Evidence Brief

Analysts, investors and operators who have to defend a conclusion in a meeting. If the first question you get is "where does this number come from", the brief is built for that question.

• 15–25 pages on one target, plus one or two layers of its supply chain
• 25–40 anchored claims; each one carries document, page, quote and confidence
• report.html renders from the graph, so a claim without a source cannot appear
• evidence.jsonl is the same content as data: query it instead of re-reading

1. Every edge in the knowledge graph stores its page reference at extraction time
2. Low-confidence edges are re-checked against the source text before they reach the report
3. Anything that cannot be anchored is either dropped or printed as an open question

A 30-minute scope call fixes the target and the question. One section is written as a free sample. Five business days later you get all three files and a 30-minute read-back.

$750 for the first three briefs, then $600–$1,200 depending on scope. Priced against the alternative: an expert call runs $700–$1,500 an hour.

1. How is this different from an AI report generator?

   A generator produces prose. This produces a graph whose edges keep their page references, so the report can be audited line by line and the data can be re-queried later.

2. What if the answer contradicts what you expected?

   The outcome field is written before the analysis starts. If the sources do not support it, the brief says so — a negative finding that is anchored is still a finding.

3. Can you cover a private corpus?

   Yes. Filings, permits, internal memos and interview notes all work as long as they are documents with pages.