The documentation has accumulated to the point where it doesn't naturally sort itself for a first reader. Architecture Decision Records are numbered chronologically — in the order decisions were made, not in the order you need to encounter them. Reading straight through means confronting distributed-systems reasoning before you know what a `Piece` contains. The answer keeps arriving before you've understood the question.

There's now a guided path that reorganises the material by conceptual dependency: nine phases, starting with motivation and working through data model, language choices, addressing, pitch representation, the plugin system, rendering, and the frontend. Each phase builds the vocabulary the next one requires.

Think of it as an adjunct to the Librarian — where the Oracle answers specific questions, this path builds the context that makes those answers land correctly. It's now the canonical entry point to the technical documentation. It's here.

Open Settings, switch the locale. Every open window follows: menus, the About dialog, Settings itself. English to German to French to Danish, pausing on Greek. Then it repeats.

This is chrome – wiring things up, giving Ooloi a face.

The obvious next step seemed to be collaboration. The architecture has supported shared backends since day one; the collaboration ticket and ADR systematise how they are handled, add authentication, authorisation, piece invitations.

But you can't test the collaboration paradigm just on application settings. Theme preferences, locale, autosave interval – these are local. They don't live on a shared backend. To verify that the pattern works, you need an actual piece: something that lives on a server that one or more clients connect to.

So a sequence of work comes first.

Undo and redo for settings changes. Every setting change already records the old value alongside the new; the undo stack accumulates and replays inverse events. No piece dependency. This establishes the mechanism before anything more complex exists.

Then creating and opening pieces — not in a naïve single-user way, but written from the start for an environment where the backend may be shared. Even locally, the model has to reflect that reality.

Then undo and redo for piece content: a single authoritative stack on the server, shared across everyone editing simultaneously. The undo gesture in the frontend inspects which tier holds the more recent operation and routes accordingly.

Then the Piece Settings editor, with the Instrument Library window.

Here is where the collaboration pattern is actually demonstrated. Consider adjusting beam thickness or the default appearance of piano pedal extension lines: ordinary engraving decisions. Now add a collaborator with the same editor open on the same piece. Your changes appear in their view instantly. Theirs in yours. Both of you can undo, on the same data, with the same shared history.

This is not a special case requiring special handling. If the preceding layers are in place, this is what falls out of the existing architecture. But it should be done as early as possible, before additional layers complicate things, so now is the right time.

The Ooloi documentation has grown considerably. ADRs, guides, READMEs, newsletters, blog posts – there's a lot of it, and not all of it is easy to navigate. So I've built something to help.

The Ooloi Librarian is a small RAG system – Retrieval-Augmented Generation, for the uninitiated – that answers questions about Ooloi's architecture and design directly from that documentation. It retrieves the relevant chunks, passes them to Claude Haiku as context, and returns a grounded answer with source citations. If nothing relevant is found, it says so rather than making something up. This is enforced in code, not by prompting.

It's live now in the Ask the Librarian section of the Documentation page. Try it. Ask about the accidental rendering algorithm, or why Ooloi uses STM, or what the timewalker is. The answers come from the docs, not from the model's general knowledge.

The backend runs on AWS – API Gateway, Lambda, Bedrock – rate-limited and CORS-restricted to this domain. Roughly four cents per thousand queries, and nothing at all when idle.

And we also have a new Guide: MIDI in Ooloi. It explains why Ooloi takes MIDI input but produces no MIDI output, and why that is a considered choice rather than an oversight. In 2026, there are cleaner ways to control virtual instruments than routing through MIDI, and the guide explains what Ooloi does instead.

Deeper technical details here: 0044-MIDI-Input-Library-and-Boundary-Architecture.

​Magnus Johansson asked in the comments of 'Surface Tension' whether Ooloi would have a separate piece window, as Igor Engraver did. I told him that the first version of it already existed on screen, and that I'd write a post explaining why it's there. This is that post.

Magnus isn't a casual observer. He's almost certainly the person alive today who knows Igor Engraver most deeply – including me, at this point. He translated the software into Swedish. He still uses it as his preferred notation tool, describing it as 'like playing an instrument': muscle memory, not nostalgia. He was interviewed by Realtid about what happened to NoteHeads. When I come to reimplement Flow Mode, Magnus is the person I will be consulting, because he carries it in his hands in a way that no documentation can replace.

When Magnus asks about the piece window, it's not a feature request. It's the question of someone who understands, at the deepest level, why it was there the first time.

The window is empty for now, and there's no reason to pretend otherwise. On a commercial software blog you'd never show this: you wait until the feature is complete, populated with convincing content, ready to impress – and then you present it. This blog has never operated that way. It's a development record, transparent about what exists, what doesn't, and why decisions were made. You're welcome to follow the journey.

​The piece window isn't here because Igor Engraver had one. Ooloi isn't Igor 2.0, and I want to be precise about this: the two systems share no code, no architecture, no technical lineage. What they do share is a set of convictions about how human beings and creative software should interact, because the person who designed both is the same person, with the same ideas – and those ideas haven't changed. They've deepened.

Igor had the piece window because the semantic model required it. Ooloi has it for exactly the same reason. Igor had Flow Mode because a musician entering notes shouldn't have to think about the software. Ooloi will have it for exactly the same reason, and Magnus will be part of getting it right. What Ooloi takes further is everything underneath: the concurrency model, the rendering architecture, the collaboration layer, the foundations that Igor never had the chance to build correctly.

The window is empty now. Not for much longer.

I've published a guide to Ooloi's frontend architecture – long, precise, and detailed, written for architects and developers who want to understand the system from the inside.

It covers window lifecycle, event architecture, the rendering pipeline, fetch coordination, localisation, collaboration, and the testing model. Together with the Timewalking Guide and the Polymorphic API Guide, it forms the main conceptual gateway into Ooloi's internals.

Worth flagging for experienced frontend developers: this architecture inverts several instincts. The frontend holds no semantic authority. It is computationally strong and structurally subordinate; that subordination is precisely what makes distributed collaboration, transport independence, and first-class plugin composition possible without architectural compromise. The guide explains why. I think the argument is sound, though I'm aware it might feel distinctly odd, or even regressive, to a developer whose instincts were formed in React or a similar ecosystem.

If you work in notation – as a musician, engraver, or editor – and find this kind of writing opaque, current large language models are genuinely useful here. Feed the guide to one and ask what it means for your daily work. It's a reasonable use of the technology, and the answers tend to be surprisingly good – when they're not inventing things I'd never claim ;)

Worth stating plainly: the subject here is art, and the technical rigour exists in service of that, not despite it. This blog documents the reasoning behind Ooloi as it develops – for future contributors, researchers, maintainers, and critics. Systems like this have a lineage and a psychological architecture as much as a technical one. I want that visible. This blog has a voice because it was written by someone who considers authorial invisibility a failure mode rather than a virtue.

