See it in action
The story of a.deepcellfile.
From a spreadsheet without reasoning to four surfaces an agent can navigate — and back again from any one of them. About three minutes, or skip ahead with the tabs.
| Q1 2025 | Q2 2025 | Q3 2025 | Q4 2025 | |
|---|---|---|---|---|
| Revenue | $2.0M | $2.4M | $3.1M | $3.8M |
| COGS | $0.8M | $0.9M | $1.1M | $1.3M |
| Gross Margin | 60% | 63% | 65% | 66% |
| OpEx | $1.5M | $1.6M | $1.7M | $1.8M |
| Net Income | ($0.3M) | ($0.1M) | $0.3M | $0.7M |
through FY2029
| Data Center growth | +89% |
| Total revenue | 392.6 |
What you just saw
What's in a .deepcell file.
Chapter one
What the file records
01
Spreadsheets store answers. The thinking is lost.
Every model carries reasoning — why this growth rate, why this scenario, why this allocation. The file preserves none of it. Then the deck retypes the numbers and the memo retypes the deck, so the two places the reasoning is finally presented are also the two places it dies. Six months later the formulas remain and the why is gone.
02
Every draft of your thinking, kept as part of the file.
A rough estimate becomes a driver model becomes a unit-economics model — and your AI's contribution becomes the next version, not a separate document. The file knows how the logic evolved, so anyone (or any agent) opening it later can pick up where you left off.
03
A value is never a number. It's a coordinate.
$2M means nothing alone. $2M of Revenue, in Q1 2025, in the Budget scenario, for Dept A — that's a fact. .deepcell stores every value with its full context, so you can pivot, roll up, or filter without rebuilding the model.
04
The thesis behind the rating, not just the rating.
A BUY at $280 is a conclusion. Beneath it sits a thesis, the evidence supporting it, the assumption it depends on, the bear case, and the counter that refutes the bear. .deepcell stores all of it as a connected map of reasoning — every claim labeled, every argument you can follow end to end, every contradiction out in the open.
One file holds three different graphs. They answer different questions — and only the third one crosses all four surfaces.
Dependency
What is this number computed from?
Formulas, item to item. This is the Relationships view.
Argument
Why does this conclusion hold?
supports · refutes · depends_on · contradicts. This is the Reasoning view.
Citation
Where else does this appear?
Any deepcell: reference. The one that reaches the grid, the deck, the memo and the claims at once.
Chapter two
How the four connect
05
Four ways to read one number.
The grid computes it, the slide shows it, the thesis rests on it, the memo cites it — four views of one document, not four documents to keep in agreement. And the square has a floor: the grid points at nothing, because a cell doesn't cite. It is what gets cited.
06
One address, and the same one everywhere.
deepcell:cell/Total_Revenue[FY2027E] is the deck's binding, the memo's link, the claim's anchor and the URL — one string, sixteen kinds of target. Because everything has a name, an agent goes straight to it instead of searching for it.
07
Ask the number who's using it.
Forward links are free — they are written on the element. The useful direction is backwards, and it was impossible until one address existed: two reverse indices, one keyed on a render coordinate and one on a URI string, could never be joined. Now a single question — where else does this appear? — spans prose, slides and reasoning, and answers for all three.
08
Change it, and watch what moves.
The cell recomputes. The slide re-renders. The memo sentence re-renders, because it was never a typed numeral. And every claim resting on it raises its hand — derived on read, never stored. Ask backwards first, and you know who cites a number before you touch it.
What this replaces isn't a format. It's four files.
The model in one app, the deck in another, the memo in a third, and the argument nowhere at all. That set disagrees with itself by Friday, and nothing in it can tell you which part moved.
| Today | One .deepcell | |
|---|---|---|
| .xlsx | the model | the grid |
| .pptx | the deck, retyped from the model | a slide that binds to the model |
| .docx | the memo, retyped from the deck | prose that links to the model |
| — | the argument, in someone's head | a typed claim graph |
| RAG | retrieves fragments of all four | one file, one address space |