What is a phase?

Focused research workflows

A phase is a focused workflow for a particular research task. Theo may invoke a phase automatically based on your request, or you can invoke one manually from the phase selector in the upper-right area of the interface.

Each phase is a research service developed by FirstPrinciples. A phase combines phase-specific tool calls, agents, subagents, databases, and structured workflows. This orchestration is a core value Theo adds beyond a general-purpose chat interface: the service can gather context, execute specialized research work, preserve its outputs, and connect them to the broader research process.

Automatic invocation

Describe the research goal in chat. Theo may answer directly, gather more context, recommend a phase, or invoke the appropriate phase(s).

Manual invocation

Ask for a phase by name, for example, “Run a Literature Review on recent surrogate modelling methods."

Alternatively, you can select a phase in the right-hand panel and enter a task-specific prompt.

Use manual invocation when you know the exact workflow you want. Otherwise, automatic invocation is recommended.

Available phases

Use the table below to understand possible workflows based on your goals.

Problem Identification

• Finds a clear research gap, limitation, or problem statement and checks candidate gaps against recent literature.

• Produces candidate research directions and a selected problem. A chosen direction can be carried directly into Literature Review.

Literature Review

• Collects, organizes, and synthesizes research papers to explain what is known, which methods exist, and where open problems remain.

• Can begin from a selected Problem Identification direction. Search depth can be adjusted by changing paper and web-source limits.

Question Formulation

• Turns a research direction into precise questions and hypotheses, with goals, assumptions, and evaluation criteria.

• Can build on Problem Identification and Literature Review artifacts so the formulation stays connected to earlier work.

Methods

• Transforms a research question into a structured method plan with a protocol, validation strategy, and next actions.

• Can plan paths based on simulation, symbolic regression, or symbolic computation and can use defined starting and ending points.

Python

• Generates Python code for implementation, data processing, simulation, visualization, and computational experiments.

• The generated code appears in the Python phase card, where it can be reviewed and run. It can use context and artifacts from earlier phases; the current workflow supports a single .py file and selected mounted files.

Math Computation

• Solves, derives, or verifies mathematical results through a natural-language interface and records the work in a persistent notebook.

• Theo can route math work to OpenAI mathematical reasoning, SageMath, or Mathematica, depending on the problem and available workspace configuration.

• SageMath is suited to exact symbolic algebra, calculus, differential equations, linear algebra, number theory, independent checks, and interactive mathematical plots.

• Mathematica can be requested for advanced symbolic or numerical computation. Name the preferred system in your prompt when you want to control the route.

Analysis of Results

• Interprets experimental or computational results from a required hypothesis and any available assumptions, model, dataset, or raw findings.

• Produces a structured report covering the result summary, conclusion, limitations, possible failures, new research opportunities, and generalization.

Citation Verification

• Checks whether cited papers are real and whether each paper supports, contradicts, or does not address the claim associated with it.

• Classifies references as exact matches, minor metadata hallucinations, or major hallucinations and can suggest corrected citation details.

• Optional checks can identify uncited claims and search the report’s other references for supporting evidence.

Summary

• Builds an editable session-wide summary from selected phase artifacts, including directions pursued and paths not taken.

Paper Radar

• Surfaces recent arXiv papers that closely match a research interest before or alongside a deeper literature review.

• Expands the query, combines searches, reranks candidates, and applies a relevance check. Results include a 0–10 topical relevance score and a short reason.

• You can adjust retrieval mode, arXiv categories, recency window, relevance threshold, and result limits. Author citation counts are supporting metadata and do not control discovery.