Ooloi is now a running desktop application on macOS, Windows, and Linux. It opens, it responds, it remembers where you left it. A concrete thing, like any other program.

But that's not enough. The next step is teaching it to talk. Ooloi's frontend and backend have always been separate components communicating through a defined protocol; in standalone mode, they simply appear as one and the user needn't think about it. But this disciplined split personality was designed that way for a reason, and it's now time to act on it.

ADR-0036 lets the two halves run apart — so one Ooloi can connect to another, or to a centralised server. You invite a collaborator; a secure connection opens. They disconnect; the application returns to working alone. No restart, no mode switching.

The separation was always there. Now it becomes visible.

​I am, as of this writing, sixty-four years old. I have been programming for over fifty years. I say this not as credential but as context, because having been there changes what you see.

I was there for the AI winter. I watched a field that had promised the moon retreat into a decade of funding cuts and reputational collapse, dragging Lisp's standing down with it. The language didn't deserve the association; it was punished for the sins of researchers who had over-promised. I kept using it anyway, because the language hadn't changed, only its fashionability. This was the first time I learned that technical merit and cultural standing are largely unrelated.

I was there for the object-oriented conquest of the 1990s. I watched as C++ and then Java established OOP as orthodoxy, complete with design pattern catechisms and UML diagrams that looked more like religious iconography than engineering documentation. CLOS, which had solved the problems these languages were struggling with years earlier, was invisible because it ran on the wrong platform and belonged to the wrong tribe. The industry chose what was marketable over what was good. This was the second time I learned that lesson.

I was there for the web's arrival, which changed everything about deployment and nothing about the fundamental problems of building complex software. I was there for the enterprise Java years, when 'architecture' meant XML configuration files and dependency injection frameworks and a peculiar conviction that more abstraction was always better, regardless of what it cost in comprehensibility. I was there for the functional programming renaissance, which rediscovered principles that Lisp had embodied since the 1950s, often without acknowledgement and frequently with a missionary zeal that replicated the very dogmatism it claimed to reject.

I am here for the current AI explosion, which rhymes with the original AI boom in ways that make my skin prickle. I have seen this arc before: extraordinary technical promise, sweeping claims about imminent transformation, and a market that moves faster than the underlying science. Time will separate what is durable from what is fashionable.
​
None of this is complaint. It is observation. When you have been programming for fifty years, you develop a particular relationship with trends: you have seen enough of them arrive, dominate, and recede to know that the arrival tells you very little about the value and the dominance tells you nothing about the longevity. What remains after the trend passes is what was true about it. Immutability was true about functional programming. The JVM was true about Java. Interactive development was true about Lisp all along.

The thing that has remained true for me, through every cycle, is the freedom principle. The conviction that the programmer should be trusted, that the tools should serve the maker rather than constrain them, that the ability to reshape your instruments is not a luxury but a precondition for doing serious work. I did not learn this from a manifesto. I learned it from a mainframe in 1973 and a Lisp prompt in 1979, and everything since has confirmed it.

Ooloi's plugin system is, in a specific sense, the culmination of everything this essay describes.

The core engine is written in Clojure and architecturally closed. I've written about this elsewhere as the 'monastic core' principle. The freedom is at the perimeter: any JVM language can write plugins, and the plugin API speaks each language's idioms natively. A Java developer writes Java. A Kotlin developer writes Kotlin. Nobody is forced to learn Clojure or to understand the internals. The benefits of the architecture propagate without requiring conversion.

This is the CLOS principle, expressed architecturally. The system trusts the developer. The developer works in their own language, with their own patterns, and the underlying architecture handles the coordination. The monastery uses Latin internally because Latin is precise; but when teaching farmers improved agriculture, you speak their language. This is not compromise. It is proper boundary design, and it is the lesson of fifty years in the workshop tradition.

When I sit at this keyboard today, fixing a macOS menu bug or stripping debug symbols from a jlink runtime, I am using skills accumulated across the full span of what I've described. The mainframe taught me that serious work requires serious infrastructure. Lisp taught me that freedom is an engineering condition, not a slogan. CLOS taught me that the system should trust the programmer. The loss of Igor Engraver taught me that good architecture survives the business that funded it, if someone is stubborn enough to rebuild. Clojure taught me that immutability changes what's possible. The whole genealogy is present in every design decision, whether or not it's visible.

I don't know if Ooloi will succeed, whatever that is. I know the architecture is right. I know the foundations carry what needs to be built on top of them.

​And I know where the freedom comes from, which is what I wanted to write down.

After nineteen months of building systems that nobody can see – temporal traversal, accidental algorithms, transducer pipelines, STM coordination – I have spent the past fortnight making menus behave.

Menus. On three operating systems. macOS wants an application menu in the system bar. Windows wants Exit in the File menu and About in the Help menu. Linux wants Quit in the File menu and Settings in the Edit menu. None of them is wrong, exactly, but none of them agrees with the others, and they all have opinions.

This is normal engineering. It is, after the intensity of the semantic work, refreshingly boring.

Dark mode and light mode. A splash screen that shows each component as it initialises. Window state persistence so the application remembers where you left it. An About dialog with a background of engraved music, rendered through the same windowing path as everything else. Translation keys for every user-visible string, verified at build time against PO files. Fixing the JVM's 64KB static initialiser limit because 185 multimethod symbols turned out to be too many for a single class. Stripping debug symbols from the jlink runtime to bring the application bundle down from 213 MB to 173 MB.

None of this is interesting in the way that a deterministic accidental rendering algorithm is interesting. And that's the point.

Ooloi now has a face and a body. It boots, shows what it's doing, presents platform-appropriate chrome, responds to commands, and shuts down cleanly. The experience of seeing it is strange – like hearing a recording of a piece you've been composing in silence. You know every voice, every structural decision, but the thing sitting there on the screen, with its dark window and its OK button, is suddenly external. It exists in the world rather than in your head.

​In a blog post from some time ago, Penitenziagite!, I wrote about linguistic impedance in software: the way systems shout incomprehensible demands at contributing developers, forcing them to think in the wrong language. That post was about architecture as translation – the scriptorium at the centre speaking Latin for precision, and the gate speaking the vernacular so people can actually participate.

This post is about governance.

Specifically: why Ooloi is open source, but not open in the way some people expect. And why it cannot be.

​There is a persistent belief among developers that open source implies a particular governance model. In this vision, design is open, debate is open, architecture is open, decision-making is open, and even the rewriting of foundations is treated as normal community activity. The project becomes a bazaar: a bustling marketplace of patches and opinions, competing visions and negotiated direction.

This model works for some kinds of software. It fails for others. Music notation software belongs firmly in the latter category.

