Case study · First critical-environment deployment
A provenance-governed analysis a critical-facilities operator acted on
Provenance is usually argued in the abstract. This is a short account of it doing real work: an independent engineer applied a provenance-governed method to a live preventive-maintenance (PM) workload problem in a federal critical-facilities environment, and management acted on the result.
2,774 assets assigned · 970 returned · 1,804 kept · ≈35% reduction in planned PM workload, across 11 work orders — accepted by management and executed the next day. Single cycle, self-reported; not an audited validation.
The problem
For a single monthly cycle, 2,774 individual assets were assigned across a set of PM work orders to one technician, against a fixed pool of labor hours. The standardized per-task time estimates in the system of record were unreliable — below the level of work actually being delivered. The question was concrete: what is the largest share of that load that can be done to standard within the hours available, and what must be reallocated?
The method
Count the work twice; never conflate the counts. One work order is not one unit of work — it may cover hundreds of assets or two. The analysis compared both the number of orders and the per-order asset counts.
Distrust the standardized estimate; compare to source. Load was re-derived from primary records, and reported times that could not reconcile with the tasks performed were flagged.
Clean the data before deciding. Thousands of non-standardized records were normalized first — a defensible decision needs a defensible dataset. Every intermediate number stayed traceable to its primary source, so the conclusion could be independently reconstructed.
The result, and what provenance added
The reallocation was accepted because it was traceable, not asserted — and it was executed: management redistributed the returned work the following day. Two things a spreadsheet would not have done: every figure could be walked back to a primary record; and the authorship and full evidence state were fixed by an Ed25519-signed manifest and a timestamped anchor before the tool was shared — so the method's origin travels with it wherever it goes ("provenance over possession").
Calibration
Single monthly cycle; self-reported; a supervisor acknowledged the reallocation in writing. This is not an independent audit of whether the retained inspections were performed correctly. The author's verbal figure in the meeting (~40%) was higher than the documented, source-verified figure (~35%); that gap is reported, not hidden. No sensitive operational detail is disclosed.
Full case study, with method and limitations: 10.5281/zenodo.21236070 (Zenodo, CC BY 4.0). This is the applied companion to the Lindsey Provenance Discipline — its first documented critical-environment deployment.
Try the tool. The workload command behind this is available as an empty, offline app — open PM Workload Command. Bring your own PM/CM work; nothing is embedded. Author-stamped (Ed25519); verify at the discipline.