A short tour of the model: load data, read the findings, pull one apart, inspect it through a lens, and export the result.
Open the workspace and load data — paste CSV, TSV, JSON or plain text into the panel, drop a file, or pick a sample. Ellery detects the format, types every column, and immediately opens the Case File.
There is no chart to configure first. Analysis happens on load, on your device.
The Case File is a ranked list of findings in plain language: extremes, outliers, concentration, availability, distribution, and central tendency. The most important finding leads as a headline. When the data is calm, Ellery says so rather than inventing drama.
Findings are the product. Visualizations are just ways to inspect them.
Some findings are drawn from other findings rather than from the records directly — "concentration risk", say, which stands on the finding that one category leads the metric. Those are marked from N findings in the list, and their trace opens with a Rests on layer naming what they are built from. Each of those is a finding in its own right, so you can keep pulling until you reach the records.
Every finding is an object you can follow down. Click one and the Case File descends through the reasoning until it reaches fingerprinted ground:
fingerprint of the snapshot.From a trace, See it plotted spotlights the exact rows in any lens — the visualization becomes a microscope, not a destination.
The lens bar at the top of the workspace switches between views of the same Case File:
A traced finding stays lit as you move between lenses, so you can view the same evidence through each.
Any lens exports to MP4 or PNG for slides, to a self-contained HTML file, or to an interactive link that carries the live board. In PowerPoint, Google Slides, and Keynote, drop the MP4 in and set it to loop — the in-app Embed in slides guide has the exact steps.
There is no model in the truth path. A finding is a pure function of evidence and policy, so the same data under the same policy always yields the same findings — and the same snapshot fingerprint. That reproducibility is the whole point: trust attaches to a re-runnable procedure, not to whoever produced the chart.