Real-world testing: run fixture3 on a matrix of real projects

[tartakovsky]

What

Run fixture3 against real projects (not just the self-suite and the fake-project) and capture every failure mode.

Why

The self-suite proves fixture3 can verify itself. The fake-project proves it can run an external workflow. Neither tells us how well it survives real-world fixture trees, real CI environments, real review flows. Every new real project exposes assumptions baked into the manifest schema, the diff format, or the doctor checks.

Suggested matrix

• A Next.js project using fixture3 for snapshot testing route handlers.
• A Python service using fixture3 for end-to-end output snapshots.
• A Rust workspace using fixture3 alongside cargo test.
• An ML pipeline (Python or Rust) where the suite output is large JSON.
• A monorepo with multiple fixture3.yaml files in subdirectories.
• A repo that wants fixture3 to gate code review without blocking.

What to track per project

• Did the manifest schema cover the project's needs, or did we need a feature we don't have?
• Were the JSON diffs readable in code review?
• Did doctor catch real misconfigurations or miss them?
• Was the approval flow ergonomic, or did it require workarounds?
• Did the reducer help, hurt, or get ignored?

Done

A document listing every project, what worked, what didn't. Each ergonomic gap becomes a follow-up feature-request issue.