Explicit strategy specification
Research logic is converted into structured, deterministic parameters rather than executed as unrestricted conversational code.
Reproducible by design
Research Methodology
The principles used to turn a conversational strategy idea into auditable quantitative research.
The research pipeline
Every result follows the same deterministic path — no step is skippable, and each leaves evidence behind.
01
Describe
A plain-English strategy idea in your AI chat — rules, assets, period, costs.
02
Specify
Translated into a structured, deterministic specification — never executed as free-form code.
03
Validate
Fields, indicator warm-up, look-ahead risk, and data coverage are checked before anything runs.
04
Execute
Orders fill on the next tradable bar — the engine never trades on a price it has already seen.
05
Grade
40+ metrics plus an 11-factor robustness scorecard that never rounds up missing evidence.
06
Audit
A self-verifying bundle — hashes, assumptions, formulas — a third party can independently re-derive.
The principles behind every result
Nine commitments, grouped by what they protect — each one is enforced in code, not policy.
Deterministic by construction
Ideas become validated, structured specifications — never free-form code.Honest simulation labels
Synthetic options use model-based repricing and are never presented as historical option-chain fills.
Transparent assumptions
Each retained research record exposes the key assumptions behind the result, including reporting currency, benchmark, cost model, data range, strategy hash, engine version, and dataset version.
Proven, not promised
The engine is pinned against known-correct answers and published statistical methods.Independent validation suite
Every release runs an automated golden-reference test suite that pins the engine against known-correct results: next-tradable-bar execution (no look-ahead), the documented drawdown and cost formulas, FX conversion that never uses a future rate, deterministic dataset hashing, portfolio accounting, and Black-Scholes call-put parity. Result metrics are additionally pinned against independently hand-computed golden answers across daily to annual frequencies.
Grounded in the peer-reviewed literature
The significance and overfitting tests implement published methods, not house heuristics: the Probabilistic Sharpe Ratio, Deflated Sharpe Ratio, and Minimum Track Record Length (Bailey & Lopez de Prado, 2012–2014), and the Probability of Backtest Overfitting via combinatorially-symmetric cross-validation (Bailey, Borwein, Lopez de Prado & Zhu, 2014). Strategy-capacity estimates use a square-root market-impact law. Naming the source lets a reviewer check the maths.
Robustness before confidence
Parameter sweeps, Monte Carlo simulation, stress analysis, and rolling out-of-sample testing help expose fragile results. The Probabilistic and Deflated Sharpe Ratios (Bailey & Lopez de Prado) quantify whether a result survives correction for sample length and the number of strategies tested.
Auditable afterwards
Every result can be re-derived, challenged, and reviewed outside the original chat.Reproducibility metadata
Stored runs preserve a strategy hash, assumptions, engine version, and dataset version so results can be traced and reproduced.
Reviewable evidence trail
Results are designed to leave an audit trail: retained research, assumptions, charts, statistics, and PDF, Excel, or CSV exports can be reviewed outside the original chat workflow.
Stated limits of validation
Passing automated checks proves the engine computes what it claims — not that a strategy is economically sound. Integrity reports flag short samples, low trade counts, indicator warm-up, stale or missing prices, and potential overfitting. Material strategies require independent replication and review before any capital decision.
See it applied to a real result.
The public sample is an actual SPY 200-day trend backtest (2015–2025) rendered by the same engine — every metric, chart, robustness grade, and assumption on that page was produced exactly as described above.