Who Pays the Regulator? What FDA User-Fee Budgets Can—and Cannot—Show

The companies regulated by the Food and Drug Administration help finance parts of its work. That is a documented feature of federal policy, not a discovery that requires speculation. In February 2026, the Government Accountability Office reported that FDA funding had shifted from predominantly general appropriations toward a nearly equal mix of appropriations and user fees over fiscal years 2008–2024. The agency’s responsibilities expanded during that period, and GAO also described staffing and operational challenges. GAO’s review of FDA responsibilities and funding.

The difficult question is what follows from that arrangement. A funding table can establish how much of a defined program was supported by fees. It cannot, on its own, show that a particular reviewer changed a scientific conclusion, that a company purchased an approval, or that every regulatory activity has the same financial dependence. Conversely, saying that fees are authorized by law does not eliminate legitimate questions about incentives, accountability and the balance of public priorities.

Those questions become more useful when separated. Who pays? Which activities may use the money? What performance commitments accompany it? What evidence would show whether a decision was improperly influenced? Each question has a different documentary trail. Combining them into one percentage produces an argument that can sound stronger than its evidence.

This article follows those trails through actual FDA financial reports, performance documents and independent oversight findings. It also uses original, explicitly hypothetical ledgers to show how misleading percentages and premature performance conclusions arise. It makes no allegation about an individual employee or company, and it does not offer advice about whether to use a medical product. The subject is how to evaluate the institution’s funding and the claims made about it.

1. “User fees” is a family of programs, not a single account

FDA publishes separate financial reports for prescription drugs, medical devices, generic drugs, biosimilars, animal drugs and other fee programs. Its reporting overview distinguishes collections, related expenses, carryover and the personnel and operating uses of funds. A report for one program should not be substituted for a financial statement covering the entire agency. FDA’s user-fee financial reporting overview.

The first step in reading a claim is therefore to identify the program. PDUFA refers to the prescription-drug user-fee framework. GDUFA concerns generic drugs; MDUFA concerns devices; BsUFA concerns biosimilars. These names are not interchangeable descriptions of one pool of money. The legal definitions and reporting boundaries determine what belongs in each program’s figures.

A congressional explanation of the four human medical-product programs describes two funding streams: general appropriations and fees paid by regulated industry. It also explains that the authorizing framework and annual appropriations have different roles in permitting collection and expenditure. The version cited here is the April 2024 report, used for that framework rather than as a source of current-year dollar totals. Congressional Research Service, FDA Human Medical Product User Fee Programs.

For a reader, the distinction is concrete. A statement about the prescription-drug review program may not describe food inspection, tobacco regulation or every function within a drug center. Even a center-level figure and a legally defined program figure can have different boundaries. The answer to “what percentage is industry-funded?” depends on which activity is in the denominator.

An honest table heading should name the program, fiscal year, accounting measure and funding source. “User-fee obligations in the PDUFA program, fiscal year 2025” tells the reader what is being measured. “Industry money at FDA” leaves too much unstated. Before debating whether a percentage is high or low, establish that the numerator and denominator refer to the same activity and period.

2. Read the actual 77-percent figure with its denominator attached

FDA’s fiscal year 2025 PDUFA financial report lists $1,359,676,908 in user-fee obligations and $398,956,452 in non-user-fee obligations, totaling $1,758,633,360. Its table eleven rounds the user-fee share to 77 percent. That is a share of the obligations recorded for the PDUFA program, not a percentage of all FDA funding and not a measure of approvals obtained by fee-paying companies. FY 2025 PDUFA Financial Report, table eleven.

Dividing the reported user-fee amount by the reported total gives approximately 77.31 percent. This is our arithmetic on those two published values. The calculation does not add information about the scientific merits of any application. It tells us the composition of a specified financial total.

The word “obligations” also matters. It describes commitments of funds under the report’s accounting, not necessarily cash paid out during that same year. Replacing it with “revenue,” “profit” or “payments to reviewers” would change the claim. A financial measure should keep its original label unless the writer explicitly reconciles it with another measure. CBO’s explanation of obligations and outlays.

The denominator is especially important when the number travels. A social-media graphic might retain 77 percent while dropping “PDUFA program,” “obligations” and “FY 2025.” The number would remain recognizable but no longer support the broader claim. Readers should be able to recover the table and reproduce the fraction without guessing which total the author used.

