By that time point (2000s), Glide itself wasn't that much relevant: most game engine relied on high level API (DirectX 3D as you mention, and also OpenGL: Quake3 had already been out for 1 year, and "mini GL drivers" that serve as an adaptation between high level OpenGL and low level such as Glide were all the rage).

Very few engines had Glide-specific optimizations.

API exclusivity wasn't playing a role anymore.

But failing to have distinguishing features that attract users did play a role.

VSA-100 in Voodoo 4/5/6 had a few interesting new feature (*vastly improved Voodoo 1, 2, and 3's ability to do "pseudo 22bit": more than 65k colors in 16-bit modes and to avoid error and dither amplification in 16 modes; new FSAA with a rotated grid that both allows better edge anti-aliasing and circumvents the need for anisotropic filtering; motion blur for supporting games; better texture compression, etc.) but these failed to attract users' interest (I suppose most users didn't even understand those features), whereas Nvidia managed to gain more users' interest with better number in 24/32bit modes benchmarks and some marketing around T&L (despite T&L not being that much actually used in games of that era -- better CPUs with SIMD achieving similar scene-processing in practice)

The next generation would have been more interesting with 3Dfx plans to add programmable pipelines with Rampage and Sage ("Spectre" graphics cards) instead of fixed pipelines (i.e.: to add geometry and pixel shaders in the parlance of high-level API like D3D and OpenGL) but it never reached market, only prototypes existed when 3Dfx folded.
Though some of that development eventually helped the GeForce FX

Well, look at Apple's iPhones...

Speaking the Sega Dreamcast, the whole snafu of "Katana" (the actual NEC / SuperH + PowerVR) vs "Black belt" (competing prototype with a 3Dfx gfx card) also did cost quite some money to 3Dfx and accelerated their demise.

The main problem is that PowerVR works in a way that is completely alien when compared to everything else (outputting fully-rendered tiles, when everyone else is drawing polygons one-by-one on a pair of frame and depth buffers).

And tile-based-rendering's performance boost is less significant when there are more transparency layers in a scene which is where most of the industry was heading. (So the advantage that Kyro II had would have melted away with successor cards).

Speaking of TBR, 3Dfx had acquired GigaPixel to get patents to such technologies, did toy with software-based hidden surface removal (HSR) that worked in some OpenGL engines (Everything built on id tech 3 / Quake III engine) and was hoping to add hardware HSR with TBR in the next iteration of Rampage/Sage ("Spectre 2" / Mojo) but never got there before folding.

(And again, more transparency layer in scenes would eventually have made the performance gains less significant in future games).

---

*: Even the venerable Voodoo 1 "Pseudo 22bit" 16bit modes are different from everyone else on the market.

All competition used pure 16bit math pipeline. When combining multiple pixels (multi-texturing, transparency, effects, etc.) rounding error accumulate. It's even more visible when Bayer dithering is used (with an Intel integrated graphics core back then, this was visible on, e.g., Quake III's logo which had multiple translucent flames effect. The errors accumulated with each layer and the logo ends up looking like a checkered board.

In contrast, when in 16bit mode, all 3Dfx chips run at "pseudo 22bits" internally and on their video output, and combine the value of 4 sources pixels (hence the 2 extra bits per channel), less error accumulate on translucent layers, less visible dithering on the output.
On Voodoo 1 and 2, this is simply done by using 4 horizontally adjacent source pixels, giving the characteristic "slightly horizontally blurred" visual effect that is typically for early 3Dfx cards.
On Voodoo 3, the 4 sources can be also arranged in a square and there's more logic in how to combine them (I suspect something like conditional blur, but never read any full description)
On Voodoo 4/5/6, the chip introduces multiple buffers (2 per chip, up to 4 buffers on the most common dual-chip Voodoo 5) (equivalent to OpenGL's accumulation buffers), which introduced tons of cool efffects (introduce an offset between the buffer, and you get both antialiasing on the edges and the same result as anisotropic filtering by using plain trilinear on texture surface; render each buffer at a different time increment, and you get motion blur; etc.) - but this also allows even less blur when pick the 4 sources pixels for "pseudo 22bit" output (just pick 4 pixels at each same coordinate on each 4 buffers)

3dfx also did a couple of pretty stupid things: at a time when most 3d companies were just making chips and collaborating with 3rd party graphic cards manufacturer, 3dfx decided to "cut the middle man out" and produce their own first-party cards instead. Cue in all 3rd party graphic cards manufacturers switching to use Nvidia's chips instead (and also some trying this new ATI that was starting to enter the 3d chips market).

Yup, if only there was some small Central-European country that could have demonstrated that direct democracy could work at the nationnal level.
But no, nope, there absolutely none, nothing to see here, move along...

