Skip to content
MoreProofMethodology

Start with the task. Check the result.

I use a small, reviewable workflow to decide whether an AI tool is useful enough to expand.

  1. Understand the job

    Identify who needs the output, where the information lives, and what can go wrong. Example: the incoming account owner needs objectives, promises, and risks in one handoff brief.

  2. Build a version people can inspect

    Make the inputs, output, and review step visible. Example: the REACH workflow explains its scores before a person confirms the account update.

  3. Test before expanding

    Check normal cases and failures, record limits, and keep a clear handoff. Example: MotionProof checks an animation's structure and browser behavior before packaging it.

Example

Evidence

An example of the checks

MotionProof release evidence with its real context, not proof of every project.

Evidence traceClaim → judgment → proof
MP-001 / Release systemPublic

Motion that earns promotion.

Claim
Separated proposal from authority: any model, agent, recipe, or human can propose a candidate, but only deterministic structure, quality, browser-render, motion, payload, and accessibility gates can promote it.
Receipt
The release gate passed 52 automated tests, strict validation for 27 of 27 production Lotties, Chromium rendering for 27 of 27, a clean consumer-package install, a certified five-artifact smoke bundle, and a three-tool MCP handshake.

SRC / https://github.com/HowdyDooToYou/lottie-animation-pipeline/blob/master/docs/public-release-audit.md

View public source audit