Relationship discovery and entity analysis across three related CSV tables.
Technically, anything is a vertical integration.
Experimental tools built on Polymodus. Savant is the first one ready to use.
Data investigation
Savant
Drop in CSV, Excel, JSON, or related tabular files. Savant profiles the data, connects fields across tables, groups recurring row patterns, surfaces anomalies, and keeps the source records one click away.
What it handles today
Measured on our development machine with fixed datasets and correctness checks for the relationships, profiles, time axes, and review candidates Savant is expected to find.
First useful profile while the complete worker analysis continues in the background.
Complete open and discovery with 6 relationships and 72,982 possible duplicate-entity pairs.
Worker and WASM ready after the first browser initialization.
How the benchmark is measured
Snapshot from 3 Oct 2026. per-metric median of 3 wall-clock runs from performance.now() in headless Chromium. The messy dataset contains five tables with duplicates, conflicts, temporal activity, and planted bursts; the scale dataset uses exact cross-table identifiers.
- Browser
- Chromium 151.0.7922.34
- CPU
- AMD Ryzen 7 9800X3D 8-Core Processor
- Memory
- 31 GiB
- Package
- aa149c7c2413…
Each published timing is the median of 3 runs. The benchmark fails if expected tables, rows, relationships, time axes, profiles, or review candidates disappear.
Start with raw files. Follow the evidence.
Explore unfamiliar data
Open full tables alongside column distributions, frequent values, missingness, and value clusters.
Connect related tables
See which fields appear to refer to one another, inspect the evidence, and resolve uncertain links.
Find patterns and anomalies
Group similar rows across several fields, surface standouts, and open the exact records behind them.
Resolve records and trace change
Review possible duplicate entities, inspect conflicts and provenance, then follow the same data through time.