Methodology & Benchmarks
Skeptical traders require verifiable technical proof, not marketing hyperbole. This document provides a transparent, auditable specification of our calculation engines, backtest data provenance, intrabar fill assumptions, statutory Indian fee models, and latency benchmarks.
1. Performance Benchmarks Specification
TradeSim cites distinct latency metrics across different layers of its architecture. To avoid confusion, each measurement is explicitly defined below with its scope and reference test harness.
Client-side Abstract Syntax Tree (AST) validation, parameter schema parsing, and local Sentinel tool serialization in browser runtimes.
Go-based vectorized backtest engine execution time for single-symbol multi-year daily bars (1,000+ bars with technical indicators and position state tracking).
In-memory order book state delta replay and intrabar micro-path traversal per bar sequence in the native backtest daemon.
2. Backtest Reliability & Data Provenance
A backtest is only as trustworthy as the market data feeding it and the fill assumptions applied. Here is our exact data treatment:
Historical Coverage & Corporate Actions
Historical market data spans NSE equity cash symbols (NIFTY 50, NIFTY Next 50, NIFTY 500) and major index derivatives (NIFTY, BANKNIFTY, FINNIFTY) from 2018 to present. Prices are adjusted for splits, reverse splits, bonus issues, and extraordinary dividends using exchange-standard adjustment factors.
Intrabar Fill Simulation (The Micro-Path Model)
Standard candle backtests suffer from lookahead bias by assuming stops or limits fill at the most favorable intrabar price. TradeSim enforces a deterministic synthetic Open → High → Low → Close (or Open → Low → High → Close depending on candle direction) micro-path.
- Bearish Candles: Simulated trajectory checks Open → High first; if stop-loss is triggered before target, the trade is marked stopped out regardless of where the bar eventually closes.
- Bullish Candles: Simulated trajectory checks Open → Low first, preventing phantom wins.
- Tick Depth Replay: Where Level-2/3 tick archives are loaded in the Replay Console, orders are filled against recorded bid/ask book liquidity.
3. Indian Statutory Transaction Costs (To the Exact Paisa)
A strategy that appears profitable on gross return frequently turns negative once real-world friction is applied. TradeSim models authentic Indian regulatory taxes and broker costs before calculating net PnL:
| Fee Component | Equity Intraday | Futures (Index / Stock) | Options (Index / Stock) |
|---|---|---|---|
| STT / CTT | 0.025% on Sell turn | 0.02% on Sell turn | 0.0625% on Sell premium (0.125% exercised) |
| GST (18%) | 18% on (Brokerage + Txn + SEBI) | 18% on (Brokerage + Txn + SEBI) | 18% on (Brokerage + Txn + SEBI) |
| Stamp Duty | 0.003% on Buy turn | 0.002% on Buy turn | 0.003% on Buy premium |
| Exchange Txn Fee | 0.00297% (NSE) | 0.00173% (NSE) | 0.03503% on Premium (NSE) |
| SEBI Turnover Fee | ₹10 / crore (0.0001%) | ₹10 / crore (0.0001%) | ₹10 / crore on Premium |
| Slippage Modeling | Configurable: Default 0.05% for liquid index / 0.15% for midcaps / bid-ask spread depth | ||
4. Deterministic AI Architecture & Refusal Invariants
We reject the misleading term “Zero Hallucination” as applied to entire systems, because generative AI models inherently risk misunderstandings or missing assumptions. Instead, TradeSim enforces an absolute architectural separation:
All numerical computations—including Black-Scholes pricing, Greeks (Delta, Gamma, Vega, Theta), payoff matrices, span margin estimates, and equity curves—are executed exclusively by native compiled code (Go and TypeScript). LLMs never compute mathematical results.
The Sentinel copilot interprets trader intent, selects appropriate deterministic tools via strict JSON-schema contracts, and formats insights. If inputs are illiquid or strike data is missing, Sentinel is hardcoded to refuse and flag the gap rather than fabricate answers.
5. Edge Ledger vs. Platform Security Audit Logs
To eliminate any confusion between trader-facing memory and administrative security logs:
Trader DNA / Edge Ledger
A trader-facing behavioral performance ledger. It stores your personal simulated trades, journal notes, emotional tilt patterns, and discipline scores. Its purpose is self-auditing and coaching within your workspace.
Platform Security Audit Trails
Administrative infrastructure logs capturing system-level security events (login attempts, password resets, data exports, account deletions). Maintained for security monitoring and incident response in compliance with data protection laws.
Questions regarding our methodology?
Contact our engineering and research team for benchmark reproduction guidelines.