Alpaca platform landscape 2026-09-26: beyond NautilusTrader and LEAN (two-family research + Tavily Research pro) - #322

Scope

• Lane: lane:shared, as evidence for the trading lane's backtesting-engine, execution-broker and market-data-reference rows.
• What: evidence only. It answers the user's 2026-09-26 question: "Is there only nautilus_trader and LEAN? What do other Alpaca users use for SOTA practice, historical data end to end?"

Answer: no. The alternatives are:

• LEAN with the official Lean.Brokerages.Alpaca plugin;
• Lumibot 4.6.1 (GPL-3.0);
• ml4t backtest/live/data (MIT);
• QuantRocket (commercial);
• vectorbt and PyBroker (Alpaca as a data source only);
• the official Alpaca MCP server, skills and CLI;
• several hosted or no-code partners.

On merit, nothing reviewed overturns NautilusTrader (destination) + LEAN (oracle) + alpaca-py (paper). The end-to-end gap is data: survivorship-free universe, delistings, dated identity. NautilusTrader has no upstream Alpaca code (RFC #3374 is open).

SOTA sources

• Alpaca: https://alpaca.markets/integrations (17 partners, read 2026-09-26), https://alpaca.markets/quantconnect
• QuantConnect: https://github.com/QuantConnect/Lean.Brokerages.Alpaca, https://www.quantconnect.com/docs/v2/lean-cli/live-trading/brokerages/alpaca
• Lumibot: https://github.com/Lumiwealth/lumibot (v4.6.1)
• ml4t: https://github.com/ml4t/backtest, https://github.com/ml4t/live, https://github.com/ml4t/data
• QuantRocket: https://www.quantrocket.com/alpaca
• vectorbt: https://github.com/polakowo/vectorbt
• PyBroker: https://github.com/edtechre/pybroker
• Alpaca MCP server: https://github.com/alpacahq/alpaca-mcp-server
• Alpaca skills: https://github.com/alpacahq/alpaca-skills (HEAD 39111abe)
• NautilusTrader: [RFC] Alpaca Markets Integration nautechsystems/nautilus_trader#3374
• Tavily Research: tavily-cli 0.1.8, tvly research run --model pro. Its 32 cited sources are kept in the artifact.

Files

evidence/artifacts/alpaca-platform-landscape-20260926/:

• README.md: the answer, the recommended trials and the catalog gaps.
• landscape.json: 30 ranked platforms, 8 catalog gaps, a facts refuter's corrections, and the raw returns from Claude Opus ×2 and GPT-6-Astra.
• tavily-research-pro.json: the Tavily Research pro report. This is the first use of the Research endpoint here. The key was held only in the kernel keyring, and UUID-shaped ids are hashed.

Evidence-class table

Commands

python3 kernel_keyring.py exec tavily_api_key TAVILY_API_KEY -- tvly research run --model pro --json --citation-format numbered -o <private> "<question>"
Workflow wf_9dec7824-293: research:alpaca-official, research:alpaca-github (claude-opus-5-5 max), research:alpaca-gpt6 (gpt-6-astra max, codex --search exec read-only), evaluate (opus max), refute-facts (sonnet max)
python3 scripts/validate.py   # passed (after hashing UUID-shaped ids)

Hand-off

The trading-lane coordinator (e7) owns the rows. They will fold this in with the local-model results. The recommended free trials are ml4t-backtest's lean preset and Lumibot 4.6.1 on the frozen SPY one_zero parity case, scored by compare.py.

🤖 Generated with Claude Code