Baibursai
Knowledge work

Research Analyst

Takes a question and a pile of source material and returns a cited brief.

Research Analyst is the heaviest agent in the catalogue and the one that replaces the most human hours. Give it a question and a set of documents, and it returns a structured brief with findings, supporting evidence, contradictions it found, and a citation against every claim.

Citations are not optional

Every claim in the output carries a pointer back to the specific passage that supports it. If the agent cannot cite a claim, it does not make the claim. This is the single most important property of a research tool and the one most commonly missing from them.

It reports disagreement rather than resolving it

When two sources conflict, the agent surfaces the conflict with both citations rather than silently picking one. Real source material contradicts itself constantly, and an analyst that smooths that over is worse than useless because it hides the thing you most needed to see.

Cost and scale

At $0.045 a run this is roughly ten times the price of the lighter agents, because it processes an order of magnitude more text. A run across forty documents that would take an analyst most of a morning costs four and a half cents. The economics are not close.

Where it struggles

Numerical reasoning across many tables is the weak spot. It reads tables accurately but is less reliable at arithmetic spanning several of them. If the answer depends on summing figures across documents, verify the arithmetic.

What it can do

  • Answer a question across many documents at once
  • Cite every claim to a specific passage
  • Surface contradictions between sources
  • Produce an executive summary and a detailed body
  • Identify gaps where the sources are silent
  • Rank sources by how directly they bear on the question
  • Extract and normalise dates, figures and named entities
  • Return structured findings, not just prose

Inputs and outputs

It takes

  • A research question
  • Source documents (PDF, DOCX, HTML, text)
  • Output format preference
  • Depth setting

It returns

{ "summary": ... "findings[]": ... "citations[]": ... "contradictions[]": ... "gaps[]": ... "source_ranking": ... }

Fit

Good for

  • Market and competitor research
  • Due diligence document review
  • Literature reviews
  • Policy and regulatory analysis

Not for

  • Live web research
  • Arithmetic-heavy financial modelling
  • Anything needing real-time data