Skip to content
Research

Research with a paper trail.
Every claim points to a source.

Pick a field, write the question, choose how deep to go. It reads primary sources, argues with its own findings and hands you four files you can audit.

Get early accessSee pricing
Method

Five steps, every time,
in the same order.

Every research run gets the same workflow, whatever the field.

  1. 01

    Scope the question

    Question, audience, date range, region and definitions are fixed before any searching starts.

  2. 02

    Gather primary sources

    Official data, filings, legislation and peer-reviewed work first — read, not skimmed from snippets.

  3. 03

    Cross-check evidence

    Each material claim is compared across independent sources before it enters the findings.

  4. 04

    Analyse and challenge

    Assumptions, contradictory evidence, data gaps and confidence are labelled, not smoothed over.

  5. 05

    Write the report

    Executive summary, findings, alternatives, limitations and the questions still open.

Choose the depth

  • FocusedFocus on the central question and strongest primary sources.
  • DeepInvestigate independent subquestions and cross-check key claims.
  • ExhaustiveExplore competing explanations, conduct a second critique pass and explicitly document unresolved uncertainty.

Instruction added to the brief for each depth, quoted from the app.

01 — Finance

Finance.
Filings first, every metric dated.

Company, macro and sector questions answered from filings and official series — with fiscal periods, currencies and accounting definitions lined up before anything is compared.

“For finance align fiscal periods, currencies and accounting definitions; distinguish observations, estimates and scenarios.”

What it reads

  • SEC EDGAR — company filings and structured financial facts
  • FRED — official economic time series
  • World Bank — country and development indicators
evidence.csvshape, not a finding
claim,evidence,source,type,period
C1,segment revenue,10-K (SEC),observation,FY [year]
C2,margin trend,10-Q (SEC),observation,Q[n] [year]
C3,rate path,FRED series,observation,[range]
C4,valuation range,model,scenario,—
Template · Finance

Company deep dive

Business model, financials, valuation and risks.

Research [company]: business model, revenue drivers, financial statements, competitive position, valuation scenarios and material risks. Use filings and date every financial metric.

Template · Finance

Macro and markets

Compare indicators, regimes and competing explanations.

Investigate [economic question] using official economic series, definitions, time ranges and alternative explanations. Deliver a data-backed brief and sources.

Template · Finance

Sector intelligence

Market structure, competitors and structural change.

Analyse [sector] in [region]: market structure, competitors, regulatory changes, unit economics and evidence-backed trends.

02 — Science

Science.
Primary papers, competing hypotheses.

Map a research front from the papers themselves: who measured what, with which method, and where the evidence still disagrees.

“Map current research on [topic], prioritising primary papers, reproducible methods and competing hypotheses.”

What it reads

  • OpenAlex — scholarly works, authors and citation networks
  • Semantic Scholar — paper discovery and citation graphs
report.mdshape, not a finding
## Research fronts
1. [front] — [method], [n] primary papers [C1][C2]
2. [front] — contested [C3] vs [C4]

## Competing hypotheses
H1 … supported by [C1]
H2 … supported by [C3], contradicted by [C4]

## Open questions
Template · Science

Scientific landscape

Research fronts, methods and open questions.

Map current research on [topic], prioritising primary papers, reproducible methods and competing hypotheses.

03 — Education

Education.
Study design before conclusions.

Evidence reviews, curricula and teaching briefs that separate study design, population and outcome — so an effect in one setting isn't sold as a rule.

“For education distinguish study design, population, outcomes and limitations.”

What it reads

  • OpenAlex — scholarly works and citation networks
  • Semantic Scholar — paper discovery
  • The web and your connected sources
evidence.csvshape, not a finding
claim,study,design,population,outcome,strength
C1,[study],RCT,[grade/age],[outcome],peer-reviewed
C2,[study],quasi-experimental,[setting],[outcome],peer-reviewed
C3,[study],observational,[setting],[outcome],limited
Template · Education

Systematic evidence review

A transparent search strategy and study comparison.

Review the evidence for [education question]. Define inclusion criteria, search strategy, populations and outcomes. Compare study quality and identify gaps.

Template · Education

Curriculum research

Learning objectives, sequencing and assessment.

