Research workflow profiles

On this page

Use the closest profile for your task. Use these profiles with the tooling guide and contribution standards. Keep only the records the task needs. Skills, vendors, and manual methods are interchangeable implementations when they meet the agreed requirements. No agent subscription is required.

A small source lead can be a link and one sentence explaining its relevance, marked unchecked. It needs no full task form. For substantive work, the short research task record can help. Record planned checks before they run. Record actual checks afterward, including failures, omissions, and pending human review.

Source check#

Inputs: A specific claim or question, the proposed source, and the passage or identifier to check.

Outputs: A short note with source identity, passage or page, access date, and what the source supports, qualifies, or contradicts. If unavailable, retain it as an unchecked lead and state the missing material.

Checks: Compare the identifier, title, authors, and version with the real record. A person reads the relevant content and checks its context. A valid DOI alone does not establish support for a claim.

Evidence synthesis#

Inputs: A bounded question, intended review type, search boundary, eligibility rules, and relevant sources or databases.

Outputs: Dated queries and search limits, source records, selection reasons, and an evidence table. Include a synthesis with conflicting findings and limitations. Retain screening and extraction records appropriate to the chosen method.

Checks: A person checks source support and whether the search and selection fit the question. Reconcile records, reports, and studies before counting them. Apply the reviewer arrangements and reporting guidance required by the selected method. An empty search does not establish novelty.

Existing-data analysis#

Inputs: The question, accessible dataset version, reuse terms, data dictionary, and analysis plan or clearly labeled exploratory aim.

Outputs: Input manifest, reproducible scripts or notebook, environment and parameters, tables with defined units, figures, and limitations. Document cleaning decisions and deviations.

Checks: Check structure, missingness, transformations, and applicable model assumptions. Use independent calculations or known cases for material results. A person assesses method fit, interpretation, and uncertainty. Report untested assumptions; distinguish exploratory findings from planned tests.

Simulation or replication#

Inputs: A reference method or model, the question, input conditions, assumptions, and the result or behavior to compare.

Outputs: Editable model and code, versioned configuration, execution instructions, result files, and a comparison against the stated target. Preserve differences from the reference and failed attempts.

Checks: Check implementation against the method. Assess numerical stability, convergence, sensitivity, or conservation where applicable. A person reviews assumptions and comparison conditions. Distinguish a reproduced calculation, benchmark agreement, and agreement with observations. State which was examined.

Research writing#

Inputs: Intended audience and output type, the approved scope, source notes or results, contribution credit, and relevant venue requirements.

Outputs: Editable draft, references, linked evidence for material claims, figures with source files, limitations, and an assistance record.

Checks: A person checks meaning, source support, numerical consistency, attribution, and whether conclusions stay within the evidence. Technical claims retain their underlying checks. Follow the venue's applicable reporting and disclosure requirements. Record human acceptance separately from external peer review and publication.

View source on GitHub