Run MMath Experiments Remotely

This page is the lightweight browser guide for the GitHub Actions runner behind the AbTweak restoration. It uses the same curated experiment vocabulary as the local harness, so you can trigger safe runs, reports, and trace workflows remotely and download the results as artifacts.

Backend GitHub Actions

Free remote trigger, hosted SBCL execution, logs, and artifacts.

Scope Curated only

Named `run`, `report`, `trace`, and preset experiment sets only.

Current focus Hanoi-4 support

The strongest solve-oriented line remains `isbm + weak-POS + stack + Left-Wedge`.

Updated March 22, 2026

This guide is build-synced from the main repository.

Run single experiment Run experiment set Open repo guide Back to project page

How It Works

1. Pick a workflow

Choose either the single-experiment workflow or a preset experiment-set workflow in GitHub Actions.

2. Select a curated name

Use only the predefined harness names listed below. No arbitrary shell or Lisp input is accepted.

3. Download artifacts

Each run publishes a compact summary plus plain output, JSON output where available, and trace artifacts for trace workflows.

Single Experiment Inputs

Use the Remote experiment workflow with:

kind = status | run | report | trace name = curated harness name, or "-" for status

Curated `run` cases

blocks-sussman-abtweak blocks-sussman-tweak hanoi2-abtweak hanoi3-abtweak hanoi4-5peg-isbm-weak-pos-lw robot2-abtweak stylistics-abtweak

Curated `report` names

benchmark-status hanoi2-historical hanoi3-historical hanoi4-historical hanoi4-solve-candidates hanoi4-score-sensitivity wide-domain-sweep

Curated `trace` names

hanoi3 hanoi4 hanoi4-isbm-weak-pos-lw hanoi4-legacy-1991

Preset Experiment Sets

`status-snapshot`

Runs the short status summary plus the benchmark-status report. Best for a quick “where are we now?” check.

`hanoi4-focused`

Runs the current narrow `hanoi-4` comparison set: solve candidates, historical-control report, and score-sensitivity report.

`publication-surface`

Runs the lower-Hanoi publication-facing reports and the benchmark-status report.

What You Get Back

Job summary

A short markdown summary appears directly in the GitHub Actions run view.

Machine-readable output

When the underlying harness supports it, the artifact bundle includes JSON alongside the plain run or report output.

Trace artifacts

Trace workflows include the copied trace directory so you can inspect frontier, score, and lineage outputs offline.