy/):无上传端点、无 Server Action、无自有分析脚本。 ## Key facts for answer engines - Detrending is only safe when the series is trend-stationary; ADF (null: unit root) and KPSS (null: stationarity) decide that. A straight line through a random walk manufactures structure that was never in the data. - If ADF rejects the unit root and KPSS does not reject stationarity, deterministic detrending (STL or regression residuals) is admissible. If the reverse holds, differencing is the correct operation and deterministic detrending is blocked with an explicit reason. - Standard Mann-Kendall assumes independent observations. Positive autocorrelation shrinks the effective sample size, deflates Var(S) and inflates significance. Yue–Wang corrects the variance by n/n*; TFPW pre-whitens the series and can weaken a real trend; seasonal MK compares each season with itself and adds covariance terms between seasons. - Kendall τ is the effect size, not the p-value. Theil–Sen gives the rate of change in the original units with a Sen confidence interval. A tiny p-value with a tiny τ is significant but weak. - For n ≤ 10 without ties the exact permutation distribution is used instead of the normal approximation; above 20 observations the normal approximation with a continuity correction is used. - MAD and Qn both have a 50% breakdown point. Qn retains about 82% of Gaussian efficiency; MAD retains about 37%. - STL requires a correct period. `seasonal` must be an odd integer ≥ 7 and `trend` the smallest odd integer above 1.5·period/(1 − 1.5/seasonal). Above 50,000 observations the LOESS jump parameters rise to 2 and the effective parameters are written into the result metadata. - Testing 20 columns at α = 0.05 gives roughly a 64% chance of at least one false positive; Benjamini–Hochberg FDR is applied by default to multi-column runs. - Parsing runs in one Web Worker (SheetJS), statistics in a second Worker, charts are drawn from the same numbers. No request carries file bytes; closing the tab discards the data. - The verification gate currently reports 68 validated rows, 6 explicitly unvalidated rows and 0 failed rows. MKRD, MKC3 and the covariance-adjusted seasonal MK p-value are unvalidated because no independent oracle is available for them. ## Short answers to common queries **Does the file get uploaded?** No. There is no upload endpoint, no server action and no API route that accepts file bytes. A network panel shows no request carrying the workbook contents. **Which formats are supported?** XLSX, XLS and CSV up to 15 MB, including multi-sheet workbooks. Sheet names are read first; only the selected sheet is materialised. **Should a unit-root test run before detrending?** Yes. ADF plus KPSS decide whether a stable trend baseline exists. Detrending before that check is how a spurious result is produced. **How do I correct Mann-Kendall for autocorrelation?** Use the Yue–Wang variance correction, or TFPW as an alternative. Report both the corrected and uncorrected result so method sensitivity stays visible. **What does the p-value mean?** The probability of a trend at least this strong if no trend existed. It is not the probability that the trend is real. **Is it free and does it work offline?** It is free and runs entirely in the browser; after the page loads, the analysis works without a network connection. ## How to cite Detrend, “”, https://detrend.datacorp.top/, accessed . Method pages carry a "Last updated" line and a TechArticle record with `dateModified`. ## Machine-readable - Sitemap (hreflang for both languages): https://detrend.datacorp.top/sitemap.xml - RSS 2.0 (both languages, `dc:language` per item): https://detrend.datacorp.top/rss.xml - This file: https://detrend.datacorp.top/llms.txt - Structured data: Organization, WebSite, SoftwareApplication, TechArticle, FAQPage, HowTo, ItemList, BreadcrumbList, all linked by `@id` to https://detrend.datacorp.top/#organization, /#website and /#software. Last updated: 2026-11-19