Notices by Alex Russell (slightlyoff@toot.cafe)

@evan Ah yes, "but the fascists promised the trains would run on time" (nevermind that they mostly already did, and now they carry a whole lot more fascists from place to place).

@evan I don't mean to suggest that there aren't ready and available critiques of old-Twitter, or that the wickedly inefficient SV-consensus status-quo-ante of engineering divorced from outcomes and even goals is desirable. But, much as it pains me given my work and priors, we should understand both as superior to net increases in power granted to fascists. They will not be neutral actors.

I get why Republicans and the press are failing, but why are there no protests outside D elected's offices? The minimum they must do is fight, and voters have to demand they at least try. Why isn't that happening?

The older I get, the more it seems that the rate at which we mint new programmers has sheered away from the rate at which we inspire people to want to learn about the computers they use.

Like, reading the deepseek paper, my overwhelming thought is "so wait, the secret sauce was...profiling the workload? Then deciding to program to the available hardware?"

I stare at the results of people carelessly composing UI systems without the faintest concern for how they will work in practice, but somehow imagined that wasn't how it's going in the rest of the industry. Woof.

A lot of people who seem to have Big Opinions about how much it costs to build a browser only have toy and/or trailing-edge systems in mind, and it's BAD for the discourse.

And let's just take a quick moment to examine the usual response of *"React can be fast enough!"*

Yes, good craftspeople don't blame their tools. They also don't bring shit tools to the job site.

What's being proposed concretely is the high-cost, low-confidence path based on little more than reckons. This is a demand that you to spend *more* to get *less*. And that's the optimistic version!

And I don't know how to say this any more directly than this: any frontend engineer who brings React or Angular on premises (without an honest bakeoff & guardrails), in 2025, is *de facto* bad at their job.

Software cultures that indulge in ignorance of constraints facilitate magical thinking, which eventually erodes the foundations out from underneath even conservatively-constructed experiences if left unchecked.

This is bad for users, but also business. A great shame of frontend's lost decade is that we lost the ability to adapt because so much of the community was high on frameworkism.

The correct response to realizing computers are fast is not to make your software slow, because:

a.) you won't benefit as much as you hope
b.) if you break pro-user norms, so will every other site/app/library, and your thing will feel slow even if it's "fine" in isolation
c.) HW bounty is not evenly distributed, so your product becomes less usable non-linearly below some resource floor

Pretending constraints don't exist is not engineering, it's bullshitting.

Apple continues to lead on top-end chip frequencies, in part because it pays more for access to the smallest process nodes...but that's not what's really going on. The big story for the past 10 years is that Android SoC vendors (Google very much included) have *sucked* at keeping their chips fed with enough data to retire work quickly.

Why? Because they were afraid to trade cores for cache, no matter how much it hobbled their phones.

I've been banging on for years about how the low-end of the mobile market has been just totally f'd in terms of the properties that make chips *actually* fast (process shrink + aggressive memory hierarchy optimisation; not core counts), and for this year's PIG post, I'm making EVEN MOAR CHARTS:

This process laundering of low ambitions has two goals, and only one beneficiary.

The goals? To preserve the status of the folks in the room as deliverers of progress in the minds of folks who aren't, and to insulate them from challenge by processes that would falsify their blame to the credulous.

The beneficiary? The implementer with the least interest in spending money to fix platform problems.

One of the worst failure modes of web standards is that many folks are easily convinced that we should accept the rate of progress that large, old WGs deign to bless as, somehow, the natural or correct pace – and often cite fig leafs like invited experts to claim that they are in touch. It's tragedy repeated until it comes back as farce.

That web developers keep falling for the same shit, year after year, is proof that there is such a thing as too much patience. All of the platform's core languages are defended by similar processes, often with no relationship to the technical merits of proposals or the market urgency for solutions. But people keep going to meetings, which looks like doing the job. When they deliver late, they can claim earlier proposals were "bad", safe in the knowledge survivorship bias will cover their tracks

@cbirdsong @daringfireball Just want to make sure that we're on the same page here: this website – as pictured in the most widely used browser, on the most widely used OS – is complaining about the state of the mobile web?

@andydavies @AmeliaBR The biggest indictment for me was always that corp policy generally rejected devices more than a few years old because the Android and Pixel teams couldn't be arsed to keep supporting those boards and architectures. Much of *that* was downstream of failing to own enough of the SoC IP (drivers, etc.) to do a good job.

Clown show.

@andydavies @AmeliaBR You can still likely find docs I wrote outlining how unserious this all was. Not buying Imagination's IP when it was up for sale...not buying Intel's modem group...not starting on custom SoCs until it was way too late, then continuing to fuck up single core perf...

*Sigh.*

The thing folks at the low-end are mostly missing out on is performance, which is dominated by combined L1 + L2 + L3 + SLC on-package caches, with process-shrink-driven frequency scaling not far behind:

For the Performance Inequality Gap series, I'm gathering new data points in charts that I'd only talked to in prose in previous years. One of those is the difference between the various market segements looks at by price, with the Flagships clocking the new, unlocked cost of the fastest chip in each line, while the lower-end lines capture the mid-tier and low-end segments from various manufacturers.

The blue line that has never bumped above $375? That's worldwide average selling price.