Why
Coaches and ICPC-style teams practise across several judges, and no single site shows the picture: who has solved this week's problems, who only did it after the editorial, who virtual-participated as a team. Squadforces watches the judges for you. You assign contests and problems, and it shows one matrix of members × problems with how each was solved.
It never runs code or hosts problems. It is an observer of Codeforces, AtCoder, Kilonova and CSES, so it works with the practice people already do on those sites.
Features
- Four judges and plain links. Codeforces (regular contests, gyms, EDU problems), AtCoder, Kilonova problem lists, CSES. Anything else can be added as a link and ticked off by hand.
- Solve classification that respects how you solved it. Live, virtual, upsolved (solved after taking part) or standalone practice; the most recent accepted submission decides, so re-doing a contest virtually updates it correctly.
- Team virtuals. Codeforces team submissions in gyms count for every member, exactly as Codeforces shows them.
- Paste links, get items. A bulk box detects the judge, contest-or-problem and gym/EDU variant from each URL; unknown links are kept, not rejected.
- Groups anyone can run. Users create their own groups (up to 5 each, with a member limit of their choice, up to 10) and manage them: assignments, hints, members. The admin can manage every group.
- Invite links, not open sign-up. Registration needs a link. The admin makes platform links (create an account, optionally straight into a group); group owners make group links (an existing user joins). Links expire, have a use limit, can be revoked, and are stored only as a hash. Nobody is added to a group without accepting a link (the admin can always add anyone).
- Hints and notes per problem. A group's owner (or the admin) has separate buttons to add hints, solutions and notes; members leave notes, each with a self-reported time-to-solve. Hints and notes are two switches, set per group.
- Resources per assignment. A third switch, "Allow resources", adds a collapsed Resources dropdown under the bulk add on each of the group's assignments: the owner (or admin) collects links or text in Markdown and can edit entries afterwards, and every member reads them.
- CSES solves that count. CSES has no API, so the member's own browser reads their solved tasks (a bookmarklet on cses.fi) and sends them to the app, which checks them against the public submission count of the account and stores them like any judge's submissions. No CSES password ever reaches the app.
- Stars on problems. A fourth switch, "Allow stars", gives the owner (or admin) a clickable star on every problem, grey until set and then yellow. Members see the starred problems as a read-only yellow star and nothing on the rest, so it works as "do these first".
- Leaderboards. Each group ranks its members by new problems solved from its assignments in the last 30 days, and the home page shows the most hardworking accounts overall (users see users, students see students, the admin both). A problem counts once, at its first accepted submission; re-solving it or having solved it earlier adds nothing. It is a single query on the submission store.
- Accounts to follow. The admin can create an account with no password: it can't be signed in to, but it can join groups and appears in the leaderboards, for following someone's activity.
- Profiles. A compact page: a chip per judge (handle, stored submissions, freshness), streaks and a heatmap of this year so far across Codeforces, AtCoder and Kilonova, with complete earlier years back to the first stored submission, on demand, and the last 20 submissions in a dropdown with links to the judge.
- Stays current by itself. Histories load when a handle is saved and refresh every two hours; only new submissions are fetched.
- Contest recommendations. A page that picks recent contests suited to a rating and division, graded by problem difficulty — Codeforces and AtCoder as two separately-ranked tabs, each with its own contest database. The member picker only lists accounts of your own kind linked to that judge, defaults to yourself when you qualify, and its "submissions" filter (not attempted / attempted, no AC / solved something) checks whichever member is picked.
- Accounts done properly. PBKDF2 password hashing, login throttling, signed sessions that end everywhere when a password changes.
How it works
graph LR
subgraph Judges
CF["Codeforces API"]
AC["AtCoder Problems + history"]
KN["Kilonova API"]
end
Store[("Submission store<br/>every submission, per user")]
Results["Results per item<br/>(what the matrix shows)"]
UI["Server-rendered UI<br/>FastAPI + Jinja2"]
CF -->|"only what's new"| Store
AC --> Store
KN --> Store
Store -->|"platform module<br/>derives results"| Results
Results --> UI
Store --> UI
A local copy of every submission. Judges are only asked for a user's submissions, once. They are mirrored into a table and afterwards refreshed incrementally: everything newer than the newest stored submission, plus anything still being judged. Every contest and problem status is then a database query, so adding the tenth assignment that mentions a contest costs no judge calls. A 15,000-submission history loads in seconds and refreshes in about one.
One module per judge. Everything specific to a judge (its links, how to fetch, how to display, how to derive results) lives in app/platforms/<judge>.py behind one Platform interface. Recognising where a link comes from is a single switch in app/problems.py. Supporting a new judge means one new file, one registry line and one if; templates never branch on the platform.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
