Research and Roadmap

The current areas of focus are listed below. This list will be continually updated and is presented in terms of overall priority:

Dual-state Cycle

1. [COMPLETE] Segregate TP handling by symbol instead of basket portfolio to optimize turnover, compound rate, and reduce slippage. Hypothesis exiting early and more frequently could decrease losses occurring due to enter/exit churn in sideways markets.

   • Demonstrated lower margin utilization in short term but eventually was on par with basket strategy

   • Did not show any meaningful improvement to equity curves under live trading conditions

   • Conclusion - this approach sacrificed 'runners' violating the rule of taking losses early (Tactical) but letting profits grow (Trend) in the system.

   • Potential future next steps - consider the same segregated TP handling by symbol but only when a given symbol runs to a significantly larger profit objective instead of looking constrain margin / avoid loss of unrealized gain.

2. Hurst processes for complementing entropy thresholds to optimize periods of flat/highly oscillatory conditions. Correlation study between market oscillation and system maximum adverse excursions per P&L epoch.

   1. [COMPLETE] step 1 start with a trivial 7 day high/low lookback to ensure slow/flat/choppy/whip-saw conditions remain risk-off. In the example below all symbols are disqualified for risk in the current conditions.

3. Optimize 3T core TRAIN/INFERENCE components to operate largely in memory with low level optimizations

4. Create framework for tagging deployed model version allowing for testing multiple models in parallel

Infrastructure, Operational and Knowledge Architecture

1. [IN PROGRESS] Migrate supporting components to GitHub with C4 diagrams and maintain via coding copilot ensemble

2. Migrate roadmap/research prioritization to GitHub issue tracker

3. Expand data layer to a more scalable solution (ex: Google Spanner)

4. Integrate additional DEX and high TPS blockchain technologies

5. Expose platform telemetry via external dashboards

Auditable Contributions

1. Automate contributions to Peter's Pence, operational architecture, bug bounties, repayment of initial funding and draw down prior to dispersal of any generated funding Holy See - Proof of Concept

   Current account Beneficiary: “Obolo di San Pietro” (Peter’s Pence) FinecoBank S.p.A. (Multi-currency bank account - EUR, USD, CHF, GBP) IBAN: IT 52 S 03015 03200 000003501166 BIC/SWIFT code FEBIITM1 or FEBIITM1XXX (if 11 characters are required) for SEPA countries FEBIITM2 or FEBIITM2XXX (if 11 characters are required) for non-SEPA countries For acknowledgement, https://trisagion.xyz will be stated in the reason for payment

2. Consider layer 1 blockchain for project tokenization, settlement layers, KYC use cases

3. Deploy $TTT token where during each P&L epoch, profits are rolled into market buys on the primary token pair. In addition to market buys, $TTT liquidity can be utilized to fund positions and increase compound rate of portfolio over time.

4. Voting for additional causes and platform changes

Distributed ML Clustering

Framework	Core ML Library	HPO Strategy	Primary Distributed Backend	Decentralized Cluster Support	Distributed Ensemble/Bagging Method
Auto-Sklearn	Scikit-learn	Bayesian Optimization, Meta-Learning	Dask	via Dask client	Parallel evaluation of models during HPO. Ensemble is built post-hoc on the head node from discovered models. 11
FLAML	LightGBM, XGBoost, etc.	Cost-Frugal Optimization (CFO)	Ray, Spark	via Ray init	Parallel HPO trials (n_concurrent_trials). Ensemble is built post-hoc. 4
TPOT	Scikit-learn	Genetic Programming	Dask	via use_dask=True	Parallel evaluation of pipelines in each generation. Final pipeline is not an ensemble by default. 5