There were five days between a surveillance recommendation inside S&P and a committee decision about what to announce. The dates were 7 and 12 February 2007. Reading the company's later agreed statement of facts, that interval becomes more interesting than the familiar observation that experts make mistakes. What happened between noticing a risk and making it available for others to act on?^[1]

Some vocabulary helps, but not a whole course in structured finance. Residential mortgage-backed securities, or RMBS, turn payments on pools of home loans into securities. A collateralized debt obligation, or CDO, can organize debt assets, including mortgage-backed securities, into further claims. Tranches have different positions in the allocation of payments and losses. Evaluating one requires attention to the arrangement, not merely counting the assets underneath.^[2]

A credit rating offers a comparatively compact assessment of credit risk. It does not guarantee repayment, establish a fair purchase price or promise that a buyer will be available when an investor needs to sell. The SEC's investor bulletin distinguishes those limits expressly. A few letters can be useful without answering every question someone brings to an investment.^[3]

This is not a story about why we should stop relying on expertise. Refusing every other person's analysis can leave us more exposed to persuasion, not less. The more difficult question is what reliance should permit: access to reasons, an understanding of uncertainty and a way to examine the judgment without having to become its author.

When an opinion becomes a condition

A rating begins as an assessment. Other institutions can give it a further role. The Senate Permanent Subcommittee on Investigations' 2011 report described regulations and institutional requirements that relied on credit grades. In those settings, a judgment could help determine which investments were eligible, rather than simply inform someone considering them.^[2]

That is not a claim that every pension fund had to buy AAA securities, or that every downgrade forced every holder to sell immediately. The institution, rule and period matter. The narrower point is sufficient: a limited assessment can acquire consequences through arrangements outside the rating agency. Identifying those arrangements is part of understanding its power.

There are understandable reasons to use a common scale. Requiring each institution to invent its own language for credit risk would be costly and would not guarantee better analysis. A shared measure makes comparison easier. The difficulty begins when the measure silently acquires responsibilities beyond the question it was designed to answer.

If a credit opinion becomes a substitute for assessing an investment as a whole, somebody's judgment has been omitted. Several participants may each say that someone else already checked. That can make a chain of apparently careful procedures remarkably hard to question: where, exactly, was the underlying assessment examined, and who remains responsible for using it here?

Payment introduces another relationship. The Senate investigation identified issuer-paid ratings, market-share pressures, model weaknesses and inadequate surveillance resources among its findings about inaccurate ratings. This was a congressional investigation, not a uniform judicial verdict on every agency or every rating.^[2]

An issuer paying for a rating creates a conflict worth addressing. It does not logically establish that each resulting judgment was manipulated. The conflict tells us what to investigate: whether commercial considerations entered analytical decisions, what happened to dissent and how existing assessments responded to new information. A diagnosis of incentives is a reason to look for evidence, not a replacement for it.

From a warning to a public action

On 7 February 2007, surveillance staff recommended public negative watch for subordinate tranches from approximately thirty RMBS transactions, and internal watch for tranches from about twenty more. On 12 February, a committee including new-issue staff decided on public negative watch for eighteen tranches from eleven transactions. These are S&P's agreed facts, not a reconstruction of an imagined meeting.^[1]

Transactions and tranches are different counting units; internal watch, public watch and a downgrade are different actions. Keeping them distinct does not soften the question. It makes the difference between a recommendation and a decision identifiable, instead of covering it with a larger but inaccurate claim.

A difference is not inherently improper. A committee might have better information or a well-founded objection to an analyst's proposal. What matters is the reason for the change, who made it and whether the explanation can be examined. The existence of disagreement is not the failure; losing the ability to investigate its resolution would be.

Hindsight needs restraint in both directions. We should not give participants in February the knowledge of the eventual crisis. Nor, once contemporary warnings are documented, should we pretend that every concern arose only afterwards. A fair inquiry returns people to the information available at the time, rather than making them omniscient or conveniently uninformed.

