2026-10-01 20:29 - Opinion
Much, much, much has been said about "AI" lately. Largely to mean LLMs (or Large Language Models). I've been a mostly late adopter and am still getting to grips with what it means, from a user's perspective.
I just watched the "Our Own Devices" (wonderful YouTube channel) video Metronomes: The Mechanical Maestro. At some point it started to cover the general theory of pendulums. But then explained that a pendulum as a time keeper for music would need an impractically long pivot arm to be useful, moved on to the invention of the modern metronome whose secret was a second weight on the opposite side of the pivot as the main weight, and moved on again more quickly than my brain could keep up with the new interesting facts. Put a pin in that.
I'm of the particular generation that really got to grow up alongside computers. Home computers were born roughly at the same time as me, and started to become common just as I was growing up. I used them extensively through late childhood, teenage-hood, and into adulthood. And even more, lived and grew with the online revolution. I can clearly remember the optimism of the early internet, and I felt it at the time. These days, it's long gone.
It's tough to pin down exactly why. I've thought about this before and never come up with a truly compelling explanation. Part of it is definitely that the early internet was still a rough place. To be there, you had to want to be there, it took effort and there was little if any return besides the intrinsic value. As a result the early internet was full of people that shared information because they wanted to, because they were passionate. This lasted for a while, but as their combined effort gathered more and more value, more people showed up to extract that value. The infinite and dirt cheap reproducibility of digital data is a wonderful thing, at its best. But eventually more participants meant a more and more diffuse landscape, and more value to be extracted from the audience itself. Commercial concerns crept in. And really took over.
Things passed a zenith. When is a matter of opinion, but certainly it's happened. The dark side of digital data's easy spread came into view: if attention is valuable (by way of advertisements), the incentive is to beg, borrow, and steal that attention any way you can get it. And if the internet enables instant global communication, then it's the scummiest actors who will fill it up, in that pursuit of monetized attention.
Which leads us to the internet of today. Aside even from the outsized effect that tiny slab-of-glass computer replacements have had, even the "classic internet" is a scarred husk of its former self. It's oft said that Google (Search) isn't what it once was, but I honestly believe that is mostly due to the fact that the "ten blue links" don't work anymore, when the majority of targets for those links are low value. The internet's success was its own demise. Some of us remember how mind blowing "Google" was when it was new. But "when it was new" is the key phrase. Back then quality content abounded. Google felt magical because it was standing on the shoulders of a giant internet-sized pile of knowledge. These days it's more a pile of slop.
Which gets me back to the metronomes. The video left me curious yet I didn't turn to a search engine, I turned to "an AI". (These days I regularly rotate around which one, specifically, I go to. Usually for no reason besides variety.) And I asked it about things. And I got an answer. Quickly. With no flashing or blinking colorful attention grabbers, right there right then. Just the answer I wanted, and again very quickly the first answer I found got me to the next question to ask, and soon to the crux of the issue.
That's very much what the early internet felt like.
2026-08-24 10:44 - Making
A while back the spokes on one of the larger rear wheels of my lawn mower started breaking. It was gradual, so it took quite a while for me to realize exactly what was going on. By the time I really understood, five of the six spokes were compromised somehow. At first I tried super glue plus baking soda to make a repair. Super glue is strong but brittle, and broke off rapidly. Then I tried hot plastic staples with epoxy as a filler. Well, epoxy is also pretty strong but brittle, broke quickly, and the staples pulled out easily as well. I then went through two rounds of trying to melt in 3D printer filament as a sort of weld filler. This went much better overall, but I never got a good bond between the original and filler plastic. Pictured above is the final attempt, just after re-installing the wheel, before using it. By the time I was done only one of my repaired spokes was still solid. The rest mostly had the filler plastic still attached, but not all the way across the fracture, so they provided very little structure if at all.
As previously mentioned, late last year I got a new(-to-me) 3D printer with a large build volume. It has a 300x300 mm build plate, and this wheel has a 277mm diameter at its widest point: it can fit! So I designed a replacement. It took around 18.5 hours to print, with settings intended to both end up strong and print reliably.