A useful correction is not to replace an overstated claim with an equally sweeping reassurance. The program’s substantial reliance on user-fee funding is real. The appropriate conclusion is that this defined activity depends heavily on that funding source, creating a legitimate reason to examine the accompanying rules, priorities and safeguards. Whether the arrangement produced a particular harmful decision is a separate empirical question.

3. How a correct program percentage becomes an incorrect agency percentage

Consider an entirely fictional regulator with two programs. Program A has $80 million in fee-funded obligations and $20 million in general-funded obligations. Program B has $10 million in fee-funded obligations and $90 million in general-funded obligations. These are invented amounts, not FDA data.

Program A is 80 percent fee-funded. Program B is 10 percent fee-funded. The combined regulator has $90 million in fee-funded obligations out of $200 million total, so its combined share is 45 percent. Saying that the regulator is 80 percent fee-funded would use the correct percentage for the wrong scope.

Now change the relative sizes. Keep Program A at $100 million total, but make Program B ten times as large: $100 million in fees and $900 million in general funds. The combined fee share becomes $180 million divided by $1.1 billion, approximately 16.36 percent. The simple average of the program percentages is still 45 percent, but it no longer describes the combined funding composition.

The reason is weighting. Each program’s share contributes in proportion to the program’s total size. Averaging percentages without the underlying amounts can give a result that is mathematically neat and substantively wrong. A comparison of agencies, centers or years should retain the amounts used to construct the percentages.

There is a second trap when programs move between reporting categories. If a mostly general-funded activity is transferred out of an agency, the remaining agency’s fee share can rise even if no existing fee program collects another dollar. The change in the percentage would partly reflect a changed boundary rather than an increased dependence within every continuing activity.

For a real comparison, create a bridge between the two scopes. List activities added, removed or reclassified, then show a like-for-like total where possible. If the information is insufficient, say the comparison is not fully comparable. This is more informative than treating every increase in a fee share as the same kind of institutional change.

4. A fee rate is not the amount collected, and neither is a guarantee of approval

For fiscal year 2027, FDA lists a PDUFA application fee of $4,600,753 for an application requiring clinical data and an annual program fee of $416,857. These are different fee categories. The agency’s current page identifies the rates for FY 2027 rather than leaving readers to reuse the previous year’s schedule. FDA’s prescription-drug user-fee rates.

A rate schedule does not tell us how many payments will be made, whether particular submissions qualify for exceptions, or when money will arrive. Multiplying one displayed rate by every application mentioned in a news story can therefore produce a revenue estimate with an invented base. The correct unit might be a qualifying application, a product-related program assessment or another specifically defined category.

The July 30, 2026 Federal Register notice establishes the FY 2027 rates for October 1, 2026 through September 30, 2027. Its calculation includes an operating-reserve reduction of $135,690,987, based on estimated reserves above the applicable upper threshold. The notice is a fee-setting calculation using information available at that stage, not the final financial report for the year being forecast. FY 2027 prescription-drug fee notice.

That example complicates a simple story in which every annual fee movement reflects a change in workload or influence. A rate can move because the funding formula includes reserves, expected volumes, costs and specified adjustments. Understanding the calculation is necessary before assigning a causal explanation to the change.

A fee invoice also is not an approval letter. The funding mechanism and the scientific decision are separate parts of the process. A writer investigating a specific application would need its review record and regulatory outcome, not merely evidence that a fee was owed. That distinction does not settle whether incentives operate; it identifies what evidence is missing from an accusation based only on a payment.

5. Follow money through a ledger instead of calling every balance “cash on hand”

Here is a second original hypothetical. A review program begins the year with $30 million in total carryover. It receives $110 million in net collections, records $2 million in recoveries of earlier obligations and incurs $105 million in new obligations. Its ending total carryover is $37 million: 30 plus 110 plus 2 minus 105.

The $2 million recovery is important. It represents funds released from a previous commitment in this example, not another $2 million paid by industry during the year. Leaving it out would make the reconciliation fail; labeling it a new fee collection would misdescribe the source of the change.

Suppose $7 million of the ending balance is unavailable for obligation and $3 million is set aside for potential refunds. The amount remaining after those two deductions is $27 million. The gross balance, available balance and balance net of the refund set-aside answer different questions. All three can be valid numbers, but they cannot be used interchangeably.

