I compiled fifty years of hardware history into your browser

Half a century of consumer hardware — CPUs, memory, storage, displays — is a story of overlapping exponential curves and the eras they define. But that story usually lives in scattered spec sheets and wiki tables, where you can read a number but never feel the trajectory. silicon·eras is my attempt to make it something you can move through: an era-banded chart from 1975 to 2026 that you pivot, index, brush, and link. It runs entirely in your tab — pure Rust compiled to WebAssembly, no backend, no data fetch. This post is the build log.

The data lives in the binary

The interesting decision is where the dataset goes. silicon·eras has no API and makes no network request after the wasm loads — because the entire dataset is compiled into it.

A long-format dataset.csv is the single source of truth. A build.rs step parses and types it at build time, serializes the result to JSON in OUT_DIR, and the app pulls it in with one include_str!. The parser is rerun-if-changed-keyed to the CSV, so editing a row — say, a PR that fixes a launch price — flows straight through CI to the deployed app with no extra step and nothing to keep in sync at runtime. Genuine parse failures surface as cargo::warning= lines instead of silently shipping bad data.

The trade is deliberate: a slightly larger wasm in exchange for zero latency, zero backend, and a build that fails loudly if the data is malformed. For a fixed historical dataset, that's the right trade.

One CSV, one chart, many views of it

The workspace splits the concerns: silicon-parse turns the CSV into typed records, silicon-model holds the domain types, silicon-viz does the geometry, and the Dioxus app crate is the UI. The first view — "trajectories" — is an era-banded chart with a handful of orthogonal controls, each of which changes the question rather than the data:

• log / linear — see order-of-magnitude jumps, or absolute deltas.
• absolute / indexed — raw values, or everything normalized to a common start so you compare rates.
• era schemes — reslice the same fifty years by different definitions of an "era."
• pins, plateaus, and an era brush — mark a moment, or focus a span.

All of it is reflected in the URL, so any particular view is a link you can share and land back on exactly.

It borrows the page's theme, not its own

silicon·eras ships no theme toggle of its own. Every color is a var(--viz-*) token, and the whole palette keys off the host page's data-theme attribute. So when you flip this site's light/dark switch, the explorable recolors with it — the site owns the theme, the app just inherits it. It's the same trick the rest of the site's embedded Rust apps use, and it means one toggle controls everything.

What it is, and isn't, yet

This is the first view of an ongoing explorable, and it's honest about its limits: it's desktop-first (the chart wants ≥880px of width), and on a phone it falls back to a read-only summary rather than pretending a dense multi-axis chart works at 375px. More views will follow the same shape — CSV in, typed at build time, embedded, pivoted in the browser.

Explore it live at /silicon-eras — real Rust compiled to WebAssembly, the whole dataset embedded, nothing leaves your tab. The source is on GitHub.