The problems Ooloi addresses – temporal coordination, concurrent structural mutation, global layout decisions, semantic correctness across voices – are not decomposable into independent opinions. They cannot be negotiated piecemeal. They require conceptual unity, which is another way of saying they require architectural closure.

A bazaar cannot produce that. However, a cathedral can.

​A closed semantic core is often misread as a claim of finality: finished, complete, unchangeable. That is not what I mean.

What is closed is the space of permissible reinterpretation. New features will arrive, new capabilities will arrive, refinements and extensions will arrive – but they must fit within the established model rather than redefine it.

This distinction matters because most architectural failure in open projects does not come from lack of innovation. It comes from uncontrolled semantic drift: the gradual erosion of guarantees as each contributor reinterprets the foundations to suit local convenience. Ooloi explicitly forbids that drift.

​Ooloi's correctness depends on architectural decisions that only work as a coherent whole.

Immutability is treated as a semantic guarantee rather than a performance trick. STM is not merely one possible concurrency implementation among several; it is the concurrency model, because the alternatives import failure modes that Ooloi is explicitly designed to eliminate. Structural sharing is not optional; it is a prerequisite for the kind of scale where a full orchestral score can be edited in real time without the system grinding to a halt. Computation is staged because global layout decisions are otherwise intractable. Piece state is single-authority because reconciliation architectures – the kind where multiple sources of truth must be merged – create exactly the class of bugs that have plagued notation software for forty years.

These are not interchangeable components. They are constraints, and constraints that reinforce each other. You cannot selectively replace one of them. You cannot introduce mutability 'for performance'. You cannot swap out the concurrency model. You cannot dissolve authority boundaries. Each such change collapses guarantees that are only meaningful because the other constraints hold.

The cathedral stands because the architecture must stand.

​Here is the paradox: by closing the core, Ooloi makes everything around it genuinely open.

Because the centre does not move, APIs stop being provisional. Semantics do not drift. Performance characteristics remain predictable. Plugin contracts remain valid across versions, which means plugin authors can build with confidence rather than anxiety about the next breaking change.

Plugins become first-class citizens rather than tolerated afterthoughts. Any JVM language can participate without translation into a single house dialect. Boundaries stay explicit. Failure modes stay contained.

This is the same pattern that made TeX durable, SQLite coherent, and the JVM stable enough to host entire ecosystems. A fixed centre enables freedom at the edges.

​Ooloi is open source in a precise, literal sense: inspectable, forkable, extensible, durable.

It is not open in the sense of open-ended architectural debate. It is not open in the sense of consensus-driven semantics. It is not open in the sense of renegotiating authority with each new voice that arrives.

The core is not a public workshop. It is a composition. Collaborative compositions exist – devised theatre, collectively improvised music – but they are specialised forms suited to particular purposes, not universal methods. Ooloi's architecture is not that kind of work. Its structure must be fixed before others can build upon it.

The bazaar is not inside the cathedral; it is outside, at the gate, where plugins flourish without threatening the foundations.

The Penitenziagite! post argued that architecture should not shout incomprehensible demands at its users. This post argues the complement: governance should not shout incomprehensible demands at its contributors.

Ooloi's structure is explicit. There is the cathedral/scriptorium: the core, architecturally closed and semantically fixed, where the hard problems are solved. There is the library: canonical plugins and reference implementations, where patterns are demonstrated rather than debated. And there is the gate and the bazaar beyond it: the plugin ecosystem, open and multilingual and creative, but separated from the core by a boundary that preserves the integrity of the whole.

This is not a contradiction. It is a design pattern.

A stable centre enables a flourishing perimeter. A closed semantic core enables an open ecosystem. A cathedral enables a bazaar – but only if the gate between them holds.

This is the very first window Ooloi has opened. Ever.

Not a console command or REPL experiment - a window created through the infrastructure completed yesterday (ADR-0031), using the new UI Specification Format (ADR-0042) that amongst other things makes all UI elements accessible to plugins as first-class citizens.

Another quiet decision: localisation from the start. ADR-0039 establishes that every UI string is translatable from day one, using GNU gettext's standard .po format. The tooling ecosystem is mature: Poedit, Crowdin, Transifex, Weblate, Lokalise, POEditor, OmegaT. Translators know these tools. Workflows exist. Nothing proprietary, nothing invented here.
​
Not retrofitted later. Built in.

Well, that was quick! The multi-tiered frontend event infrastructure is complete and implemented. One evening's work added the Event Router, the Rendering Data Manager, and the Fetch Coordinator. This implements ADR-0031 and shows that having the implementation architected in detail before writing the code saves a lot of time. I'd expected to be busy with frontend events for some time.

Now that that's out of the way comes what it enables: windows, dialogs, menus, notifications. I'll be working with JavaFX, cljfx, AtlantaFX: what's called 'UI chrome'.

It's concrete work without semantic subtleties for a change: does the window open? Does the NordDark theme apply correctly? Does the dialog render? Aesthetic decisions instead of semantic ones.

I expect these blog posts will turn quite practical in tone for a while, as this is standard engineering, plumbing, connecting things, turning detailed ADRs into working code. Quite relaxing, as a matter of fact.

And I can now, after 18 months, finally give in to that desire to actually open a window. Fresh air!

Today, hierarchical key signature overrides were completed. Global key signatures can now be overridden at the instrument level as well as the staff level, as many times within a measure as required, with fully deterministic accidental handling. The updated ADR-0035 documents the algorithm.

This sounds narrow. It isn't. The ability to handle this correctly, without manual intervention or heuristic guesswork, required everything underneath it to be correct first. The last piece going in means the semantic aspect of the engine is complete.

Eighteen months of invisible work, ended.

The focus now shifts to the frontend event system, then to opening windows, etc. The infrastructure phase succeeded by disappearing; the next phase will succeed by appearing. Different disciplines, different anxieties.

I find it difficult to articulate what this moment feels like. The closest analogy I have is from composition: finishing a score and waiting for the first rehearsal.
​
A carefully prepared potentiality, awaiting concretion. 

For decades, orchestral playback has been organised around an absence of trust.

The score was not considered precise enough to stand on its own, so it had to be translated. MIDI became the intermediary. DAWs became the place where music was made to behave. Templates grew until they resembled orchestras-in-storage. Machines were added not because the music required it, but because the protocol did.

Ooloi takes a different position.

The score is treated as the authoritative description of the music. Pitch, timing, articulation, phrasing, microtonality and humanisation are not annotations awaiting interpretation elsewhere. They are part of the structure. Playback does not correct the score. It follows it.