Research and design a curriculum for [audience/topic]. Map learning objectives, prerequisites, evidence-based teaching methods, assessments and accessibility.

Template · Education

Learning science

Turn evidence into an actionable teaching brief.

Compare approaches to [learning challenge], separating effect sizes, study settings and practical tradeoffs.

04 — Policy

Policy.
Enacted rules apart from proposals.

Regulation read from the legislation and the regulator, with jurisdiction and effective dates on every line — and what is merely proposed kept in its own column.

“Research [policy question] for [jurisdiction]. Use official legislation and regulators, cite effective dates and separate proposals from enacted rules.”

What it reads

  • Official legislation and regulators, via web research
  • Your connected sources
sources.csvshape, not a finding
url,title,publisher,published,accessed,status,claim
[official url],[act],[legislature],[date],[date],enacted · effective [date],C1
[official url],[rule],[regulator],[date],[date],in force,C2
[official url],[bill],[legislature],[date],[date],proposed,C3
Template · Policy

Policy and regulation

Jurisdiction, effective dates and implementation.

Research [policy question] for [jurisdiction]. Use official legislation and regulators, cite effective dates and separate proposals from enacted rules.

05 — Business

Business.
Dated evidence for real decisions.

Competitive research from primary product evidence, and decision dossiers that test assumptions and name what is still uncertain.

“Inspect primary product evidence, document features and pricing dates, and identify underserved needs.”

What it reads

  • Product pages, docs and pricing, via web research
  • Your connected sources and project files
methods.mdshape, not a finding
# Method
Criteria: [cost, risk, time to value]
Options compared: [A], [B], [C]
Evidence: product pages and pricing, dated [date]

# Assumptions tested
…

# Remaining uncertainty
…
Template · Business

Competitive research

Product evidence, positioning and customer needs.

Compare [products/companies] for [audience]. Inspect primary product evidence, document features and pricing dates, and identify underserved needs.

Template · Business

Decision dossier

Evidence, options, risks and a decision framework.

Build a decision dossier for [decision]. Define criteria, gather evidence, compare alternatives, test assumptions and identify the remaining uncertainties.

Output

Four files,
each one auditable.

The guidance asks for the same handoff on every run. Open each to see its shape.

# Executive summary
Three findings, each with its confidence
and the sources behind it.

## Findings
1. … [C1] [C3]
2. … [C2]

## Alternatives
## Limitations
## Unanswered questions

Primary data,
read directly

SourceWhat it holdsAccess
SEC EDGARUS company filings and structured financial factsPublic API; identify the client and respect fair-access rules
FREDOfficial economic time seriesAPI key required for direct API integration
World BankCountry and development indicatorsPublic data API
OpenAlexScholarly works, authors and citation networksFree account and API key; daily included usage
Semantic ScholarPaper discovery and citation graphsAPI key and quota depend on endpoint and usage
Your models and the webReasoning, source discovery and research helpersUses your existing model configuration

Rules it follows
on every run

  • “Prioritise primary sources, official data and peer-reviewed work.”
  • “Read and compare evidence, not search snippets alone.”
  • “Every material factual claim needs a supporting source.”
  • “Label assumptions, contradictory evidence, data gaps and confidence.”
  • “Do not invent citations, figures, interviews, live market prices or APIs that were not used.”
  • Quoted from the guidance every research run receives.

Questions

Which fields are built in?
Ten templates across five fields: Finance (3), Education (3), Business (2), Science (1) and Policy (1). Each fills the composer with a brief you can rewrite — or skip the templates and ask anything.
Does it make up citations?
Its instructions forbid it: no invented citations, figures, interviews, live market prices or APIs. Every material claim needs a source in sources.csv, and evidence.csv shows what supports or contradicts it.
Which data sources need a key?
SEC EDGAR and the World Bank are public APIs. FRED, OpenAlex and Semantic Scholar need an API key for direct access. Web research uses your existing model configuration.
What do the depth settings change?
Focused stays on the central question and the strongest primary sources. Deep adds independent subquestions and cross-checks key claims. Exhaustive explores competing explanations, runs a second critique pass and documents unresolved uncertainty.
Which model does the research?
The one you pick in the composer — any of the chat models. See the models

Ask the hard question.
Get the paper trail with it.

Get early accessSee pricing