Pick a subject on the left.
Click a node to walk to it · drag to pan ·
dashed edges were inferred, solid ones were stated
Review copy · RyanAI LLC · synthetic data only
1,009,300 records and 1,239,467 connections across twenty-five businesses, generated with known ground truth and sharing nothing between them. Search it, open any subject, read the brief. Every timing on this page was measured on your machine just now, not quoted from ours.
What this page is. The corpus is real output from the generator that ships with the product, at a scale nobody would hand-check. The brief is rebuilt here in JavaScript from the same graph, so you can poke it without installing anything.
What it is not. These milliseconds are your browser's, on a copy of the data held in memory. They are not the engine's numbers. The engine's numbers are below, and every one of them is reproducible from a command in the repository.
One laptop, single process, live PostgreSQL 16. Not this page.
| Measurement | Result | Conditions |
|---|---|---|
| Source indexed | 15,699,911 rows / 2,124 MB | PostgreSQL 16, one laptop |
| Graph produced | 900,000 nodes / 800,000 edges | 911 MB on disk |
| Rows summarised, not copied | 15,000,000 in 3 s | aggregated in the database |
| Re-index, nothing changed | 0.12 s | watermark per table |
| Peak memory, whole import | 278 MB | streaming cursors |
| Largest graph built | 4,400,000 nodes | 17.4 min, 3.9 GB on disk |
| Memory at 250k vs 4M rows | 190 MB vs 194 MB | flat, by design |
| Throughput cost of scale | 8.2% per doubling | index depth, a slope not a wall |
| Reading 4,000,000 rows | 43 MB | 657 MB before the fix |
| Brief | 0.4 ms median | on the 900,000 node graph |
| Recall | 63 ms median | on the 900,000 node graph |