(BAI® — LIVE PILOT EVIDENCE)
Arcade Star Constructions LLC
A scale-validation pilot that processed a multi-format tender corpus and separated quantified work from items requiring manual review.
- 1,557
- tender files ingestedAn 8.16 GB corpus spanning the issued tender package.
(BAI® — 01)
Recorded results
- 682 / 0
- CAD conversions / recorded read errors682 DWG files were converted to DXF; the run recorded no read errors.
- 1,466
- Bill lines across 15 sections979 machine-quantified; 487 routed to manual review with a stated reason.
- 46 / 47
- Door and window marks matchedItem-universe coverage against issued schedules—not quantity accuracy.
- 216 / 165
- Findings / RFIs216 documented findings supported a 165-item RFI schedule.
(BAI® — 02)
Evidence basis
Pilot-recorded internal evidence drawn from the build artefacts described in the Construction Take-off Suite dossier. It is not an independent third-party certification.
- The corpus included 2,325 PDF pages and 682 CAD files.
- Nine sampled section quantities were independently re-derived; all nine matched the persisted calculations.
- The issued client bill contained no numeric quantity cells, so an engine-versus-client quantity benchmark could not be computed.
(BAI® — 03)
Read with scope
These constraints travel with the metrics. They are part of the result, not fine print.
- 01
The quantity benchmark was correctly declared NOT_COMPARABLE because the issued bill was unquantified.
- 02
The 46-of-47 result measures mark coverage only. It must not be presented as quantity accuracy.
- 03
The bill was not fully automated: 979 of 1,466 lines were machine-quantified and 487 were explicitly assigned to manual review.
- 04
The nine matching re-derivations were an audit sample, not verification of every bill line.
- 05
The internal dossier identifies the contractor as Arcade Star Construction. The public-facing company name on this page follows the supplied website brief.