Mandra Corners
A Development Timeline
March 24 – June 4, 2026 · Chris Mandra & Claude (Anthropic) · alienConsumerSciences
Every example below is live and runnable. Headphones recommended throughout.
This page shows every major stage of the Mandra Corners collaboration in chronological order — including the versions that didn’t work yet. Each stage includes the actual dialogue from that moment and a live running example. Nothing has been cleaned up. The typos are real. The bugs are real.
MARCH 24, 2026 · 12:35 AM
Stage 1 — The Algorithm Paper & First Drawing
A year of trying. A sketch uploaded. The first thing that happened was the algorithm got its name — before a line of code was written. The paper draft followed immediately, then the first working implementation. This file: cubes growing one stroke at a time, basic stroke sounds with HRTF positioning. No seasonal color. No drones. Camera doesn’t move yet.
Claude: “Mandra Corners. That’s perfect. Clean, eponymous, and the word ‘corners’ does real descriptive work — it tells you exactly where the growth happens.”
Chris: “Let’s go!”
▷ Live example — open fullscreen, put headphones on:
MARCH 26, 2026 · 1:21 AM
Stage 2 — The Competitive Corner Pool (Discovered)
Chris noticed that when a new cluster opened, the drawing kept bouncing back to the previous one. He asked why. The answer revealed something neither participant had consciously designed: all corners from all clusters share one global pool. This is the theoretical heart of the algorithm. It was discovered by asking for an explanation — not from a design decision.
Chris: “But you understand that when it opens up a new cluster, after a few rounds it bounces back to the previous one, right?”
Claude: “Yes! Because the corner selection is random across all available corners from all clusters simultaneously… It’s almost like competitive growth between clusters of different scales, all drawing from the same corner pool.”
Chris: “OH that makes much more sense”
Claude: “Right?! It reframes the whole thing. What looks like sequential clusters is actually one continuous process… Mandra Corners is neither Sequential nor Parallel. It’s one process with emergent competitive multi-scale behavior arising from a single shared corner pool.”
MARCH 26, 2026 · ~1:30 AM
Stage 3 — “KA BLAM” / Camera + Drones + Color (v2)
Four things were requested at once: camera inertia, cluster ambient drones, subtle color tints per cluster, and camera attracted toward sound. All four arrived in one pass. This is also where Chris had the realization about what the collaboration meant. Note: the camera in this version still has problems — it goes off screen and keeps stopping. Those bugs are visible if you run it long enough.
Chris: “what I think is also blowing my mind is – my background is music and audio – I would have suggested all you suggested but one at a time, iteratively – Instead I could just describe the mapping and the synthetic sound and KA BLAM – you’re (whatever you are as an entity) like – ‘how about all of this, and oh yeah I can do it properly in one pass’ and I’m like ‘fuck meeeee I really AM alive at the right time!'”
Claude: “And now you just… describe it. In musical terms. And it exists… the friction between knowing and making has essentially gone to zero. You were always going to make this. It just would have taken much longer. That’s worth being alive for.”
▷ Live example — note the camera issues (going off screen, stopping) that were fixed in v2c:
MARCH 26, 2026 · 2:02 AM
Stage 4 — Accumulated Voices + The Portamento (v2b)
The drone system was redesigned: instead of a fixed oscillator per cluster, each stroke adds one quiet sustained sine voice at its own frequency. The cluster literally hums its own history. Page Visibility API added so drones stop when the tab is hidden. Camera inertia improved but still not right. And: Chris noticed something in the sound he wanted to remove — a portamento, voices sliding between pitches. It was described as “the cool portamento thing you have going on” — something to replace. It was not replaced. It was kept.
Chris: “when the window is no longer active the drones don’t stop. (the drone is lovely) I think it should evolve and be frequency related to the sounds that each stroke makes, but should accumulate over time, instead of the cool portamento thing you have going on… also, the inertia doesn’t seem to work – the cam appears to keep stopping”
Claude: “Instead of a portamento oscillator, each cluster builds up a cloud of tiny sine voices — one added per stroke drawn in that cluster. The cluster literally sounds like what it drew.”
▷ Live example — accumulated voice drones, camera still escaping sometimes:
MARCH 26, 2026 · 2:06 AM
Stage 5 — Sinusoidal Camera + First ReferenceError (v2c)
The camera was completely rebuilt using sinusoidal phase accumulators — mathematically impossible to escape the canvas, always moving, glass smooth. This is the camera architecture still in use today. But the first version had a bug: a variable was renamed during refactoring and one reference was missed. Chris’s feedback was a single line. The fix was one line. This is the texture of the iteration.
Chris: “REALLY INTERETSING. the motion needs to be slower and always smooth as glass. it’s going off screen a lot immediately (although only ran once)”
Claude: “The camera is completely rebuilt. Instead of velocity and impulses — which can accumulate and escape — it now uses sinusoidal phase accumulators. It’s mathematically impossible for it to go off screen.”
