Products · Deep research
Multi-angle research run live and cited in the body, rendered as the document you would actually circulate rather than a summary somebody has to rebuild. Concrete claims are checked against two independent sources before they land.
The trail
So the method is on the page rather than behind it. This is what runs between the question you type and the document you download, in the order it runs.
01
A question is decomposed into separate lines of enquiry - the state of play, the numbers, what is coming - because one search phrasing finds one corner of a subject.
2-8
angles per wave
02
Every angle dispatches concurrently rather than in turn, so a wave costs the wall clock of its slowest leg instead of the sum of all of them.
1 wave
not one after another
03
Each leg walks a chain of independent search providers and stops at the first that answers, so one provider being dark or rate-limited costs a hop rather than the finding.
4 deep
providers behind each leg
04
Angles overlap on purpose, and the overlap is collapsed on the canonical URL before anything is read - so a source found three ways is read once and counted once.
≤ 20
unique sources returned
05
Surviving sources enter one running ledger, and the claims in the document cite into it by number. The bibliography is not assembled at the end; it is what the writing points at.
[n]
cited in the body
Read one properly
Three research reads the platform produced end to end - a brand, a competitive set and a market. Page through any of them and the sources are in the body, not a bibliography bolted on the back. Then download the file it was rendered from.
Brand Deep-Dive
PDF · page 1 of 7
Switch deliverable
How deep
How wide the wave fans out, and how much of each source comes back to be read. A deeper run is not a longer document; it is a better-founded one.
A fact you need standing up before a meeting.
A narrow fan-out and snippet-level reading. Fast, and honest about being a first pass.
The default, and what most reports are run at.
A wider fan-out with full-page extracts pulled back, so claims are read in context rather than off a search result.
A read somebody will be held to.
The widest fan-out at the same full-page depth, up to the twenty-source ceiling, and the overflow is kept on disk for the agent to page back into.
Who needs it sourced
Nobody asks where a figure came from until the moment it matters, and by then the tab you found it in is closed. These are the desks where that moment arrives on a schedule.
The work is not finding the number, it is being able to stand behind it a quarter later when somebody builds a plan on top of it. A read whose sources are still attached can be re-run and compared; one assembled from browser tabs has to be rebuilt from memory.
A diligence read is graded on what it missed. What matters is the trail: which sources were consulted, what they actually said, and which claims were corroborated rather than repeated. An answer with no trail cannot be checked, so it cannot be relied on.
Committing a budget to a category means defending the read behind it to a finance team that was not in the market. The useful artefact is one where the evidence sits under each conclusion, in a document that can go into the pack rather than into a conversation.
Anything published carries the organisation's name, so an unattributed statistic is a liability with a delay on it. Being able to show the origin of every figure before it goes out is the difference between a briefing and an exposure.
Get started
Researched live, verified against independent sources, and delivered as the document rather than as a chat reply.
Keep exploring
The same research delivered in whichever format it belongs in.
The same research, voiced as a two-host episode.
A data-led page built on the research instead of a file.
Ongoing, sourced intelligence across your market.
How McLeuker's cited research compares to Manus.
From a research question to a finished deliverable.