Revised and Withdrawn Figures

What changed

An earlier run of the analysis measured a median of 80.8 on one cohort. The metric was re-derived on a corrected, larger sample and the median moved. The figure that stands now is shown beside this row with its sample size, read straight from the canonical dataset. Anyone holding the earlier number should use the current one. A public post quoting the earlier number is not corroboration for the current one, and should not be treated as a second source.

Recorded 2026-07-26.

What changed

This was flagged for review because it was shown on the benchmark page but was not carried in the canonical machine readable dataset, so an automated check had nothing to compare it against. The derivation has since been traced: it was first computed on 2026-04-07 as a billable hour ratio with a median of 96.7 on 654 companies, then re-run on a larger cohort of 963 companies, which moved the median to 97.1 and made the denominator explicit. It is now carried in the canonical dataset with its definition and sample size, so it is verifiable there. The figure did not change as a result of this review. One thing worth keeping straight: this is not the same measure as industry utilization, which divides by all paid hours including drive time, admin, and training. Always state the denominator when quoting it, because without it the number reads implausible.

Recorded 2026-07-26, review closed 2026-07-26.

What changed

We could not establish where this figure came from, and it is not something Level measures. The underlying records carry employees but no termination dates, so no turnover rate can be computed from them at all. The Bureau of Labor Statistics reports hires and separations for construction as a whole and does not break either out by occupation, so there is no national by-role rate to cite instead. Anyone quoting a by-role trades turnover rate has estimated it. The figure is withdrawn and is deliberately not restated here as true. What we do stand behind on labor churn is the industry level hires and separations data on the contractor page, sourced to named BLS series that anyone can pull.

Recorded 2026-07-26.

What changed

The claim was computed from one job size tier that does not behave consistently with the tiers around it. That tier works out to roughly $2,200 of revenue per logged labor hour against about $143 per hour for the tier next to it. A 15x gap between two adjacent tiers points at material, equipment, and subcontractor dollars sitting against very few logged hours, not at a real difference in economics. The tier's own table also implies a different share than the one that was published. So the claim is withdrawn until the tier is re-derived with the non labor cost components separated out. The job size buckets that do behave consistently are published on the contractor benchmark page, and the finding that survives is the inversion between job count and revenue share, not a single headline percentage.

Recorded 2026-07-26.

What changed

This came from a company level COST variance, actual dollars against budgeted dollars, that was relabelled as an HOURS variance without being re-derived. It contradicts the labor hours measurement Level publishes now, which is computed per job against budgeted hours and verified against the job level data. The dollar figure is withdrawn. The hours measurement on the contractor benchmark page is the one to use, and it says something different: the median job lands on budget while a long tail runs far past it.

Recorded 2026-07-26.