Mix-adjusted index: each run is compared with the median rate for the same machine, bar-size mix, fabrication-class mix, and similar run size — weight bands of under 300, 300–1,000, 1,000–3,000, 3,000–10,000, and over 10,000 lbs. Run size gets its own control because it is the strongest rate driver the shop has measured: large runs spread setup and scan-gap time over far more weight, so without size banding a shift toward bigger runs would read as a performance jump (the planner compares against similar-size runs for the same reason). A score of 100 means expected performance; 85 means 15% below that comparable baseline. Runs with no usable rate estimate (no measurable active time, or a scan-artifact rate of 20,000+ lbs/hr) still count toward weight, piece, run, and estimated-hour totals but are excluded from rates, baselines, and indexes. Baselines are always computed from the trailing 52 weeks of the archive, so switching between 12, 26, and 52 weeks changes which runs you see — never what “expected” means.

Weeks: runs are grouped into Monday–Sunday weeks by Eastern-time production day, matching the shop calendar. The newest week is usually still in progress; it is drawn hollow on the chart and starred on the heatmap. The four-week comparisons on the tiles and signal cards use complete rolling 28-day windows ending yesterday (Eastern) — the tiles state their exact dates, and the time-range selector never changes them; it only widens or narrows the chart, heatmap, and exception evidence below it. If a nightly sync fails, the newest days in those windows can be empty until the next sync.

Evidence threshold: a baseline requires at least five comparable runs. Size banding makes comparable groups thinner, so more runs — especially run sizes that are new for a machine — show “no baseline” until enough same-size history accumulates. That is deliberate: the tool declines to score a 6,000 lb run against 200 lb history rather than issue a wrong index. A run’s own rate is included in the median it is measured against — that pulls the baseline slightly toward the run itself, so a slow run’s index reads marginally higher, never lower, than a comparison against its peers alone, and each exception’s comparable-run count includes the run being measured. Exceptions are shown below 75 and prioritize meaningful runs of at least 250 lbs and 15 estimated active minutes. Impact is the gap below the 75 threshold multiplied by the run’s weight, so a heavy run just under the threshold can outrank a light run far below it. The table lists at most the 12 highest-impact exceptions for the current selection — the counter above it always shows the true total, ordering by impact or completion date reorders those same rows without swapping in others, and each row shows how many comparable runs back its baseline. Exceptions cover the entire selected time range; check the completed date before treating one as a current problem. Active time and rate are scan-derived estimates—not machine utilization or OEE.