Marketlens vs PolyOrderbooks

Marketlens stores a full L2 snapshot about every 60 seconds and, between snapshots, every individual price change with a millisecond timestamp and its parent snapshot ID — so any instant reconstructs exactly by construction. Its public pages report 1,054,008 markets, 43.0B order book rows, 294M trades, and 229M candles, with recording complete (not rolling) since March 1, 2026.

PolyOrderbooks records a fixed 250ms interval grid: every order book row carries a wall-clock stamp, so querying by calendar time is the primary interface and there is no reconstruction step. The grid is coarser than an event stream but dense enough that final-minute moves in 5-minute markets are fully visible, and the archive covers markets long before March 2026.

Both are legitimate; they answer different workflows. Event-delta replay is the strongest tool for exact tick reconstruction; interval grids are the simplest for calendar-time research and settlement studies.

• Marketlens — REST at api.marketlens.trade/v1, an open-source pip SDK (pip install marketlens), an MCP server for Claude and Cursor, bulk Parquet exports, and an offline backtesting engine with realistic execution (latency, queue position, fees).
• Marketlens free tier — 180 req/min, 2M data rows one time, markets open in the last 7 days, one key, no card.
• Marketlens paid — Pro $39/mo (5B rows/mo, 5 keys) and Scale $199/mo (50B rows/mo, 25 keys); rows are metered per data row returned.
• PolyOrderbooks — REST-first 250ms books/prices/metrics, an official Python SDK on PyPI, an MCP server, CSV and JSON downloads, plus Backtest AI for plain-English strategy tests.
• PolyOrderbooks free Starter — 250ms L2 with 3 days of history, same 250ms resolution on every plan; paid windows to 30–120 days for crypto, 30 days for sports.

Marketlens is explicit that its archive is complete only from March 1, 2026 — anything before that date on its own account does not exist. That is seven-plus months of depth at the time of writing.

PolyOrderbooks records from well before that: crypto coverage reaches back to 2023-01-03 and the archive is the basis for published settlement research (for example 298 resolved BTC five-minute contracts and 17,880 one-second snapshots). If your study needs a year or more of history, that decides it.

Neither is a substitute for the official live API. For live state, use Polymarket's own endpoints for free.

Choose Marketlens when tick-exact replay of the period since March 2026 is your requirement and you want its SDK and backtest engine as the delivery mechanism. Its free tier and $39 entry price make it cheap to evaluate.

Choose PolyOrderbooks when your research needs the longer record — pre-2026 markets, sports, or the full crypto breadth — or when a strict 250ms calendar-time grid is the interface you want to write against.

Reproduce the recorded August 2026 BTC event — the ladder going 0.044 to 0.996 in one second on a 26-second family collapse — and the two approaches are both capable but feel different. Marketlens, with its millisecond deltas on top of ~60s snapshots, replays the exact tick sequence from any instant after March 1, 2026. PolyOrderbooks serves the same second straight off the 250ms grid, no reconstruction step, back to 2023.

The pre-March-2026 boundary is the decider for research: any study spanning the earlier archive, or any sports/threshold series outside Marketlens's window, simply has no tick-level source elsewhere. The published settlement research (for example 298 resolved BTC contracts, and the reprise studies) is built from material that predates Marketlens's guarantee, so a longer-record requirement pushes the 250ms grid by construction.

Where the two genuinely differ daily is cost shape: Marketlens meters rows at $39 and $199 tiers against 180 free requests and a 2M-row trial, which suits bursty live replay; PolyOrderbooks prices by time window with the same 250ms resolution on the free Starter plan, which suits calendar-bound series work.

Choose Marketlens when you need tick-exact determinism since March 2026 and its SDK-plus-backtest engine is your delivery mechanism; its free tier and $39 Pro are honestly priced for that. Choose PolyOrderbooks when the requirement is the longer record — pre-2026 crypto, the 700k+ sports universe, thresholds, and metrics — or when a strict calendar-time 250ms grid is the cleaner interface.

Neither replaces the official live API; both documentation pages say the same thing in different words, and this page agrees. Use the official endpoints for live state, and pick an archive by what your shortest reproducible study actually requires.

• Full provider matrix
• Crypto data
• Order book data
• Provider comparisons