And here it is, intact after the first mow! I gave it eight spokes, beyond the original's six, and I made each spoke quite beefy. Only time will tell how long it lasts, but so far so good.
2026-06-27 17:33 - General
The quick opening context: I had some spare computer memory lying around. After RAM prices were almost perfectly flat from early 2023 to mid 2025 things got wild. The common narrative is that many companies, in a drive to build new AI systems that they can charge money for, drove demand for memory up significantly. So from April 2025 to April 2026, RAM prices nearly tripled. This reminded me of the sticks I had lying around unused. Why are they unused? Because they're registered ECC memory, which won't work in a typical PC. Several parts of the PC including the CPU must be specifically built to work with ECC memory, and for some reason CPU makers only put this support in the types of CPUs designed for servers, not for standard PCs. But: I've built servers, and really what makes a server is the use it's put to. "Server" hardware can be used as a desktop. So I had a mission.
Technically some "workstation" PCs are designed and built with ECC support as well. But with my experience building used-as-a-server type servers, I knew better what to look for and leverage there. I searched for both workstation and server parts, and I got lucky, finding a Supermicro X10DRi board being sold with dual Xeon E5-2680 v4 CPUs. It fully supports the memory I already have, and I got lucky that the power supply in my existing PC already had full support for dual CPU power connections. Asking price was $150, I haggled down to $130 because it turns out the seller was not including the heat sinks in the pictures. The pictured ones had fans integrated, they actually sold it with passive heat sinks. In a typical rack mount server case, this is fine. It's not what I wanted, though, so $20 off.
This is an E-ATX board. It rather has to be, to fit both processors. The ATX standard has existed for many years, but in more unusual cases like this dual processor board makers will exceed that. So this motherboard won't fit in any case. Again I searched for a while to find a nice solution. I got lucky, again on marketplace, to find a local PC builder with a "scratch'n'dent" case selection. Their largest case was just big enough! I had to remove two (of the twelve it came with!) fans to fit in the big motherboard, and make some of my own mounting holes that didn't line up, but it fit.
That was it, almost complete: motherboard and CPUs fit into a case, transferring power supply and disks from my existing PC, and voila! Almost. The motherboard did not come with an IO shield, but I found a compatible one already modeled online, and it fit and works great.
The eagle eyed among you will notice what I hadn't yet at this point, but did shortly: Being a server class motherboard, it does not include audio. Servers don't need to make sound! So I looked around, again. There is at least one compatible dedicated sound card still being made, but it's bottom of the barrel quality. For only slightly more I got a used Sound Blaster, which included the extra feature I wanted: support for the front panel audio connector. I've got memories of not only using, but very carefully selecting the model of discrete sound card I used back in the '90s. However, it's been built into the motherboard and a non issue for many years, until now.
I also had to figure out how to mount the disks in this cheap case. It was probably supposed to come with some sort of tray or sled for this purpose, but the scratch'n'dent model at least did not. I modeled it this time, for everything to line up with holes that already existed. A couple spacers to hold the 3.5" hard drive off the bottom, and curved arms to go from its side mount holes up to the 2.5" solid state drive. Then I had to address the CPU cooling. I first tried just hoping: the case is full of fans pushing a lot of air around, just like a real server case, right? No. Really using the CPUs would heat them up very quickly. So first, I tried to leverage the fans more directly: I designed and 3D printed a duct or baffle to direct two of the 120mm fans' output directly over the CPUs.
I was pretty happy when I first got this design to work. I managed to leverage my new-to-me large format 3D printer for a legitimately large part. It used the top two front case fans, and attached with their screws, directing that air right over the two passive CPU coolers. This was an improvement, but after only a few minutes of heavy load the high temperature alarm would go off, when one of the CPUs reached 95 °C. Not good enough. But I've got some old unused CPU coolers, and in fact two with identical fans, so I set to a new 3D modeling challenge: little ducts to directly attach these fans, one to each passive cooler. The first CPU is easy, being close to the edge of the motherboard where there's already some free space. The other, however, is crammed right between all the memory slots and quite close to the first CPU. So I started with a basic model of the motherboard, the CPUs with their big metal coolers, and all the RAM slots. Building from there, and taking into account the size and shape of the fans I had, to put in.
