iiinova — where your digital self keeps evolving

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 13/10/2026

Opus 5.5 can apparently write code for my 45+ year old Atari 400.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 04/10/2026

Decided to work through all the Messier objects over time. Here are three I didn't have in my catalog before.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 30/09/2026

My own data determining the brightness of the supernova in NGC 7331 across the month of September, showing its flash as it fuses to Ni56, and its slower decay through Co56 and Fe56.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 30/09/2026

"Brains To Go", a short film created by me and claude code, entirely from scratch, in less that twenty four hours.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 29/09/2026

Yesterday my AI brief included links to a couple of projects that used AI to create animated videos. I thought I would give it a whirl. I brainstormed the idea of "Brian Wagon", the disembodied brain mascot for my blog at brainwagon.org. A couple of hours later, I had this video.

001 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 24/09/2026

Looks like the supernova in NGC 7331 may be beginning to fade. New data tonight acquired with my Seestar S50. #astrophotography #seestar

020 2

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 18/09/2026

NGC 6992, the Eastern Veil, blinked against Pan-STARRS to check which wisps are real. Every filament survives the blink. Caveats all round: 42 min is short, the framing is poor — the arc runs off both edges and NGC 6995 missed the frame entirely — and there […] [Original post on mastodon.social]

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 18/09/2026

SN 2026aaiv in NGC 7331, four nights with a Seestar S50. V = 13.24 → 12.64 → 12.38 → 12.17, Sep 8.2 → 18.2 UT. Measured on linear restacks of the raw subframes — not the camera's own stack, which is sharpened and reads ~0.75 mag too bright this close to a […] [Original post on mastodon.social]

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 13/09/2026

SN 2026aaiv in NGC 7331, three nights with a Seestar S50. Measured on linear restacks of the raw 10 s subframes, calibrated against APASS stars in the same frame: Sep 8.20 UT: V 13.24 ± 0.27 Sep 11.23 UT: V 12.64 ± 0.25 Sep 13.24 UT: V 12.38 ± 0.15 Still […] [Original post on mastodon.social]

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 12/09/2026

Today's random project, I made a little display out of a CYD (Cheap Yellow Display) board that has an ESP32 with wifi. It monitors my Claude Code subscription so I can now when I am about to run out of tokens.

A CYD (Cheap Yellow Display) acting as a monitor for my Claude Code subscription.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 11/09/2026

I took a Cheap Yellow Display and made a little monitor to keep track of my Claude Code usage. It probes every 15 minutes, and updates its little bar graph.

Trashbot keeps an eye on my Claude Code usage, and when it resets.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 11/09/2026

30 minutes of exposure on SH2-101, the Tulip Nebula. The frame also contains the bright blue optical companion to the strong X-ray source Cygnus X1, the first object to have been widely accepted as a black hole. Nifty! #astrophotography

SH2-101 and Cygnus X1, shot with my Seestar S50 and aligned/processed using custom code written by Claude Code.

011 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 11/09/2026

New post: the supernova magnitude I posted on Tuesday was wrong. The Seestar S50's own stacked FITS is contrast-stretched and sharpened, and next to the nucleus of NGC 7331 it made SN 2026aaiv read about 0.75 mag too bright. Restacking the raw frames myself […] [Original post on mastodon.social]

Three panels of the galaxy NGC 7331, north up. Left: September 8, 2026, supernova SN 2026aaiv marked just southeast of the nucleus, V = 13.25. Middle: September 11, the supernova visibly brighter, V = 12.63. Right: the difference image, Sep 11 minus Sep 8, where the galaxy and field stars cancel out and the supernova remains as the only bright spot.

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 10/09/2026

New post: vibe woodworking with GLM as my shop assistant. A rolling cart for a dozen IRIS storage bins, designed by arguing with GLM-5.3-Flash in opencode — a parametric 3D tool with an auto-generated cut list, and no shelves: the bins' tapered bodies hang […] [Original post on mastodon.social]

Three.js render of a rolling OSB cart holding three columns of four translucent storage bins

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 09/09/2026

Found a 2.7GB AVI on a backup disk that I barely remember shooting: 14 seconds of the moon from May 2020, taken with a $60 SVBony SV105 at prime focus on the 6" f/3.9 Newtonian I built more than 20 years ago. Stacked the best 43 of 435 frames using lucky […] [Original post on mastodon.social]

The waxing crescent moon of 26 May 2020, stacked from video and sharpened, with eighteen lunar features circled and named, including Mare Crisium, Mare Fecunditatis, Cleomedes, Geminus, Endymion, Atlas, Hercules, Proclus, Taruntius and Vallis Rheita.

