Total cycle time is computed as days between the 4-Way Confirmed Date/Time (present on every job) and the Major Tasks Completion milestone date.

Stage cycle times use consecutive milestone dates: 4-Way Pass → WRB Pass → Base Install Complete → Major Tasks Completion. Because intermediate milestones rolled out partway through the dataset, stage statistics are based on smaller samples.

Building Type is derived from the "Building Flag" column: 0 = single-family (SF), 1 = multifamily (MF). Multifamily jobs run noticeably longer than single-family across every division.

Cost correlation. The labor cost vs. cycle time analysis uses Pearson r (linear correlation, sensitive to outliers) and Spearman ρ (rank correlation, robust to outliers and non-linear relationships). Jobs with zero or missing labor cost are excluded from this section. The scatter plot is sampled to 800 points for readability and the X-axis is clipped at $50,000 to keep a small number of very expensive jobs from compressing the visible cluster — extreme cost outliers are still included in all summary statistics.

Time trend. Jobs are grouped by Major Tasks Completion month. Months with fewer than 5 completed jobs are excluded from the chart to avoid noise from sparse periods. The "Smoothed" view applies a 3-month centered moving average to make the underlying trend more readable; the "Monthly" view shows the raw month-by-month medians. Period comparisons use rolling 90-day windows anchored to the most recent Major Tasks Completion in the dataset, so the "12 months ago" comparison is the same 90-day window from one year prior — apples to apples.

Time Range filter. The "Last 30/60/90 days" filters select jobs by Major Tasks Completion date relative to the most recent completed job in the dataset (the anchor date shown at the top of the trend section). The filter applies to every section except "Cycle Times Over Time," which always shows full history — that section's whole purpose is to look backward, so filtering it by recency would defeat the point. Sample sizes get small quickly when you stack filters: a 30-day window on a single division's multifamily jobs may only have a few dozen records, so treat those medians as directional rather than precise.

Outlier handling for summary statistics: jobs with negative cycle times (Major Tasks Completion before start) and jobs with cycles longer than 365 days are excluded from medians/means but remain visible in the per-job table with a flagged status. Jobs with WRB Pass dated after Base Install Complete suggest milestone-entry errors and are excluded from the WRB→Base stage calculation but kept in totals.

Median vs. mean. Cycle-time data is right-skewed: a few long-running jobs pull averages well above what most jobs experience. The median better represents a "typical" job.