Could you please remove all references to lemmy-status?

• Wayback Machine
• February - actual status page
• July - up for sale
• July - casino

I see at least uptime and response-times badge svgs being included in the sidebar.

I also see it in a head script, which I assume lemmy software generates automatically for preload or something.

We thought updating Oh Deer would take a week. Nine months later, we’d end up rebuilding our multiplayer hide-and-seek game from the ground up. Enjoy!

We share how fixing performance turned into rebuilding almost everything: the Hunter, the maps, the NPC deer, the multiplayer systems, and even a tiny animated screen on a TrailCam. Along the way, we learned better ways to make games and discovered how much we still had to learn.

The new Oh Deer brings:

• 10-player matches: one Hunter and up to nine player deer.
• 100 NPC deer to hide among.
• Reworked maps, improved visuals, and better performance.
• Upgraded TrailCams and deer abilities.
• A completely reworked night phase: survive 40 seconds while your friends hunt you as Wendigos.

In Love Me Not you are a flower that climbs out of a cave. The cave has been cursed by a witch, so the level is regenerated above you whenever you fall.

This is similar to the procedurally generated levels in infinite runners like Doodle Jump, except that instead of erasing the entire level on a failure, Love Me Not only erases the most recent sections. This style of level generation is novel to me. For the purposes of this post I’ll call it spool generation.

I often simplify a player’s time with a game into a “novice period” and an “expert period”. These are usually characterised by different goals. In Love Me Not a novice’s objective is to escape the cave by any means necessary, while an expert’s objective is to escape the cave as fast as possible.

Spool generation had some natural effects on the novice period:

[expert period] I wanted to find a scoring system that could bring spool generation back into the decision space of an expert player.

Love Me Not on itch.io

[…] about some Celeste game-feel tricks. I don’t think that we invented any of these techniques, but they are important parts of the moment-to-moment experience of Celeste.

1. Coyote Time
2. Jump Buffering
3. Halved-Gravity Jump Peak
4. Jump Corner Correction
5. Dash Corner Correction
6. Semi-Solid Popping
7. Lift Momentum Storage
8. Wide Wall-Jump Window
9. Even Wider Super Wall-Jump Window
10. Stamina Refunds

You might have noticed a pattern! All are centered around widening timing or positioning windows, so that everything is fudged a tiny bit in the player’s favor. I think that this is a big reason why Celeste can feel kind even though it’s very difficult — it wants you to succeed.

In this video I explain the methodology and techniques I used to make the physical-yet-arcadey physics in my game Parking Garage Rally Circuit, including a (possibly misguided) observation about Mario Kart that influenced my entire process.

Today, we're releasing version 0.116.0 of Nu. This release brings a major custom completion rework, new tui commands for building terminal interfaces, interactive debugging, more tutor lessons, and a faster shell.

While load testing one of our Rust services at work, we ran into something that took us way longer to figure out than we'd like to admit: memory that shoots up under load and just stays there.

Though glibc is the default allocator and works well for the majority of applications, it can retain significant allocator-owned memory for bursty workloads due to arena reuse, fragmentation, and cached free chunks.

Allocator switch, and this article was mentioned in Lemmy Release v0.19.20 - Reduced Memory Usage

This release significantly reduces memory usage for the Lemmy backend. Metrics on production instances show a reduction up to 10 times. Here you can see the statistics from a few different instances. Read on for a technical explanation below. Note that the change only affects x86, there is no difference on ARM (e.g. Raspberry Pi), see here for details.

Clara explores the concept of metacognitive awareness, the role it plays in learning, common traps in your thinking, and how using LLMs affects it in surprising ways.

Goes into research about learning, and more specifically about metacognitive awareness.

Timestamped link to where significant content begins.

Metacognitive traps:

• Forming: Right question, wrong answer - right problem understanding, wrong solution approach
• Dislodging: Failing to change an approach that is not working even upon realization
• Assumption: Wrong question, right answer - solving the wrong problem
• Location: So close, but so far - almost done, but realization that crucial steps were skipped in the beginning
• Achievement: Band-aid on a broken bone - a lot written, small fixes in hopes that it'll start working

LLM don't solve any but introduce three new metacognitive traps:

• Progression: Falling Behind Without Noticing - using AI-generated code beyond own understanding
• Interruption: Distracting pop-ups - like generated-code auto-completion suggestions
• Mislead: Following bad advice

Before these sections, I find "pauses are not bad" particularly noteworthy as well. I will be sending this to a Junior colleague.

The htmx team is very happy to announce the release of htmx 4.0.0! This is the culmination of 8 months of work (plus a game) and we are very happy with the results.

Note that we are not marking 4.0 as latest in NPM because we do not want to force-upgrade users who are relying on non-versioned CDN URLs for htmx. Instead, 2.x will remain latest and the 4.0 line will remain next until some point in early 2027. The website, however, will reference 4.0.

We hope you enjoy htmx 4. htmx 2 will continue to be supported indefinitely so don’t feel any pressure to upgrade.

For five days in March 2026, a single stolen token let one group poison five software ecosystems including a package downloaded 95 million times a month. The attack started with a misconfigured GitHub Actions workflow, and ended with backdoored code sitting inside CI/CD pipelines around the world.

Today, two people behind this attack were arrested. TeamPCP, allegedly operated by these threat actors, started attacks in late 2025 running opportunistic cloud exploits, then pivoted in early 2026 to targeting the software supply chain itself.

Flare’s Emerging Threats Team wanted to share examples of some of the techniques that can be used to unmask one operator behind TeamPCP, using the Flare platform to trace a single alias across the accounts, credentials, and infrastructure that connected back to a real identity. There were other independent investigations on this topic, including by Brian Krebs. Below, we walk through who TeamPCP is and share the process of deanonymizing threat actors.