The Evidence Base Matrix: One Question Across Many Documents

Some questions are really the same question asked many times. What is the termination clause in each of these forty contracts? What revenue figure does each of these portfolio companies report? What compliance position does each of these policy documents take? Asked in a normal chat, one document at a time, this is slow, tedious, and easy to do inconsistently, because the question drifts slightly each time you retype it.

The Evidence Base matrix is built for exactly this shape of work. Instead of one answer to one question, it runs the same question down a column of documents and lays the results out as a grid: documents as rows, questions as columns, one answer in each cell. It turns a repetitive manual task into something you can see all at once, compare across, and review systematically rather than one scattered chat at a time.

Why a grid beats a chat for this

A chat is linear. You ask, you get an answer, you ask again, and the results pile up in a conversation you have to scroll back through to compare. For a single question that is fine. For the same question across thirty documents, it is a recipe for missed rows and inconsistent phrasing, because each answer was produced in slight isolation from the others.

A grid makes the comparison the point. Every cell answers the same question against a different document, so the differences between rows are immediately visible: the contract that is missing a clause, the figure that is out of line with the rest, the document that takes a different position from its peers. The structure does the work of surfacing the outlier that a linear chat would let you scroll straight past.

Every cell is still grounded and reviewable

A grid of answers is only useful if each answer is trustworthy, so each cell is grounded the same way a single answer would be. The cell draws on the actual document behind its row, cites the specific passages it relied on, and can be opened to check the evidence rather than taken on faith. A grid that could not show its sources would just be a faster way to produce unverified claims, which is the opposite of the point.

Cells that fail to produce a reliable answer are flagged rather than filled with a confident guess, and they are not charged for. The grid is honest about where it could not answer, which matters, because a blank that admits it is blank is far safer than a cell that quietly invented something to avoid looking empty.

Where it fits in the workflow

The matrix sits on top of the same Evidence Base and source-integrity checks as the rest of Qonera, so the documents behind the grid are audited for staleness and contradiction before any cell runs. And the grid feeds into the same review and sign-off layer: a reviewer works through the cells, checks the flagged ones, and a named person approves before the analysis is used. Producing the grid quickly does not skip the step where a human stands behind it.

For investment research comparing figures across a portfolio, for consultancies checking assumptions across a document set, for any team asking one question across many files, the matrix turns a slow manual sweep into a structured, reviewable artifact. See how it fits the broader review process on the workflow page. The value is not just speed. It is that the comparison becomes something the team can see, check, and defend, instead of a stack of separate answers nobody can line up side by side.