The fictional ledger illustrates a distinction present in FDA’s real reporting. The FY 2025 PDUFA report shows total ending carryover of $413,940,928, but $335,089,933 after subtracting unappropriated amounts, and $305,308,933 after the further refund set-aside. A headline should not describe all three as the same freely usable balance. FY 2025 PDUFA Financial Report, tables eight and nine.

Neither a large balance nor a small one supplies a complete operational diagnosis. A reserve might support continuity when collections are uncertain, while an excessive or persistently growing balance might raise questions about forecasting and fee levels. The evaluation requires the restrictions, expected uses, expenditure pattern and relevant reserve policy. A number without those features can be made to support opposite stories with equal ease.

6. A payment can belong to one cohort and arrive in another year

FDA’s FY 2026 version of its five-year PDUFA financial plan distinguishes cohort-year collections from collections recorded when payment is received. Its example explains that a fee due in one fiscal year but received in the next can be assigned to the earlier cohort while appearing in the later year’s net collections. The plan also separates actual values from estimates. PDUFA VII Five-Year Financial Plan, FY 2026 version.

Use a small invented case to see the difference. A fictional program assesses a $4 million fee due in Year One. The payer sends $3 million in Year One and the remaining $1 million in Year Two. A cohort view attributes the full assessed-and-collected amount to the Year One fee cohort once payment is complete. A receipt-year view shows $3 million arriving in Year One and $1 million in Year Two. These are different classifications of the same payments, not necessarily conflicting accounts.

Add a $500,000 refund in Year Two and the reconciliation needs another field. The analyst must establish which table nets the refund, which cohort it belongs to and which period records the cash movement. One cannot simply subtract a refund from whichever total makes a preferred trend look stronger.

For a careful comparison, retain a payment identifier, the fiscal year originally due, the receipt date, the amount, any refund and the report’s treatment. This is a proposed research schema, not a claim that FDA publishes every transaction in that form. Its purpose is to show the information needed to reconcile a disagreement between cohort and receipt-year totals.

The same issue appears when comparing a financial plan with a later financial report. A plan may show an estimate for a future year, an earlier estimate for a recently completed year and a revised actual value in adjacent columns. Selecting the largest number from one column and the smallest from another is not trend analysis unless the labels are preserved and the comparison is justified.

The question to ask is not merely whether two numbers differ. Ask whether they measure the same thing, use the same cutoff and have the same status. A forecast error, a timing difference and an accounting reclassification are distinct findings. Good reporting identifies which one the available evidence supports.

7. A review deadline is not a promise to approve

The PDUFA VII commitment letter sets review-and-action goals, including a 90-percent goal for specified application categories within defined timeframes. Its definitions explain that the completed action can identify deficiencies rather than grant approval. An on-time review action and an approval are therefore not the same outcome. PDUFA VII performance goals and definitions.

This distinction is central to evaluating a fee-funded review system. Timeliness can matter to patients, developers and the public without requiring the agency to approve an application that does not meet the relevant standard. Conversely, a fast action is not automatically evidence of a thorough review. Speed and substantive quality need separate measures.

Imagine one hundred fictional applications with applicable goal dates. Eighty-five receive timely approvals, ten receive timely letters identifying deficiencies, and five receive late actions. The on-time action rate is 95 percent. The timely approval rate is 85 percent. Calling the 95-percent figure an approval rate would combine different outcomes and overstate what it measures.

Now imagine another fictional program with ninety-five timely approvals and five timely deficiency letters. Both the first and second programs could report high timeliness, but their approval patterns differ. Without information about the applications’ merits, neither pattern establishes that one program is more rigorous or more captured. The case mix could differ substantially.

A meaningful performance assessment would retain the application category, the starting point of the clock, any permitted extension, the action date and the action type. It would then evaluate the substance of selected decisions separately. The deadline metric is a useful operational measure; it should not be forced to carry an evidentiary meaning it was not designed to support.

8. An unfinished cohort can look better—or worse—than its final result

FDA’s FY 2025 PDUFA performance report distinguishes final FY 2024 review-goal results from preliminary FY 2025 results. For the preliminary measure, applications still pending within their goal date are excluded from the current-performance denominator. The report warns that longer review periods leave more work unfinished at the cutoff, making preliminary results less informative about final performance. FY 2025 PDUFA Performance Report, reporting methodology.

