Version control

What changed

Each corpus release will ship with a git-style diff of the underlying data — so downstream researchers can reproduce or roll back their benchmarks.

Preview · illustrative example
diff --corpusvN…vN+1
+812
Added fragments
−63
Removed fragments
~148
Re-annotated
+12
New sources
+lexicon"захисний шар арматури" — new canonical term, 42 occurrences
+sourceДБН В.1.2-7:2008 — seismic classification, 47 fragments
+speech+3.2 hours from Zakarpattia region
~nerreclassified 38 "material" spans → "structure" (finishing)
−sourceRemoved 12 fragments from a retracted paper
+lexicon"двокамерний склопакет" — brand-qualified variant added
~nertightened span boundaries on 61 "measurement" entries
+estimate+14 field-estimate samples from Dnipro projects
−corporateRemoved 51 proprietary lines by partner request
~speechRe-transcribed 7 audio samples after annotator feedback

Releases after v1.0 will produce this diff automatically from the dataset-versioning snapshot table. The example above is illustrative, not real release history.

Collaboration

Join BUDOVA

We are looking for researchers, construction professionals, and language specialists to participate in the project.

Supported by
Microsoft AI for Good Lab