Case 03 DoorSet-CSV-Extract · governed extraction
Fix it at the source, not downstream.
Topology was being reconstructed after the fact, by correlating two manually exported files. That correlation was fragile, the exports were a recurring monthly field task, and every new site paid the cost again at onboarding.
The fix was not a better correlation algorithm. It was to stop needing one.
replaced by a schedule
removed per site
emitting degraded data
01 — The problem
Correlation is what you do when extraction failed you.
The legacy path depended on two XML exports pulled by hand from the access-control system, then correlated downstream to reconstruct which hardware belonged to which door.
Every part of that is a liability. The exports were manual, so they were a recurring field task and a single point of human failure. The correlation was inference — reconstructing a relationship that the source system already knew, from files that had lost it. And inference fails quietly: a mis-correlated door produces a plausible diagnosis about the wrong hardware.
A fragile correlation step doesn't announce itself. It produces output that looks right.
Everything downstream — diagnosis, ranking, the customer's report — inherits that error with no indication anything went wrong. The cost surfaces at the door, in front of a technician.
02 — What was built
Read the topology from the system that owns it.
A governed extraction utility that pulls configuration and topology directly out of the access-control database on a schedule, into a dated CSV. The relationship between door and hardware is read from the system of record rather than inferred from its exports.
Replaces two monthly hand exports outright. The recurring field task disappears rather than being made faster.
Door-set structure is reconstructed where the truth lives, removing the fragile downstream correlation step and the class of silent error it produced.
If the source schema changes, extraction blocks rather than emitting data that would be misread downstream. Absence of output is a legible failure; wrong output is not.
Roughly 30 to 40 minutes of onboarding effort removed per site, and the monthly task eliminated entirely — but the durable gain is the removal of a silent failure mode, not the time saved.
03 — Why it belongs in the suite
The upstream tool decides what the downstream tools can prove.
This is the least visible tool in the suite and the one everything else rests on. The diagnostic engine's accuracy claim is only meaningful if the topology it reasons over is correct. The pseudonymization tool's fidelity guarantee is only meaningful if the structure it preserves was right to begin with.
Extraction is where a data pipeline either earns or forfeits the right to make claims about what comes out of the far end.
The same DoorSet-extract CSV model that this tool produces is the input model DPT-Workbench moved to in v3.0.0.
That is what makes the suite a suite rather than a list — the tools agree on the shape of the data, so a change in one is a governed change across all of them.
Absence of output is a legible failure. Wrong output is not.