It took several false starts in CAD, learning to use some new features and figure out how to get them to do what I needed. All while trying to maximize the effectiveness: keeping a wide open path for airflow while squeezing in amongst all the obstructions. And with only one tiny exception (missing the nub for one of my attachment straps) it all worked first try!
After running a full 100% CPU usage (all cores/threads) stress test for over 15 minutes, CPU1 stabilized at 70 °C and CPU2 at 86 °C. (Both wiggling plus or minus one degree from there over the course of several more minutes. When idling: below 40 °C.) CPU2 has a more tortured path for its duct, and also experiences some of the exhaust of CPU1, based on their relative positioning. Anything in the 70 to 90 degree range is acceptable, at full load. And being intended as a desktop PC, it won't spend a lot of time at full load.
Used motherboard and CPUs: $130. Scratch'n'dent case: $40. Used sound card: $27. Memory: I already had it. Disks, power supply, graphics card: transferred from previous PC. Now I have a dual-CPU, 28 core/56 thread (14/28 each) monster with 80GB of RAM. Which lights up a lot more than I'd prefer, but thankfully the default mode is just a very slow and gradual fade between various colors. Not too bad, definitely beating the modes that do crazy flashing of colors. When I got the case, I assumed I'd be finding a way to defeat the lights, but they're gentle enough to have grown on me.
I didn't realize until I was actively building it: this case has no provision for an optical drive. I suppose it's mostly nostalgia that makes me want to have one, because to be honest I haven't used the one in the previous PC for ... well I can't remember, definitely a few years. I've got USB optical drives for that rare but eventually likely case.
2026-06-18 16:13 - Making
All things being equal, I'd live a "no gifts between adults" life. But among other things, my sister wants to set a good example for my nephews, so last xmas I ended up with a 3M headlight restoration kit. Something I've heard good things about, had been thinking about for a while, but never quite bubbled high enough on the list to buy it myself.
Yes, last xmas. I knew I didn't want to bother with this during the middle of winter so I set it aside. Perhaps for a bit too long, but yesterday I decided it was time! In the meantime I figured out that removing my car's headlights is just a screw or two and a connector, so I ended up doing that. Unfortunately first I dropped the driver bit and got it lost inside the car. Thanks to the borescope which I had misplaced for some time but recently located again, I eventually located that. After fighting through one rusty screw, I had the lights out. I washed them and left things to dry. Before that, the car looked like:
One of the closeups is blurry, but I can't go back and fix that! The front on view shows things quite clearly: no yellowing, but lots of cloudy blotches over most of the surface. A.k.a "headlight cataracts". Can I improve things? Like mentioned, I started by removing them so I could work more easily, and guarantee no damage to the paint nearby. Once they were inside and washed (and dry again) I got to work. The kit is simple: some sanding pads (500, then 800, then 3000 grit) and some special wipes with a clear coat to apply.
Things got worse before they got better. The "cataracts" are especially visible here. I assume there's a bunch of plastic dust ground/stuck preferentially into that less-smooth surface. I worked my way through all the sanding steps and these didn't disappear. I decided to start over because I could do that before applying the clear coat, but not after. I spent a while with the heaviest (500) grit that came with the kit and saw progress, but very slowly. I got bold and pulled out the finest (320) grit I had for my palm sander and used that. It worked pretty well, leaving a very frosted surface but I could see the cataracts being worn away. Each following grit got better and better. After the final (3000) grit, things were already quite acceptable.
Before we continue, here's a shot of the car "blind" — with no headlights installed. Anyway, I proceeded to apply the clear coat as the instructions said: two parts are included, a first and second coat. And indeed, things got even better still!
