BrainDrive Evaluation Process

BrainDrive Development
1

@DJJones

Here is a summary doc of the full evaluation process for the backend of the software factory for discussion on Monday.

Dave J, we’ve assembled a reusable evaluation-documentation system for BrainDrive products and capabilities. Before sending more individual package questions, I wanted to give you the complete process and show where your Build Gate and implementation responsibilities fit.

The best entry point is:

It walks through the full lifecycle:

Product specification → evaluation plan → evaluation kit → Build Gate → candidate evaluation → AI qualification → preassigned human release authorization

Governing documents

  • Software Factory Workflow — the overall build and evaluation lifecycle.
  • Accepted Evaluation Protocol — the currently accepted rules for valid evaluation evidence, version pinning, independence, reruns, and AI qualification.
  • Build Gate proposal — the proposed always-on engineering, security, safety, contract, and regression gate. This is still a proposed skeleton and is specifically awaiting your acceptance or revision.

Reusable authoring and derivation documents

Evaluation-kit templates

Formal evaluation and reporting

What will be published versus kept in the execution workspace

The intended boundary is:

  • BrainDrive Docs holds accepted reusable specifications, plans, protocols, procedures, templates, and reusable kit materials.
  • Working repositories retain candidate-specific manifests, raw evidence, run logs, calibration attempts, reports, holdouts, and release-authorization records.
  • Every product-affecting release requires Dave W’s review by default.
  • Explicitly implementation-only work may name you as the release authority.
  • Mixed or uncertain work requires both.
  • Human authorization happens only after all-green AI qualification and cannot override failed or incomplete evidence.

For your review, the most important immediate item is the Build Gate. We need you to accept or revise its required checks, commands, runner, evidence record, and relationship to the exact candidate being evaluated.

The other documents are shared so you can see the complete system around that gate. We’ll submit the reusable process set and each product-specific evaluation package to BrainDrive Docs through pull requests, with Dave W reviewing before anything merges.