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 08/09/2026

I normally develop my astrophotos in Siril. This time I handed Claude Code the raw stacked FITS straight off my Seestar S50 and asked for three things only: flatten the background gradient, denoise, stretch to bring out the spiral structure. Left is the […] [Original post on mastodon.social]

Side-by-side comparison of M33, the Triangulum Galaxy. Left panel: the Seestar S50's own in-camera JPEG, showing a grey sky with a visible gradient and only a faint smudge of galaxy. Right panel: the same frames reprocessed from the raw FITS stack, showing a dark neutral sky, blue spiral arms, and pink HII regions including NGC 604.

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 08/09/2026

Caught the new supernova SN 2026aaiv in NGC 7331 from the backyard — Seestar S50, 371 x 10s. Then, with a lot of help from Claude Code, I did something I've never tried before: actually measured it. V = 12.5 ± 0.2, which at NGC 7331's distance is an absolute […] [Original post on mastodon.social]

The spiral galaxy NGC 7331 photographed with a Seestar S50 telescope. A yellow circle labelled SN 2026aaiv marks the supernova, a star-like point just below and right of the galaxy's bright core. A smaller companion galaxy is visible to the lower left.

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 31/08/2026

Pondering whether I should touch up/repaint the TIPPETT VFX logos on this carrying case I got at the Tippett Studio estate sale, or leave it scuffed and used. What do you all think?

Carrying case labelled with TIPPETT VFX I got at their estate sale.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 28/08/2026

I was trying to tech myself to be able to compute the day of the week for any date (well, really only the dates in the years 1900-2099) using the "odd-11" variant of the doomsday algorithm. I had claude code help me create this little quiz application […] [Original post on mastodon.social]

You can see that I'm beginning to get better, it's taking me about 25 seconds, and my accuracy is improving.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 27/08/2026

Hmm. The stealth "ox-alpha" model has apparently been revealed as glm-5.3-flash (and is no longer free). Also interesting is that Qwen 3.8 Flash has been made the roster on openrouter. Hoping we see the open weights for that one soon. mvandewettering.com/openrouter-reco…

OpenRouter Price Watch Live pricing from the OpenRouter models API, ranked cheapest first. Prices are USD per TM tokens.
Built by Mark VandeWettering with Claude Code, to understand what OpenRouter actually
charges and how the costs compare — not just between model families, but between slugs
inside one family, which can differ several-fold. Why a mode id is a menu and not a price.
explains what sits behind these numbers.
MW Claude 4 & older WM Claude5 WM DeepSeekV3&R1 WM Deepseekv4 Wl Gemma3 HM Gemmas HM GIM52&older ¥ GM53 Ml GPT-4&o-series WM GPT-5
MW Grok3aolder Ml Grok4 Ml Kimik2&older Ml Kimik3 Ml Nemotron3aolder Ml Nemoton3s Ml Qwen36 Wi Quen3s
#0 show :free Ml show :batch filter substring of model i, blend ratio. 3 in: 1 out Refresh Resetchangebascline Resetview
updated 421:06 PM. 417 models on OpenRouter
MODEL INPUT OUTPUT CACHEREAD BLENDED ESTLAVG CONTEXT PROVIDER
2-2i/gln-5.3-Flash | mutimedsl LM 53 Fish 0.075 0.250 0.015 se.119 s0.158 wm za
quen/quen3.B-Flash mutimodsl Quen Fash 0.160 0.470 0.016 s0.237 50.237 WM Alibaba
quen/quen3.8-27b muimossl Quen3s 276 0.425 vs.ex $2.55 vox 0.085 50.956 $1.05 Alibaba
~z-ai/gln-latest GLM Latest $1.40 $4.40 50.260 52.15 m
z-ai/gln-5.3 GLM53 $1.40 $4.40 50.260 52.15 $2.15 wm za
quen/quen3.8-2.4t-a95b Quenl® 247 A955 $2.00 $6.00 0.250 $3.00 $3.18 WM Deepinfra
quen/quen3.8-max | mulimodsl Quien38 Max $2.00 $6.00 0.250 $3.00 53.00 Ababa
> How to read this table, how a slug is priced, and how the numbers are computed
Source on Github data from the OpenRouter models API

100 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 26/08/2026

Today, in "things that I'm older than" news: I'm apparently older than Fantastic Voyage.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 06/08/2026

Today's minor bit of time saving for AI: tinkered together a gadget that added modifier objects to a 3D print file that sets infill around holes to 100% automatically. Now, I can largely just model the holes, and worry about reinforcing them during printing later, if at all.