It doesn't quite show on camera; the closeup on the right probably does the best job. It's not quite like brand new but it's very good! If I look close and with light at just the right angle, I can see I didn't quite get all the sanding marks out. But who's ever going to look that closely? Cindy definitely got a few years taken off her face today!
2026-05-05 11:10 - General
Where I'm living now is only a few miles from a major airport. Overall I'm surprised and pleased that the noise is no big deal at all. But I do notice patterns some times, which got me curious. When I lived in NYC a coworker had set up a system with an antenna that gathered data about planes flying nearby, and I thought about doing something similar. Long story short: this is ADS-B (Automatic Dependent Surveillance–Broadcast). A nifty modern technology.
To my limited understanding, and simplified a bit: air traffic control one relied upon radar to know where planes were, and voice communication over radio. In the ADS-B system the plane instead broadcasts its own position, speed, etc. Both air traffic control and other planes can receive this data and make much more well informed and rapid decisions based on it. And, anyone can receive this data! I already had an "RTL-SDR" device — "Software Defined Radio". This common cheap variety is designed for radio and/or TV reception, but being software defined, it can do many things. I plugged it in, using the antenna it came with, installed the right software, and bam I could see planes nearby.
So now I was interested. After looking into it, and waiting to get the one connector adapter I needed, I made my own antenna. Old TV coax cable connects up to a coupler, which is threaded into a hole in the bottom of an old can. The can functions as a sort of "ground" as I understand it. Then a carefully trimmed inner core of another old TV cable sticks out of that, aiming for exactly a quarter wavelength of the 1090 MHz signal.
The stock antenna was not tuned for this signal frequency, and this one I made was an improvement. But not huge. This signal does not penetrate objects well, so the house blocks a lot of it in most directions besides the window. It did let me sometimes spot planes over a hundred miles away though!
So now I can hear a plane, look at a map, and see in real time exactly where it is. No big surprise: lined up with the nearby runways for takeoff or landing. But not always! I believe the planes are directed certain ways based on weather conditions, especially wind. Prevailing winds locally seem to usually keep the planes rather far away. When they're close, it's mostly a landing approach. And now that I've got this system, it can be fun to watch...
Like when I spotted this plane clearly making a back and forth grid pattern. Had to be collecting imagery, right?
2026-04-11 09:29 - General
I have the good fortune to currently be living very inexpensively in a home that's been in the family for seventy years or so. An older, barely-post-war, home. I don't know for sure but I suspect that thermal insulation is minimal to nonexistent. In looking for my own more permanent home, concentrating on an at least slightly newer one, guessing that more insulation will be one of the things that implies, has been on my mind. But how much of an issue is that really?
Here's my actual natural gas usage from September 2024 through August 2025. (Actually, a 5 day moving average for a slightly less spiky graph.) You can clearly see that in May through September usage is both very low and very flat: this is just the hot water heater, stove/oven, and clothes dryer. This non-heating period averages 0.354 therms per day. We can also clearly see on the graph, starting gradually in October and lasting through April, a much higher usage over the winter, when the house is being heated. This year, the peak was on the 25th of January, 10.93 therms used in one day — almost exactly a non-heating-month's usage!
On my most recent bill natural gas cost $1.51 per therm. If I subtract the average non-heating daily usage, then sum each day's usage for the month, then factor in this dollar cost, I can find out my actual heating cost:
| Month |
Therms |
Dollars |
| October |
19.87 |
$29.99 |
| November |
52.54 |
$79.26 |
| December |
164.25 |
$247.78 |
| January |
240.62 |
$362.97 |
| February |
171.91 |
$259.32 |
| March |
91.19 |
$137.57 |
| April |
37.78 |
$57.00 |
Or: $1,174 to heat this 1,300-ish square foot house for one season. From what I can look up pretty quickly, this isn't too far out from typical for the region. Perhaps I need to adjust my expectations. Either way, it's nice to have this one real, carefully gathered, data point. These 778 therms used compare to the 2025-2026 season (so far) at 768. Distributed differently month-to-month but very similar overall.