For that reason, Ooloi has no MIDI output in the core. Audio is produced entirely by frontend plugins: VST/AU instruments, SoundFonts, synthesis engines. The backend never allocates channels, never bends pitches, never streams audio. It manages musical structure. Plugins turn that structure into sound. The instruments run inside Ooloi, not in an external DAW waiting for MIDI.

Pitch is represented symbolically and exactly. A note is not a frequency approximation or a MIDI pitch with heavy makeup but something like "C###4+50". A sustained chord can contain a continuously glissing inner voice without splitting staves, duplicating instruments, or consuming additional channels. There is no pitch-bend choreography, no controller bookkeeping, no library-specific workaround logic. The DNA soup is gone.

For readers coming from notation, this restores the score's authority. Slurs, dynamics, accents, register and phrasing are no longer hints for later repair. They are the performance.

For readers coming from virtual instruments, this architecture removes entire categories of work that have become normalised:

- No permanent DAW templates holding a full orchestra 'just in case'
- No slave machines preloading instruments that never play
- No CC drawing to approximate phrasing already present in the notation (the work that currently keeps Hans Zimmer's assistants employed)
- No channel allocation strategies for divisi or microtonality
- No waiting for templates to load before hearing a single note

Because the semantic model captures what is actually written, playback plugins can analyse the score and load only what is required. If a piece contains no contrabassoon, no contrabassoon need exist in memory. If a technique is missing in one library, another can be invoked for that passage alone. Routing, balance and reverberation can follow from structure, not from global assumptions.

This is why large template-based setups become unnecessary. Not because of optimisation tricks, but because they were compensating for a semantic gap between notation and sound. The architecture closes that gap.

Do DAWs still matter? Yes, but later. Mixing, sound design, video synchronisation and final delivery remain DAW territory. What changes is that the DAW is no longer required to make the music behave like music.

Ooloi does not replace the DAW.

It removes the need for the DAW to do a job it was never meant to do.

​A few weeks ago I wrote a post explaining computer science terminology to musicians. This is the inverse: an explanation of music notation for programmers who assume the domain is simpler than it is.

That assumption is common. It's also expensive. Most notation software has been written by very competent engineers who underestimated the problem. The result is systems that work impressively until they don't – at which point they fail in ways that seem arbitrary to users but are, in fact, architectural.

This post explains why.

​Here is a concrete example of why music notation is harder than it looks.

Question: Does this F need a sharp sign?

Answer: It depends on:

1. The key signature (established possibly pages earlier)
2. What happened earlier in this measure in this voice
3. What happened earlier in this measure in other voices on the same staff
4. What happened earlier in this measure in other staves (in keyboard music, an F♯ in the left hand may affect an F in the right)
5. Whether there's a tie from the previous measure
6. Whether this is a grace note (different rules apply)
7. Publisher conventions (some always mark cautionary accidentals, some don't)
8. Historical period (18th-century conventions differ from 20th-century)

A naive algorithm checks each note against previous notes in the measure. This is O(n) per note, O(n^2) per measure. Fine for simple music.

Now add multiple voices. The F in the soprano might be affected by an F♯ in the alto, depending on conventions. You need to track alterations across voices. If voices cross staves (as in keyboard music), the interactions multiply.

Now add simultaneity. What happens when two voices sound the same pitch at the same instant but with different accidentals? Yet another edge case difficult to catch when imprisoned in the Jail of Recursive Descent.

Now add grace notes. A grace note at the beginning of a measure might be notated at the barline but sound at the end of the previous measure. Does its accidental affect the next measure? Does the next measure's key signature affect it?

The dependency graph for accidentals is not a simple sequence. It's a directed acyclic graph with edges determined by temporal position, voice membership, staff assignment, and notational category. Correct handling requires topological sorting with domain-specific edge semantics.

Most notation software handles this with heuristics: rules-of-thumb that cover common cases and fail on uncommon ones. The failures aren't bugs in the usual sense. They're consequences of architectural decisions made decades ago when the problem was less well understood.

Ooloi handles this through what I call the 'timewalker' – a transducer that processes musical events in semantic order (not storage order) allowing full temporal context accumulation. The accidental algorithm becomes stateless over its input stream: given the same sequence of musical facts, it produces the same decisions. Deterministically. Every time.

The complexity is still O(n · v) for n events and v voices, but the constant factor drops dramatically because traversal and processing are separated. The transducer handles navigation; the algorithm handles meaning.

​Here is where most notation software fails silently.

A modern computer has multiple cores. Users expect responsiveness: type a note, see it appear immediately. But notation isn't like a word processor where inserting a character affects only nearby layout. Changing one note can alter accidentals throughout a measure, reflow an entire system, or invalidate cached engraving calculations across pages.

The traditional approach is to serialise everything. One thread handles all mutations. The UI blocks or queues. This is why notation software freezes when you paste a large passage, or why collaborative editing remains essentially unsolved in this domain.

The problem deepens when you consider what 'correct' means during an edit. If a user changes an F♮ to an F♯ in measure 4, the accidental state of measure 5 depends on this change. But if the rendering thread is currently drawing measure 5, what does it see? The old state? The new state? Some inconsistent mixture?

Mutable state with locks doesn't solve this; it creates a new category of bugs. Deadlocks. Race conditions. Heisenbugs that appear only under specific timing. A very difficult work environment indeed, and one where diminishing returns becomes a real impeding factor.

The functional approach is different. If musical data is immutable, the question dissolves. The rendering thread has a consistent snapshot – always. The editing thread produces a new snapshot. There's no moment of inconsistency because there's no mutation.

This is what Software Transactional Memory provides in Clojure. Transactions compose. Reads are always consistent. Writes are atomic. The system guarantees that no thread ever sees a half-updated score.

The performance implications are substantial. Ooloi's STM handles heavy contention without the coordination overhead that serialised systems pay constantly. The bottleneck shifts from synchronisation to actual work.

Real-time collaboration becomes architecturally possible. Two users editing the same passage don't corrupt each other's state; they produce alternative futures that the system can merge or present as conflicts. This isn't a feature bolted on afterwards. It's a consequence of representing music correctly from the start.

​Programmers sometimes assume that once you have the musical data, displaying it is straightforward. Calculate positions, draw glyphs, done.

This underestimates the problem by roughly an order of magnitude.

Consider accidental placement in a dense chord. Multiple accidentals must avoid colliding with each other, with noteheads, with stems, with ledger lines, and with accidentals in adjacent chords. The placement rules aren't arbitrary – they encode centuries of engraving practice optimised for readability.

This is a constraint satisfaction problem. In the general case, it's NP-hard. Real systems use heuristics, but the heuristics are subtle. Gould's Behind Bars – the standard reference on notation practice – devotes many pages to accidental stacking alone.

