Built by Hannah Levinson • more at hrlevinson.com
A reading log for everything that isn't a book — essays, articles, reports, poems, chapters. Log what you read, see your own record, and follow what other people are reading.
Live: https://everything-counts.vercel.app/
Real accounts, real data: sign up, claim a handle, log what you read, follow people. Postgres (Supabase) with row-level security behind a thin data-access layer.
- Log a piece by pasting a URL. Title, author, source, and image are pulled from the page's metadata.
- Your logs form a chronological record (the profile).
- Logs are public by default, so they also appear in a feed for the people who follow you.
- When several people log the same piece, their logs pool on one work page ("four people you follow read this").
- No books. Goodreads already does books. This is for the reading that isn't tracked anywhere. The unit is the discrete piece — a chapter counts, a finished book doesn't.
- The log is the signal, not a review. Written "takes" are optional and minor. Most people log what they read without reviewing it. Discovery is built on who read what, not on commentary.
- Public by default. Logging a piece shares it. No separate share step. A private option is planned but deferred.
- Canonical-URL clustering, not full deduplication. Logs resolve to a shared work by normalized URL. Cross-edition/reprint matching is a later migration; the data model points at a work ID rather than a raw URL so that stays a migration, not a rewrite.
- Next.js (App Router), TypeScript, Tailwind, deployed on Vercel.
- Supabase: Postgres + Auth, with row-level security (you can only write your own rows; reading is public by product design).
- A data-access layer (
getLogsByUser,getFeed,getWork) sits between the app and storage; it is the only module that knows the storage is Supabase. - One server route fetches link metadata (OpenGraph) from a pasted URL.
- Built and deployed in phases: empty deploy first, then logging, profile, feed, work pages, then real accounts.
A private-log toggle (the flag exists in the data model), rich deduplication, a reading metric, multimedia (video essays, podcasts), and an algorithmic feed. Each was left out deliberately; the architecture leaves room for them.
Swiss-brutalist: black rules, one accent color, large grotesque display type, square corners. Space Grotesk for structure, a serif for the reading itself.