{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,30]],"date-time":"2026-07-30T13:04:34Z","timestamp":1785416674478,"version":"3.56.0"},"posted":{"date-parts":[[2026]]},"group-title":"SSRN","reference-count":0,"publisher":"Elsevier BV","license":[{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"unspecified","delay-in-days":0,"URL":"https:\/\/www.uspto.gov\/ip-policy\/copyright-policy\/copyright-basics"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"abstract":"<jats:p>Construction AI is starved for supervised data because annotation requires the same scarce expertise the models are meant to encode. We present the design, guarantees, and verification of an instrumented takeoff environment \u2014 the measurement canvas of a production Division-9 (finishes) estimating system \u2014 in which the geometry an estimator draws to measure a job, the finish condition they assign to price it, and the corrections they make to bid it are captured as complete supervised training rows at zero marginal annotation cost. Each row aligns render-invariant normalized geometry, a professional classification label, accepted quantities, the CAD linework and PDF text beneath the region, a correction trail, and machine-readable provenance flags. A two-tier capture design (provisional rows on autosave, certified rows at the commit gate) banks work even from sessions that never reach commitment. The architecture is code-complete and verified by 312 passing backend tests, including pixel-level geometry-invariance regressions. We are explicit about scale: at first drafting (2026-07-04) the captured corpus was approximately zero rows; after the first week of production capture (2026-07-11) the instrumented counters report 681 rows (23 certified, 658 provisional) across 4 projects \u2014 the flywheel is built and beginning to spin, but these are early counts, not a dataset claim. The contributions of this paper are the instrumentation thesis, the system design, and its tested guarantees; corpus-scale results are future work. Prior fine-tuning results from the same research program (e.g., a 31B LoRA reaching 13.5% median error predicting held-out bid totals) derive from a separate document corpus and are cited here only as motivation for the data-centric approach.<\/jats:p>","DOI":"10.2139\/ssrn.7118624","type":"posted-content","created":{"date-parts":[[2026,7,30]],"date-time":"2026-07-30T12:35:16Z","timestamp":1785414916000},"source":"Crossref","is-referenced-by-count":0,"title":["Markup-as-Label: An Instrumented Takeoff Environment that Converts Estimating Work into Machine-Learning Supervision"],"prefix":"10.2139","author":[{"ORCID":"https:\/\/orcid.org\/0009-0000-9771-4920","authenticated-orcid":true,"given":"Michael","family":"Edlin","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"78","container-title":[],"original-title":[],"deposited":{"date-parts":[[2026,7,30]],"date-time":"2026-07-30T12:35:17Z","timestamp":1785414917000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.ssrn.com\/abstract=7118624"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":0,"URL":"https:\/\/doi.org\/10.2139\/ssrn.7118624","relation":{},"subject":[],"published":{"date-parts":[[2026]]},"subtype":"preprint"}}