Render, don’t build

A page is produced when it is requested, from the files as they are at that moment. No build step, no compile, no generated site to deploy, no prebuilt copy of anything.

The choice

Static-site generators build the whole site on every change; compiled systems resolve references into a baked graph on publish. Both buy fast reads with a second copy of the truth — one that can be stale, half-built, or out of step with its source.

CTXR renders on request: an entity is read, its references resolved, the template applied, the page sent. The machine views are cached against the live layer, but a cache is disposable and never consulted for truth.

What it buys

One copy. There is no built output to keep in sync with the source. Change a file; the site has changed.

No waiting. Publishing moves a file from the draft layer to the live one, in the time it takes to rename. The next request renders the new state.

No cascade. A compiled system must recompute everything that embeds a changed entity — work that grows with the site and fails when interrupted. In CTXR a person’s name changes in one file, and every page that references it shows the change on the next request.

Small and inspectable. No build machinery is a large part of why the engine is small, can be audited, and is replaced by replacing one folder.

What it costs

Work per request rather than per change: a few file reads and a template every time a page is served. For the sites CTXR is made for, that is a few milliseconds, behind a content delivery network in any case. Millions of pages and a change a second would want a different trade; hundreds of pages and a change a day is where rendering wins.

The lesson behind it

An earlier engine we built compiled, and accumulated exactly this debt: interrupted cascades leaving stale indexes, embedded copies growing with depth, limits that measured the source while the weight sat in the baked artefact. Rendering on request removed that whole category.