The TweakPad blog
105 articles on JavaScript, TypeScript, React, Next.js, databases, data formats and the tools developers use every day.
39 more articles published on Medium, on benchmarks and performance.Browse them
Developer Tools8 min readThe Caret Range Changes Meaning Below Version 1.0
A caret range allows every future minor, unless the version starts with zero. It then narrows, and those dependencies quietly freeze.
The TweakPad Team
Web Development8 min readYour noindex Stops Working the Moment You Block the Page
Adding noindex and a robots.txt block together is worse than either alone. Google never reads the rule on a page it cannot crawl.
The TweakPad Team
Backend & Databases8 min readPostgres Logical Replication Does Not Send You the Old Row
Logical replication sends the new row and, by default, no before-image at all. What Postgres puts in an UPDATE, and what it leaves out.
The TweakPad Team
Backend & Databases8 min readYour Token Count Is Not a Property of Your Text
A token count depends on the language, the model and the content. The same text varies 15x by language and 30% between model versions.
The TweakPad Team
Backend & Databases8 min readLate Data Is Dropped by Default, and Nothing Tells You
A watermark declares that late data should not arrive. Some does, and Flink drops it by default with no error. What the docs really say.
The TweakPad Team
Backend & Databases8 min readThe BM25 Versus Embeddings Argument Measures the Wrong Gap
Two benchmarks agree across five years. Choosing BM25 or embeddings shifts recall a little. A reranking stage shifts it three times more.
The TweakPad Team
Backend & Databases8 min readBinary Quantization Works on Your Embeddings, or It Does Not
Binary quantization keeps recall on contrastive embeddings and fails on others. The outcome belongs to your model, not to the method.
The TweakPad Team
Backend & Databases8 min readParquet Row Groups Decide Whether Your Filters Skip Anything
Parquet row groups only skip data when the layout cooperates. Why unsorted writes make min/max statistics useless, and how to fix it.
The TweakPad Team
Backend & Databases9 min readStructured Outputs Guarantee a Grammar, Not Your Schema
Structured outputs constrain replies to a grammar, and several JSON Schema keywords never reach it. What the guarantee really covers.
The TweakPad Team
Backend & Databases9 min readExactly-Once Semantics Stop Where Kafka Stops
Exactly-once semantics cover a read-process-write loop inside Kafka. Your database write sits outside it. What the guarantee really says.
The TweakPad Team
Backend & Databases8 min readLost in the Middle Is an Argument for Retrieving Less
Lost in the middle is usually read as a reordering problem. The paper's own numbers argue harder for retrieving fewer documents.
The TweakPad Team
Backend & Databases8 min readTruncating Embeddings Works Without Matryoshka, Up to a Point
Truncating embeddings supposedly needs Matryoshka training. A 2026 study across 24 tasks found plain truncation holds until you cut 80%.
The TweakPad Team
Backend & Databases8 min readIceberg Partition Evolution Changes the Future, Not the Past
Partition evolution in Iceberg rewrites nothing. Old data keeps the layout it was written with, and every later query plans both layouts.
The TweakPad Team
Backend & Databases8 min readYour Airflow DAG May Not Have a Data Interval Any More
The standard fix for execution_date was to use the data interval. Airflow 3 allows runs without one, so those fields can be missing.
The TweakPad Team
Backend & Databases8 min readSemantic Chunking Wins on the Datasets Built to Flatter It
Semantic chunking beat fixed-size only on synthetically stitched documents. On real ones, fixed-size won every document retrieval test.
The TweakPad Team
Web Development4 min readCSS Anchor Positioning Is Baseline. You Can Probably Delete Floating UI
Tooltips and popovers that flip when they hit the viewport, in pure CSS. What anchor positioning replaces, and the cases it still cannot cover.
The TweakPad Team
JavaScript4 min readTemporal Is In ES2026 and You Still Cannot Ship It Without a Polyfill
Temporal hit Stage 4 in March 2026 and shipped in Chrome, Edge and Firefox. Safari has none of it. What that means for using it today.
The TweakPad Team
Web Development4 min readYour Hex Colours Are Leaving a Third of the Screen Unused
Modern displays render Display P3, but hex can only describe sRGB. How oklch() and color() reach the rest, and how to fall back safely.
The TweakPad Team
Backend & Databases4 min readQR Phishing Rose 146% in One Quarter. What to Check Before You Scan
A QR code hides its destination by design. How quishing bypasses email filters, and how to inspect a code before it reaches a browser.
The TweakPad Team
Web Development4 min readView Transitions Went Cross-Browser. Here Is What That Costs You
Firefox 144 completed the set at 90% support. How the snapshot model works, why one CSS rule animates a whole page, and the traps.
The TweakPad Team
AI & Prompt Engineering4 min readEvery WebMCP Tutorial Shows the Wrong Object. The Spec Says document
WebMCP lets a page expose tools to browser agents. It is a Draft Community Group Report, not a standard, and the API already moved.
The TweakPad Team
Backend & Databases4 min readSunrise 2027 Does Not Kill the UPC. It Puts a URL in the Barcode
GS1 asks every retail checkout to read 2D codes by 2027. It is a commitment, not a mandate, and the interesting part is Digital Link.
The TweakPad Team
Web Development4 min readCSS contrast-color() Is Baseline. It Only Ever Answers Black or White
It picks readable text against any background automatically. But it has exactly two answers, and on mid-tones neither is comfortable.
The TweakPad Team
AI & Prompt Engineering5 min readScaled Content Abuse: Google Bans Purpose, Not the Tool
The scaled content abuse policy names purpose and volume, not the tool. What the wording prohibits, and the test that settles it.
The TweakPad Team
Also on Medium
Benchmarks and deep dives published on Medium, where they stay canonical.
Data Science Collective8 min readdateutil.parser.parse Is 293× Slower Than the One-Liner in the Standard Library.
Six parsers, three timestamp formats, 100,000 values each — and the one that could not parse a standard access log at all.
- programming
- python
- performance
- data science
Read on Medium(opens medium.com in a new tab)
3 Python Regex Engines, 5 Patterns.
re is the fastest engine on some patterns and the only one with an unbounded worst case. Both halves measured on the same corpus.
- python
- technology
- software development
- programming
Read on Medium(opens medium.com in a new tab)
Member-only story on Medium
Python 3.15 Has 4 Ways to Defer an Import.
Measured on 3.15.0rc2 across 26 installed packages: the option that needs no code change fails silently until first use.
- software development
- python
- coding
- programming
Read on Medium(opens medium.com in a new tab)
Member-only story on Medium