A bounded engagement · scope agreed before work starts

Reproducible Workflow Sprint

Turn one existing scientific workflow into a command-line tool another engineer can run, inspect, and maintain.

Buy a working workflow—not a follow-on proposal.

For materials and semiconductor teams with an existing computational model, a scientific owner, and a result that is difficult to reproduce without its author. The deliverable is one agreed, runnable workflow. A larger project is optional.

Price and duration are agreed after inspecting the code and estimating delivery capacity. The homepage’s general four-week / approximately £40k guidance is not a quote or schedule for this offer.

What you bring

  • Existing runnable code or a reference implementation, with the rights and access needed to work on it.
  • Representative inputs, reference outputs, and the supported hardware and environment.
  • A scientific owner who can explain assumptions and approve comparison tolerances.

If the starting workflow does not run or reference cases are unavailable, feasibility is a separate agreement. Its deliverables, stopping conditions, and payment treatment must be agreed before work begins. We do not promise a finished workflow before that uncertainty is resolved.

What we deliver

  • One CLI-first computational workflow in an agreed reproducible environment.
  • Documented inputs, units, assumptions, outputs, and limitations.
  • Automated repeatability checks and comparisons against customer-approved reference cases.
  • A runnable artifact, source code under agreed terms, tests, and execution instructions.
  • A technical walkthrough and a stop/go recommendation for further development.

A graphical interface is not included by default. The recommendation does not replace the runnable deliverable or oblige you to buy further work.

Acceptance is observable

  1. Before starting, agree the workflow boundary, supported environment, reference cases, numerical tolerances, and treatment of randomness.
  2. A designated customer engineer runs the workflow in the agreed clean environment using the delivered instructions.
  3. Reference cases meet the agreed tolerances. Invalid inputs and failed runs produce understandable errors.
  4. Outputs retain the configuration and version information needed to repeat the run.
  5. The handover is complete, outstanding defects are recorded, and software acceptance is documented separately from scientific acceptance.

Scientific acceptance remains with your scientific owner. Trebuchet is responsible for the agreed software-engineering deliverables. Matching reference outputs does not independently validate the physics.

Explicit boundaries

New scientific models, guaranteed physical accuracy, extensive integrations, production deployment, and ongoing support are excluded unless separately scoped. Ownership, licences, confidentiality, and changes to scope are agreed before work starts.

This is a proposed engagement structure, not a claim of previous customer outcomes or validated market demand.

Inspect the relevant work

FeCIM: reproducible simulation studies
Fractal Forge: shipped-product evidence

Start with one workflow.

Bring the existing code, one reference case, and the result another engineer needs to reproduce.

Discuss your workflow