This section tracks the publicly visible history of Parameter Golf. It is intentionally separate from the much richer local experiment history already visible in this repository’s own logs and commits.

Because the public record is still thin, the goal here is not to pretend there is already a rich leaderboard culture with dozens of disclosed methods. Instead, this section tries to do three things well:

  1. record the hard public facts that actually exist
  2. separate observed archetypes from likely future strategy families
  3. explain how the challenge is already evolving conceptually from “tiny model” thinking toward artifact-aware co-design

What is publicly visible so far

As of 2026-03-19, the upstream public repository exposes only a very small number of concrete run records:

  • one main-leaderboard baseline under the 10-minute / 16 MB track
  • one notable unlimited-compute non-record run that stays under the same artifact cap

That is enough to establish some early reference points, but not enough to support strong claims about which strategy family will dominate the challenge.

Pages in this section

What this section is careful about

This section avoids a few common mistakes:

  • treating a baseline as if it already defines the full frontier
  • inferring hidden implementation details that the public record does not disclose
  • confusing challenge-implied strategy families with strategies that have already been publicly validated
  • reading one unlimited-compute run as if it automatically predicts the main leaderboard

How to use these pages

Read them in this order if you want a compact orientation:

  1. Public runs for the factual record
  2. Submission archetypes for the currently visible patterns
  3. Likely strategy families for the lanes the rules seem to favor
  4. Conceptual evolution for the bigger picture

4 items under this folder.