On 10 July, S&P put 612 RMBS classes on negative watch; large-scale downgrades followed on 12 July. Through the 10 July announcement, it had continued issuing or confirming CDO ratings without adjusting existing CDO criteria for anticipated negative RMBS actions. The statement records a connection between two sets of judgments that had not been made in those criteria.^[1]

This moves the issue beyond whether a warning reached an outside investor. Where an assessment depends on assets undergoing reassessment, how should the dependent judgment respond? Knowing something in one part of an organization is not enough. The organization needs a way for the knowledge to reach decisions that rely on it, with responsibility for making that connection.

There are real reasons not to alter an assessment at the first sign of trouble. Evidence needs checking, standards need consistency and hurried revisions can produce another kind of error. Delay is not automatically concealment. But needing time is an explanation that can be examined, not an exemption from explaining what happens while the clock runs.

During that interval, what continues to be represented to the public? What limitations accompany the existing opinion? Which dependent assessments are reconsidered, and by whom? These are not demands for perfect foresight. They ask how an institution manages uncertainty it already has reason to recognize. An opinion can remain uncertain without leaving its users unaware of the uncertainty.

The Justice Department announced a $1.375 billion civil settlement with S&P and its parent on 3 February 2015, involving nineteen states and the District of Columbia. It reported S&P's acknowledgment that business concerns affected rating-model decisions. The agreed facts go beyond untested allegations; the resolution was not a criminal conviction, nor does every prosecutorial characterization become a finding after trial.^[4]

The scope matters. S&P's statement is not an admission by Moody's or another agency. This episode is not a complete explanation of the financial crisis. Lending decisions, product design, investors' own choices and other institutional conditions do not disappear because one decision chain becomes documented. A case can establish something important without establishing everything.

Not another infallible judge

Replacing the rating provider is an obvious possibility. A different analyst may offer a better assessment. Competition can also mean competing to satisfy the party bringing business. Counting providers will not tell us which kind of competition is occurring. We have to ask what earns the business and what makes an unwelcome judgment sustainable.

Changing who pays is likewise worth examining, but is not a complete answer by itself. The SEC bulletin notes potential conflicts in subscriber-paid arrangements too, arising from investors' positions and trading interests.^[3] A different source of revenue changes the relationship; it does not abolish the need to examine influence, access and the ability to seek review.

More immediate questions follow from the documented episode. Was the original recommendation preserved? Was a departure from it explained? Could a decision-maker be questioned by someone whose objective was not winning the next transaction? Were assessments depending on the affected assets reviewed as well? Records can answer such questions without anyone first identifying a person incapable of error.

That is a different ambition from making the public vote on every model. Technical disagreement has substance, and protecting analysis from commercial pressure can itself require considerable expertise. Openness should make reasons more examinable, not allow the loudest objection to dictate a calculation. Asking for a method and the reason it changed is not pretending that every reader can independently rebuild it.

It is also important to distinguish a model's error from interference with its use. An honest assessment may fail because its assumptions are poor. A well-designed procedure can be weakened if inconvenient findings do not reach decisions. The remedies are not identical. Better data will not by itself protect a suppressed objection; a protected objection will not by itself repair a faulty model.

Reviewers need scrutiny in turn. What can they obtain? What time and resources do they have? Can they require a reasoned response, and how can their own mistakes be corrected? Moving a judgment outside the original firm changes its location. It does not remove its dependence on evidence or its responsibility to answer questions about what it does.

Nor should the people using a rating vanish behind the agency's authority. They may reasonably employ a common assessment, but they still decide how much work to ask it to do. Treating a bounded credit opinion as a universal permission is a further decision. The existence of expertise does not make that use inevitable, or relieve its users of explaining it.

For an ordinary reader, the practical lesson need not be suspicion of all professional language. It can be a more discriminating form of reliance. What question does this opinion answer? What has it left outside? What would cause it to change? Which arrangements should be reconsidered if it does? These questions do not make expertise unnecessary. They help keep its usefulness from becoming an excuse to stop thinking.

Those two February dates leave a modest but consequential space in the record: a recommendation inside and a differently scoped action outside. We need not inflate that space into the whole history of the crisis to take it seriously. An institution's authority should not rest only on its ability to judge the world. Its own decisions must leave somewhere for a question to land.