Now multiply by every other engraving decision: beam angles, stem lengths, tie curves, slur routing, dynamic placement, tuplet brackets, ottava lines, system breaks, page breaks. Each interacts with the others. Change a stem length and the beam angle needs adjustment; beam ends need to follow sit-straddle-hang rules and might both move. Change a beam angle and adjacent voices may collide.

LilyPond's engraving quality comes precisely from treating this as a serious optimisation problem. Its page-breaking algorithm runs a variant of the Knuth-Plass line-breaking algorithm (developed for TeX). Its spacing uses spring-based constraint systems. The 'simple' rendering layer contains some of the most sophisticated code in the system.

Interactive notation software faces an additional challenge: these calculations must be fast enough for real-time editing. You can't spend ten seconds optimising layout every time the user adds a note. The tension between quality and responsiveness has driven much of the architectural evolution in this field.

​A programmer encountering music notation for the first time will find countless apparent inconsistencies. Surely these could be rationalised?

No. Or rather: not without destroying information.

A fermata on a note means 'sustain this note'. A fermata on a barline means 'pause here'. A fermata on a rest means something slightly different again. These distinctions emerged from practice. Composers use them. Performers understand them. A system that normalises them into a single 'fermata event' loses musical meaning.

Grace notes in Bach are performed differently than grace notes in Chopin. The notation looks similar. The performance practice differs. A system that treats all grace notes identically will produce correct output that performers find misleading.

Clefs change meaning by context. A C-clef on the third line is an alto clef; on the fourth line, a tenor clef. Same symbol, different staff position, different meaning. This isn't legacy cruft – orchestral scores routinely use multiple C-clef positions because different instruments read different clefs.

Mensural notation (pre-1600) has entirely different rules. The same notehead shapes mean different things. Accidentals follow different conventions. Barlines serve different purposes. A truly comprehensive notation system must handle this or explicitly disclaim it.

The temptation to clean this up, to impose regularity, to treat historical conventions as technical debt – this temptation must be resisted. These conventions encode musical knowledge. Destroying them in the name of consistency destroys the knowledge.

​All of this serves a purpose that has nothing to do with computers.

Notation exists to communicate with humans. Specifically: performers who bring interpretation, physical constraints, and centuries of inherited practice to their reading.

This isn't sentimental. It has engineering consequences.

A page turn in the wrong place forces a pianist to stop playing or memorise a passage. This is a functional failure, not an aesthetic preference. The page-breaking algorithm must allow this to be modelled.

Awkward beam groupings obscure metric structure. The performer reads beats through beam patterns. Wrong beaming means wrong emphasis. The beaming algorithm must model this based on the currnent time signature.

Unclear voice leading forces performers to decode the texture before they can interpret it. The system must make voices visually distinct in ways that match musical distinctness.

Engraving choices that are mathematically equivalent may not be performatively equivalent. Two layouts might use the same space and avoid all collisions, but one reads naturally and the other requires active deciphering. The difference matters.

This is why notation is not a solved problem. The specification is 'communicate musical meaning to human performers'. That's not a format spec. It's a design constraint that touches every decision in the system.

Music notation looks like a data format problem. It is actually a meaning representation problem with a millennium of accumulated context.

The systems that fail – and most fail in some way – do so because they treat notation as something to be processed rather than something to be understood. They model the symbols without modelling what the symbols mean.

The systems that succeed take the domain seriously. They represent time exactly. They handle multiple hierarchies without forcing false containment. They process context correctly even when context is non-local. They treat engraving as a real problem, not a rendering detail. They preserve historical conventions because those conventions carry meaning.

This is why Ooloi's architecture looks the way it does. Rational arithmetic isn't ideology; it's the only way to represent tuplets exactly. Immutability isn't fashion; it's the only way to guarantee that accidental decisions are deterministic. Transducers aren't showing off; they're the cleanest and most efficient way to process a semantic stream with accumulated context. STM isn't over-engineering; it's the only way to achieve both correctness and responsiveness without sacrificing either.

The domain dictates the tools. Not the reverse.

We have a new year, and it's time to make plans. Ooloi's architecture is now closed, and the work enters a new phase. What follows is a road map, not a schedule – no dates, no promises – but a logical progression.

These stages are different in nature from what has gone before. Conceptually simpler than the deep semantic work (what the music is, how it's represented and manipulated) or the infrastructure (server/client transports and roundtrips, monitoring, certificates). Everything below is comprehensively architectured and implementation-ready. The thinking is done; what remains is execution.

To implement what was described in the last blog post deterministically, ADR 0028, on the hierarchical rendering pipeline, and ADR 0037, about measure distribution, both have been updated. The rendering pipeline now has six stages rather than five.

The additional stage enables the behaviour discussed in the previous post, but also allows slurs, lyrics, dynamics, and other elements that require vertical space to adapt dynamically without introducing layout jitter.

What follows is a brief tour of the six pipeline stages. Each has a clearly defined role, a fixed direction of information flow, and no hidden feedback paths. Together, they transform musical structure into pages without iteration.

If you are a musician, you should be able to follow how the music moves from essential meaning to visual expression. If you are a developer, the architectural choices should be readable between the lines. In both cases, the ADRs contain the full technical detail.

Each measure is processed independently and turned into engraving atoms: indivisible musical units that already contain noteheads, augmentation dots, accidentals, stems, articulations: everything that belongs together. A note or a chord is an atom, including everything at a fixed distance from that note or chord. Once an atom is computed, its constituent parts never move relative to each other, though they can be adjusted manually or converted by the user to pure graphics for Illustrator-like manipulation.
​
From these atoms, the system derives three facts about the measure stack: a minimum width, where atoms are simply lined up after one another (guaranteeing that nothing ever collides); an ideal width, based on proportional spacing from rhythmic density; and a gutter width – extra space required only if the measure begins a system. Courtesy accidentals caused by system breaks are accounted for here as space requirements, not as symbols. Ordinary courtesy accidentals are already part of the atom.

All measures in a stack contribute a rhythmic raster: the sequence of atom positions implied by their rhythmic content at ideal spacing. Stage 2 takes these per-staff rasters and aligns them into a single, stack-wide rhythmic grid, so that 'the beat structure' is shared across staves rather than each staff carrying its own local spacing. Because the system has both the rasters and the atom extents from Stage 1, it can adjust the aligned positions so that collisions are avoided while still preserving the intended proportional structure.

With rhythmical alignment complete, the score collapses to a one-dimensional sequence of measure stacks, each described only by stack minimum width, stack ideal width, and possible stack system-start gutter. System breaks are chosen using dynamic programming, and a scale factor is computed for each system so that ideal proportions are preserved as well as possible within the available width. At the same time, system preambles (clefs, key signatures) are fully determined. Apart from these, the stage produces only break decisions and scale factors; no measure content is positioned yet.

