Last week's comment about flipping the hierarchy turned into something unexpected. The work finished in four hours instead of the estimated week. Details in When Four Hours Suffices.

That structural change enabled the second step: optimising the timewalker itself. Two passes – non-consing traversal, then push-based transducer – cut the implementation from 1,461 lines to 647 while improving both speed and memory usage. The timewalker is Ooloi's data processing foundation; getting it right matters.

A question kept coming up on VI-CONTROL: 'Do I need to learn Clojure to use this?' No. The core is Clojure because that's what solves the hard problems. But plugins work in any JVM language – Java, Kotlin, Scala, whatever. First-class citizenship, no penalties. Penitenziagite! explains why the vernacular matters.

Wagner-scale benchmarks validated the architecture at extremes: 1,000 measures, 29 staves, 520,000 pitches. Sub-second traversal, 172 KB files, linear hardware scaling. The M3 results show 2–3× improvement over 2017 hardware – which means typical edits complete in under 2 milliseconds. A Wagner-Sized Benchmark has the numbers.

Nando Florestan pointed out that the anti-rebranding clause made Ooloi proprietary in practice, however good the intentions. He was right. Pure MPL 2.0 now, no amendments. Poetic Licence covers the correction and what open source actually means for this project.

Next: planning the frontend client's event architecture. That's the preparatory stage for windowing, which leads to drawing actual music.

The invisible foundation is validated. Soon, things will become visible. But first, some architecting.

/ Peter Bengtson