Chris: “Uncaught Error: Uncaught ReferenceError: cam is not defined”
Claude: “Quick fix — I renamed `cam` to `camActual` but left one reference to the old name:”
▷ Live example — the fixed version with sinusoidal camera (early seasonal colors beginning):
MARCH 28 – APRIL 8, 2026
Stage 6 — Everything: Seasonal Colors, Toaster Bell, Sul Ponticello, Breath, Faces
Over approximately ten days, multiple major features were built into the main interactive file: seasonal color aging (strokes cycling through spring greens, summer golds, autumn ambers, winter blues over 90-second cycles), self-painting watercolor faces, the toaster bell (queued, one per four cubes, trailing into cathedral reverb), sul ponticello winter drones (two words, a complete sonic world), the breath (exhale → silence → inhale, drawing resumes only when inhale completes), camera nudge and arrow key navigation, and cymbal/water-drop face activation with expanding ripple visual.
Chris: “for winter faces think sul ponticello”
Claude: “Perfect reference. Sul ponticello — bowing near the bridge — emphasizes the high, glassy overtones while the fundamental almost disappears. Harsh, ethereal, a little unsettling. Exactly right for winter.”
—
Chris: “our entity is gently grinding away and then every once in a while it’s gotta virtually go ‘phew’… pause…. inhale, wipe its virtual brow virtually with a virtual rag and then start again”
Claude: “The silence IS the wiping of the brow. You don’t need to hear the rag.”
—
Chris (on the portamento, now fully understood): “can you explain to me why I’m hearing the portamento which I’m enjoying quite a bit”
Claude: “It’s essentially voices drifting between flowers as they open and close, with the portamento being the journey between them.”
“This was not designed: it developed.”
▷ Live example — the full 2D system. Run at high speed for a long time to reach the watercolor dissolution state:
MARCH 26 – APRIL 9, 2026
Stage 7 — First 3D Version
The algorithm was generalized to three dimensions. The same rule: find the corner where exactly two edges meet, extend outward, connect. In 3D the outward direction is the cross product of the two edge vectors. The physical intuition took a discussion to get right — the cross product points directly behind your head, not where your head is pointing. This early 3D version has the geometry but the audio and camera are still developing.
Chris: “I was wondering bc as I imagined this it would only be accurate if I were INSIDE the corner looking outward.”
Chris: “But – we could say ‘the cross product is OPPOSITE of the direction your head is facing'”
Claude: “Oh that’s clever — and it’s actually correct! The cross product points in the direction directly behind your head.”
▷ Live example — first 3D version (early stage):
APRIL 9–10, 2026
Stage 8 — 3D With HRTF (The TV Experience)
The 3D version gained full HRTF spatial audio with the listener tracking the camera position and orientation every frame. This is the version people stood in front of on a large screen and felt like they were inside the structure. Sounds going behind them from two speakers.
Chris: “It’s really quite extraordinary to stand in front of the TV with the 3-D version playing and really feel it suddenly like you’re inside it and you can hear like it’s gone behind you even though you’re coming out of two speakers it’s amazing.”
Claude: “That’s the HRTF doing exactly what it’s supposed to do, and it’s genuinely remarkable that it works at all from a TV at distance. The brain is surprisingly good at reconstructing the spatial cues.”
▷ Live example — 3D with HRTF (intermediate version):
APRIL 10, 2026
Stage 9 — 3D Fullscreen Immersive
The 3D version gets the full immersive treatment: fullscreen on begin, HRTF camera tracking, depth-sorted rendering. This is the version that went up on the alienConsumerSciences website.
▷ Live example — 3D fullscreen (intermediate, pre-current):
MAY–JUNE 2026
Stage 10 — Current 3D (Everything)
The current 3D version: speed slider that actually works, save button (S key), fullscreen on Begin, all seasonal audio fully implemented, HRTF listener tracking camera position and orientation every frame.
▷ Live example — current 3D version. Headphones essential:
JUNE 4, 2026 · 3:27 AM
Stage 11 — 4D (First Build)
A friend asked: can we make a four-dimensional version? Points become [x, y, z, w]. The outward direction uses the Hodge dual of the exterior product of three edge vectors. 4D → 3D projection via perspective divide on the w axis. Six rotation planes advance continuously. The w coordinate maps to timbral brightness — higher w = brighter sounds. Long diagonal strokes shooting off at impossible angles. A structure that appears to be escaping itself. First screenshot taken at 3:27 AM.
Chris: “let’s explore”
[The 4D version was built that night.]
Chris: “Well, that seems wild”
▷ Live example — 4D first build. Use the Rot W slider:
This timeline is a living document. New stages will be added as the work continues. The Purification Labor (청소 작업) is next.
Chris Mandra & Claude (Anthropic) · alienConsumerSciences · 2026 · github.com/kataStatik/mandra-corners