The QAI research method

How QAI research is built

Every QAI research surface — the finance boards, company dossiers and fund profiles, and the Insights market maps — is researched, argued and assembled with QAI's own proprietary corpus and Research Framework, run by the QAI Financial Harness over two Anthropic Claude frontier models, Fable 5 and Opus 5. Every load-bearing claim is cited and checked against its primary source through a four-gate quality pass, and every thesis names the observation that would prove it wrong.

The method produces two different things. A finance board argues a position — bull, base and bear, with a named falsifier. An Insights surface maps a market — the AI Stack atlas places companies across 11 layers from energy to applications, and each sector radar sources every company row it shows.

Last reviewed . The method is described here; the underlying corpus is refreshed continuously and every figure carries its own vintage on the page that renders it.

The QAI stack

What makes it QAI's, not a model wrapper — our own proprietary data, research framework, and orchestration engine.

QAI proprietary data

The corpus

A proprietary research corpus QAI builds and maintains — fundamentals from primary filings, a mapped value chain, estimated private valuations, sourced company rows behind every market map, and a vintage stamp on every figure, refreshed by the daily worker feed.

QAI Research Framework

The method

QAI's own analyst method — argued bull / base / bear with explicit assumptions and a named falsifier on every call, every load-bearing claim cited and verified against its primary source through a four-gate quality pass.

QAI Financial Harness

Orchestration engine

The engine that runs the pipeline end to end — routing the models, verifying every claim against its source, and keeping the whole atlas current.

The models

The frontier reasoning layer QAI runs its data and framework through — Anthropic's Claude models.

Fable 5

Claude frontier model

The deepest reasoning — argues the bull, base and bear theses, stress-tests the moats and dependencies, and sizes the risk on every name.

Opus 5

Claude frontier model

Breadth across the whole map — sweeps the primary sources, structures the fundamentals, and drafts the first pass of every dossier, board and market map.

How Insights is built

QAI Insights is a peer to QAI Finance, not a section of it. A board argues a position and carries a book; an Insights surface maps a market and says where the white space is. Same corpus, same models, same evidence discipline — a different question.

The AI-Stack atlas

The value chain, end to end

An 11-layer value chain from energy to applications. A company appears once per layer and category it plays in, sized by market cap and wired by who supplies whom. Each placement carries the company's role — a conviction and a one-line stack thesis — not its position; the bull and the bear live on the company's own dossier, written fresher and at more length.

The market radars

One map per sector

A market map of one sector, segment by segment. Every company row carries its headquarters, what it does, its traction and a source link. A segment carries its count and, where a credible published figure exists, a third-party market estimate with the year, scope, source and caveat that make it readable.

The landscape read

Published, not generated live

Per segment: whether it is saturated, contested or open, a verdict, the white space, the incumbents that matter and the entry risks. It is produced at research time and published as part of the corpus — so the reading is reviewable and gated like any other claim we make, rather than generated unseen on each request.

Each surface is one published corpus document, read by key — research, then corpus, then publish. Nothing reaches a reader that has not been through the same gate as the rest, and a segment with no credible published market figure says so rather than borrowing one. Because scopes differ per segment, those figures are context for a segment, never a series to add up.

What it builds

One method, five surfaces. The same models and the same discipline produce every part of the atlas.

Research boards

Argued bull / base / bear theses over a sector — explicit assumptions, named falsifiers, and cited evidence.

Company dossiers

Financials, SWOT, moats, supplier–customer ecosystems and a valuation range for every public and private name.

Fund profiles

Holdings, strategy and performance for the funds, families and figures whose flows move the market.

The AI Stack

The 11-layer value chain from energy to apps — every company placed, sized by market cap, and wired by who supplies whom.

Market radars

Sector maps of who is building what — every company sourced, every segment sized where a credible figure exists, and a read on where the white space is.

The declaration

This is AI-generated research — and we say so plainly. The models do the work; the method keeps it honest.

  • Every load-bearing claim is cited and verifiable.
  • Theses are falsifiable — each names what would prove it wrong.
  • Figures are checked against primary sources, not model memory.
  • A market map states where no credible figure exists rather than inventing one.

The finance surfaces are analyst work product for a qualified professional. Not financial advice — always verify against the primary source.