Yellow cylinders mark regions where infill is set to 100% to reinforce the area around holes.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 15/07/2026

Polonius has evaluated over 1B checker positions. Will this result in stronger play when all is done? We shall see.

polonius nn server - up 987m, evaluated over 1B checker positions...

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 06/07/2026

Bernd Mayländer celebrates his historic P1 finish in the British Grand Prix! #grandprix

Okay, it's AI. I thought it was funny to make one of the near legendary safety car drivers standing on the P1 podium.

120 2

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 04/07/2026

I'm at a point in my #telescopemaking project where I need to cut a hole in my tube for the eyepiece. But... could I find my hole saws? After an hour, I found them in the bottom of a random box, covered in dust. I decided that I should create a little shelf […] [Original post on mastodon.social]

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 27/06/2026

Years ago, started "milhouse", my conventional Nchecker-playing engine that used α-β search. I did a brief excursion into making a similar program ( α-β search, but with self play to learn the evaluation function implemented with neural nets). It was somewhat […] [Original post on mastodon.social]

® # Good morning — the overnight grade is in, and it's a dominant result.
gen-276 champion vs simplech — 1,772 games (80@ sims, balanced)
[ew [wo [v [am
- Zero losses in 1,772 games. simplech could not beat the champion a single time — it won 463 and drew the rest.
- High precision: SE +1.16 pp, so 61.U% * 1.2.
— The deadline harness worked exactly as designed: 12 chunks, stopped itself at 86:42 (8 min before the 06:50 cutoff)
when there wasn't time for a 13th.
What this means vs the old champion
The pre-E19 champion (gen ~172) scored ~28-29% vs simplech at the same 800 sims (from the 2uh post-mortem) — i.e. it
lost most of its games to simplech. The new gen-276 champion scores 61.U% and never loses. On the external scale
that's roughly a +238 Elo swing (% -158 » +80). Even allowing for some cross-session apples-to-oranges in the old
U4o-game baseline, the qualitative jump — from losing most games to literally zero losses over 1,772 — is unambiguous.
The whole arc, closed

100 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 24/06/2026

Combining my experiments with #telescopemaking and #ai, I used Claude Code to help me make a small application that can output DXF patterns for all the major flat/cut from plywood items to make a basic Dobsonian telescope. It will automatically layout finger […] [Original post on mastodon.social]

Interactive view of all the parts that it will lay out for the simple Dobsonian telescope.

100 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 23/06/2026

Some more #onshape cad, the rough model for the next few bits of a Dobsonian telescope project I'm working on. No details yet, just mucking around with how it looks. Trying to use 1/2" ply rather than 3/4" ply, and will add some gussets to reinforce bits. #telescopemaking

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 19/06/2026

Crazy #ai programming project of the day: a prototype application (front end/back end) that allows LLMs to play the old board game Diplomacy. The python diplomacy package diplomacy.readthedocs.io/en/stable did all the heavy lifting for move […] [Original post on mastodon.social]

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 15/06/2026

#3dprinting of the day: some of the old telescopes I have seem to be lacking anything to cap the focuser so that dust can't get in. In the old days, I used to use 35mm film canisters, but who has those anymore? So, I made some quick replacements. I could […] [Original post on mastodon.social]

Use these to plug up the 1.25" focusers of common telescopes.

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 12/06/2026

Tiny progress on my Monte Carlo Tree Search checkers engine which I coded with Antigravity, and some help from Claude's latest Fable engine to debug an issue.

First draw between Polonius, my checkers engine based upon Monte Carlo Tree Search and Martin Fierz's reasonably capable "simplech". Record for the 32 game series is 30 losses, 2 draws, which is progress, especially considering the tiny amount of network training has been done. It will be interesting to see where it's at in the morning.

