The experiment that changed
how we approached validation.
One of Dhanvi's early strategy-generation experiments produced approximately 24% return in historical backtesting. Rather than treating that result as proof of future profitability, it became a reason to build a substantially more rigorous research and validation architecture.
Could Dhanvi turn an idea
into a testable strategy?
This early experiment explored whether Dhanvi could move beyond market analysis and produce systematic trading logic that could be represented as machine-executable rules and evaluated against historical data.
From research to measurable behavior.
The experiment established the initial foundation for Dhanvi's computational strategy lifecycle.
Generate and evaluate systematic trading strategies from structured inputs.
Convert trading logic into machine-executable rules without ambiguity.
Run strategies against historical market data under reproducible conditions.
Simulate entries and exits using a deterministic backtesting engine.
Measure strategy behavior and performance numerically across test periods.
Compare experimental strategies and identify candidates for deeper research.
One strategy generated during this early research phase produced approximately 24% return in historical backtesting. The result was interesting enough to justify deeper investigation — but not sufficient to establish that the strategy would perform similarly in unseen or live markets.
The referenced ~24% result was produced in historical backtesting during an early Dhanvi research experiment. It does not represent live investment performance, returns earned using real client capital, or expected future performance. Historical and simulated results are subject to limitations including overfitting, market regime dependence, transaction costs, data quality, execution assumptions and other factors. Past or simulated performance does not guarantee future results.
24% wasn't the conclusion.
It was the beginning.
A strong historical result raises more questions than it answers.
These questions pushed Dhanvi toward a more rigorous research architecture. Rather than publishing an unverified return as a product feature, the team re-architected validation to actively challenge and test strategy fragility.
From backtest result
to research system.
Explore the eight verification stages developed to validate subsequent research hypotheses.
Specialized agents contribute different forms of market and company intelligence.
Specialized agents contribute different forms of market and company intelligence.
Redefining the validation paradigm.
How our internal validation requirements transformed after Experiment 003.
The goal changed from finding an impressive backtest to finding strategies that survive attempts to invalidate them.
A historical return is not enough.
Standard research vulnerabilities that explain why backtested metrics do not equate to live viability.
A strategy may fit historical noise rather than a persistent relationship.
Behavior observed in one environment may not persist in another.
Fees, spreads and slippage can materially alter simulated results.
Historical data and preprocessing choices can affect outcomes.
Backtest fills may differ from achievable live execution.
A small or concentrated sample can produce misleading conclusions.
What we know / What we are not claiming
Because surviving metadata for this early milestone is limited, we state boundaries explicitly.
- ~24% approximate historical backtest return
- Produced during an early Dhanvi strategy experiment
- Generated inside Dhanvi's experimental research environment
- Used historical market data
- Helped motivate more rigorous validation architecture
- Live-account return
- Return on investor capital
- Guaranteed future performance
- Proof of strategy robustness
- Proof of live profitability
Different experiments.
Different questions.
Each published study documents a different phase and question in Dhanvi's development.
- 5 agents
- 459 opportunities evaluated
- 33 trades selected
- +1.89% paper result
- RELIANCE.NS
- 72 daily bars
- 2 trades
- -0.88% backtest result
- Historical strategy research
- Early research milestone
- Limited historical metadata
- ~24% backtest return
Dhanvi publishes both positive and negative experimental results. The purpose of this research log is to document how the system evolves, not to present a portfolio performance history.
More experiments will be published here.
As Dhanvi evolves, this research log will document new system tests, strategy experiments, validation results and failures.