Sourcing un-conflated US Equities MBP data for retail algorithmic engines: Architectural and licensing constraints

External question — Quantitative Finance Stack Exchange Author: Sush Original post: https://quant.stackexchange.com/questions/85751 License: CC BY-SA 4.0 — https://creativecommons.org/licenses/by-sa/4.0/ Adaptation: HTML converted to plain text; contact email addresses removed. I am building a headless, Java-based execution engine designed to calculate short-horizon alpha (specifically Order Flow Imbalance (OBI) and order book depth dynamics across 10 depth levels) for liquid US Equities (NYSE / NASDAQ). To accurately compute sub-second OBI and track liquidity sweeps, the engine requires a streaming, low-latency market depth feed delivered via direct sockets/APIs (e.g., WebSockets, raw TCP, or C++/Java SDKs) without desktop GUI dependencies. The Technical & Financial Bottleneck: Conflation in Retail Feeds: Standard retail broker APIs (such as IBKR TWS API) buffer and conflate Level 2 / MBP depth into 100ms–250ms snapshots. This periodic sampling drops intra-interval order additions, cancels, and executions, distorting micro-second volume delta and queue tracking. Institutional Non-Display Licensing: Direct, raw streaming TCP feeds (e.g., Databento, MayStreet) provide un-conflated L2/L3 data, but trigger exchange non-display fees and vendor platform base fees starting around $1,500–$ 3,000+/month. This creates a steep barrier for individual capital validation. Questions: Are there any market data vendors or clearing APIs offering raw, un-conflated MBP-10 depth for NYSE/NASDAQ equities under a Non-Professional schedule (targeting under $300/month)? From a quantitative research perspective, how do retail strategies typically account for or filter out the noise/lag introduced by 100ms conflated broker snapshots when modeling short-horizon order book imbalance? Is direct, un-conflated equity depth structurally locked behind institutional non-display licensing rules, making conflated snapshots an unavoidable architectural constraint for retail-tier US equity algorithms?