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 06/06/2026

Hey #3dprinting experts. I was making a little side bracket to hold some eyepieces for a telescope project, and it looks like it will work fine, but there was a small pair of defects in the first layers, which were highly localized, spoiling an otherwise very […] [Original post on mastodon.social]

Small bracket/shelf to hold four eyepieces.

Focus isn't great, but these are the two defects.

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 05/06/2026

Decidedly mixed feelings about the release of Toy Story 5. It'll be the first one where I don't appear in the credits. I was interviewing for a lead position on the film when the layoffs happened in 2024. Would have loved to have completed my own personal […] [Original post on mastodon.social]

100 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 05/06/2026

I bought two of these www.seeedstudio.com/XIAO-ESP32S3-Se… little xiao sense boards, which have an ESP32S3 and a tiny camera on board for just $14 each for a robotics project, and coded up a quick little camera streamer web server. I […] [Original post on mastodon.social]

Temperture at idle, then 320x240, then 640x480. The increased resolution maybe causes a 1-2 degree C jump.

100 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 05/06/2026

This demonstrates the fuzziness. The SeeStar S50 has a 50mm aperture, with about 2.5 arcseconds per pixel. The ISS was about 50 arcseconds in size, so the ISS is about 20 pixels in size. I tried to track it and register it to try to get a better image of the […] [Original post on mastodon.social]

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 05/06/2026

Transit of the ISS this morning across the solar disk. Recorded video with my SeeStar S50, and then used some ffmpeg/numpy magic to stabilize and show the progression of the station. It took about 0.72 seconds for the ISS to cross the sun. Focus wasn't the best. #astrophotography

Transit of the ISS across the solar disk. 2026-06-05, from El Sobrante CA

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 03/06/2026

Today in #3dprinting: I made this little holder for a chuck key for my new drill press. It has a space for a magnet in the back, so I can just slap it on the drill press, and now the key will always have a good place to live.

Design for a magnetic drill chuck key holder.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 22/05/2026

Sigh. Academic publishing is such a scam. I know I had a copy of this paper back in the day, but all attempts to locate it now have failed.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 02/05/2026

The Gemma 4 E4B model has definite limitations, but it's kind of shocking how well this works, running the model entirely locally (and using my own agent harness).

Demonstration of "99", my own agentic harness using a locally hosted Gemma 4 model, which has image understanding as part of its model.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 01/05/2026

I updated my agent "99" to accept image attachments, which can be described by the gemma 4 models, which include image understanding.

Barbara Feldon as Agent 99, from the TV show "Get Smart"

A picture of Pepita, one of my little black cats.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 30/04/2026

Today's AI exploration: writing my own very minimal coding harness. It's actually not that difficult, just a couple hundred lines of Python to implement the basic loop, tool calls, a SOUL.md file, subagents... My suspicion is that a large amount of code in […] [Original post on mastodon.social]

Minimal agent invokes a subagent to write code.

110 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 27/04/2026

I had a package of 8 punches of various sizes from harbor freight, still in their blister pack, when I needed one today. But, I didn't have an place to store them. So, I made this very simple 3D print, which I will eventually attach magnets to so I can attach […] [Original post on mastodon.social]

Space for all 8 punches, with the sizes clearly marked on the outside.

010 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 24/04/2026

Tinkering with AI coding some more, I decided to try to make a live ISS viewer that displays the ISS + the Earth in their current positions. Took me a couple of hours to get to this point: the app works on both desktop and mobile devices reasonably well […] [Original post on mastodon.social]

My applications view of the iss, trailing it in orbit.

100 1

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 22/04/2026

Digging through old photos, I found this curious still life, which I believe was shot using a primitive camera I built as part of Eben Ostby's class on primitive photography as part of Pixar University. It was constructed from foam core and used a simple lens […] [Original post on mastodon.social]

Toys from my desk, shot with a primitive camera using a magnifying glass as an objective, constructed from foam core and shot onto a paper negative.

000 0

Mᴀʀᴋ VᴀɴᴅᴇWᴇᴛᴛᴇʀɪɴɢ @brainwagon.mastodon.social.ap.brid.gy · 19/04/2026

Simplified the design of a Crayford focuser body for #3dprinting. Uses less hardware, is easier to print, and sturdier. It would be great if I could do it without any supports, but I think I am satisfied with it as it is. Inspired by […] [Original post on mastodon.social]

Focuser body, designed without the need for locknuts, which actually makes it sturdier.

020 2