By late June 2026 the first generation of NOVAQ had reached its ceiling. One process and one analytical database served everything, and a heavy calculation could stop the whole application for over a minute.
The second generation grew from a lessons register: each live incident received a number and became a rule. The old system kept running while the new one grew beside it.
Fidelity first
Clean code was carried over verbatim and large engines were imported and given new data, so nothing could drift in translation. An early rewrite of the analytics had produced different numbers, which set the rule for everything after: the target is a bit-level copy with better data handling. Fidelity tests ran old and new code over the same data and required matching output.
The result
All 52 strategy providers passed native fidelity checks. Two days before the migration, a parity audit found seven critical gaps in automated protection, and all were closed. On 10 July the cutover completed on the third attempt: 254,425 historical trades were mirrored with parity checks, followed by a test trade through the whole chain.
What it gave the project
Services that fail separately, restart on their own and share one database made everything that followed possible: AI agents, parallel analytics and the audit swarm.
Limits
Parity at cutover covers what was tested. A later check found old code still pointing at frozen stores of the first generation, and a class test now fails on any new read of those stores.