Seventeen mount files become one

We have never refactored this codebase. Seventeen micro-frontends and eleven services got built one at a time, each correct on its own, none of them looking sideways at the last. That accrues a specific kind of debt — not bugs, but repetition with no home — and the only honest way to find it was to stop guessing and measure.

So we built a knowledge graph of the source: 4,330 nodes and 6,011 directed edges over 490 files, for zero tokens, because code is parsed structurally rather than by a model. Then we verified every candidate it surfaced against the actual files, because a graph tells you where to look and not what is true.

What it found, and what we did

Eleven of the seventeen mount.ts files differed from each other by exactly one line — the exported function name. The other six differed only in comments. Four hundred and six lines of it, none carrying distinct behaviour. They now share one factory in a new @augment-it/federation package, and each member is down to twelve lines.

The nervous question is whether sharing code costs a micro-frontend its independence. It does not, and that is structural rather than hopeful: the federation host declares no shared block, so a workspace import is inlined into each remote's own bundle at build time. One source of truth in the repo, seventeen independent artifacts, every remote still booting with every other remote down. We proved it — record-collector's built stylesheet is byte-identical before and after, same content hash.

It also puts the repo's most fragile constraint in one place. theme.css must load before a member's app.css, or that member's var() references resolve against tokens that do not exist yet. That rule was previously restated in seventeen file headers and enforced by nobody.

Two hundred and fifty lines that nothing imports are deleted: an OrgCreate.svelte whose five siblings are all wired up, and two orphaned bundles.ts modules kept alive by a comment that pointed at them without importing them.

pnpm build now works, for the first time in this repo's history. It called turbo run build, and turbo was never a dependency and was never installed. Turbo also had no job here: its whole purpose is ordering a dependency graph, and no package in packages/ has a build step — they export raw source, bundled by each app. The task graph had zero edges. turbo.json also used a Turbo v1 key renamed in v2, and declared Next.js output paths in a repo that emits dist/. It is gone; the root scripts use pnpm's own recursive runner.

The one import cycle is broken. capabilities.ts and searches.ts pointed at each other. It was type-only and therefore harmless at runtime, but it made the module graph lie about which direction the dependency runs.

Three tsconfigs became one. The apps carried three byte-distinct variants in an 8/7/2 split — drift, not intent. Each app's config is now four lines.

The finding that changes the roadmap

We expected the frontend to be full of copy-pasted CSS waiting to be deleted. It is not. Of 1,392 distinct selectors across 5,434 lines of app CSS, 1,386 are unique to a single app. The six shared ones are keyframe stops and media queries.

There is almost nothing to deduplicate. There are seventeen independently invented vocabularies that have never had to agree. That inverts the component library from a consolidation job into greenfield extraction plus migration — worth knowing before starting rather than after. The token layer underneath is already working: 1,519 token references against 490 hardcoded escapes.

Verified, not assumed

19 packages build. 1,494 files pass svelte-check with zero errors. Every test suite passes except the end-to-end group, which needs Docker running and was skipped. Net −466 lines across 65 files.

One caveat we are keeping visible: "nothing imports this" is a candidate list, not a verdict. Two components first looked dead and were not — they are imported through package subpath exports that static analysis cannot follow. Everything deleted here was confirmed by hand afterwards.

Full backlog, including the deferred items and why they were deferred: context-v/refactors/Structural-Refactors-Surfaced-by-the-Codebase-Graph.md.