How to run serious research in CharmIQ — designing briefs that produce useful output, choosing the right model for the job and working with data sources without drowning in them.

The quality of your research output is determined almost entirely by the quality of your research brief. This is the part most people skip or underinvest in and it's why they end up with generic output that requires heavy editing.

A brief designed by a domain expert produces dramatically better research than a brief written by a generalist. Which means: before you hand anything to a researcher Charm, have a smarter Charm design the brief.

That extra step — having the brief reviewed before execution — is the difference between a research output you can use and one you have to rebuild.

When working with structured data sources, the instinct is to paste raw data into the document and ask the Charm to analyze it. That's usually the wrong move.

 of your data source — what the schema looks like, what fields are available, what the export format contains — and ask it to generate the queries you need.

The Charm understands query logic natively. It's better at figuring out 

 than at processing a raw CSV dump you've pasted in.

Not all research tasks are the same. The model matters.

 is your default for collaborative research work — developing the brief, working through the analysis, iterating on findings.

 handles the weight when your source material is large. If you're working with a long document, a big reference file, or a context that would strain other models, use Gemini.

 (o3 or o4 Mini) is a different category entirely. It's designed for systematic, multi-step internet research — the kind of exhaustive competitive audit or market research task where you want every relevant source surfaced, not just the obvious ones.

Deep Research is slower and more expensive. Start with o4 Mini for most tasks; escalate to o3 when the depth genuinely warrants it. A useful calibration exercise: run a research task you've already done through Deep Research and see what it catches that your current workflow missed.

The same principles from working with Reference Materials apply here with extra force: don't load everything at once.

Research involves a lot of source material. The temptation is to attach all of it and let the Charm sort it out. That produces averaged output — responses that try to account for every source and land clearly on none.

This keeps the Charm's signal clean and your document readable.

 — The full pattern for sequencing researcher Charms, critic Charms, and execution Charms.

 — Designing the domain expert Charms that write your research briefs.

 — Managing Reference Materials dynamically so context stays clean.

How to run serious research in CharmIQ — designing briefs that produce useful output, picking the right model, and staying in control of your sources.

Task	Model
Iterative thinking, brief development	Claude Sonnet
Large documents, big context windows	Gemini Pro
Exhaustive multi-step internet research	OpenAI Deep Research

Research & Data

The Brief Is the Work

Show the Schema, Not the Data

Picking the Right Model for Research

Managing Research Context

Patterns

Next Steps