Scoring versions and what changed

Summary

The checks and the way they add up to the dimension and quality scores are versioned. Whenever a formula changes, the scoring version goes up. The version is printed on every benchmark release, so you can always tell whether two numbers were calculated the same way.

Version 3, October 2026

• Quantities match geometry now counts towards Consistency. The check compares the quantities declared in the file with the object's own geometry. Its score is the share of values that agree, minus the share that is wrong or in the wrong unit, so those count twice. A file with 96% agreement and 2% gross errors scores 94.
• It is scored once at least 10 values could be compared, and weighs three quarters of a normal check.
• Files that declare no quantities, or fewer than 10 that could be compared, are not affected: for them the check stays informational. Missing quantities already show under Quantity coverage on the Information page.

How the check works is explained in The quantity check.

Version 2, September 2026

Version 2 calibrated the scores on the reference corpus of 4,138 unique models, so that a score means the same thing across very different files:

• Hygiene checks weigh less when passed. Basic checks that nearly every file passes count for a quarter of their weight when the file scores 95 or more on them, and full weight when it doesn't. Passing the basics should not lift a weak model; failing them should still hurt. These are: units defined, parse errors and header warnings, spatial containment, spatial hierarchy, spatial structure warnings, elements with geometry, geometry that can be processed, duplicate GlobalIds, negative areas or volumes, quantities on objects without geometry, units match their quantity, header metadata coherent, openings filled consistently, material hygiene and file topology integrity. In What we check most of them are marked Gate.
• The dimensions weigh differently in the quality score. Information counts double, Geometry and Consistency three quarters, Validity and Structure once. Reusable data is what makes a model valuable to the next person.
• The quality score is limited by the data. The quality score can be at most 35 + 0.7 × the Forensic BIM score™. A tidy model without usable data can't score high. The Scores page says when this limit applies (Limited by its data), and the scatter on the Benchmark page draws it as a dashed line.

Version 1 was the scoring before this calibration.

What a new version does

To your reports

A report is scored when it is shown, from the analysis results stored for the file. A report you open after a scoring change therefore shows the new scores, without uploading the file again. A standalone HTML download is a snapshot: it keeps the scores it had when you downloaded it.

To the benchmark

Benchmark releases are built offline with one scoring version. After a scoring change we build a new release with the new version, because percentiles are only meaningful when your file and the corpus are scored the same way. Until the new release is active, the Benchmark page shows This release was built with scoring v2; this dashboard uses v3. Rankings may be off until the benchmark is rebuilt. See Benchmark: peer groups and percentiles.

To the valuation

The scoring version applies to the quality checks and scores. The valuation has its own method, described in Valuation methodology.

Frequently asked questions