Software needs a score, not just a recording.
For fifty years the source of truth for software has been its code. In the age of agents, code is increasingly the output rather than the input, and it is a poor place to keep intent. Score Markdown is the format Maestra uses to write software down: a readable document that says what an app does and why, precise enough for an agent to build from and plain enough for anyone to review.
Code records what was played. It forgets what was meant.
Read any mature codebase and you can see exactly what it does and almost never why. The decisions, the constraints, the thing the customer actually asked for: all of it lived in conversations that ended.
When a person wrote the code, that loss was survivable, because the person remembered. When an agent writes the code, nobody remembers. The intent has to be written down somewhere durable, or it is gone.
A score is Markdown with a few conventions.
Score Markdown is ordinary Markdown. Headings describe the parts of an app; prose describes behaviour; a small set of fenced blocks declare the things that must be exact: data shapes, roles, methods, and the rules an understudy is graded against.
The score is what everyone reads.
The agent builds from it. Reviewers approve changes to it rather than to thousands of lines of diff. Security reads the roles block. A new owner reads the whole thing on their first morning.
Because the score is regenerated from the app on every deploy as well as written by hand, it never drifts far from reality. When it does, the difference is itself a finding.
Scores make software portable between minds.
A score can be handed to a different agent, a different model, or a different team, and the app can be rebuilt from it. That is what it means for software to outlive the people and tools that made it.
Every Maestra app carries its score.
It is how the agent knows what to build, how your team knows what was built, and how the app stays understandable long after the conversation that started it has ended.