⚡️ Satisficing vs Optimizing Agents in Peer‑to‑Peer Electricity CDAs

Agent‑based P2P electricity market simulator with a 5‑minute continuous double auction (CDA). We compare optimizing agents to lightweight satisficers, measuring both market welfare and per‑agent compute.

Summary

• In sufficiently thick markets (e.g., N ∈ {200, 500}), a simple limited‑search satisficer achieves near‑optimizer welfare (≈100–103% of optimizer’s normalized welfare in our runs) while using ≫ less per‑agent compute.

Repository Structure

• p2p/market/: order book and clearing (price‑time priority; maker‑price rule).
• p2p/agents/: agent implementations (optimizer; satisficers: τ‑band, K‑search/greedy; baselines: ZI‑C, learner).
• p2p/sim/: experiment runner, metrics, aggregation, overlays, plotting, theory checks.
• outputs/: generated CSVs and figures (created by scripts below).
• docs/: LaTeX paper (docs/paper.tex) and references.

Requirements

• Python 3.11
• Install dependencies: pip install -r requirements.txt
• Pinned (CI/repro): pip install -r requirements-lock.txt
• Optional (local paper build): a LaTeX distribution with pdflatex + bibtex.

Table of Contents

• Quick Start
• Reproduce Main Results
• Outputs
• Figure Gallery
• Paper
• Data & Integrity
• Citing
• License

Quick Start (smoke test)

• Run a tiny smoke: python -m p2p.sim.run --smoke --intervals 2 --agents 4
• Run tests and static checks: pytest -q && ruff check . && mypy p2p

Reproduce Main Results (v4)

• Full pipeline (experiments → aggregation → overlays → figures):
  • bash scripts/run_final_v4.sh
  • This archives any existing outputs/ to outputs_YYYYMMDD-HHMMSS/ and regenerates everything.
• From existing outputs only (overlays + figures):
  • bash scripts/repro_figs_v4.sh
  • Add --full to rerun the entire pipeline first: bash scripts/repro_figs_v4.sh --full

Outputs (where to look)

• Aggregated CSVs: outputs/analysis/**
• Final figures: outputs/analysis/figs_final/, including:
  • frontier_overlay_cda.png — normalized welfare Ŵ vs per‑agent ms.
  • ratio_to_optimizer.png — R_W = W_sat / W_opt vs per‑agent ms (0.90–0.98 band shaded).
  • connector_overlay.png — Ŵ axes with optimizer→satisficer connectors annotated by R_W.
  • frontier_and_ratio.png — two‑panel small multiples (planner‑bound and ratio views).

Data & Integrity

• Each experiment directory contains a manifest.json (config, seeds, environment, git hash).
• Overlays and figures are derived from these manifests; analysis keeps all runs and reports bootstrap CIs across seeds.

Citing

• If you use this code or results, please cite the accompanying paper:

  • Om Tailor, “Satisficing Agents Achieve Near‑Optimal Welfare with Orders‑of‑Magnitude Lower Compute in Peer‑to‑Peer Electricity Markets,” 2025. (See paper.pdf.)
  • A DOI will be added here once a release is archived (e.g., Zenodo).

  BibTeX (placeholder):

  @misc{tailor2025satisficing,
    author = {Tailor, Om},
    title = {Satisficing Agents Achieve Near-Optimal Welfare with Orders-of-Magnitude Lower Compute in Peer-to-Peer Electricity Markets},
    year = {2025},
  }

License

• MIT

Releases (for dataset/DOI)

• Package artifacts for a release: bash scripts/package_release_artifacts.sh
• Produces dist/ tarballs: manifests, analysis CSVs, and figures, plus SHA‑256 checksums.