2026-04-10 21:43 - Making
I'm not totally sure where and when I first got it, but I know I've got a DYMO LabelManager Plug N Play Label Maker. A nifty device; unlike your typical label maker with a low quality built in keyboard and tiny display, this is a small and sleek unit with neither. But it does have a USB port (original USB A, i.e. "printer USB") for plugging into the computer. Do so and it presents as a USB drive with the simple and portable software to control it.
For some reason however, just like a more typical/portable label maker it uses batteries. A built in, rechargeable, lithium battery. And they go bad
. I know in the past, I had problems with mine. Plug it in and it would say the battery was too low to use, but charging. It never charged, though, almost. If I tried several times I could get it to work. Now however, it never worked. I had the realization that I could do something about that though!
Not only do I have a 3D printer, I've got a collection of "street lithium". Even disposable "e-cigs" have rechargeable lithium batteries in them, plus they end up littering the streets just like cigarette butts. People like me will gather and harvest these battery cells. The original pack I'm replacing says it's "4.81Wh" on the side (and 7.4V which means it's made up of two cells internally), and I found a pair of 2.4Wh cells of appropriate size. First, I crudely connected these bare cells to the label maker with alligator clips. And it worked!
So I designed, more quickly than I expected, a compatible shell to fit in the same space. It's the same overall shape, plus critically two things: it's got notches to mate with a tab in the device (for polarity protection), and it's exactly the right length to snap in against the spring terminals. It doesn't look great; I've been tuning and tweaking the 3D printer and finding the edges of how fast I can drive it. Not quite this fast, for a high quality result. And that was mostly it! But once I got this far I decided to go a smidge further.
I manually hooked the original battery pack up to both charge and discharge it. Via experimentation I noticed that the pack would cut itself off under most charging conditions; this really seemed to be triggering over voltage protection. If I charged very gently I could get it to a sort-of reasonable voltage, but discharging it from this point gave roughly 7mAh &emdash; far less than the 650mAh advertised on the side. I decided I had nothing to lose and started cutting the pack open. Inside as expected I found two cells (though of different dimensions than I expected) and, under their outer wrapper layer, a BMS (battery management system) board, which is of course meant for a two cell pack like this. I harvested it for my replacement. After installing this, which fit in the slight spare space I had already left, I wrapped the replacement in just enough polyimide tape to keep things tucked in neatly.
When I put this replacement pack in, the software says not charging. But that's because, as the display shows, it knows this pack is already fully charged! It fully works as a replacement. With the added layer of tape at least, it's a smidge big, and takes some effort to remove. I'm not going to store this replacement inside the unit. I don't know what kills the batteries, but that might contribute.
So why the battery at all? Well, your more typical label maker will run off of six alkaline cells: nine volts nominal, closer to six and a half as they're nearly empty. This lithium cell will provide around nine to six and a half volts or so, nearly the same range. So it will work. But of course, the thing is designed to plug in via USB, and can't work otherwise. And to charge this battery via the USB connection! I think the hint is the full size USB A style plug. I this device seems to be from 2010, i.e. before the "smart" "phone" market really changed USB. Back when you might only get half a watt, possibly 2.5 watts, from a USB port. Potentially not enough to drive the motors and thermal printer head? So a very low powered boost circuit will slowly charge the battery, and the battery will definitely have the power to run the printer when that's needed. And the software will monitor the battery, refusing to print when it's low. That's my theory.
2026-03-10 17:39 - Making
Back in 2016 I got my first 3D printer. It's a minor hobby for me, but I keep using it every once in a while. It came with, as was very typical at the time (and not unusual still today) a version of Marlin for its controller software. In fact, part of the assembly/installation instructions involved downloading the source for it, tweaking the source, and compiling it. As far as I can tell, from a comment in configuration.h it was Marlin version 1.0.3 — but there is no such version? The vendor patched version was probably based on 1.0.2, from 2015. This is fine. But is fine good enough? In the middle of 2024 I looked into newer Marlin releases. Apparently I ended up with 2.1.1.2. There were a variety of new things. I especially appreciated additional controls available with the built-in screen, and overall I liked the result. For a while. It took me a long time to figure it out, but there was a serious problem.
