EigenScore
EigenScore
EigenScore is a competitive platform for quantitative problems: timed contests where you solve probability, statistics, optimization and market-related problems by writing Python, judged automatically against test data. Results feed an Elo-style rating, so your profile tracks how you stack up over time.
Create an account, open the contest on the Contests page and hit Register. When the contest window opens, the problems unlock: read, code, submit — as often as you like until the clock runs out. After the window closes, submissions are finalized, standings settle, and rated contests update your rating.
Yes — everything in the problem archive is open for practice at any time. Practice submissions never affect your rating.
Python (currently 3.13). The exact environment — interpreter version and the preinstalled scientific packages — is shown next to the editor on every problem page. Your program reads input from stdin and writes its answer to stdout.
Run executes your code against the public sample cases only and shows you everything: output, expected output, and stderr (including full tracebacks) — it is your debugging tool. Submit evaluates against the hidden test data and shows verdicts only, never outputs or error streams. That opacity is deliberate: hidden tests are what keep results meaningful.
AC accepted · WA wrong answer · TLE time limit exceeded · MLE memory limit exceeded · RE runtime error (your program crashed) · OLE output limit exceeded. Time and memory limits are stated on each problem (typically 2 seconds and 256 MB per test case).
No — submissions are judged in order in an isolated sandbox. Under load it can take a moment. If a submission seems stuck for a long time, reach out via the contact form.
Classic judge format: your program must produce the correct output for every hidden test case, either exactly or within the tolerance stated on the problem. Full points or nothing; the sample cases are public and can be run from the editor.
There is no single correct answer. Your output is first checked for validity, then scored by a problem-specific scoring function described on the problem page; better solutions earn more points and invalid outputs score zero. The sample panel shows an example output together with the score it would receive, so you can see how the scoring behaves before you submit. In contests, the score shown while the contest is running is provisional; the final score is computed on a separate held-out set when the contest is evaluated after it ends.
You are given data and must produce forecasts or estimates that are scored against a truth your program cannot see, by a scoring rule stated on the problem page. Prediction problems have no public samples (some provide a training file instead), so Run is disabled — Submit shows your score directly. As with optimization problems, contest scores are provisional until the contest is evaluated on a held-out set after it ends.
An Elo-style rating updated after every rated contest: it rises when you place above what your current rating predicts and falls when you place below it. Only rated contests move it — practice and unrated contests never do.
Rating bands carry titles (from Newbie up to Legend), shown as the color of your username across the site. New accounts are Unrated until their first rated contest is settled.
One account per person, your own work during contests, no probing of the judging infrastructure or hidden test data. The full rules live in the Terms of Service — the short version: compete honestly.
At the bottom of your own profile page. Deletion anonymizes your account: login, e-mail and submitted code are removed; contest results and rating history remain in anonymized form. Details in the privacy policy.
Via the contact form, ideally with the problem name and what you believe is wrong. Confirmed judging errors in contests are corrected and standings recomputed.