Strategy Study Design

Design a Strategy Study

Strategy Study Design is where you define how Arizmic should test one exact saved Strategy version. You choose the market data, time window, execution assumptions, values to compare, and result detail to retain.

A Study does not change the saved Strategy. It records the version and settings used so the completed result remains understandable after the Strategy changes.

Select the Strategy

Open Studies -> Study Design and choose Strategy under Study Level. Select the saved Strategy and confirm its displayed version. If you arrived through Create Study in Strategy Library or Composer, Arizmic carries that selection into Study Design.

Enter a clear Study Name. Use Presets when you want to save or restore a reusable setup. A Preset stores design choices; it is not a completed Study and does not contain Results.

Choose the Study mode

Use Solo Run to check one fixed setup or establish a baseline. Use an optimization only when you have a deliberate set of values to compare.

Grid, Random, and Optuna

• Grid checks every declared combination. It is useful when each combination is intentional and the total remains manageable.
• Random samples from declared choices. Set a seed when you need the same sample again.
• Optuna allocates a trial budget through its selected supported sampler. Choose the objective, direction, sampler, maximum trials, seed, workers, retained count, and supported pruning options.

See Study Types, Optimization, and Walk-Forward for the complete method comparison, every Optuna sampler and pruner, the walk-forward variants, and guidance on when each choice is useful.

The estimate in Readiness is the authoritative candidate count. Assets, execution sets, folds, and other fan-out choices can change the total.

Bind market inputs

In Market Inputs, choose the Dataset and the portion the Study should use. Confirm its asset, data shape, available coverage, start, end, and session rules.

A Strategy with several named inputs shows each required input separately in Strategy Inputs. Bind every input to compatible data. When the Strategy uses a Signal output, Signal Sources shows the required Signal and source binding. These are dependencies of the saved Strategy, not additional trading rules added by the Study.

If the Dataset does not cover the requested window, shorten the window or choose another Dataset. Missing rows are not a neutral result.

Set execution assumptions

Use Execution Assumptions to define how candidates are simulated. Available controls depend on the Strategy and selected fidelity and can include the execution model, starting capital, fees, slippage, and sizing.

These values change what a result means. A candidate tested with fixed contracts, no fees, or bar-based fills is not interchangeable with one tested under different assumptions.

Walk-forward settings

When walk-forward is enabled, define the supported training, validation, step, purge, and embargo windows. Arizmic shows whether the selected data can form the requested folds. A fold without compatible coverage is not silently treated as valid.

Choose Study axes

Study Axes contains the values an optimization may vary. Axes can come from Strategy parameters, Signal parameters, supported bindings, or other declared controls.

Keep the search space deliberate:

1. Start with values that have a clear operational meaning.
2. Include a baseline when the control supports one.
3. Avoid ranges so broad that neighboring candidates answer different questions.
4. Check the final candidate or trial count before queueing.

For a Solo Run, keep the selected values fixed. A Solo Run produces an individual Run Output rather than a candidate leaderboard.

Validate, estimate, and queue

Review Readiness after every material change. It combines the Strategy, data, dependencies, execution path, Study axes, and requested output.

1. Resolve every blocking finding.
2. Select Validate where shown.
3. Select Estimate Study.
4. Review the candidate count, preparation requirement, fidelity, and resource estimate.
5. Accept the current confirmation when the estimate is reasonable.
6. Select Queue Study.

Changing the draft makes the previous estimate stale. Estimate again before queueing.

The top status area shows active work. Select it for current progress. System & Jobs -> Jobs owns detailed history, cancellation, retry, and recovery.

Troubleshooting

• No Strategy is available: save a Strategy version first.
• The Dataset is incompatible: choose data with the required schema, level, asset, and coverage, then prepare it when the selected path requires prepared data.
• A Signal source is unresolved: sync or repair the Signal and confirm its required inputs.
• The estimate is too large: reduce axes, assets, execution sets, folds, or the trial budget without changing the question you intended to test.
• Queue Study is disabled: resolve Readiness findings and create a current estimate.
• The Study fails: open the active status or System & Jobs -> Jobs, read the first failed phase, correct its owning input, and queue a new Study.

Next

When an optimization finishes, continue to Strategy Study Results. For a Solo Run, open its retained Run Output under Results.