Proven where it's hardest.
Client names are withheld under confidentiality. Every figure below is measured, not asserted — and verifiable in conversation.
A 25-year-old core system, reconstructed from evidence.
A global testing, inspection and certification company runs its business on a core system built 25 years ago — largely undocumented, its original architects gone. Every modernisation attempt had stalled at the same wall: nobody could say with confidence what the system actually does.
We reconstructed that missing ground truth forensically: ~110,000 lines of embedded SQL decoded; the workflow engine mapped across 24 classes, 184 states and 557 transitions; ~466 database tables attributed to their functions; behaviour reconstructed across 71 certification standards. Crucially, the reconstruction was validated against ~24.5 million production log rows — documentary claims tested against operational reality, not taken on trust.
The deliverable is a living evidence platform: 300+ cross-linked documents and ~400 graded facts, every one source-linked, with automated integrity checks on every change — tuned from 780 false positives to zero across 230 files.
~110,000 lines decoded
24 / 184 / 557 workflow map
~24.5M log rows validated
400 graded facts
The same method, portable to a new enterprise in a day.
For a top-3 German automotive OEM — in partnership with an established systems integrator — the same evidence-graded corpus method is being applied in a second enterprise context: an agentic software-development-lifecycle platform prototype, where grounded, source-linked knowledge about an application drives the generation and validation of its tests.
The method was bootstrapped onto this engagement in approximately one day, as an independent corpus with zero cross-client content carried over — evidence that the discipline is a repeatable capability, not a one-off effort.
~1 day method bootstrap
0 cross-client content
Details are factual; identities withheld under confidentiality agreements.