Lossless Base tick data & sandboxed backtesting, settled in USDC over x402.

1 · Decoupled, observational by design

Our pipeline is a read-only observer. It never participates in the trading loop, never places capital, and can be killed at any moment with zero impact on data integrity. It captures every meaningful tick on the WETH/USDC 0.05% pool and writes one immutable file per UTC day. Liveness guaranteed — a heartbeat row is written at least every 10 seconds even in dead-zone markets, so there are no silent gaps.

2 · The Pessimistic Multi-Collision Rule

Our backtester refuses to flatter itself. When a single price window touches both the take-profit and the stop-loss, the intrabar path is unknowable from one bar — so the engine always books the loss. This is the academically defensible default (it is what production-grade engines like Backtrader and vectorbt use under pessimistic flags) and it structurally eliminates the look-ahead bias that inflates naïve backtests.

Combined with pessimistic intrabar fills (a buy layer only fills when low ≤ L_k ≤ open), this means a reported edge has already survived the harshest reasonable accounting.

3 · Parametric Matrix Computation

The $25 backtest tier is not a single run — it is a native matrix engine that expands your strategy across every combination of parameters in one sandboxed job and ranks the survivors. Four engine properties make the numbers it returns defensible:

Because every cell runs the identical pessimistic accounting, the leaderboard ranks only edges that have already survived the harshest reasonable execution assumptions. The builder below assembles a valid request payload against the live blockquant.backtest_matrix.v2 schema — adjust a control and watch the JSON regenerate in real time.

copy{ }

Use add to stack a slider value into a sweep dimension — the live cell count is the product of all three arrays, capped at 1000 by the engine. POST the payload to /api/v1/matrix/backtest with your X-Payment proof; see The Agent Manifest for the full settlement flow.

4 · Lossless integrity — strings, not floats

Uniswap V3 stores √price as a uint160 (sqrtPriceX96), routinely ~25 digits — far beyond the 2⁵³ integer ceiling of a 64-bit float. Spreadsheet exporters silently corrupt it to 3.76e24. We never do. The value is captured as a string at the source and carried un-truncated through every boundary, so you can reconstruct price algebraically to full precision. Block references are EIP-55 normalized for unambiguous, checksum-safe logs.

5 · ?format=human — clean base-10 arrays for Excel & Pandas

Bots want the lossless integer; humans want a number they can chart. Append ?format=human to the tick-stream endpoint and we compute the clean base-10 price (USDC per ETH) on the fly from the canonical sqrtPriceX96 — price = (sqrtPriceX96 / 2⁹⁶)² × 10¹² — and return a flat JSON array that drops straight into a spreadsheet or a DataFrame:

pd.json_normalize(resp.json()["rows"]) yields a tidy, typed frame — no string-to-float coercion, no scientific-notation surprises, ready for resampling and indicator work.

6 · Pricing

All prices settle in USDC on Base Mainnet via the x402 protocol. See The Agent Manifest for machine-readable schemas and payment flow.