An original example shows why that qualification matters. Suppose a fictional cohort contains one hundred applications. At the reporting cutoff, forty have been acted on within their goals, five have passed their goals without a timely action, and fifty-five remain pending but are not yet overdue. The current on-time rate among the forty-five classified cases is forty divided by forty-five, or 88.89 percent.

It would be wrong to report 40 percent as the final timeliness rate, because the fifty-five pending cases have not yet had a chance to complete within their goals. It would also be wrong to declare that the final result must be 88.89 percent. The remaining cases can change it.

If all fifty-five remaining cases finish on time, the final rate becomes 95 percent. If all miss their goals, the final rate becomes 40 percent. Those are arithmetic bounds for this invented, fixed cohort, not a statistical confidence interval and not a forecast that either extreme is likely. If the cohort definition later changes, the calculation must be revised accordingly.

This example can also explain why two reports show different numbers without either being erroneous. One report may stop at the end of the fiscal year; another may include later actions on the same submission cohort. The later report may be more complete while still using the earlier cohort’s label. A researcher should track both the cohort year and the measurement cutoff.

A comparison between programs needs similar caution. A program with many short deadlines may have far fewer pending cases at year-end than a program with long deadlines. Their current percentages may therefore have different degrees of completeness. Before attributing the difference to funding or management, inspect the pending share and the distribution of review types.

The useful reporting unit is not a lonely success percentage. It is the number received, number resolved on time, number overdue, number still pending within goal and the date at which those categories were measured. That compact record allows readers to distinguish a current operational result from a final assessment.

9. Restricted funding cannot be moved wherever a headline says the need is greatest

The PDUFA financial plan states that fee funds may support defined costs of the human-drug review process, with allowable and excluded costs described separately. The prescription-drug fee stream is not presented as unrestricted money for any agency activity. PDUFA VII Five-Year Financial Plan, scope of obligations.

That restriction changes how a staffing or inspection problem should be analyzed. A balance in one fee program does not demonstrate that another activity can immediately hire staff using that balance. The researcher has to establish the permitted purpose, the available appropriation and the relationship between the proposed work and the funded program.

Consider a fictional organization with $12 million available in a legally restricted review account and a $4 million shortfall in an unrelated inspection account. At the organization-wide level, assets exceed the shortfall. At the account level, the inspection activity may still lack usable funding. An argument that the shortage is fake because the organization “has the money” would omit the restriction.

This does not mean restrictions are beyond criticism. A policymaker can ask whether they produce an appropriate balance across activities or whether the law should change. But that is a proposal to change the funding rules, not evidence that the existing balance was automatically available for another purpose. Reporting should keep the diagnosis and the proposed reform distinct.

The current framework also involves non-user-fee spending conditions. The April 2024 CRS report describes statutory conditions intended to prevent fees from simply replacing general appropriations, while FDA’s later financial reports document whether the applicable conditions were met for the reported year. A historical framework description should be paired with the relevant annual evidence rather than treated as proof of compliance indefinitely. CRS explanation of funding conditions.

A small hypothetical makes the policy issue visible. Suppose a program needs $100 million to perform a defined workload. One model supplies $60 million from fees and $40 million from general funds. A second supplies $90 million from fees and $10 million from general funds. Both provide the same total on paper, but they distribute the financing burden and dependence differently. If a legal condition requires a minimum general-funded contribution, the second model may not be permissible even though its arithmetic balances.

The example is not a calculation of FDA’s actual spending trigger. It shows why a total alone cannot establish whether a funding arrangement complies with its rules or remains resilient under a change in collections.

10. A budget request is evidence of a proposal, not proof of what Congress enacted

FDA’s FY 2027 congressional budget justification requests $7.227466 billion in total program funding, including $3.921404 billion in user fees. The document also adjusts its comparison for a proposed transfer of the National Center for Toxicological Research to CDC. These are features of the administration’s request and its presentation; they should not be described as a final enacted budget or an accomplished organizational transfer merely because the document is official. FDA FY 2027 congressional justification.

An official document’s status is part of the evidence. A request can reveal priorities, assumptions and proposals. An enacted appropriation establishes a different legal and financial stage. An annual financial report records still another stage: what happened during operations under the applicable framework. They are not successive names for the same number.

In the request, dividing the user-fee amount by the total yields approximately 54.26 percent. That is an original calculation of the requested funding mix. It is not the fee share of actual FY 2027 obligations, and it cannot be compared mechanically with the FY 2025 PDUFA obligation share because the scope, year and accounting measure differ.

