Shipyard Solver Lab

A reproducible optimization workbench for packing shipyard blocks, validating placements, scoring runs, and generating the technical report trail.

Submission package ready OGC 2026 prep lane

Numbers below are sample-instance and public OGC example results. They prove the solve-validate-package loop. They are not OGC leaderboard scores.

--best sample score
--placed blocks
--block coverage
--compactness
--benchmark runs
--official example projection

One-Sentence Pitch

judge hook

For optimization competitors who need proof before official data opens, Shipyard Solver Lab turns the shipyard puzzle into a reproducible solve-validate-package loop.

Who

Builders entering OGC who need a working algorithm pipeline before the official scoring window starts.

Pain

Optimization entries fail when the solver, checker, benchmark, report, and submission zip are not reproducible.

Judge Signal

This project runs the public official checker, matches the public greedy example with a single-file candidate, improves robustness variants, and builds the candidate package.

First-Time Guide

start here

What this does

It tries ship-block placements, checks if they are valid, and keeps the best run.

How to use it

Read the score cards first, then check the yard layout and solution table.

What to trust

The current screen proves the pipeline on sample and public example data. It is not a leaderboard claim.

30-Second Review Path

review fast

1. See the loop

Confirm the product is a solve, validate, score, report, and package workflow.

2. Check proof

Read the official checker, portfolio candidate, and package cards before judging the layout.

3. Know the limit

The app proves readiness on sample and public example data, not a final leaderboard result.

Yard Layout

valid placement proof

Solution Table

generated from JSON
Block Yard Position Size Rotated

Claim boundary: this lab demonstrates the solver pipeline on toy data, a public OGC baseline example projection, an exact official-checker smoke test, a checker-validated public-example portfolio candidate, and deterministic public-example-derived robustness variants. It does not claim leaderboard rank, training-instance performance, or final submission readiness.