The workspace

DeepGarden Canopy

Ask a hard question. Get a checked answer. Canopy is the part of DeepGarden people actually open: type the question you would have emailed the data team, and get the finding, the view that suits it, and the record of where every number came from.

Request a demo See how it works

What people do inside Canopy

Three capabilities of one workspace, not three products. None of them appears on its own order line.

Don't let another question turn into a goose chase

01

Ask Canopy a question

Type the question the way you would say it out loud. Nothing to learn, nothing to file, and no waiting on someone else's queue.

02

Canopy creates a plan

The question is read against your business definitions, narrowed to exactly what you are cleared to see, and turned into a plan before anything runs.

03

Canopy works its magic

A validated read-only query executes against your sources. The numbers come back from your systems, deterministically — not from the model.

04

Canopy answers with citations

The finding, the view that suits it, and every figure re-checked against the rows that came back before any of it reaches you.

The longer version

A question becoming an answer

A question, the clarification it needed, the answer, and the query behind every figure.

Health plan · market intelligence

AskWhere is Aetna gaining or losing MA market share fastest in Ohio, and to whom?

Clarifying question

Two readings are available. Which did you mean?

Share of Medicare Advantage enrolled, by countyselected
Share of all Medicare eligibles, by county

Across Ohio counties, Aetna's MA share moved most in Cuyahoga, where it fell over the six months to June 2026, from to .

Calculation

Share shift is the difference between the two period endpoints, computed on MA enrolled rather than on eligibles.

share_shift = (aetna_enrolled_end / ma_enrolled_end) − (aetna_enrolled_start / ma_enrolled_start)

Endpoints are June 2026 and December 2025 enrollment
Plans are crosswalked to 2026 contracts, so consolidations are not read as growth
Counties where CMS suppressed enrollment are excluded from the county total and flagged below

The share went predominantly to Anthem, which gained over the same period in Cuyahoga.

What this cannot tell you

Two Ohio counties — Vinton and Noble — are excluded: CMS suppressed Aetna enrollment there at or below ten members in at least one endpoint, so a share shift cannot be computed. They are listed in the evidence rather than counted as zero.

Query evidence

Dataset: CMS MA enrollment by contract, plan and county
Rows: 10
Duration: 412 ms
Executed: 2026-07-14 09:22:07
Status: succeeded
Mode: read-only

View assembled from the result (pp)

Cuyahoga−2.4
Franklin−1.1
Hamilton−0.7
Summit−0.5
Montgomery+0.3
Lucas+0.5
Stark+0.8

Representative interface. The block structure, verified-value chips and evidence records mirror the product; the figures shown are illustrative rather than customer data.

Ask

Someone types the question they would otherwise have emailed the data team, in the words their business uses, across as many connected systems as it needs. When a question could mean two things — gross or net, the parent or the contract — Ask says which readings it can see and asks you to pick, because guessing produces a fluent answer to a question nobody asked.

What you do with it

Ask across several connected sources in one question
Follow up without restating the context
Get a clarifying question instead of a guess
Get a plain unknown when the data cannot answer

View

A trend arrives as a chart, a ranking as a sortable table, a single fact as a number with its evidence attached. Nobody picks a visualization; the shape of the answer follows the shape of the question. Anything worth returning to pins into a shared workspace with the live query underneath — unlike a dashboard estate, these come from questions people actually asked, produced fresh rather than stored waiting to go stale.

What you do with it

Charts, tables, comparisons and single figures, chosen automatically
Pin a result into a shared workspace
Re-run a pinned view against current data
Executive and analyst display modes on the same result

Evidence

A good analyst never hands you a number on its own. Click any hard figure and it opens its own record: the executed SQL, the dataset, the tables touched, the row count and the timestamp, plus the exact result cell it was read from. None of it is reconstructed afterwards — the record is written in the same run as the number, so a figure in a deck from three weeks ago can still say what it was standing on.

Before an answer is shown, a second pass compares every figure against the rows that came back. Values that reconcile are marked verified; anything that does not is flagged rather than published.