The main missing point on your list:

- popular initiatives and regularly held referendums (i.e.: the general population voting on most decisions, and bringing new decisions to be voted upon).

i.e.: what a *direct* Democracy (a.k.a. the only *true* Democracy in the sense of directly giving political power (cratos) to the general population (demos) - not merely putting, in decision-making roles some "representatives" who will be then subject to legalized bribes... huh... sorry to "lobbying")

You might notice there's a certain overlap between "country that follow direct democracy" and "country which avoided the last two rounds of world war" (despite rampant fascism at the same time in the rest of the continent around). These might be at least somewhat linked.

Noun-genders have little (but not exactly zilch) to do with emotions, and a lot (but not everything) to do with the very complex and messy evolution of language.
(Some very old language, including reconstructed proto-indo-european, had two genders..."animate" and "inanimate".
then a messy lineage of sometime reassigning(*) them to- or co-using them along with- "masculine" and "feminine",
some time adding back a neutral on top of that to bring back inanimate and/or try to (partially**) assign objects(***),
then reshuffling stuff around.
English notoriously started putting all objects and inanimates into neutral, though there are a few exception where older noun-genders were kept around. I knew for ships for sure (as pointed out: because "navis" in Latin) I wasn't sure if submarines fall in the same category.

Given the answers I got: apparently yes, it's customary to use "she" for submarines too, as a sub-type of ship.

(*) modern French has notes of that (that were carried through latin). Normal noun-genders are masculine and feminine, but concepts tend to mostly feminine, agents tend to be mostly masculine. (But of course, French being French, it has fractal-pile of exception all the way down)
(**) German, lots of slavic languages (including my own Bulgarian) have neutral, but not all objects are neutral
(***) all neutrals aren't object: children are neutral in several language (bg: "dete" is neutral (to). I've heard that's the case in Finnish too?)

The word used in a lot of slavic languages is masculine (something like: korab) (Imagine it in cyrillic. But, well, /. and unicode...)

BTW: So is the word used in French (le bateau, le sous-marin. Even "le navir" shifted to masculine gender-noun, even if in Latin navis was feminine).

It's the English where I wasn't sure if the "ships are feminine" extends to include submarine.

Yup.
Those where Model M, and specifically the ones that established several standards, such as the PS/2 mini-DIN connectors and cementing the 10x-keys keyboard layout that the first model Ms introduced to PC AT two years before (bringing the Function key to the top compared to AT's model F, keeping the inverted T arrows unlike previous crosses on terminal keyboards) and which is still the standard for "full" keyboards to this day (with only minor recent additions of windows, context menu, and optionally media keys)

Same.
Started with a model M as a kid (though not really knowing back then), that my dad brought from work for our first *own* 386DX workstation (previously I we had only borrowed Toshiba laptops with 8086 and 286)
Eventually keyboard died after decade, went through a couple of cheap no-names (told you I didn't know much as youth)
Then eventually learned about Unicomp still making them, buying one and having the "oh, this is the keyboard of my childhood! so much better!" experience.
Then after the last model started failing (fun fact: since COVID-19 I work a lot with W aste W ater sequencing. The "W" started being unreliable, probably due to broken platsic stud at that location), I splurged on the New model F (IBM-style mechanism, but on a modern layout 104/105 like modern Ms, not the older F layout).

That, and:
- the fact that the general public doesn't understand the implications of having stuff left on "shareable"
- in general settings that affect privacy should be "opt-in", not "opt-out", because not every single person online will be attentive to the thousand of tiny settings hidden behind dark patterns that needs to be flipped so their privacy isn't utterly violated.

The UE could step in for the point 2 above.
Reporting on such unintentional leaks could help raise awareness of point 1 above.

Well, it did pass lots of loopholes with flying colors.

Did. The pressurized space part of the Deepsea Challenger (as of the Trieste before her -- does English also use she for submarines?) was a pressure sphere at the bottom of the vessel.

You're confusing the whole submarine including her outer shell and couple of tons of non-pressuized technical equipement, with the tiny pressure sphere within her where James Cameron sat.
See diagrams on her wikipedia page.

I think the confusion is due to the fact that the pressure sphere is embed within the hull of the vessel and not very visible, whereas on the Trieste it was clearly distinct and attached under the hull.

Possible? Probably.
Easy? Even less than when using spheres.
Economically feasible? It's definitely not cheap given the failed attempt of proving the opposite by Titan.

Given that sentence I might presume you mixed up and meant to say the "F" (capacitance) is the better buy over the "M" (membrane) on the grounds of "more longevity"?

Also:

Correct me if I am wrong, but the original PC shipped with a type F, right? The membrane type M came later as a cost-reduction, right?
(So you basically managed to kill the F tank with your typing :-D )

So I guess, you're not the kind of guy who adds after-market solenoids for extra "oomph"! XD

From what I've been checking the project, it seems to me that:
- What eventually disappears is old stock from previous iterations of the project (e.g. the square machined case, before the current cast one) and small side projects (the split keyboard, the extra numpads, etc.) if all you want is to "get a model F", there's regularly something
- I am under the impression that the guy over plays a bit the "take them while they are still here!" marketing strategy.

