LogN

English · Português

Competitive programming practice with the mechanics of a contest, on your phone.

It is not a video course or an exercise list. You join a match with three lives and a clock, and every accepted answer raises a balloon: the ICPC metaphor, where a wrong attempt costs a penalty and the scoreboard freezes near the end. A session takes minutes, not an afternoon.

How it works

Tracks of topics in programming, linked as a graph, not a list: a topic can open more than one path, and paths can meet again further on. Each node opens with accumulated XP, and XP comes from accepted answers. The curriculum itself (topics, statements and answer keys) lives in the content repository.

Six challenge formats, because knowing how to program has different parts:

format	what it measures
Find the bug	reading someone else's code and pinning the defect to one line
Fill in the blank	picking the right expression among plausible mistakes
Dry run	simulating the state step by step and predicting the result
Match the complexity	the two axes of complexity, time and space, kept apart
Weigh the trade-off	the cost of each approach, with sentences instead of O(n)
Tag the pattern	recognizing the technique that solves it, before writing code

Truly offline-first. All the logic lives in a Rust core shared across platforms; the client only draws. Answering without a network loses no progress: events go into a hash-chained queue, and the server validates the chain instead of rewriting history.

Mistakes teach. Every wrong answer opens a card that explains what is wrong in that code, not generic advice. It is the game's only feedback surface, and it is where the work goes.

Platforms

The launch is on Android, on Google Play (ADR 0022). The iOS client shares the same core and comes to the App Store later; the waiting list is on logn.sh.

Architecture

A three-layer monorepo. Go + PostgreSQL for sync, auth and persistence; challenges live in a JSONB column with CHECK CONSTRAINTS that reject, at INSERT time, a challenge with no possible answer. Rust/Crux is the mind: model, rules, match engine, offline queue. Jetpack Compose and SwiftUI are dumb layers: they send events and receive a view model. The bridge is bincode over native FFI, with generated types.

Structural decisions are in docs/architecture/decisions/.

The content is not here

This repository has the app; it does not have the track. Statements, answer keys and explanations live elsewhere, and a clean clone brings up a backend with the right structure and nothing to play.

The split is by nature: schema, constraints, judging engine and screens are engineering, and they are all here. The curriculum is something else.

To run with your own content, write a migration that fills skill_nodes and challenges in backend/schema/migrations/. The constraints tell you right away if a challenge has no possible answer.

License

Code under the Apache License 2.0; see LICENSE. Use, modify and redistribute it, commercially too.

The LogN name, the symbol and the visual identity are not under that license. A derivative is welcome; it just needs another name and another look. Details in TRADEMARKS.md, and third-party attributions, including the IBM Plex fonts, which are OFL, in NOTICE.