Craft a testable idea before chasing a result
A strategy begins as a hypothesis: when a condition appears, take a defined action under defined risk rules. Koval lets you craft that hypothesis as a visual graph. Market data, indicators, conditions, entries and exits become connected blocks you can inspect before running a test.
A return number is only one part of the answer. Inspect when a decision was made, what price could be used next, which costs were charged and what happened to open positions. Read the drawdown and the trades that created it. Keep the first result as a baseline before refining the rules.
How the current backtest model works
Koval's current fixed-cost model uses historical OHLCV: each candle's open, high, low, close and volume. These bars summarize a period. They do not reveal every trade or the order in which prices occurred.
| Assumption | What it means for your test |
|---|---|
| Decision timing | Market entries fill at the next bar's open after the signal. A completed candle cannot authorize a fill earlier in that candle. |
| Protection timing | Protection becomes active from the bar after the entry fill. This delay matters when interpreting stops and targets. |
| Execution costs | Configured spread, slippage and commission are applied by the fixed model. They are chosen assumptions, not historical account fees or measured fills. |
| Margin | The fixed model uses notional divided by leverage. It does not model exchange margin tiers or liquidation. |
| Ending policy | By default, an open backtest position is valued at the last close. Stopping a paper session closes its paper position. Match ending policies before comparing results. |
The current app does not model funding, historical fee tiers, partial fills, liquidation, latency or order-book queues. Spread and slippage are fixed assumptions, not a reconstruction of changing liquidity. Read the backtest accuracy guidefor the consequences of these limits.
Replay the evidence, not a fresh download
Market data can change after a run. New archived Koval backtests retain the normalized candles and settings used for execution, alongside recorded results. Their charts use those retained candles. That gives a particular experiment a stable reference.
Verify replay creates a separate comparison run. It requires matching software identity and complete evidence, and compares recorded outputs. A replay bundle can also be exported for offline verification with the matching software environment. The original experiment remains separate.
Older runs without archives remain readable but can be unverifiable. Missing or damaged evidence, incompatible software and incomplete audit records cannot establish verification. A matching replay demonstrates agreement under the recorded inputs and model; it does not prove exchange fill accuracy.
Put the strategy on the anvil: live paper testing
Historical data tells you how a frozen rule set behaved under historical conditions. Live paper trading lets that graph encounter newly closed market candles. Koval evaluates the graph and simulates orders and fills with paper capital. You can follow the session and review its saved results.
Live refers to the incoming market data. Capital and execution are simulated. The fixed-cost paper form uses explicit assumptions; older or legacy sessions can use different profiles. Differences in input windows, warmup, starting state, costs and ending policy can all change the result.
Freeze the graph for your observation period. Changing rules halfway through makes it harder to tell whether the original hypothesis held up. Treat missing or delayed data as a reason to inspect the experiment, not a reason to trust a clean-looking summary.
Implemented, in development and unavailable
Reviewed September 13, 2026 against the app implementation using engine and Backtrader packages 0.12.0. “Implemented” describes the development product; it does not mean public hosted access is open.
| Capability | Status and scope |
|---|---|
| Visual backtests and live paper | Implemented with the current fixed-cost model, supported feeds and simulated capital. |
| Archived backtest replay | Implemented for evidence-complete archived runs with matching software. Older runs may be unverifiable. |
| Paper-session diagnostics | In development: an initial internal release compares recorded sessions after stopping, shows differences and feed issues, and covers one timeframe. Full protective-order comparisons and automatic recovery remain pending. |
| Binance and WhiteBIT data | Supported public market data and Koval paper simulation, depending on the market and feed. Data support is not exchange execution support. |
| Exchange sandbox | App sandbox promotion is blocked pending validation. Binance futures testnet infrastructure exists upstream. WhiteBIT has no verified public sandbox for Koval and exchange execution is disabled. |
| Higher-fidelity execution | Not available in the current app: historical funding and fee evidence, partial fills, liquidation, latency and order-book reconstruction. |
How to make the next experiment more useful
- Write the hypothesis and freeze the first version of the strategy.
- Choose a market, timeframe and test window; record warmup and cost assumptions.
- Inspect trades, drawdown, open positions and costs alongside return.
- Verify the archived baseline where complete evidence is available.
- Use separate runs to test nearby parameters, higher costs and untouched periods. Automated robustness tooling is not implied.
- Observe new market data in paper mode without changing rules during the observation window.
- Document what you learned and what the model still cannot answer.
The MIT engine and separateGPL-3.0 Backtrader adapterare public. The hosted application is proprietary. Neither a profitable backtest nor a paper or replay match establishes future performance. Koval never places real-money orders.
Keep building your testing process
- How to backtest a trading strategy without code
- Design a readable visual crypto strategy
- What backtesting, paper and sandbox each validate
Bring your next idea to the forge.Join the Koval waitlistfor one email when the hosted app opens, with pricing and access details.