Skip to content
ParadigmPIRXSOONDatasetsPricingHow it worksApplied workAboutContact
Start an engagement
hello@staticsignal.co.uk
Applied work

Engagements.

Some problems are not solved by a dataset. When a decision has to be made on a production line, the work runs from the capture rig to the operator’s screen, and the parts have to hold together: evidence produced by one stage is what the next stage is judged on.

We build these end to end. Clients are anonymised; each engagement is described by its scope, its taxonomy, and the evidence it produces.

What an engagement covers

Every applied engagement runs these seven stages. What changes between clients is the decision being made and what it costs to get it wrong; that sets how much weight each stage carries, not whether it happens.

Capture
Imaging geometry, illumination, and the constraints the decision actually depends on. Measured before anything is trained.
Taxonomy
Derived from the policy the decision is judged against, not from what happens to be easy to label.
Data
Constructed and verified on the same instrument we operate as a standalone service.
Model
Built to abstain. A prediction that is not supported is refused and recorded, not rounded to the nearest class.
Interface
The operator’s screen, keyboard-first, in the operator’s language.
Evidence
What the system records at the moment of decision, so the record exists before anyone asks for it.
Acceptance
Phased, with criteria agreed per stage before that stage starts.
ENG-01 · 2026In delivery

Industrial Vision System

European device refurbishment operator

Vision-transformer-based computer vision for automated laptop quality assessment. A defect taxonomy derived from manufacturer warranty policy (Customer Induced Damage). PASS/FAIL classification with explicit confidence scoring, abstention threshold, and evidence capture.

Timeline · stage 04 of 06

  1. 01Scope definition
  2. 02Taxonomy design
  3. 03Pipeline build
  4. 04 — current stageModel training
  5. 05Evaluation
  6. 06Delivery

Scope

  • 01GradingVision-transformer-based quality grading across the defect taxonomy
  • 02TaxonomyPolicy-derived grading taxonomy (Customer Induced Damage)
  • 03ClassificationPASS/FAIL with calibrated confidence scoring
  • 04AbstentionExplicit threshold: uncertain predictions are refused, not guessed
  • 05PipelineDeterministic data pipeline with full version control
  • 06EvaluationReproducible framework with evidence capture

Applied engagements are scoped individually and invoiced per stage, with acceptance criteria agreed before each stage begins.

Start an engagement