anorthernman.

Two oscillators, one detuned a hair from the other. That is the entire instrument.

Two oscillators at the same pitch (220Hz, an A), one nudged six cents sharp. They cannot agree, and the disagreement is audible as a slow beating — the drift is not a bug in the code, it is the code.

It is the 6. At six cents the beat is a slow tide, one pulse every couple of seconds. At 60 — ten times bigger — the two notes stop fighting each other and start sounding like two separate notes, which is a different instrument. At 0.6 the drift is so slow you will check whether it is running. It is; give it a minute.

North is six of these, not two, each drifting independently rather than sitting at a fixed offset, plus a feedback delay so the room keeps a little of what already happened. The extra voices and the delay are refinement. The instrument — two oscillators, one detuned — is entirely in the code above.

Twenty lines was the promise. Here they are, with the one number actually worth touching.

Paste that into the p5.js editor and press run. Draw a short line, turn a little, repeat — draw() fires again every frame, so “repeat five thousand times” happens on its own in about ninety seconds.

It is turn. At 0.15 the line wanders like someone thinking while they walk. Push it to 1.5 — ten times bigger, not ten per cent — and the wander becomes a scribble that still won’t cross its own path cleanly. Drop it to 0.015 and you get a lazy river. Same twenty lines. Three different pieces. The only thing that moved was one float.

Then break it, per pass two’s five questions. Resize the window mid-draw and watch it recentre wrong. Hold the mouse still — it was never listening to the mouse, which tells you where the actual input is: nowhere, on purpose. Leave it running and it eventually walks off every edge, because a rule that only turns has no reason to come back.

The tools behind series two, in one list, so nobody has to go hunting for a name mentioned in passing.

Most of that afternoon gets spent getting over the belief that you need an idea first. You do not need an idea. You need a rule small enough to say out loud.

“Draw a line from where the last one ended, turn a bit, repeat.” “Play a note every time the number is prime.” “Make the page darker for every second nobody moves the mouse.” Each of those is a piece. None of them is interesting yet. Interesting arrives with the second half of the method, which is: run it, change one number, run it again.

Change the number by a lot. Ten times bigger, not ten per cent bigger. Most of what gets called taste in this medium is really a willingness to try the absurd value, notice that it is better, and keep it.

Three rules to steal, if you would rather not write your own.

Then publish it with a number and a date. Keep the one that failed and label it as failed. A body of work is mostly a record of attempts, and the tidied version is always the less useful document. The kit list for all three starters is on 0168.

Start with the bad news, because it is structural. This work runs on something, and the something gets discontinued.

Flash went in 2020 and took a decade of the medium with it. Java applets went earlier. Shockwave, QuickTime VR, the browser plugin as a whole idea. Museums that collect this work are in the emulation business whether they like it or not, and a fair share of the canon now survives as video documentation of an interactive piece — which is roughly a photograph of a conversation.

Which is why a link list in this field has to be dated and corrected in public rather than quietly tidied. This one was checked on the date above. When an address dies it is struck through and left here, not removed.

• Rhizomerhizome.orgThe field’s institutional memory since 1996, and the people doing the serious preservation work.
• Net Art Anthologyanthology.rhizome.orgA hundred works restored and given context. Start here if you start anywhere.
• Whitney Artportwhitney.org/artportA museum commissioning for the browser since 2001.
• Ars Electronicaars.electronica.artLinz, since 1979. The archive is the useful part, not the festival photographs.
• transmedialetransmediale.deBerlin. More argument, less spectacle.
• V&Avam.ac.ukCollects the design and the machinery — the part everyone else throws away.

Reading. Manovich, The Language of New Media, for the vocabulary; be patient with it. Christiane Paul, Digital Art, for the map. Reas and Fry, Processing, for the practical way in. Fuller, Software Studies, when you want the argument rather than the survey.

House note: this series is a working guide, not a syllabus. It will be corrected in place, with the corrections kept visible, as is the custom here.

You cannot read this work the way you read a picture, because there is no single state to read. Eight seconds in front of a generative piece gets you eight seconds of an argument that runs for ever.