The abstract decisions from system formation are applied. Each atom is positioned horizontally by scaling its ideal spacing according to the scale factor of its system. Because atoms were defined with purely relative internal geometry, they can be moved without recomputation. This stage is a pure projection from musical proportions to absolute coordinates.

All connecting elements are resolved against the fixed atom positions: ties, slurs, hairpins, ottavas, lyric extenders. They adapt to the horizontal geometry they are given and never influence spacing upstream. System-start decorations live here as well: courtesy accidentals now appear in the gutter space that was explicitly reserved earlier.

​Actual system heights emerge at this stage as a consequence of what must be drawn – slurs, ties, lyrics, pedalling indications, hairpins – each claiming vertical space as required to avoid collisions. In this stage, braces, brackets, and instrument name indications are also computed for the left margin, as all information required to determine system grouping and extent is now available.

Systems are distributed across pages using dynamic programming, this time over the actual system heights produced by the previous stage. As in Stage 3, the distribution is optimal with respect to the available space. Once page breaks are chosen, layout is locked and no earlier stage is revisited. At this point, all rendering decisions have been made. The score can now be drawn directly from the cached paintlists.

Standard practice in many house styles is to show a courtesy accidental at the start of a new system when a tied note continues across a system break. In traditional notation workflows, this is handled manually: the engraver adds or forces the accidental after layout.

That manual step has consequences. A courtesy accidental has width. Adding it can push spacing past a threshold, move a system break, or reflow measures. A small local correction can trigger a cascade elsewhere, sending the engraver back to the same passage again. This is not a matter of care or skill. It is a side effect of heuristic layout combined with post-hoc correction.

Ooloi takes a different path.

Courtesy accidentals that exist because of a system or page break are derived, not edited. Ooloi has user settings to control preferred behaviour – such as whether these accidentals appear at all system breaks, page breaks only, or not at all. The required space is known before layout begins. System and page breaks are explicit decisions, and the distribution pass incorporates these accidentals exactly once.

Edit the score and the layout is recomputed.
Change a preference and the layout is recomputed.

There is no iterative correction loop, no heuristic adjustment, and no manual cleanup phase. Layout-dependent notation is a deterministic consequence of the score, its breaks, and the engraver’s stated preferences.

I keep an internal development log. It's not public, partly because publishing it would create expectations about sequence and timing that I have no intention of meeting. But something happened to its structure recently.

For over a year, the log tracked numbered phases. Phase 1 through Phase 6, each with clear boundaries, some of them with sub-phases. The numbers implied I knew what came next.

After Phase 6, I was supposed to implement Phase 7 (event-driven client architecture) and Phase 8 (windowing system). Opening a window was 'weeks away'. That's what the plan said.

That's not what happened.

Something – veteran instinct, fifty years of pattern recognition too deep to articulate – said: complete the semantic layer first. Not the plan. Not what 'agile' methodology recommends. But the foundations needed to handle real musical complexity before anything was built on top of them.

So instead of windows, I implemented the full semantic layer. Time signatures, key signatures, remembered alterations – the systems that determine what accidentals print and when. The hard problems in notation software, the ones that have resisted clean solutions for forty years.

I've written about that work as it happened. What I hadn't written about, until now, is what it revealed about the plan itself.

​The development log no longer uses numbered phases for future work.

This isn't abandonment of structure. It's recognition that the work has entered a different mode. The section heading now reads 'Future Stages' with a note: 'Not numbered phases – sequencing may shift based on what the work reveals.'

That sentence took eighteen months to earn. The plan implied I knew the sequence. The reality is that the architecture now tells me what it needs. The foundations are proven; what follows is revelation, not construction. And I now need to listen.

​For eighteen months, the work was architectural: designing possibilities, extending the system, proving it could handle what I asked of it. That mode of thinking was adaptive. It built what now exists.

That mode of thinking is now dangerous.

The guardrail, however, is surprisingly simple. A single principle: if rendering something changes the semantics of what's being rendered, the logic is wrong. Not 'possibly wrong' or 'worth reconsidering'. Wrong as in 'don't go down that path; retrace your steps entirely'.

This sounds obvious, but it's deceptively easy to violate. The instinct to adjust, to consider alternatives, to make something 'look better' by feeding visual decisions back into musical ones – these reflexes served construction. They now threaten what's been built.

The architecture is now closed. It’s no longer fluid in any sense.​ Rendering reveals; it doesn't renegotiate. I may return to this topic, because the implications run deeper than they first appear.

Two weeks ago, the remembered alterations algorithm proved that accidental rendering could be solved deterministically. I wrote about needing to breathe, to get my head around it.

Today, I formally specified measure distribution as ADR-0037. The Knuth-Plass algorithm – the same dynamic programming approach TeX uses for paragraph breaking – applies directly to distributing measures across systems. Not through cleverness, but because Ooloi's five-stage rendering pipeline decouples concerns that traditional architectures leave coupled. By the time the distribution algorithm runs, collision detection and vertical coordination are already resolved. What remains is a one-dimensional sequence with scalar preferences. Textbook problem. Textbook solution.

The algorithm itself is not news. Knuth-Plass has existed since 1981.

The news is that this is the second time.

Two problems the industry treats as inherently heuristic – requiring manual correction, special cases, user-facing options to tune approximations – collapsed into straightforward algorithms. Same architectural properties both times: immutable data, rational arithmetic, explicit stage boundaries, semantic determinism before layout.

Once might be coincidence. Twice suggests the architecture itself is doing something.

For engravers, if this holds: no accidental weirdness. No measure distribution cleanup. No jitter when editing – changing bar 47 won't mysteriously reflow bar 12. Layouts that are globally optimal, not locally adjusted. Proportional spacing preserved by construction. Adjustments for taste, not necessity.

I keep writing 'if this holds' because the proof is mathematical, not yet empirical. The algorithm is sound; the question is whether real scores – Gesualdo madrigals through Berg's Orchesterstücke or Ligeti's Requiem – expose edge cases the staged pipeline doesn't fully decouple. The remembered alterations work is tested. Measure distribution needs equivalent validation.

But the pattern emerging is harder to dismiss than any single result. If two 'impossible' problems turn out to be straightforwardly solvable, how many others?

I don't know yet. That's the honest answer.

Roland Gurt's comment last week – about spacing instability in the programs he uses, where editing one element mysteriously reflows unrelated staves on following pages – started me thinking seriously about this problem. Thanks, Roland.

Happy New Year, everyone.

Today I discovered that measure distribution over systems and pages, thanks to Ooloi’s architecture, can be solved deterministically using the Knuth-Plass algorithm. 

Wow.

Just verified that the Ooloi development and application environment installs and runs on macOS, Windows, and Linux. All tests pass on all three platforms. This is exactly as unsurprising as it sounds when you're running on the JVM.