Pictured above is the very same simple box shaped object (a spacer for my silverware tray, so that each utensil fits neatly into the space available, rather than rattling around in empty space). On the left is one I printed while Marlin 2 was running on my printer. This printer uses a GT2560 control board, with the main processor running at 16MHz. Despite being a supported target, something about all the nearly ten years of changes between versions was too much: Marlin did not run well enough on my printer. When making probably any, but especially noticable in long straight moves like the faces of these boxes it would jitter periodically, where the X/Y translation motion would speed up briefly, go slower for a while, and repeat. This leaves very obviously visible artifacts in the objects being printed.
It actually took me quite a while to connect the dots and figure out that it was the Marlin upgrade that was the problem. I think I had generally smaller objects without such long straight lines, where this problem was most visible, for a while. But eventually figure it out I did. Despite it not directly supporting my model of printer (with going back to the original Marlin version as the backup plan), I took the leap of trying out a different type of printer control software: Klipper.
Like mentioned above, Marlin does a lot of complex math inside the printer and the newer software was (it seems) trying to do more than my printer could handle. Klipper splits this in two: the computer built into the printer essentially only does the direct controls of the hardware, the minimum possible. The complex math involved in deciding exactly when to (e.g.) turn which motors happens elsewhere: on another computer. This is almost definitely a more powerful computer. And since it's a regular computer, it's much easier to build and try features into this software.
It took some time, starting in October 2025, to build up a Klipper configuration from scratch for my unsupported printer. I could thankfully leverage other GT2560-based printers as a base. Then the most useful thing for me: Klipper has a lot more features for tuning and tweaking built in. I used them to find the mechanical limits of what my printer could do. Once tuned, it worked at least as well as ever, probably even much better than new.
I used it this way for a while with a small renaissance for me, it was new and exciting again, at least a little bit. Eventually, however, things took a rough turn again. The printer would randomly crash and cut out. Powering off and on would make it work again, usually. Then even that didn't seem to help again. After a short while I finally started looking into it. At first I checked the power supply, but it was outputting a perfect twelve volts the whole time. Then I looked at the control board for the printer and rapidly found the issue: the power connector. It was terribly black and charred. I think it wasn't quite good enough. Close but not quite, so with too much contact resistance it slowly built up heat over time until it failed too far. Once I replaced that connector the machine again was good as new. Maybe better? Long ago I added some LED strips, to make things bright enough to monitor with a camera. As best I can recall they would always dim when the heaters, especially the large heated bed, kicked in. Now that I've replaced the power connector they don't dim anymore — or so little that it's hardly perceptible.
This was a bit of a roller coaster. I had noticed the problem just before holiday season, and didn't figure it out for a month or two simply due to the timing of when I even tried to. Something about fixing it made me happy and I put a few extra dollars in: I upgraded to a magnetic flexible build plate. Which by the way was a great improvement! And to celebrate this I took the time, like I had long intended, to design an organizer insert for my primary toolbox. Before, somehow the tool I needed seemed to always be on the bottom.
Here's the result of that. I split the design into four parts, one for each corner. Primarily because the 200 mm diameter print area of my delta printer definitely didn't fit the whole thing as one piece, and possibly not two. Turns out to have been a good thing: the first piece I printed, in black, used up the last of that roll of filament. One of the other corners (bottom right in this picture) was a failed print: a kink in the filament made it stop feeding, 80 or 90 percent of the way through. And then after finishing the fourth quarter, I again almost perfectly used up that roll of filament, the last I had on hand.
Somehow very shortly after that I was browsing Marketplace, as I sometimes do for curiosity and entertainment. I found not one but two very reasonably priced 3D printers. Both around an hour drive away, but very near each other. I reached out to both sellers and very quickly was setting up appointments with both of them!