A useful research worksheet would give each observation a status field: requested, enacted, estimated, collected or obligated. It would also preserve any scope adjustment. If the worksheet combines a proposal with an actual value, the chart should identify the difference visibly rather than placing both under an undifferentiated heading such as “FDA spending.”

This is not bureaucratic fine print. A proposal can receive extensive coverage and never be enacted in its original form. A reader who finds it months later may reasonably mistake a confident headline for a report of the outcome. Status-aware citations prevent a document about what someone wanted to do from becoming evidence that it was done.

11. Institutional dependence and a person’s financial conflict are different investigations

FDA’s guidance on public disclosure of advisory-committee financial interests describes disclosure of the type, nature and magnitude of interests when a waiver permits participation. The guidance concerns particular members and matters, not the percentage of an agency program financed by user fees. It also distinguishes the disclosure process from other screening questions. FDA guidance on financial-interest information and waivers.

A fee-funded institution and an adviser with a relevant personal investment are therefore different units of analysis. One question concerns the organization’s resource dependence and incentives. The other concerns an individual’s relationship to a particular matter and the applicable participation rules. Evidence sufficient to discuss the first does not automatically establish the second.

Imagine a fictional advisory meeting about Product Q. A researcher finds that the review program receives industry fees. That fact does not establish whether Member R owns shares in the sponsor, has a consulting relationship, has been recused or has received a waiver. Those questions require the meeting-specific records and a correct understanding of the member’s role.

Now suppose a public record shows a relevant financial interest and a documented waiver. The existence of the interest is not the same as proof that it was hidden. The waiver is not the same as proof that the advice was unbiased. The proper analysis examines what was disclosed, the basis for participation, any limitations and the member’s actual role in the meeting.

The same care applies to an absence of a public document. Failure to locate a waiver could mean no waiver was required, the member did not participate, the document is located elsewhere or a disclosure was missing. A researcher should investigate which explanation fits before asserting concealment. Search failure is a prompt for a more precise question, not a completed finding of wrongdoing.

An evidence packet for that fictional meeting would contain the roster, agenda, relevant disclosure or waiver, participation limitations, transcript or minutes and the question the committee actually addressed. It would not substitute a list of fee-paying companies for the required individual evidence. If any of those records are unavailable, the report should identify the gap rather than fill it with a presumed motive.

12. There are documented transparency problems; describe the actual finding

In March 2026, GAO reported that FDA had not finalized required guidance explaining certain advisory-committee conflict-of-interest determinations and had not publicly shared the interim approach or the approach for guest speakers. GAO made three recommendations; HHS concurred. The recommendation page still listed all three as open when checked on October 29, 2026. Those are documented oversight findings, not an inference from a funding percentage. GAO’s advisory-committee transparency report.

This finding should neither be inflated nor minimized. It concerns transparency and guidance about how determinations are made. It does not, by itself, establish that every committee member had a conflict, that every decision was affected, or that a particular drug was improperly approved. At the same time, the absence of such sweeping conclusions does not make the recommendation trivial. A public process is harder to evaluate when its decision rules are not accessible.

The disclosure guidance discussed above and GAO’s finding address different functions. A document explaining what financial-interest information to publish is not necessarily the missing document explaining how the agency decides whether a conflict exists or participation is appropriate. Treating the existence of one guidance as a complete answer to the other criticism would skip the question GAO actually asked.

A careful follow-through would examine whether FDA subsequently publishes the relevant guidance or interim policies, what those documents cover and whether GAO closes the recommendations. An open recommendation is a status in an oversight process, not a permanent label that should be repeated without checking. Conversely, a newly posted document would need to be read before calling the issue resolved.

For readers trying to assess “capture,” this is a more actionable line of inquiry than a slogan. It identifies a specific process, an independent finding, requested corrective actions, the agency’s response and a status that can be checked later. It leaves room for criticism while keeping the criticism tied to evidence rather than insinuation.

13. To investigate influence, define an outcome that could actually test the hypothesis

Suppose an analyst observes that a program’s approval rate rose while its fee funding increased. The two trends might justify investigation. They do not establish that the funding change caused the approval change, because the applications, legal standards, staffing, product mix and observation window may also have changed.