Expectation is not the same thing as verification. This check removes a class of future problems before they arise. There are no platform assumptions embedded in the codebase. If you can run a JVM, you can run Ooloi.

Roland Gurt, a musician and regular reader here, recently asked whether I could explain some of the computer science terminology that keeps appearing in these posts – terms like functional programming, immutability, and transducers – in layman's terms, and preferably from a musical point of view.

I was genuinely glad to be asked.

At this stage of Ooloi, nothing has yet been drawn on the screen. What has been happening is work on problems that sit underneath notation, problems that have been considered 'mostly solvable' at best since the 1980s. The deterministic accidental system I recently wrote about is one such problem. It looks abstract because it has to be. It deals with musical meaning before anything becomes visible.

This post is an attempt to explain why these terms keep appearing, and what they actually refer to, without turning this into a programming tutorial. It's written for musicians who are curious, not for programmers who already know the vocabulary.

​One reason Ooloi can attempt this at all is that it's a new system.

Most notation programs in use today have roots in techniques from the 1980s and 1990s. Internally, they tend to look surprisingly similar. They represent music in roughly the same way and rely on the same architectural assumptions: object-oriented design, mutable state, and large codebases written in C or C++.

Once a system has grown for decades on top of those assumptions, it becomes extraordinarily difficult to rethink its foundations without breaking everything at once. We saw Finale collapse in 2024 for precisely this reason.

There have been only two real departures from that lineage. One was Igor Engraver in the 1990s. The other was Dorico from 2016. Dorico has much in common with Igor Engraver in terms of musical data representation, and it made important advances. But even Dorico still rests on the technological assumptions of that era: object-oriented design, mutable state, and C++ at its core.

Ooloi puts all its money on a different assumption, namely that those older techniques fundamentally are architecturally inadequate for the problem of music notation as we now understand it. That inadequacy has consequences for users: accidentals behaving oddly, freezes and delays, limited scalability, no real collaboration, and poor use of modern multi-core machines.

​To address these problems, Ooloi uses an entirely different set of techniques.

​It also needs to be said very clearly that music notation, from a computer science perspective, is an extremely hard problem.

It's not 'a word processor for music'. That analogy has done lasting harm.

Music notation sits in the same general complexity class as compiler construction, symbolic mathematics systems, or advanced typesetting engines. It combines several difficult aspects at once: hierarchical structure, temporal meaning, long-range context, simultaneous layers, and a visual representation that's not a straightforward reflection of the underlying data.

A compiler reads a stream of symbols, builds meaning from context, enforces consistency rules, and produces deterministic output. Music notation does all of that, and then adds engraving conventions, historical practice, performer expectations, and visual clarity on top.

If you approach this domain with tools designed for relatively static documents or interactive graphics, you end up fighting the problem instead of modelling it. The result is systems that work impressively most of the time, but fail in edge cases, which in music aren't rare.

​Before any technical terminology enters the picture, there's a simple musical requirement.

If you engrave the same piece twice, you should get the same result twice.

Not roughly the same. Not 'close enough'. The same.

Most notation software quietly gives up on this requirement. Instead, they rely on rules of thumb: 'this usually works', 'this is how engravers tend to want it', 'clean this up manually if it looks odd'. Musicians have learned to live with that because, for a long time, there was no alternative.

From a musical point of view, this is deeply unsatisfactory. It means the system doesn't really know what it's doing. When things get complicated, it starts guessing. And once guessing enters the picture, everything downstream becomes fragile.

Determinism simply means that given the same musical input, the system makes the same musical decisions every time.

​Imagine packing a suitcase before a trip. You put clothes, toothpaste, shoes inside. Once the suitcase is closed and checked in, you make decisions based on what you packed. You don't buy extra clothes at the airport because you trust they're already there.

Now imagine that, while the suitcase is in transit, someone silently changes its contents. In a trivial case this is merely annoying. In a serious system – air-traffic control, medical records, nuclear command – it's catastrophic: people might die.

To reason about anything, you must be able to trust the past.

In computer science, immutability means this: once a thing has been created, it never changes. If you need a slightly different version, you don't alter the original. You create a new one. The old one remains exactly as it was.

This sounds wasteful until you realise that modern systems can do this efficiently by sharing structure internally. This is where a bit of Harry Potter magic comes in. If someone changes one shirt, they get a new suitcase with just that one shirt changed, while you still have your original suitcase, untouched and with the original shirt still inside.

In music notation this matters because the domain is full of remembered context. Accidentals depend on what happened earlier. Key signatures establish expectations. Voices interact. If earlier musical facts can silently change while later decisions depend on them, determinism collapses.

​Most people already understand from school what a function is.

If x is 100, the result is always 101. Not sometimes. Always.

That property is determinism.

Now imagine that x is not a number but a suitcase. If the function's result depends on what is inside the suitcase, you must be absolutely certain that the contents haven't changed since last time. And for this, the suitcase needs to be one of our magical suitcases. Otherwise, calling the same function twice with what looks like 'the same input' might silently produce different results. The consequences range from a missing accidental to obliterating Tokyo.

Functional programming is simply the discipline of insisting on this property at scale. Functions don't reach out and change things behind your back. They don't depend on hidden mutable state. They take inputs and return results without any such surprises.

In Ooloi this matters because musical decisions must be reproducible. If accidental decisions or tie behaviour depend on invisible state changes, the system cannot be fully trusted, even if it usually looks right.

A central idea in Ooloi is to treat the score as something that flows.

Music is read from left to right. Context accumulates. Decisions depend on what has already passed. Accidentals are remembered and later forgotten. Grace notes steal time from what they follow. Tuplets locally distort time.

This is not a static tree that you occasionally inspect. It's musical meaning - semantics - unfolding over time.

Once you accept that, you need a way to traverse this stream efficiently without constantly rebuilding lists or retracing paths.

​And this brings us to transducers.

​Imagine a high-security prison complex with multiple buildings, floors, and endless cells. Every day, a warden must traverse the entire structure to take stock of what's inside. And he must do it in the same order each day, just as the notes in a piece must be played in the same order each time.

Most of the warden's effort goes into navigation: remembering routes, tracking where he's been, writing lists just to keep his place. He does this over and over again, every day.

A transducer is like giving the warden Ariadne's thread. But again, there's magic involved: the thread weaves from cell to cell of its own accord. The route through the maze is fixed and reliable. Navigation stops being the problem. 

Better still, the warden doesn't even have to walk, thanks to more magic: the thread acts like a firehose and sends back all the information to the origin. The warden can sit in his office while the traversal happens and receive a continuous stream of information about what's found along the way. He can devote his entire attention to the meaning of the data, not how to walk the prison complex.

The crucial point is that traversal and processing are separated. The route is handled once. The real work happens as data flows past, without intermediate lists or repeated navigation.

