Previous accuracy
77.0%
1000 tasks
Executive summary
What this dataset says
Comparing 2026-08-03 with 2026-08-10 across 2 matched workflow/market groups. Task-weighted accuracy changes +11.0 pp, but the fixed-mix change is -5.0 pp.
Latest accuracy
88.0%
1000 tasks
Within-group effect
-5.0 pp
Previous-period task mix held fixed
Mix effect
+16.0 pp
Remainder of observed change
Evidence
Findings supported by the data
01Exact additive decomposition
A change in mix can mask deterioration
Fixed-mix accuracy = Σ(previous task share × latest group accuracy) = 72.0%. Within-group effect -5.0 pp + mix effect +16.0 pp = total +11.0 pp. Investigate matched groups before rewarding or penalizing the overall trend.
0290 / 120 implied errors
Prioritize error burden, not just the lowest rate
Review / US accounts for 90.0% of latest tasks and 75.0% of implied errors. Target review there; the data does not identify the root cause.
03Descriptive, not causal
What this does not prove
Accuracy must apply to the Tasks denominator. Rounded rates produce approximate error counts. These aggregated rows do not support individual-level significance tests or causal claims. No unmatched groups were silently excluded.
Composition
Latest implied errors by group
Action plan
Recommended next steps
Review the evidence before acting
Use the calculations, scenarios, and limitations in the executive summary. Validate denominators and row grain before making decisions.
Method & scope
How to use this report
Analysis performed
Patlr profiled 4 rows and 6 source fields, detected 2 record groups, and added 11 analysis-ready fields while preserving every original column.
Interpretation boundary
Results describe the submitted records and depend on their coverage and accuracy. Associations and anomalies should be validated against source context before decisions are made.
Evidence-only mode
Patlr does not add external facts. AI-assisted classifications are retained only when paired with an exact quote from the submitted text; unsupported output is left blank. Interpretations should still be validated before decisions.