What the tape can tell you.
Constrain execution with source trades. Keep source-backed quantities, inferred liquidity, and modeled expectations separate.
Return to the moment before a trade.
Understand what could fill. Account for what did.
Q044 follows trade-flow imbalance through real historical markets. Inspect the simulated trades and the account they left behind.
Shaded dates have no trade input. Account marks carry across gaps; the curve is not smoothed. Drawdown uses all saved observations.
The requested window was February–July 2026. The saved input spans July 8–28 with missing dates. This is a saved research result, not a new run of the current engine.
Three observed markets from the same run. Explore actual recorded prices and a selected pair of simulated fills.
YES price in cents per share. Gaps over six hours remain open. Markers show simulated fill prices; these can differ from the observed market price.
| SIDE | FILL TIME / UTC | FILL PRICE | LAST OBSERVED PRICE | PRICE AGE | FEE / USDC |
|---|---|---|---|---|---|
| Reading saved fill records… | |||||
Markets are selected by observation count and history span, not returns. Only the selected complete trade is marked. These examples do not attribute the shared account's PnL to three markets.
125,383 order intentions in the saved study.
Q044-RELAXED-V1 uses historical native trade observations and arrival-snapshot PML2 simulation. Each order is evaluated against a recovered arrival book; this is not the standard continuous-book replay. Market identity is retrospectively mapped.
Fees are fixed at zero. Valuation carries the last observed native trade across input gaps. Residual inventory remains; marked equity is not fully redeemed cash. No profitability or live-execution claim is made.
Small synthetic examples now live in Tutorials.
A trade print and an order book answer different questions. Keep their execution semantics distinct.
Constrain execution with source trades. Keep source-backed quantities, inferred liquidity, and modeled expectations separate.
Reconstruct snapshots, deltas, and trades. Walk visible depth and track order lifecycles, with explicit maker-queue assumptions.
Every replay connects a frozen input to an execution path, an account, and a traceable result.
Bind market identity, timestamps, and source hashes. Act only on information visible at the time.
Apply each model's order, latency, and capacity rules. Preserve missing-data states.
Reserve cash or inventory before a fill. Track fees, positions, and settlement in the shared ledger.
Bind receipts and hashes to the result. Leave valuation unknown when the evidence is incomplete.
Source-backed fills, inferred quantities, and modeled expectations are reported separately. Visible L2 depth does not establish a hypothetical order's actual exchange fill or its queue position.
Insufficient funds reject an order before liquidity is consumed. Settlement requires matching market identity, final payout, and cutoff evidence. A scheduled end time alone does not authorize redemption.
Event time and local receive time retain separate meanings. Future data cannot repair a past observation. Missing coverage stays DATA_NOT_READY; recovery must verify the original input identity.
Model definitions are read live. Q044 presents a saved historical research run with simulated execution. Synthetic tutorials have their own page. Full replays accept caller-supplied frozen inputs through the Python, CLI, or private API entrypoints.