Every Jenvu signal goes through the same 7-stage pipeline. The first six stages are deterministic — pure TypeScript over candle data — so the same input always produces the same read. Only the final narration stage uses an LLM, and even there the model is constrained to describe what the deterministic stages already detected.
The 7 stages
- Multi-timeframe candle ingestion — 1H for context, 15M for execution.
- Swing detection and structure mapping — HH / HL / LH / LL.
- BOS (break of structure) and CHoCH (change of character) detection.
- Fair Value Gaps and Order Block identification.
- Liquidity pool mapping — PDH / PDL, equal highs / lows, session sweeps.
- Premium/Discount zoning + killzone overlay + weighted scoring.
- LLM narration synced with the chart markings from stages 2–6.
Why deterministic + LLM
Pure LLM signals hallucinate. Pure rule-based signals can't explain themselves. Jenvu's hybrid keeps the analysis grounded in measurable price action while letting the agent speak to you like a human analyst.
Note
Scoring weights are tuned per XAU cross — DXY correlation matters most on XAU/USD, Tokyo liquidity on XAU/JPY, LBMA fix flows on XAU/GBP and XAU/EUR.