In Ooloi, transducers allow musical meaning to be processed in exactly this way. The system doesn't build a large structure and then analyse it afterwards. It reacts to musical facts as they appear, deterministically and efficiently.

This is what a transducer is. Then there's even more wizardry involved in that several transducers can be stacked on each other to perform even more abstract operations. The path has been abstracted away, the distractions are gone, and the flow of  meaningful musical data is all that matters.

​These terms aren't ideology, and they're not badges of virtue. They're simply the most precise language I've found for describing how musical meaning behaves when you try to handle it rigorously.

The music comes first. The vocabulary follows.

If anyone wonders what the above is, it's the deterministic solution to a hairy engraving computer science problem nobody thought was solvable. It's been like that since the 80s. Everybody has just kind of capitulated and lived with a 'it's 95% okay, the rest is impossible' type of mindset and then, as one must in that situation, used rules-of-thumb to reach that 95%. Which is a lot, and a very respectable achievement.

However, it turned out that Functional Programming and immutability - and non-consing push transducers - could solve the problem deterministically. For the first time ever.

​I think it holds. We shall see.

What it would mean for notation? No accidentals weirdness. No cleanup. Consistency. Adjustments for taste, not necessity.

The power of Clojure, of treating the piece as a transforming semantic stream, and of not working with OOP and mutable state.

Now I need to sit down and breathe for a while. Time to go into Hammock Mode. I need to get my head around this.

​When I began coding Igor Engraver around 1995, the choice of platform was straightforward. Macs were where creativity lived. Windows – clumsy, unintuitive, user-hostile – was for accountants and management consultants. I needed to escape Finale's stranglehold, and I needed the best possible foundation for professional music engraving.

That foundation was QuickDraw GX.

Apple had released something genuinely remarkable: a complete 2D graphics and typography system with sophisticated font handling, Bézier curve operations, transformation matrices, and sub-pixel anti-aliased rendering. For music notation – which is essentially complex typography with thousands of precisely positioned curves – GX was perfect. Not adequate, not sufficient: perfect.

Igor Engraver was built on QuickDraw GX from the beginning. Mac-only, by choice and by necessity. Windows didn't matter. We founded NoteHeads, shipped the software, and believed we'd eventually need to address cross-platform support. But that was a distant concern.

​AlphaMask Inc. was founded in 1999 by Mike Reed and Oliver Steele. Reed had been the tech lead on Apple's TrueType and font system. Steele had been on the QuickDraw GX development team and had led the Apple Dylan project at Apple Cambridge – the former Coral Software, where Macintosh Common Lisp originated.

The people who built QuickDraw GX had left Apple and founded a company to continue that work. When Apple made what I considered a profound mistake in abandoning GX for OS X, the GX team apparently agreed – to the point of leaving Apple entirely to focus on their superior graphics engine.

Whether I knew about Steele's Lisp background when we chose AlphaMask, I honestly cannot recall. I like to think the choice was purely on merit: AlphaMask offered GX-level capabilities in a more decoupled, portable form. It did what we needed. The fact that someone who understood both graphics and Lisp had designed the API might explain why it integrated so cleanly with our Lisp codebase, but that may simply be a pleasant historical detail rather than a decision factor.

Either way, when QuickDraw GX disappeared, I had unknowingly followed the people whose work I trusted.

​Years later, when designing Ooloi, I chose Skia as the graphics foundation. Modern, open-source, GPU-accelerated, excellent typography, sophisticated path operations, cross-platform. I chose it on technical merit, comparing it against alternatives and finding it superior.

I had no idea that Skia was founded by Mike Reed and Cary Clark – another QuickDraw GX team member – a few years after AlphaMask. Or that Google had acquired Skia in 2005 and made it the graphics engine for Chrome, Android, and Flutter. Or that billions of devices now use Skia for their rendering. Or that the internal name at Apple for Quickdraw GX was - Skia.

QuickDraw GX has had three incarnations: first as itself, then as AlphaMask, then as Skia. The same design philosophy that made GX excellent – abstract graphics model, resolution independence, professional typography – survived through each transformation. I recognised that quality in 1995, in 2000, and in 2025, without realising I was choosing the same team's work each time.

Perhaps this indicates that certain kinds of graphical excellence are simply necessary for music notation, a constant need that has persisted since the last millennium. Or perhaps I'm simply stubborn enough to arrive at the same solutions regardless of how much time passes.

​Another detail emerged from the research. AlphaMask was acquired by OpenWave around 2001–2002, and the desktop product was discontinued. OpenWave wanted the technology for mobile browsers, not for professional graphics applications. Support ended, updates ceased.

2002 was also when NoteHeads fell silent.

Whether that timing was coincidental or causal, I cannot say with certainty. Finding a replacement for AlphaMask's capabilities in 2002 would have been extraordinarily difficult – arguably impossible. The engineering effort to rebuild on different foundations would have been substantial. Perhaps the ponytailed pop zombies running NoteHeads at that point gave up when the graphics engine disappeared. Perhaps they simply declined to invest in solving the problem. I don't know if we'll ever have a definitive answer, and frankly, the question is less interesting than the pattern it reveals.

​The reassuring aspect of this circle is that it cannot break the same way again.

Skia powers the rendering in Chrome, Android, Flutter, and countless other applications. It has billions of users. It's open-source, BSD-licensed, maintained by Google and a broad community. Even if Google stopped development – which won't happen, as Android depends on it – the codebase is available, the expertise exists, and the user base is large enough that maintenance would continue.

Similarly, Ooloi runs on the JVM, which has multiple vendors: Oracle, Azul, Amazon, Microsoft, IBM, Red Hat, Eclipse. Battle-tested is a trite phrase, but it's accurate here. The JVM has been refined for nearly three decades across billions of deployments. It provides capabilities – proper concurrency models, cross-platform consistency, mature tooling – that enable much of Ooloi's architecture.

Everything Ooloi depends upon is either open-source with massive adoption or has redundant commercial vendors ensuring longevity. This isn't accidental. This is architectural design informed by what happens when foundations disappear.

Looking back across thirty years, there appears to be a unifying pattern that I wasn't consciously aware of whilst making these decisions. A consistent need for graphical and typographical excellence. A recognition of quality when it appears, regardless of who built it or where it came from. A preference for sophisticated abstractions over quick implementations.

Perhaps I've learnt something during that time about building software that endures. Or perhaps I'm simply persistent enough to keep arriving at similar solutions when faced with similar problems. The distinction might not matter.

What matters is that the circle closes. The technology that made Igor Engraver possible in 1995 has evolved, through the hands of its original creators, into the technology that makes Ooloi possible in 2025. And this time, the foundations cannot be deprecated on a whim or acquired into oblivion.