The Clearinghouse Has a Valuation Problem

In March, two Georgia athletes submitted third-party NIL deals for airline travel vouchers worth $4,400 and $4,800. NIL Go denied them. The deals fell outside the range of compensation the clearinghouse deemed reasonable for athletes in their position.

The athletes appealed to arbitration. Before the case was decided, the College Sports Commission updated its range-of-compensation model. Under the revised model, the same two deals fell inside the range. The CSC informed the arbitrator and moved to dismiss. The arbitrator denied the motion and ruled for the athletes anyway.

Nothing about those deals changed between March and June. The vouchers were the same vouchers. The athletes were the same athletes. What changed was the model.

That is the clearinghouse's central problem, and it has almost nothing to do with whether a cap on athlete compensation is a good idea. It is a problem of disclosure. The College Sports Commission now holds the most complete record of college athlete NIL transactions that has ever existed — tens of thousands of contracts, with counterparties, deliverables, and dollar figures attached. It publishes bimonthly totals from that record and essentially nothing else. Every structural weakness in the system traces back to that gap.

How the process actually works

Under the House settlement framework, any third-party NIL agreement worth $600 or more — including multiple agreements with the same party totaling $600 — must be submitted to NIL Go. Submission is the athlete's obligation. Agents can prepare the paperwork, but the athlete presses the button.

From there, the deal runs three gates.

Gate one: payor association. Is the party paying the athlete an "associated entity" — a booster, a collective, or a business whose relationship to the school is close enough to raise circumvention risk? This determination governs everything downstream. Deals from genuinely unassociated payors clear on the business-purpose test alone and never face a valuation review at all.

Gate two: valid business purpose. Is the payment tied to the promotion or endorsement of goods or services offered to the general public for profit? An autograph session qualifies. A commercial qualifies. A speaking appearance qualifies. Paying an athlete for NIL rights with no plan to use them does not.

Gate three: range of compensation. Only associated-entity deals reach this gate. A 12-factor framework benchmarks the proposed payment against historical deal data drawn from both college and professional athletes. Publicly described inputs include athletic performance, social media reach, local market size, and the reach of the athlete's institution within that market. Roster value and recruiting incentives are excluded by design.

A deal comes back cleared, in review, or needing information. An athlete holding a deal that doesn't clear has four options: revise and resubmit at a compliant number, take it to a neutral arbitrator, proceed anyway and accept eligibility risk, or walk away.

Two things about that structure deserve more attention than they usually get.

First, gate one does most of the work. The valuation question — the part everyone argues about — is downstream of a classification question, and that classification has been the single most contested issue in the entire post-settlement era.

Second, the range of compensation is not a fair market value determination. The CSC has been careful about this language. It does not calculate what an athlete is worth. It determines whether a submitted number sits inside a band of what comparable athletes have been paid. Those are different exercises with different failure modes, and the difference matters enormously once you start asking what happens at the tails.

What a year of deal flow shows

The CSC publishes deal-flow reports on a rolling basis. Stack them and a few patterns emerge.

Rejected deals are enormous relative to cleared ones. In the January–February window, the average cleared deal was roughly $10,600 and the average not-cleared deal was roughly $76,800 — about seven times larger. In March–April, the gap narrowed but held: roughly $13,700 cleared against roughly $60,800 not cleared, a ratio of about four and a half to one.

That single ratio explains more about the system than any policy statement. The clearinghouse is processing an enormous volume of small, ordinary commercial deals and clearing nearly all of them. Its enforcement energy is concentrated on a small number of very large agreements, most of them from associated entities, most of them from two conferences. In late May, the CSC reported that more than 75 percent of all submitted deals came from the Big Ten and SEC.

Deal sizes are climbing. Through October 2025, the CSC reported an average cleared deal of about $7,190. By March–April 2026 the two-month average was roughly $13,700. Cumulatively through July 1, the average cleared deal sits near $10,400.

Throughput is getting worse, not better. In the fall of 2025, the CSC reported 53 percent of deals resolved within 24 hours and 74 percent within a week. In the July 2026 report, those figures were 41 percent and 63 percent. The system is slower now than it was nine months earlier, while handling more money.

The review floor keeps rising. At launch, everything at $600 and up faced review. In April 2026, the CSC exempted deals between $600 and $2,500 from range-of-compensation review. Effective July 1, 2026, deals from $600 to $15,000 skip range-of-compensation review entirely until an athlete crosses $50,000 in total associated-entity deals within an academic year.

That last change is often described as the CSC easing up. Read against the throughput numbers, it looks like something else: a capacity decision. A system that could not clear deals fast enough stopped reviewing most of them.

Three structural problems

The model moves, and nothing records that it moved

The Georgia arbitration is the clean example, but it is not an isolated one. The CSC's own explanation was that as more deals clear, it updates its dataset, and when new data showed those deals were in range, it acted. That is a defensible thing for a valuation model to do. Models should absorb new data.

The problem is that a model which absorbs new data is a model whose outputs are only valid as of a date — and NIL Go's outputs are not dated, versioned, or published. An athlete who received a denial in March has no way to know whether that denial would survive the June model. Nobody re-ran the March denials against the June model. The two athletes who happened to be in arbitration got the benefit of the update. Everyone who took the denial and revised downward did not.

For a valuation shop, this is the most familiar failure mode there is. Any model that ingests new observations will move. The discipline isn't preventing movement — it's stamping every output with the model version that produced it, and having a policy for what happens to prior outputs when the version changes. NIL Go has neither.