First was a Monoprice Select Mini 3D Printer V2. A rather famous early inexpensive model. I didn't want this terribly much, but it came bundled with five mostly full rolls of filament, including a glow-in-the-dark roll and a flexible material roll. Just the filament was nearly worth the $65 asking price. I've given it some attention and it's definitely working, though I might try to resell it, as it doesn't do much my other printers don't, if anything.
Next up was an Artillery Sidewinder X3 Plus. Much like my original printer, this is a nearly-no-name brand that offers good price vs. performance. It's a large format printer, fitting 300x300 mm on the bed and going up to 400mm high. It's quite new by comparison, having been originally released in 2023 (I don't know about when this particular one was made but I'm the 3rd owner, so probably not super recently!). It went for $60. It did not work when I first got it home, and I suspect the seller knew this. But, after a bit of looking around, unplugging and re-plugging things: now it works great! I've made a few real prints with it and I'm very happy with it for what I spent. In fact, it can print big enough things that I might redo my toolbox organizer as one big piece! I also should redo the wiring repair from the previous owner, with the sloppy white tape visible near the top.
I haven't yet decided if I'll switch one or both of these new printers to Klipper, yet. It should improve their performance. And there's some value to having the same control and interface for them all. But both of these new (to me) printers have nice built in interfaces, which Klipper would not use at all. So I'll live with them as is for a while, and think about it again later.
2025-12-30 14:31 - General
Almost exactly three years ago I was moving and needed a vehicle. I was interested in getting an EV but not excited about paying new vehicle prices, and unsure if I could expect (post move) reliable access to charging, so an EV as my only vehicle might be a risky prospect. I eventually decided on a used plug-in hybrid, with a bit of EV only range but also a regular gasoline engine for longer trips. Overall, this has worked out great for me so far.
Not too long after getting it, with the ability to go both full EV and traditional gasoline powered, I started to wonder: what's the cost efficiency of driving on battery power vs. gasoline power? I've got some notes based on documented average data (not my car) that gives a ratio of 1.89: gasoline being almost twice as expensive as electricity per mile. I also know I carefully measured my own vehicle, but I think I did a 100% to 0% EV only trip and recorded the distance and electricity in, and I also did a significant fraction of a full gas tank, in that 0% (i.e. hybrid-only) battery state. I wrote the numbers down on paper and remember a very similar two-to-one ratio (but don't have that piece of paper anymore). So I've been charging at every opportunity, and driving EV only whenever practical ever since.
A comment online was claiming very different numbers, but in a vague hand-wavy sense. I thought it must be wrong, can't be so different from what I've measured, so I tried to prepare some actual documented numbers from my own vehicle and experience... and found that indeed, the numbers are very different today!
I might not have taken things like charge/conversion efficiency into account, or I don't know what else. But I've decided to throw together a quick tool to help me figure this out once and for all. Put in: your (marginal) electrical cost per kWh, your cost of gasoline, plus your capacity and distance values. If you've typed in one cost (per unit) and not the other, that other will be calculated as the equivalent cut-off rate. So e.g. because electrical costs are more stable than gasoline: fill in all numbers besides your cost for gasoline and it will tell you the cutoff cost. When gasoline costs less than this, it's cheaper to drive on gasoline than electricity!
If possible, you should give the full amount of electricity consumed. (I have a charger (EVSE) that will report the total energy delivered in a charging session.) There are delivery, conversion, and charging losses and you pay for all the energy lost at each step. Unit labels are for my convenience. If you'd rather type in liters or kilometers or anything else, just ignore the unit labels.
At the time I write this: local gasoline was $2.499/gal while electricity was 22.966¢/kWh. Two years ago electricity was 17.219¢/kWh (so it's gone up by 33%) while gasoline was $3.019 (so it's gone down by 21%). With the most reliable distance/capacity numbers I can look up easily (it takes a time and just the right trips to gather really accurate ones): the energy cost of electricity today is about 38% more than gasoline, to drive the same distance. Quite a difference from the 50% less I calculated a while ago, probably with some bad assumptions or rounding making that number even further off.
