ENSK
NOVAQ

Development Journal

How NOVAQ grew from an empty repository into a connected research and trading ecosystem in under six months.

Generations

Generation Period What it brought
Knowledge base May 2026 Rules first, then structured content
CML May to June 2026 Quantitative laboratory
NOVAQ v1 June to July 2026 First live engine, in one process
NovaQ From 10 July 2026 Federation of services around a shared database
AI and audit layers From late June 2026 Agents, experts, neural research and the audit swarm

Timeline

Date Event Type
14 May Vault founded, rules written before content Foundation
22 May Laboratory started with a research integrity layer Foundation
2 June Portfolio analysis found 12 macro bets Research
3 June Data and execution bridge to the trading platform Build
4 June Strategy book frozen and reproducible Build
6 June Risk layer completed, project named NOVAQ Build
8 June Strategy book armed in the live engine Build
28 June Semantic search and the first intraday AI agent AI
29 June First AI trader AI
3 July Lessons register and NovaQ plan Decision
10 July Migration to NovaQ Build
28 July 20-round adversarial loop and 10-model test Quality
16 Aug First AI expert with vault knowledge AI
31 Aug Clean corpus goes into production Build
3 Sept Second market added to neural research in simulation Build
5 Sept Audit swarm begins Quality
15 Sept Drawdown forensics separates market from execution Research
4 Oct DR/IDR claim register Research
4 Oct Independent verification of an AI brief, 253 fields AI
12 Oct Formal AI trader verdict Result
17 Oct Joint study of options positioning: size, not direction Research
18 Oct Joint study of open types and the first hour Research
19 Oct Audit of 341,992 data files Quality
20 Oct Internal cycle cut to 2.77 s Build
31 Oct Quant Sources imported into the vault Knowledge
3 Nov Inventory of about 700 studies and tests Research
3 Nov Second Quant Sources set and Dalton's live class added to the vault Knowledge

Decisions that shaped the system

  • Rules before content. The schema existed before the first note.
  • History is not rewritten. Schemas extend, and notes are not overwritten.
  • Nothing is dropped silently. Weak strategies stay on record with a reason.
  • Fidelity before invention. Ports reproduce earlier behaviour exactly.
  • Models learn only from backtested trades. Live ledgers do not enter calibration.
  • Causes are fixed in code. An alert is not a repair.