Soulhall Product handbook

Change language?

 Open Soulhall
Chapters

Run a research project

Specify the question, data, and assumptions, track the evaluation to completion, and review which conclusions it supports.

Research aims to answer a question you can verify. This walkthrough asks whether a strategy is still worth considering after trading costs increase, from assignment to review. It illustrates how to write a request; it is not research that was actually run.

Write the question to answer

Decide what evidence would change your mind before asking a tool to run. Specific comparison conditions are easier to review than “Find the best strategy.”

  1. Write the hypothesis to test, such as “This strategy still has positive net returns after trading costs increase.”
  2. Specify the strategy, instrument, period, and comparison, and which decision the result will inform.
  3. Assign it to an Embodiment in the Office’s Conversation and select Send. If using your own tool, make the same request there.

For example:

Use Edgewright to test whether this strategy still has positive net returns after trading costs increase. First list available, authorized workspace data and confirm the strategy version, instrument, and period. Keep data, strategy, period, and position settings unchanged; compare only original and increased costs. Show me the cost values for confirmation first. After evaluation finishes, deliver both evaluation IDs, output files, differences, and limitations. If data is insufficient or the engine refuses, explain why and ask before using another method.

This combines the question, comparison conditions, and deliverables. Replace “this strategy” with the actual name or ID.

Confirm which data you use

Check workspace data before deciding what you can study. Do not write a conclusion first and find data to fit it.

  1. Ask the Embodiment or tool to list available datasets and read the description of the one to use.
  2. Confirm source, instruments, period, and usage rights. Ask it to list gaps and mismatches with the question.
  3. If more data is needed, confirm acquisition and usage rights before asking the tool to upload or import it into the workspace.

You should be able to identify which data this evaluation used, not merely its filename.

Fix the comparison conditions

Keep other conditions fixed when comparing trading costs so you can attribute differences to the factor being tested.

  1. Ask the tool to list strategy version, execution method, starting capital, positions, currency, and cost assumptions.
  2. Confirm both evaluations use the same data, strategy, and period, changing only agreed costs.
  3. Retain both configurations and evaluation IDs for later review.

You now have comparable conditions. If the engine rejects an input, read the reason and fix it first. Before using another method, explicitly agree to that specific method and require results labeled “Not produced by Edgewright.” This is research practice, not a restriction your external tool automatically follows.

Confirm the evaluation really finished

A submission reply means only that work was submitted. Ask the tool to keep tracking the original operation.

  1. Retain the operation ID returned after submission.
  2. Ask the tool to check that ID’s state, distinguishing queued, running, succeeded, partially succeeded, rejected, failed, and cancelled.
  3. After success, read evaluation details and output files. For partial success, first check which parts completed and which step failed.

You receive a completed evaluation and its evidence, not just a “submitted” message.

Review before the next experiment

Check that this comparison answered the original question before changing strategies or extending the period.

  1. Compare both evaluations’ data, strategy, and settings to confirm only costs changed.
  2. Read results and output files. Ask for separate statements of “Supports the hypothesis,” “Does not support it,” and “Insufficient data.”
  3. If an Embodiment delivered a report, select Read in Results. Use Ask about it in the room for questions, naming the evaluation ID and issue.
  4. Write the conditions to change for the next study separately and retain this comparison’s evidence.

Completion means you can say what the evidence supports under these data and assumptions—not simply that a tool says “Research complete.” See Understand research results for a fuller checklist.