Write an experiment specification
Record the strategy source version, symbol and exchange, timeframe, chart type, test period, commission, slippage and order-size settings. Identify which inputs vary and which stay fixed. Check that each selected setting is exposed in the strategy and can be changed correctly.
For an illustrative moving-average strategy, fast lengths 5 through 30 in steps of 5 give 6 values; slow lengths 20 through 100 in steps of 10 give 9; stop values 1 through 5 give 5. The Cartesian grid contains 270 combinations per interval. This is a counting example, not a measured profitable strategy or runtime estimate.
Check constraints and interactions
Some combinations may violate the intended logic, such as a fast length that is not smaller than the slow length. The optimizer enumerates selected values; do not assume it understands your strategy’s economic constraints. Restrict the grid or define in advance how invalid combinations are excluded.
To inspect interaction, vary two inputs together while holding the others fixed. For example, compare every pair of two moving-average periods and two band multipliers. Record all four backtests. If the effect of changing one input differs depending on the other, that is interaction within this sample. It does not identify a universal best setting.
One-factor-at-a-time testing can still be a useful diagnostic. It does not automatically average results or prove inputs are independent; its limitation is that it can leave joint combinations unexplored.
Run a small grid before expanding
Pineify Supercharged uses a browser-based Cartesian grid and requires Advanced or Expert access. Confirm one or two collected reports against manual TradingView runs before starting the larger set. Keep the required chart session available.
Numeric ranges and discrete choices affect the total differently. A checkbox contributes two choices only when both true and false are selected; a fixed checkbox contributes one. Multiply by the number of selected intervals when comparing several intervals. Runtime depends on report calculation and browser state, so measure a small run instead of assuming a five-minute completion.
Review coverage before ranking
Check the planned combination count, successful rows, failures and no-trades cases. A timeout or missing metric must not be interpreted as a zero-return trial. Sort recorded metrics only after confirming that each row belongs to its stated input values.
Inspect trade count and drawdown alongside net profit or profit factor. Nearby settings with similar results are worth investigating, but a cluster alone is not proof of robustness. Refine only within the training sample, and keep a log of every grid you tried.
Freeze candidates and evaluate separately
Choose a small candidate set using criteria written before the search. Configure a separate test period manually, preserve costs and settings, and compare the resulting trade lists and reports. The current Pineify optimizer does not schedule automatic walk-forward runs.
If you change settings after viewing this separate period, it has become part of the research process. Further evaluation needs additional untouched data or a prospective test. Neither a larger grid nor a high-ranked historical result guarantees future performance.
Frequently asked questions
- How many inputs should I vary?
- There is no universal safe count. Consider the product of the choices, sample size, strategy logic and the number of research attempts. Two inputs with many values can require more work than several tightly constrained inputs.
- How long does a grid take?
- Measure a small run with your strategy and chart. There is no fixed minutes-per-thousand guarantee, and a faster search does not improve the quality of the underlying test.
- Does multi-parameter mean parallel?
- No. It means the experiment varies joint input combinations. Pineify applies combinations through the chart workflow and waits for reports; this guide does not promise simultaneous backtests.
- Can I export the experiment?
- The results view exports collected rows to CSV. Preserve the source version, requested dates, actual data coverage, costs and selected values alongside that file.