[ishaansathaye]

The silicon world is now interactive: paint an image, follow its computed values through a floating 3D schematic, and change memory layout or queue policy to see a real consequence. All four original brain scenes remain available through the top navigation.

PR 3 of the new stack, based on #5 (research in #4). No rejected PR is included and nothing should merge until the owner approves the experience.

Try it

Build normally and open /silicon.html (the current local review is http://127.0.0.1:4181/silicon.html).

1. Paint or choose a pattern, then Run, Pause, Step or finish with ⇥. CPU/GPU/Tensor must produce the same difference-filter result.
2. On GPU, compare Packed and Strided; the Rust address model requests different traffic for the same answer.
3. On CPU, toggle the tiny cache and rerun.
4. On Tensor, inspect the 4 × 4 operand wave.
5. On Switch ASIC, block A, inject A then B, run FIFO and change to VOQ. Packet identities persist while unblocked destinations can progress.
6. Orbit, zoom and select blocks. Reference facts, model rules and linked sources open in the same left-hand tab panel.

Six public product references include Ryzen 9 9950X, RTX 5090, H100, TPU v4 and Cisco G200/G300. Selecting a profile changes sourced reference context; these deliberately small models do not benchmark or reproduce proprietary chip internals. Lights come from model events and occupancy, and functional positions are original schematics.

Validation

Ten browser groups pass: real canvas picking and live inspector tabs, keyboard/pointer input, exact matrix results, layout traffic effects, pause/step/reset, FIFO/VOQ conservation, source navigation, 390px touch/reduced-motion layout and explicit WebGL fallback. WebGPU and WebGL rendered successfully; no browser errors or failed assets. The existing brain suite also passed 13 groups against the shared renderer. Independent visual review led to camera framing, title visibility and label-overlap fixes. 29 core tests and 258 generated-Wasm assertions passed. A Linux Chromium/WebGL CI job now builds Wasm and runs these hardware checks without downloading the anatomy corpus.

Only one compact screenshot is tracked. Generated Wasm, full-size test captures and raw research remain ignored. No deployment, server exposure or merge is included. The brain/atlas comparison is available in #7.

The strict Chromium render check exposed a portability issue: wgpu requested eight color attachments while the bundled browser offered six. The application uses two; the renderer now requests two. The original failure was reproduced and corrected, with the full suite passing on bundled Chromium using SwiftShader as well as system Chrome.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[ishaansathaye]

These foundation commits are included in merged PR #8 and are now on main. Closing this stacked PR as superseded. Live release: https://doke-6lk.pages.dev/