Maybe the situation is different where you live.
But my local experience (Switzerland) the shop at the station closes during the night and cuts it light, so you also have less light too (only lights at the pump) and no attendant (you need to pay by card from the ATM-like between the pumps) neither.

In fact the EV charging station is slightly better: It has exactly the same illumination (because on highways it's also on the same rest area, either one extra stall next to the gas pumps if there's only a gas station, or a bunch of charging stalls next to the restaurant or hotel if there is one), but because usually you have a subscription to some charging network and have roaming between a lot of networks, you don't even need your wallet. So you would have had fewer reasons to get mugged, if mugging was something that happened in any significant frequency (but then again, Switzerland, insanely low criminality, so no mugging to begin with).

I am very happy with mine. It's build like a tank.
I did a "dental floss" mod to reduce the "ping" of the springs.
And added lifted feets in the back for adjusting the angle I prefer (I went for ESC Keyboard Flip Stand).
It's one of the earlier machined "square" cases.
(New Model F now also have injection-molded cases that look like the original IBM's).

Small volume production, driven by somebody who's insanely attentive to details.
On the other hand: it seems even more sturdy than the Unicomp Model M i have and those survive decades. I totally expect my Model F to outlive me. So diveded by the number of years, it seems to be okay-ish, if you're the kind of keyboard snob to like this type of keyboard (I'm guilty!).

For now, it seems to be that it was worth the investment. (Ask me again in a decade if I am still happy)

Note: They have the traces on PCB, but last time I checked and ordered they only ship keyboard pre-configured in ANSI (US) layout of the keys, with no key caps installed. (It's the buyers jobs to add or move barrels for other key layouts - I switched ANSI (US) to ISO (European) - and then to install the keycaps and check the springs.)
(Though the guy offer to change the barrels for a fee).
(You'll need to order the extra barrels and paddles and the unusually-sized space bar and either install them yourself or ask for a customisation and pay the guy)

If you like Model M, you'll mostly like F, too.

They have both roughly the same keyboard feel (they are both using buckling spring, only the exact way the keyboard registers the key is different: "M" - paddle hits a membrane contact ; "F" - paddle change capacitance, it's solid state).

The changes are:
- sound slightly differently (slightly longer paddles, and entirely made out of metal, no plastic in the casing nor barrels so it resonates a bit more).
(sound is moddable if you - or anyone your share space with - don't like the "ping").
- unlike original IBM F, and like IBM and Unicomp M, key do bottom-out "hard" (if you have a keycap at hand, look at the stalk: it's not flat it's tappered. The original IBM "F" paddle had matching slots, so the end of the key's travel is dampened by the elasticity of the tappered end bending into the slots. The New's "F" paddles and the "M" paddles lack the slots and the key hits the bottom hard). I personally don't mind (I'm used already to stop pushing as soon as the I feel the "click" and the keyboard registers, I don't bottom out pushing at full strength), if you mind it's moddable.
- Unlike "M", both the new and original "F" are fucking heavy (entire casing is metal, no plastic parts).
- Unlike IBM and Unicomp: currently only come in USB, the PS/2 daughter board (Leiden Jar) wasn't ready yet last time I checked.
- Some complain about the USB-C connector (it's not as strongly latching as a USB B). I am happy: given the wait the keybord isn't going to move, so no risk to yank out the USB-cable.
- No built-in feet, you need to buy after market feet (e.g.: any adjustable after-market laptop feet would do, there are suggestions in the forum, I went for ESC Keyboard Flip Stand), the guy sell quite a choice of fixed-heigh feet (rubber, cork, etc.)

And last warning:
- If you end-up being as happy with yours as I am with mine (absolutely 0% buyers remorse) you might end-up wanting to buy another for your office, which is almost definitely not covered by company purchases which means you'll part with yet another chunk of change (ask me how I know :-P ).

Sending important alerts from random numbers is a big security NO-NO.

In fact, in the transaction alert case, if it's from some random number and not your bank's number, that most likely yet another phishing campaign trying to trick you into providing your info by pushing you using urgency (Omg, there might be a fraudulous transaction I need to check ASAP).

A medical appoint SMS from someone who is not your usual doctor or the official number of the usual platform their are using might also be potential information leaking.

When people need to trust completely random numbers for critical alerts, it's a big security nightmare.

If it takes Apple's SPAM folders to force those to get their shit together and setup a more secure system, then all the better.
(Though I am afraid that "more secure system" is going to be "three 3rd party buggy middleware in an 'Mobile App' trench coat that absolutely require Google SafetyNet and won't work on after market firmwares or on Linux phone or on dumb phone or etc.)