A denial is a verdict without a rationale

An athlete whose deal does not clear learns that it does not clear. What they do not receive is a reasoned explanation: which factors drove the result, which comparable set was used, what number would have cleared.

Without that, the revise-and-resubmit path is guesswork. The athlete's only real signal is the denial itself, so the rational response is to resubmit at a substantially lower number and hope. That mechanism systematically pushes deals below the band rather than to it, which is precisely the outcome critics have in mind when they describe the clearinghouse as suppressing the market.

It also makes arbitration nearly the only route to a substantive answer — an expensive, slow route that, as of the CSC's May reporting, had been used by just 21 deals consolidated into three cases out of more than a thousand denials.

The hard part of the problem is invisible

Here is where we should be honest about our own position. TNS builds valuations from public signal: production, usage, positional context, program and market factors. Deloitte and the CSC have something we do not — the actual contracts. On the raw input question, they win, and it isn't close.

Which makes the disclosure gap more frustrating, not less. Because the genuinely difficult problems in this exercise are not data problems, and anyone who has built a compensation model can name them:

"Similarly situated" is doing almost all the work. At the middle of the distribution, comparable sets are deep and a band is meaningful. At the top, they collapse. There may be a handful of true comparables nationally for an elite quarterback in a large media market. A band built on a comparable set that small is not a market estimate; it is an average of three or four negotiations, at least one of which was probably itself a circumvention attempt.

Market size and market reach are different variables and the difference is not intuitive. The reach of a flagship program in a mid-size city routinely exceeds that of a smaller program in a top-ten metro. The framework accounts for this in principle. Whether it does so well is unknowable from the outside.

Positional scarcity is not linear. The commercial market for athletes does not distribute value the way the on-field market does. A model calibrated to one and applied to the other will systematically misprice entire position groups — and no one outside the CSC can check whether that is happening.

Feedback contamination is a live risk. The model is trained on cleared deals. Cleared deals are, by construction, deals the model already approved. Absent deliberate correction, a system like that drifts toward its own priors and gets more confident while getting less accurate. Whether the CSC corrects for this is, again, unknowable.

Every one of these is a legitimate modeling challenge that a competent team could be handling well. The point is that the market is required to take that on faith.

Four things that would fix most of this

None of the following requires reopening the settlement, weakening the cap, or exposing a single athlete's contract terms.

Publish aggregate range bands. Not per-athlete outputs — distributional ones. Median, 25th, and 75th percentile cleared deal values by sport, by position tier, by payor association status, by conference. This is the disclosure a valuation model owes the market it governs, and it can be done at a level of aggregation that protects every individual deal.

Version the model and date every decision. Every clearance and denial should carry a model version and effective date. This is a one-line change with outsized consequences: it converts a black box into an auditable one without revealing a single weight.

Re-review on material model change. When the model moves enough to reverse prior outcomes, prior denials in the affected band should be automatically re-run and the athletes notified. The Georgia case established that outcome for two athletes who had lawyers and an arbitration docket. It should not require either.

Issue reasoned denials. A short written rationale — the governing factors, the comparable basis, and the value that would have cleared. This converts revise-and-resubmit from guesswork into an actual negotiation and would likely reduce arbitration volume rather than increase it.

Taken together, these are transparency measures, not deregulation. A regulator with better disclosure is a regulator that is harder to sue and easier to defend.

What's still moving

Three things could change the shape of this within months.

The associated-entity fight is settled for now, but only for now. In June, the special master overseeing the House settlement declined to categorically exclude multimedia rights companies or third-party brand sponsors from the associated-entity definition, holding that the question requires individualized review. In August, Judge Claudia Wilken upheld that reasoning. She also ordered the CSC to respond to class counsel's requests for information about how it classifies MMRs in its investigations — the first meaningful crack in the CSC's disclosure posture, and it came from a court rather than from the CSC.

The architecture is under direct antitrust attack. A suit filed in the Northern District of California in June by two current Power Four football players names the NCAA, the CSC, the four power conferences, and their senior administrators, challenging the cap-enforcement structure under federal antitrust law and state NIL statutes. It is early. It is also exactly the challenge the "black box algorithm" critique was always building toward.

Congress is one vote away from rewriting the definitions. The Protect College Sports Act advanced out of Senate Commerce on a 19–9 vote, picked up SEC and Big Ten endorsement in late July, and was revised on August 4 to adopt the House settlement's associated-entity definition, harden the revenue-share cap, count associated-entity deals against it, and add certification requirements for multimedia rights holders, sponsors, apparel companies, and vendors. It did not reach a floor vote before the August recess. It is on the September calendar.

If that bill passes in its current form, the classification question at gate one largely goes away. The valuation question at gate three does not. It gets bigger, because more deals land inside the cap and more of them need a defensible number attached.

The bottom line

The College Sports Commission is winning. It won its first arbitration. It won the associated-entity ruling and then won the appeal. It has cleared more than $355 million in deals and built the only comprehensive NIL transaction dataset in existence.

It is winning with a valuation model that no one can inspect, that changes without notice, and that issues verdicts without reasons. That combination is survivable while the CSC keeps winning. It is not survivable through a bad ruling, a hostile statute, or a single well-documented case of a model error costing an athlete real money.

The fix is not a better algorithm. The CSC probably has a reasonable one. The fix is showing the work.