about aoc-runner
Every submission is executed on one machine under identical constraints. Times are only comparable within this instance — a hardware or image change invalidates cross-era comparisons (each run records its environment).
the machine
| host | 200766e08896 |
| cpu | Intel(R) N100 × 4 |
| memory | 11 GB total (solutions get 512 MB) |
| platform | linux x64, node v22.23.5 |
| image | python:3.15-slim@sha256:65a65fc7b30b06cb6df4930fd4650697cb23f4c2698ab81fc87e4b031bc36928 |
solution contract
- Python 3.15, stdlib only — no network, no filesystem writes, 1 CPU, 512 MB.
- Read the entire puzzle input from stdin.
- Print part 1 on the first line, part 2 on the second (omit part 2 if unsolved → partial).
- Exit 0. Comparison is per-line, trimmed, exact.
- Hard wall limit 60s per attempt; output capped at 2 MB.
how timing works
- One submission runs at a time (serialized queue) — no two solutions compete for CPU.
- One untimed warmup run (populates bytecode/page caches), then 3 timed runs.
- The board records the median; the spread (min–max) is shown per row.
- Measured inside the sandbox around the
python solution.pyprocess: interpreter startup counts, container bootstrap does not. A no-op baseline is recorded on every runner boot. - Revisions are append-only: every submission (paste, upload or repo snapshot) is its own board entry, numbered per handle/day, with an author comment. Your fastest correct entry counts.
privacy / fairness notes
- Puzzle inputs and expected answers never leave the server — feedback shows only your own output and per-part correctness. This is also what blocks hardcoded answers.
- Repo submissions are frozen at the commit shown in the preview; later pushes never change an entry.
- Submissions from a GitHub repo may bundle helper
.pymodules (repo root + the day's directory, ≤50 files / 1 MB).