







All shitposting aside, though, if we go off the explanation the CA has given then some reformating script accidentally dropped most of the bytes or never requested them from the HSM or similar. A couple of years ago, we had a similar worry with Tink. In particular, we wanted to have a cheap test that catches obviously non-random values with high confidence. I added three little statistical test back then, which are run as unit test in every presubmit on everything that is supposed to look like a uniform random string. The tests are not constant time, so you cannot run them over your actual keys as is, but they are very reliable to test tooling that is supposed to handle and transform key material. They are tuned aggressively to detect non-randomness only up to 10 standard deviations, which makes the test extremely non flaky, at the cost of missing some events. But as I said, they're designed to be unit tests, that run with high frequency, so false positives would be very bad in that scenario. There are multiple ways to test this, some more principled than others, and somewhat depending on other requirements, but this is what I added to Tink back then: The first test is a z test that checks that every bit in the input string has a probability of 0.5 of being set. I.e. it sums all bits and checks whether the result is within 10 standard deviations of half of the length of the string. The second test is a cross correlation test that takes two bit strings, xors them, and runs the first test on the result. In other words it checks whether two independently drawn strings agree at half the positions. The third test is the most expensive one to run, but also the most powerful of these tests and tests the autocorrelation of a string. It rotates the string, shifting each bit by one and putting the last bit to the front, and then runs the cross correlation test on that string with the original string, for each possible shift. This test detects repetitions of sequences within a string, and will e.g. detect that a string contains only printable ASCII, although at the setting of 10 standard deviations it will require a couple of sentences before firing. https://github.com/tink-crypto/tink-cc/blob/3c94644e5b2dffacd66764da13083a97ee782d15/tink/util/test_util.h#L136
Keyboard Checker | The best online keyboard tester
Test all keys on your keyboard with this easy-to-use website. Works with any keyboard and any browser. Simply press any key to test it.
Introducing Automatic Key Verification
Signal now offers a feature called “automatic key verification” which complements the existing safety number system. Signal is always end-to-end encrypted, and automatic key verification provides an additional, streamlined way to confirm that there’s no unexpected party between you and the other ...

Jacky Kwok on Twitter / X
Scaling self-verification with DeepSeek V4 Flash beats Claude Fable 5 on Terminal-Bench 2.1, while being 11x cheaper 💰As open-source models become more capable, they can now generate large numbers of high-quality candidate solutions and verify their own outputs at very low… https://t.co/as2HtyHzzW pic.twitter.com/XzVBgr5JPz— Jacky Kwok (@jackyk02) August 17, 2026

Microsoft Keys
Back in 2021 I gave a Talk At Berkeley's Information Access Seminar that summarized two long posts from two years before that: Certificat...
Hypothesis: A new approach to property-based testing
The property-based testing library for Python
On the Risks of using LLM-Generated Tests for Regression Testing
Software is under constant evolution: developers continuously add features, fix bugs, and refactor code, and any of these changes may break existing functionality. Regression testing guards against such effects by capturing expected behavior in test cases. LLM-based test generation aims to automate this process by generating regression tests directly from the code under test. This is beneficial when the implementation is correct, but problematic when the code contains faults: the generated tests may then encode and preserve incorrect behavior. To investigate this risk, we apply LLM-based regression test generation to pull requests merged into the main branch of software projects and study the impact of the generated tests on subsequent project evolution. We distinguish between fault-revealing tests, which assert correctly implemented behavior, and fault-enforcing tests, which assert faulty behavior. Across 145 pull requests from SciPy, Qiskit, and pandas, 8%-17% of the generated tests are fault-enforcing, while only 2.4%-4.8% reveal faults. Fault-enforcing tests persist over time: after several subsequent commits, 83%-91% of them are still relevant and pass. They also accumulate: when the faults of all pull requests are combined in one codebase, 83%-92% remain enforced at the end of the commit history, and the developer-written test suite detects only 14%-30% of them. Our results reveal a fundamental risk of LLM-generated regression tests: without manual validation, they may encode faulty behavior as expected behavior, allowing bugs to persist across software revisions and largely evade developer-maintained test suites. LLM-based regression testing can thus give rise to a new form of technical debt.

Testing can be fun, actually
Writing and maintaining tests is boring. But they're also some of the most valuable code we can write. With this blog post you'll learn a criminally underrated testing technique to add to your testing toolbox that can make tests a whole lot more pleasant.

10 Best HALL EFFECT Gaming Keyboards!
Inside the keyv npm Supply Chain Compromise | Snyk
The keyv npm compromise used preinstall malware, trusted provenance, and IDE hooks to target developer and CI credentials. Learn how to detect and respond.

Unkey's Globally Distributed API key generation - with cofounder James Perkins
The sad state of property-based testing libraries
Property-based testing is a rare example of academic research that has made it to the mainstream in less than 30 years. Under the slogan “don’t write tests, generate them” property-based testing has gained support from a diverse group of programming language communities. In fact, the Wikipedia page of the original property-basted testing Haskell library, QuickCheck, lists 57 reimplementations in other languages.
npm Worm Poisons keyv, cacheable and 400+ Other Packages Across Twelve Organisations
A worm moved one byte-identical credential stealer through more than 400 npm packages in twelve organisations on 4 August 2026, including @ornikar, @deliveroo, @servicetitan, @qlik, Picsart and the keyv and cacheable family. latest still resolves to a poisoned version on most affected names, and the payload installs a dead-man switch that fires when the stolen GitHub token is revoked, so rotating credentials first triggers it.

The Test That Only Failed in the Summer
In the summer, while I was working on a feature locally, I noticed a test started failing. I initially dismissed it: I assumed there was some flakiness, and CI was green, so everything must’ve been good, right? Until it kept failing, only locally. A test that a few days before was working as normal.

The Test That Only Failed in the Summer
In the summer, while I was working on a feature locally, I noticed a test started failing. I initially dismissed it: I assumed there was some flakiness, and CI was green, so everything must’ve been good, right? Until it kept failing, only locally. A test that a few days before was working as normal.

Published part two on wiring up Standard.Site as a test of said wiring, in my hubris, and: no dice. Looks like discussion around arbitrary `rkey` values is still in flux — my “slugified post path” rkeys no longer validate, and I get no fancy preview for documents with an `rkey` > 13 characters.
Is TID the best choice for rkeys? #7
tangled.org