A music-industry legal workspace.Built by the people who needed it first.

The notation nobody teaches

A classical training buys fluency in one difficult notation system. Years of it, and a page of music reads the way a headline reads. Then you start releasing records, and you find there is a second written system running underneath the first. It is the one that decides whether any of the work pays.

Nobody teaches that one.

The early contracts went unread. Several hundred an hour to have somebody explain a deal worth less than that is not a real choice, so most people sign anyway. So did I. What that costs does not arrive at once. It arrives years later, in a royalty statement that does not match what you remember agreeing to, or in a recording you cannot license because of a sentence you skimmed.

Consulting is where it stopped being personal. Several hundred artists later, the same two clauses kept arriving with the same misreadings attached: a reversion clause worded so it could never trigger, and recoupment padded with every marketing and video cost the label could attach, uncapped, while its own margin stayed untouched. These were not careless people. They were fluent musicians and complete beginners at the only document governing their income, priced out of that fluency on the exact day they needed it.1

It is not only the artist side of the table either. A label with four staff signs upward to a distributor on terms it has no more leverage to argue with than any artist does. An agent confirms a festival on a deal memo nobody reads twice. A manager carrying twelve careers has counsel on retainer for none of them. Being the counterparty in somebody else’s deal does not mean you had counsel in your own.

One thing worth having, free

The most useful thing anybody taught me costs nothing, so take it whether or not you ever use this product. Move the negotiation into writing as early as politeness allows. On a call the more experienced party sets the tempo, and they will use it: you answer at their speed, on their ground, and nothing said is written down. In writing, the slower party sets the tempo. You can take an hour. You can take a day. You can ask somebody who knows before you answer. That is the whole advantage, and it disappears the moment you agree to talk it through.

Who is here

The company is in New York. The rest of us are spread across several countries, and all of us arrived from inside the business rather than from software. Artists who signed the bad version. Managers who found out where a term sheet is soft by conceding it first. One of us spent years certain they were owed royalties and unable to work out by whom, which is a particular kind of helplessness: not being cheated so much as being unable to find out. The audit tools exist because of that.

Nothing here came off a roadmap. Every feature was somebody’s bad week.

Why it distrusts its own output

There are a great many contract apps now. Most were built in a weekend on a general model, and they stop at reading a PDF back to you. Nearly all of them share one design assumption, which is that the model is right.

This one started at the other end. Automated assessment of contract fairness is, across the field, wrong about roughly one clause in five.2 That number became a fixed input rather than something to optimise away later, and it changed what got built. A layer sits between the model and anything you are shown, asking whether a quoted clause is genuinely in your document, and whether a figure follows from the ledger it came from. To find out whether it worked, ten known faults went into a document that had independently been declared clean, with the requirement that the layer catch all ten and raise nothing else.

The engine that is switched off

There is also an engine that grades how fair a deal is, and it has never been switched on. Its first live judgement was wrong, on real data, and would have accused a paying customer of fraud. So it sits inside the product recording what it would have said and changing nothing you see, until it earns the right to speak.3

Restraint is not a shortage of ambition. It is the only honest position for a system whose worst failure would be a confident accusation against somebody who did nothing wrong.

What we will not do

WALDHORN.AI is not a law firm. There are no attorneys on staff, and nothing the product tells you is legal advice. It was built with one: a music attorney of more than twenty years, hundreds of these agreements behind them and probably thousands, who sat with us through every clause we had misjudged. There were more of those than anyone expected.

Introductions to counsel work the same way. No fee share, no success fee, no paid placement, and no cut of anything you sign. The pilot is New York only and invitation only, and it will grow slowly for the same reason the grading engine is still switched off.4