multi-goal

Multi-objective optimization via Design of Experiments. Optimize several competing metrics at once — latency and cost and bundle size and coverage — and find the factor settings that best trade them off. A focused, standalone, host-agnostic plugin.

Extracted and extended from build-loop's single-metric optimize subsystem. multi-goal adds true multi-objective selection (scalarization, Derringer-Suich desirability, Pareto frontier) while keeping the same numpy-only DOE engine.

Why

Most "make it faster" work optimizes one number and silently regresses another. multi-goal measures every objective on every experimental run, fits which factors move which number, and selects the run that best satisfies all goals under explicit weights — no vibes, no single-metric tunnel vision.

Two modes

• DOE mode (default). Test up to 11 factors in a single experiment. 2–3 factors → full factorial; 4–7 → fractional factorial (8 runs); 8–11 → Plackett-Burman screening (12 runs). Fits main effects + interactions per objective.
• Autoresearch mode (fallback). One factor, greedy loop: hypothesize → measure → keep-if-better-else-revert.

Selection methods

| Method | What it does | Use when | |---|---|---| | scalarize | Best weighted sum of normalized objectives | You can express priorities as weights | | desirability | Derringer-Suich D (geometric mean of per-objective desirabilities) | You want each objective to clear a bar, not just average out | | pareto | The non-dominated trade-off set | You want to see all the trade-offs before committing |

Install

As a Claude Code plugin — this repo is its own single-plugin marketplace:

/plugin marketplace add tyroneross/multi-goal
/plugin install multi-goal@multi-goal

Then /multi-goal (guided flow), /doe (direct matrix), and /status are available. The host coding agent's LLM does the reasoning (hypotheses, factor confirmation); the scripts are deterministic and host-neutral.

As a Codex plugin — a .codex-plugin/plugin.json manifest ships alongside the Claude one; point Codex at the repo.

Standalone — the scripts run on their own:

uv run python scripts/doe.py detect 4          # which design for 4 factors
uv run pytest -q                                # 76 tests

Requirements: Python ≥3.10, numpy. Dev: pytest.

Quick start

# 1. which design for k factors
python3 scripts/doe.py detect 2
# 2. generate the matrix
python3 scripts/doe.py generate --factors '[{"name":"workers","low":2,"high":8},{"name":"batch","low":16,"high":64}]' --design auto --seed 1 > doe.json
# 3. run each row, measure every objective into results.jsonl, then:
python3 scripts/doe.py analyze --design doe.json --results results.jsonl \
  --objectives '{"objectives":[{"name":"latency","direction":"lower","weight":0.7},{"name":"cost","direction":"lower","weight":0.3}],"selection":"scalarize"}'

Full walkthrough and the method/math: docs/usage.md, docs/method.md.

Scripts

| Script | Role | |---|---| | scripts/doe.py | DOE matrix generation + multi-response effects analysis | | scripts/objectives.py | multi-objective core: scalarize, desirability, Pareto, baseline aggregate | | scripts/loop.py | single/few-variable autoresearch greedy loop | | scripts/suggest_factors.py | codebase scanner for factor candidates | | scripts/metric_runner.py | sampled metric / guard execution |

License

Apache-2.0 © Tyrone Ross, Jr. See LICENSE and NOTICE.