So: questions rather than glances. These five get me most of the way.

• 1 · What is the system?Say the rule in one sentence. “Draw a line, turn a little, repeat.” If you cannot get it into a sentence, either the piece is badly made or you have not watched it long enough. Usually the second.
• 2 · What is the input?Time, a seed, the mouse, a camera, a live feed, a crowd. The input is the real subject. A piece fed by a stock ticker is about money whatever the wall text says.
• 3 · What does it do when nobody is watching?Some pieces perform — they wait, they react, they need you. Others simply continue, and you are a witness rather than a participant. Different kinds of work, and the difference is usually the point.
• 4 · Where does it end?Loop, exhaustion, or refusal. A loop says the artist chose a length. Exhaustion — it uses up its material and stops — is the most honest ending there is. Refusal is the hardest to build and the easiest to claim.
• 5 · What happens when it breaks?Ask it of the piece and of the medium. Every work here breaks eventually, because the runtime it was written for gets switched off. How it behaves at the edges tells you what it is made of.

The method, such as it is: give it three minutes rather than eight seconds, and note the exact moment you first predicted it correctly. That is where the system became legible. Everything after it is the artist’s decision about whether legibility was the goal.

Then break it on purpose. Resize the window, cover the camera, hold the mouse still, leave the room. Well-made work has an answer for all four.

Two things on this site to practise on, since they cost nothing. North, on 91.5, is six oscillators drifting slowly out of tune with one another; the rule takes one sentence and the piece never resolves, on purpose. The field checker on 96.2 reads the page you are actually looking at rather than the colours anybody intended — its input is the thing you are inside.

Most art you can stand in front of. This kind you have to run. That is the whole difficulty and most of the appeal.

A painting is finished when the painter stops. A piece of new media art is finished when somebody writes the last rule, and after that it carries on without them — differently on Tuesday than it went on Monday, differently on your machine than on mine.

The definition that has held up: the computer is the material, not the tool. A photograph edited on a laptop is a photograph. A page that draws a different picture every time you open it is something else. The test is simple — take the computation away. If the same work is still there, it was a tool. If nothing is left, it was the material.

None of this is recent. Vera Molnár was writing rules for a plotter in 1968. Nam June Paik put a magnet on a television in 1965 and bent the picture into a shape no broadcaster had authorised. Lillian Schwartz made films out of borrowed mainframe time. By the mid-nineties the browser was a gallery with no invigilator, and net.art happened in it — mostly badly, occasionally brilliantly. Processing arrived in 2001 and turned “learn to program” into “learn to draw”, which is a different sentence entirely.

The family names, said plainly. Generative: the artist sets the rules, the system does the work. Interactive: the viewer completes it, and without a person in the room there is only apparatus. Participatory: many people complete it, and can see each other doing it. Data-driven: the input is a real measurement — tides, traffic, a price, an inbox. Glitch: the failure is the subject, not the decoration. Software art: the code is the piece and reading it is part of looking. Real-time: it is happening now, and no recording of it counts.

Most decent pieces are two or three of those at once. The labels matter less than working out which one the artist actually cared about.

What makes it awkward: it needs electricity, it has no edition, it ages like milk, and the thing you saw in 2004 is now a screenshot on somebody’s dead blog. What makes it worth it: done properly, the work can surprise the person who built it. That is rare in any medium and close to routine in this one.

A feed says everything is the same kind of thing, arriving in one order, and that the newest is the most important. None of that is true of the work.

So: stations. One for writing, one for things you listen to, one for things being tried, and one for the proprietor. Tune across the top. You always know which one you have open.

The drawers from the last build are still here — they have been filed into the paper you are reading, station by station. The plates run as pictures, the audio runs as records, and the writing runs as what it always was: posts, kept in order, kept in view.

Red is not a section colour. If it were, it would mean Music rather than look here.

On this build it marks what is live: the wordmark, the tuning needle, the lamp when we are on air, and one focal thing per station. Nothing else. If two decorative things on a page are red, one of them is wrong.