KATHA — engineering a reading & publishing platform end to end

Two App Router route groups keep the reading product and the writing product cleanly separated — different navigation, different concerns, one codebase. The decision this proves: route groups as product boundaries, not folder decoration.

One identity, three states: guest → reader → author. The model separates the person (account) from the pen (public writing identity) — becoming an author links a profile to the same user rather than creating a second account, which is how pen names work in real publishing.

Real authentication is deliberately deferred — but the seam for it is not. All role gating flows through one function, and the module names exactly where the future backend plugs in:

Search is a pure function from (query, books) to ranked, grouped results — no React, no DOM, no storage, and no import of the catalogue itself, which makes the whole engine testable in isolation. Matching is deliberately forgiving; every query token must hit (AND semantics), and each token scores on a descending ladder:

Quality work here is verifiable in the history rather than claimed: a dedicated “motion-safe sweep” pass, “next/font migration + motion consistency”, and “loading skeletons for all dynamic routes” all appear as commits in the v1.0 polish series. The global stylesheet carries explicit prefers-reduced-motion and focus-visible handling.

On performance, the architecture does the work: a server-first component tree, self-hosted fonts with no render-blocking requests, a search index that keeps prose out of the bundle, and a dependency tree of exactly three packages. No third-party scripts, no client-side data fetching for the catalogue.

Every architecture buys something by giving something up. These are the trades KATHA makes today — each one paired with the seam the codebase already documents for undoing it: