Notifications You must be signed in to change notification settings
Fork 0
Star 0
BranchesTags
Open more actions menu
Folders and files
NameName
Last commit message
Last commit date
Latest commit
History
3 Commits
3 Commits
docker
docker
scripts
scripts
trader_stack
trader_stack
.env.example
.env.example
.gitignore
.gitignore
LICENSE
LICENSE
README.md
README.md
fix.sh
fix.sh
oracle.py
oracle.py
requirements.txt
requirements.txt
run.py
run.py
setup.sh
setup.sh
widget.py
widget.py
Repository files navigation
A quantitative strategy validation pipeline. Find the opportune moment, or prove it does not exist.
Built by Mohamad Omar · USTechai · https://www.ustechai.com
In ancient Greek there were two words for time. Chronos is the steady tick of the clock, one second following another, indifferent to what happens inside it. Kairos is different. Kairos is the moment that matters, the narrow opening where timing itself becomes the edge. Markets run mostly on chronos: noise, drift, cost, and the slow grind of the session clock. Somewhere inside that noise, sometimes, there is kairos. This engine exists to find it, and to prove honestly when it is not there.
What it does
Kairos Engine is an end to end quantitative research platform for scalping signals in FX and metals markets. It ingests raw tick data, classifies market regime with a statistical model, evaluates an ensemble of modern time series foundation models, and puts every candidate strategy through a strict walk forward backtest against a broker cost model built from real measured spreads. Nothing graduates from this pipeline without clearing an explicit expectancy gate: profit net of every cost, evaluated out of sample, over a trade count large enough to mean something.
The platform has processed a full year of tick level XAUUSD data, fit and validated a four state Hidden Markov regime classifier, evaluated four separate time series foundation models as a forecast ensemble, and walk forward tested a validated strategy variant across 221 trades over 365 days with full per regime performance attribution. One variant passed every gate. Several did not. Both outcomes are the point.
A validation pipeline that only ever says yes is not a validation pipeline, it is a marketing document. The value of this engine is as much in what it rejects as in what it approves. Several approaches were built, tested, and killed in this repository before a single variant survived contact with real cost. This engine kills bad strategies before they cost real money, and that capability is the whole product.
Architecture
%%{init: {"theme": "base", "themeVariables": {"primaryColor": "#001621", "primaryBorderColor": "#FF4103", "primaryTextColor": "#FFFFFF", "lineColor": "#FF4103"}}}%% flowchart TD A["Dukascopy tick feed"] --> B["Regime classifier, 4 state HMM"] B --> C["Forecast ensemble, 4 TSFMs"] C --> D["Trade filters"] D --> E["Signal assembly"] E --> F["Walk forward backtest, session aware costs"] F --> G{"Expectancy gate"} G -->|PASS| H["Validated strategy"] G -->|FAIL| I["Rejected, no capital risked"]
classDef stage fill:#001621,stroke:#FF4103,color:#FFFFFF,stroke-width:2px; classDef gate fill:#FF4103,stroke:#FF4103,color:#FFFFFF,stroke-width:2px; class A,B,C,D,E,F,H,I stage; class G gate;
Loading
Every stage above is real, tested code, not a diagram of an intention. The feed layer pulls raw tick data from Dukascopy and reconstructs minute bars with genuine measured spread rather than a synthetic estimate. The regime classifier is a four state Gaussian Hidden Markov Model, refit on a rolling walk forward schedule so it never sees the future when labeling the past. The forecast ensemble runs four independent time series foundation models, Kronos, TimesFM, TiRex, and Chronos-2, and blends their output with a confidence weighted vote. The cost model prices every trade using session aware spread, slippage, commission, and swap, calibrated against live broker quotes rather than a textbook assumption, using hmmlearn for the regime fit and a pure Python Dukascopy client for data. Nothing reaches the expectancy gate without surviving all of it.
Validated findings
Approach tested Sample Result Verdict
Four model forecast ensemble, ten minute horizon 4,929 signals, 90 days, XAUUSD Predicted direction showed near zero correlation with realized direction FAIL, no exploitable edge
Mean reversion strategy, one minute bars, multiple stop widths 700+ trades per variant, 90 days Negative expectancy before any cost was applied, consistent across every stop width tested FAIL
Same strategy, fifteen minute and one hour bars 12 to 40 trades, 90 days Negative expectancy FAIL
Same strategy, five minute bars, tightest stop 221 trades, 365 days Positive expectancy before and after confirmed broker costs PASS
The validated variant cleared every gate under confirmed broker costs, with a gross to cost multiple of 16x against a required minimum of 1.5x. Gross expectancy came in at plus 246.91 pips per trade, and net expectancy after confirmed real broker spread and commission held at plus 222.91 pips per trade, across a full year and 221 trades, well past this project's own significance floor of 100 trades.
Per regime performance attribution on the validated variant, using confirmed broker costs:
Regime Trades Net expectancy per trade
Breakout 48 +1,411.6 pips
Trend 56 +243.2 pips
Chop 52 (51.1) pips
Cascade 63 (479.7) pips
Trend flips from a loser to a winner once real, confirmed costs replace an assumed cost figure. Cascade stays negative in every cost scenario tested and is the one regime this variant should not trade.
Quick start
./setup.sh --pretrain-hmm python scripts/replay_backtest.py --report-only python scripts/backtest_manual_bb_strategy.py --days 90 python scripts/cost_scenario_report.py
The setup script builds the virtual environment, clones the forecasting model repositories the ensemble depends on, and pretrains the regime classifier on historical bars. Each script documents its own assumptions in its module docstring. Read those before trusting any single number out of context.
Disclaimer
Every result in this repository is a rigorous backtest. A backtest is a historical simulation, not a live trading record, and no capital, real or otherwise, has been risked on any finding described here. Past performance in a backtest does not guarantee future results. Kairos Engine is research software, provided for educational and research purposes only. Nothing in this repository is financial advice, and nothing here should be read as a recommendation to buy, sell, or hold any instrument.
License
Kairos Engine is released under the MIT License. See LICENSE for full terms.
© 2026 Mohamad Omar · USTechai
About
Quantitative strategy validation pipeline HMM regimes, walk forward cost aware backtesting
www.ustechai.com
Topics
python
hmm
time-series
hmm-viterbi-algorithm
backtesting
quantitative-trading-tool
algorithmic-trading-quantitative
Resources
Readme
License
MIT license
Uh oh!
There was an error while loading. Please reload this page.
Activity
Stars
0 stars
Watchers
0 watching
Forks
0 forks
Report repository
Releases
No releases published
Packages 0
Uh oh!
There was an error while loading. Please reload this page.
Contributors
Uh oh!
There was an error while loading. Please reload this page.
Languages
Python 96.0%
Shell 4.0%