Sondara, a Local-First Drillhole Intelligence Workbench
Published:
Drillhole data is where a mineral project’s value is decided and where its data is most fragmented: a collar table, a survey table, assay intervals and geology logs that only make sense desurveyed together and looked at in three dimensions. Sondara does that reconstruction in the browser, keeps the data on the machine, and shows the estimate’s doubt next to its number.
Reconstruct locally
Surveyed borehole tubes, assay supports, an ore continuity shell, geology contacts and depth slicing, with orbit, section and plan views and a selection that follows the 3D scene. The source loader parses assay CSV depth and value columns into the browser-side support set; seed cases are explicit planning fixtures.
Estimate with the doubt attached
The local estimate action recomputes support-weighted values, uncertainty, cross-validation and the grade profile for the selected method. The spatial-conditioning engine, support-aware conditioning, covariance estimation and sequential simulation, is geocond, published on PyPI from its own repository on 2026-09-26; the product declares no package of its own.
Not a resource estimate
QA and QC, compositing, variogram fitting, full covariance solvers, categorical simulation and competent-person review are outside the public workbench; the learned method matrix and the bounded numerical server lane are recorded as planned, not shown as done. A public alpha with its plan as the implementation authority, on GitHub Pages plus a static mirror on the ML VPS; the lifecycle stays building.