An original fictional example isolates one of those problems. In Year One, a regulator receives twenty applications in an easier-to-satisfy category and eighty in a harder-to-satisfy category. It approves eighteen in the first category and sixteen in the second: thirty-four approvals out of one hundred applications. In Year Two, the mix reverses: eighty easier-category applications and twenty harder-category applications. It approves seventy-two and four respectively, or seventy-six out of one hundred.

The aggregate approval rate rises from 34 percent to 76 percent. Yet the category-specific rates are unchanged: 90 percent in the easier category and 20 percent in the harder category in both years. This constructed example does not show what happened at FDA. It shows that a large aggregate change can occur without a change in either category’s decision rate.

A real investigation would need defensible categories and evidence that the comparison groups are genuinely comparable. The labels “easy” and “hard” are teaching devices, not variables that can simply be assigned to actual applications. Relevant differences might include submission type, evidence completeness, disease context, review pathway or manufacturing issues. Those features would need to be defined before conclusions were drawn.

Even a well-adjusted association would not necessarily reveal the mechanism. A funding change might alter staffing, turnaround times or the mix of applications submitted. An influence claim about scientific judgment requires evidence bearing on that judgment: review disagreements, departures from stated standards, documented interventions, or another specifically defined indicator. Each would require context and appropriate safeguards against selective examples.

The study should also specify what would count against its hypothesis. If the claim is that fees encourage approval regardless of evidence, a design that counts every approval as confirmation and every nonapproval as a strategic exception cannot test the claim. A credible analysis needs an outcome definition and comparison that allow the proposed explanation to fail.

Timing matters as much as the outcome. A funding rule may begin after an application’s substantive review was largely complete. A later safety action may reflect information unavailable at approval. Counting it automatically as proof that the earlier decision was corrupt would collapse different knowledge states. The chronology must distinguish what decision-makers knew then from what became known later.

None of this means institutional incentives cannot matter. It means the question deserves a method capable of distinguishing influence from other explanations. A transparent limitation is more valuable than a confident causal story built from two unrelated trend lines.

14. Compare funding reforms by their tradeoffs, not by whether they sound independent

GAO’s user-fee design guide examines how fee setting, collection, use and review affect efficiency, equity, adequate revenue and administrative burden. Published in 2008, it is a design framework rather than an assessment of today’s FDA decisions. Its central usefulness here is to make the financing choice explicit instead of treating any one funding source as free of consequences. GAO, Federal User Fees: A Design Guide.

Consider three hypothetical reform proposals for a program that requires $100 million each year. Proposal One replaces all fees with general appropriations. Proposal Two keeps $60 million in fees and $40 million in general funding but adds a public review of spending boundaries and performance measures. Proposal Three reduces fees to $30 million without securing additional general funding. These are analytical examples, not recommendations or descriptions of pending legislation.

Proposal One changes who finances the work and removes direct fee dependence, but its operational result depends on whether the replacement appropriations arrive at the necessary level and time. A statement that it “makes the regulator independent” would still need to address other sources of political, administrative or informational influence. Changing the funding source answers an important question, not every question.

Proposal Two retains the dependence but tries to make its operation more observable and accountable. Its effectiveness would depend on the substance of the review, the information disclosed, the ability to correct problems and whether the measures include substantive quality rather than only speed. Publishing more tables is not automatically a remedy if the tables cannot answer the relevant questions.

Proposal Three leaves a $30 million shortfall under the stated assumptions. Unless workload, costs or another funding source changes, the proposal cannot maintain the same $100 million program simply by declaring a more desirable financing mix. A reform discussion should show the replacement resources or the activities that would be reduced, not hide the gap behind an appealing percentage.

A funding comparison should also examine who ultimately bears costs and who benefits from the regulated activity. The payer named on an invoice is not necessarily the only party affected economically. Determining how costs are passed through would require additional evidence; it should not be assumed from the legal obligation to pay the fee.

The most informative public account would end with three separate findings. First, describe the verified funding composition and the boundaries of the numbers. Second, identify documented oversight or transparency problems and their current status. Third, distinguish those findings from claims about particular decisions that require further evidence. That structure permits strong criticism without overstating what the records establish.

The question “who pays the regulator?” deserves a clear answer. The answer should be followed by better questions: what does the money support, what conditions govern it, which results are measured and what would demonstrate that a decision was improperly influenced? Those are the questions that turn a budget controversy into an investigation readers can evaluate.