Skip to content

Hackathon — Contributor Guide

The canonical hackathon page lives on the main site: openlithohub.com/hackathon. That page holds the live status, contract values (test-set tag, sample count, target EPE), prize specifics, and "Notify me" sign-up.

This page is for contributors who are writing a submission and need to know how the hackathon plugs into the OpenLithoHub toolchain.

How the hackathon hooks into the toolchain

The hackathon reuses the open-leaderboard pipeline. Nothing special — same CLI, same submission YAML, same CI verification. The only difference is a single track: field that routes your entry to the hackathon ranking instead of the open one.

# submissions/<your-handle>/<model>.yaml
track: hackathon-2026q3   # routes to the hackathon board
model_name: ...
code_url: ...
# ... regular BenchmarkResult fields ...

Frozen contract

The hackathon's test split, gates, and ranking rules are pinned in hackathon/2026q3.yaml. Once that file ships with status: open in a tagged release, the contract is immutable — adjustments after open invalidate prior entries.

The contract YAML is the source of truth for the live values rendered on openlithohub.com/hackathon; read the file directly if you want the exact tag, sample count, and target EPE.

Hard gates

The hackathon shares OpenLithoHub's manufacturability gates — these are the same gates documented in Benchmarks:

  • MRC violation rate must be 0.0 — check_mrc / check_curvilinear_mrc
  • DRC must pass — check_drc

If either gate fails, the submission is rejected before ranking. "I got SOTA EPE but my mask wouldn't tape out" is exactly the failure mode the project exists to prevent.

Ranking among gate-passing entries is by EPE mean (nm) ascending, with EPE max and PV-Band mean as tiebreakers (in that order).

Submission flow

  1. Build your model so it conforms to the LithographyModel interface — the BYOM Colab has the minimal example.
  2. Run the eval CLI locally against the frozen test tag (the tag value is on the main hackathon page and in hackathon/2026q3.yaml):
openlithohub eval --dataset <hackathon-test-tag> --model <yours>
  1. Open a PR adding submissions/<your-handle>/<model>.yaml with track: hackathon-2026q3. The Leaderboard Submissions guide covers the full schema and the submission PR label.
  2. The auto-leaderboard.yml workflow re-runs your model against the frozen test split. If your reported numbers match within tolerance and gates pass, the entry appears on the hackathon board within 24 hours.

Anti-cheat — what the CI checks

  • The model class must be pip-installable from code_url. Opaque binaries are rejected.
  • The CI re-runs the eval against the frozen tag. Numbers in the YAML must match within tolerance; mismatches close the submission.
  • If your training set provably contained the test split, the entry is removed. We rely on contributors to declare training data honestly and spot-check by sampling.

Eligibility

  • Anyone, anywhere, individual or team.
  • Code submitted under a permissive open-source licence (Apache-2.0, MIT, BSD).
  • Sponsors and core maintainers may submit, but are scored separately (hors concours) so the prize pool stays focused on the community.

Questions

  • GitHub Issues with the hackathon label.
  • #hackathon channel in Discord once the server launches — see the Community page.