Adaptive Market Intelligence Engine · See the Regime. Measure the Risk. Wait for the Edge.
What it is
SARVAM-369 does not replace the other engines. It reads their output and
answers one question: is this setup worth acting on at all?
It is built to reach WAIT often. An engine that produces a
LONG or SHORT on every candle is not measuring anything.
The six families
S — Sentiment: news and events, weighted by source reliability, severity, novelty and time decay
A — Adaptive ML ensemble: pluggable slot, see limitations below
R — Regime & reversal: regime label and change-point risk from TRAMSE-369
V — Volatility & value-at-risk: CVaR and tail risk from MARP-RX 369
A — Alpha discovery: multi-timeframe momentum and mean reversion
M — Market microstructure: order-book imbalance, spread and liquidity
Weights are not fixed. The regime controller reshapes them: trending markets
favour momentum and discount mean reversion; ranging markets do the opposite;
shocks raise news and microstructure and cut technical weight.
How a decision is reached
The weighted family blend produces an alpha score. That score is then
discounted by five multipliers — regime compatibility, model calibration,
liquidity quality, data quality and the risk gate — before it becomes a
probability.
A direction is only taken when all of the following hold:
Directional probability is at least 62%
Expected median move exceeds the estimated cost and slippage floor
If any one fails, the answer is WAIT — and the panel lists which one.
What SARVAM-369 does not do
It does not predict prices. Output is a probability
distribution and an 80% expected-move interval, never a target price.
It does not invent direction. It can veto a LONG or SHORT
down to WAIT; it can never turn a WAIT into a trade.
The ML slot is empty by default. No XGBoost, LightGBM or
LSTM model ships with this build. When no model is attached, that family
reports itself unavailable and its weight is redistributed — the engine does
not pretend to have a model it does not have.
It has not been backtested. There is no walk-forward
validation behind these numbers yet. Until that exists, treat the
probabilities as a structured opinion, not a measured edge.
Honest status
This is research tooling under construction, not a proven
system. No claim is made that it is the best algorithm, or that it is
profitable. Any such claim would require independent backtesting,
walk-forward testing and published live paper validation — none of which has
been completed.
Educational analysis only. Not financial advice. Crypto is high risk and you
are responsible for your own decisions.