Data Analysis

Before you begin

• A CSV, workbook, or research source
• Defined question, time range, metrics, and output format
• Network or file access required by the research

1. Select Data Analysis, reference the data, and define each important field.
2. Ask for a research plan and review its dimensions and evidence requirements.
3. Confirm the plan; research_start must receive the returned, current confirmationToken.
4. Inspect each research_delegate dimension independently: queued, running, success, or failed.
5. Require contradictions, missing evidence, and assumptions to be separated before synthesis.
6. Export the report and spot-check source rows or links.

›_Copyable prompt

Analyze @survey.xlsx user retention. Propose a plan covering data quality, segments, time trends, anomalies, business explanations, and actions. Wait for my confirmation before starting. Cite a field, row, or source for every conclusion and label assumptions.

◇

Approval checkpointConfirmation tokens expire. After changing a plan, prepare and confirm it again instead of retrying an old analysis ID.

✓

Expected resultA report with method, evidence, limitations, conclusions, and next steps, plus independent researcher states.

Select Data Analysis, reference a local dataset, and state both the local-analysis and deep-research goals before sending.

After local analysis, explicitly confirm external research and choose normal, heavy, or quick depth.

Each dimension has independent evidence counts and running or failed states while the right panel aggregates sources.

Open one dimension to inspect activity, evidence, and the result separately instead of relying on aggregate progress.

The review stage validates claims and sources per dimension and preserves an independent review state.

Charts gathers expandable local-analysis visuals while keeping them in the same session as the research execution.

What is different in Wordless

Research is not an opaque tool card. Wordless separates the confirmation token, research dimensions, per-researcher sources, and final synthesis so a failed dimension can be rerun without repeating the entire analysis.

Scenario	Example dimensions	Verification
Retention diagnosis	Quality, cohorts, trends, anomalies, explanations, actions	Claims cite fields, rows, or sources
Market research	Size, competition, users, pricing, risks, evidence	Each researcher can be expanded
Resume after failure	Rerun only failed dimensions	Successful dimensions are preserved