elizaos--eliza/packages/benchmarks/social-alpha/CLAUDE.md

Social-Alpha — Agent Guide

Trust marketplace benchmark that evaluates systems on real Discord crypto-chat data from the ElizaOS Trenches community. Measures four capabilities via the EXTRACT / RANK / DETECT / PROFIT suites and produces a composite Trust Marketplace Score (TMS). Registered in the suite registry as social_alpha.

Run

# Direct, from this directory — rule-based baseline, bundled smoke fixture
python -m benchmark.harness --data-dir fixtures/smoke-data --system baseline

# With the full Trenches Chat dataset (267k messages, requires download)
python -m benchmark.harness --data-dir trenches-chat-dataset/data --system baseline

# LLM-backed full system (requires provider keys)
python -m benchmark.harness --data-dir trenches-chat-dataset/data --system full --model groq/llama-3.3-70b-versatile

# Run a single suite
python -m benchmark.harness --data-dir fixtures/smoke-data --suite extract

# Through the suite orchestrator (resolves provider/model, stores results)
python -m benchmarks.orchestrator run --benchmarks social_alpha --provider <p> --model <m>

Available --system values: baseline (rule-based, no LLM), smart (improved heuristics), full (LLM extraction), oracle (perfect-knowledge upper bound), eliza-bridge (TypeScript agent via elizaOS bridge).

Smoke test (no API keys)

The harness falls back to fixtures/smoke-data automatically when the full dataset is absent. Running baseline against the fixture needs no keys:

python -m benchmark.harness --data-dir fixtures/smoke-data --system baseline

Test the harness

pip install -e ".[dev]"
pytest tests/ -v

Layout

Path	Role
benchmark/harness.py	CLI entrypoint (python -m benchmark.harness)
benchmark/suites/	Four scored suites: extract, rank, detect, profit
benchmark/systems/	System implementations: full_system, oracle, smart_baseline, token_registry
benchmark/protocol.py	Abstract SocialAlphaSystem interface + dataclasses
benchmark/ground_truth.py	Ground-truth generation and caching
fixtures/smoke-data/	Bundled minimal fixture for key-free runs
trenches-chat-dataset/	Full dataset (267k messages; fetched separately)
tests/	pytest suite

Notes

• Results write to the --output directory as benchmark_results_<system>.json (not committed).
• Composite TMS weights: EXTRACT 25%, RANK 30%, DETECT 25%, PROFIT 20%.
• Scored by _score_from_social_alpha_json in registry/scores.py; score is TMS / 100 (ratio, higher is better).
• Full dataset schema and download instructions: trenches-chat-dataset/README.md.

⛔ NON-NEGOTIABLE — evidence, trajectories & real end-to-end tests

The binding, repo-wide standard is AGENTS.md. Read it. Nothing in this package is done until it is proven done — a reviewer must confirm it works without reading the code, from the artifacts you attach. This applies to every feature, fix, refactor, and chore here. "Tests pass" is not proof; "CI is green" is not proof.

• Record AND read model trajectories. Capture the actual inputs and outputs of the model from a live LLM — not the deterministic proxy, not a mock: the prompt, the providers/context, the raw model output, every tool/action call, and the result. Then open the trajectory and review it by hand. A captured-but-unread trajectory is not evidence (packages/scenario-runner/bin/eliza-scenarios run <scenario> --report <out>).
• Real, full-featured E2E — no larp. Every feature ships detailed end-to-end tests that drive the real path end to end. Not the happy "front door" only: cover error paths, edge/empty/invalid input, concurrency, roles/permissions, and adversarial input. A test that asserts against a mock/stub/fixture standing in for the thing under test does not count. If the real model/device/chain/connector/account is hard to reach, make it reachable — that is the work, not an excuse to mock. If the existing tests here are shallow or mocked, fixing them is part of your change.
• Screenshots + logs at every phase, plus a complete walkthrough video/run-through of the entire feature or view, start to finish (bun run test:e2e:record).
• Manually review every artifact the change touches — never just the green check: client logs (console + network), server logs ([ClassName] …), the model trajectories in and out, before/after full-page screenshots, and the domain artifacts listed below for this package.
• No residuals. No shortcuts. The goal is not "done" — it is everything done. Clear every blocker by the hard path: build the real architecture, stand up the real model/device/service, actually test it. Never leave a TODO, a stub, a stepping-stone, or a "follow-up." When unsure, research thoroughly, weigh the options, and ship the best, highest-effort, production-ready version. Keep going until every possibility is exhausted.

Artifacts → attached inline in the PR (MP4 video, JPG screenshots, logs in <details>); attach each evidence type or explicitly mark it N/A with a reason — never leave it blank. If develop moved and changed behavior, re-capture evidence; stale proof is worse than none.

Capture & manually review for this package — benchmark / eval suite:

• A real-model run (not the mock/smoke fixture) producing the score-report JSON, with the numbers inspected and the provider/model recorded.
• The per-item trajectories the harness captured, spot-reviewed for correctness — a green harness run over mock fixtures is not a result.
• The provider matrix actually exercised, and the scoring math validated against a known case.
• Failure / timeout / partial-output handling in the harness itself.