Open beta · Every cell graded and linked
One question. Every company. All at once.
Companies down the side, questions across the top. Each cell is answered from a document, graded on how good that sourcing actually is, and linked so you can check it yourself in a minute. The cells it can't answer say so rather than guessing.
Companies
Dashed = a dossier exists. Anything else routes to EDGAR.
Questions
Each becomes a column.
A custom column searches the metric names and findings already read out of each company’s filing. It does not go and read the filing for you — where the answer isn’t already there, the cell comes back empty with the EDGAR route that would fill it. That is the honest behaviour and it is why the empty cells are the interesting ones.
The grid is a snapshot. The work is noticing when a cell changes.
Every figure above was true on the date of the filing it came from. The watchlist re-reads the same companies every Monday and tells you which cells moved.
The weekly brief.
Ten signals a week, every one carrying the state of its own sourcing.
The grade tells you what to check by hand
An AI grid that answers every cell with equal confidence is worse than no grid, because you cannot tell which answers to trust. Every cell here carries the same three-letter grade the brief uses.
- A readTaken off a filing someone opened. The link goes to that document. Spot-check it if the number matters; you won't usually need to.
- B derivedComputed or restated from a figure that was read — a ratio, a delta, a midpoint of a disclosed range. Check the arithmetic before it goes in a board pack.
- C unreadNobody here has opened this company's filings. No answer is offered. The cell hands you the EDGAR search that would find it.
That last one is the difference. A tool that fills the cell anyway is optimising for looking complete. The empty cell is the honest output, and it is why you can hand the rest of the grid to someone without a disclaimer attached.