豁免的成本：为什么规则最怕的不是被违反，而是被例外
Rewriting a rule means persuading everyone; waiving it means persuading one person. The effort saved does not vanish — it is spread out as the cost everyone pays to find out whether this time will be an exception
1. Two Kinds of Noncompliance
Breaking a rule and being granted a waiver look identical from the outside: in both cases the rule was not followed. Their direction, however, is opposite. The violator treats the rule as an obstacle to be got around. When he is caught and punished, the authority of the rule is publicly reaffirmed — punishment is a confirmation. The exempted party does not go around the rule at all. He invokes a permission located inside the rule system: someone with authority has decided that the rule does not apply this time. The rule is not denied; it is formally cited. And yet it fails at precisely the point where it was supposed to supply certainty.
This distinction is not wordplay. It explains a familiar puzzle: why some systems punish heavily and still have poor order, while others punish rarely and remain highly predictable. The difference usually lies less in enforcement than in how "not applicable" is allowed to happen.
Aristotle gives the earliest and most dispassionate account of the problem in Book V of the Nicomachean Ethics, in his discussion of equity. Law has to be written in universal terms, while particular cases are inexhaustible; where the law goes wrong because of its generality, a correction is needed. What matters is that he does not place this correction outside the law, as a damage done to it. He treats it as the completion of justice itself. Put differently: an exception is not a hole in a rule but its structural companion.
So the interesting question was never whether exceptions ought to exist. The question is: once an exception is made, whose account does the saving get booked to?
2. Where the Cheapness Comes From
Changing a rule is expensive. It requires open negotiation, reasons that can be challenged, an answer to everyone who will be affected later, and the cost of changing it back if the change was wrong. A waiver is far cheaper: it needs one person who can approve it, one judgment that need not be public, and often no written trace at all.
The crucial point is that a waiver rewrites expectations, not text. The text stands untouched; everyone is still told that the rule is thus. But those with information have already begun acting on "the rule plus some probability of an exception." From that moment the rule no longer predicts consequences — which was its only real function, since its purpose was to let strangers anticipate one another.
Every rational actor is then forced to make an additional investment: find out whether this case will be let through, who decides, what it costs, and how the last one was handled. That investment produces nothing. It is pure friction, and its returns are highly concentrated, accruing to those who hold the approval and the information. More subtly, the investment does not require exceptions to be frequent. The mere possibility that one might be granted is enough to make everyone add an extra calculation in advance.
In Soulstealers, Philip Kuhn's study of the 1768 sorcery scare that spread from local rumours in Jiangnan into a nationwide political case, the underlying subject is an extreme version of this structure: the dynamics of autocratic rule. The throne could bypass the routine procedures of the bureaucracy and apply pressure directly, while those routines were both an instrument of government and a form of protection for officials. What is worth extracting here is not the conclusion that autocracy is bad, but a description of a mechanism. When the highest authority can intervene outside normal procedure, the expectations that the whole routine apparatus generates for subordinates are rewritten.
Ray Huang's 1587, A Year of No Significance offers the long-run form of the same phenomenon from the other side: the Ming state governed through morality in place of law, lacked quantified administrative technique, and filled the gaps with unwritten discretion and ritual adjustment. A caution is in order. "Mathematically manageable" government is a contested interpretive framework, and historians have argued that the judgement rests on utilitarian and modernisation assumptions; it should not be treated as settled. But even if the overall thesis is rejected, one layer of mechanism holds: where written rules are too coarse and discretion must repeatedly stand in for them, the distance between the published rule and the operating rule becomes tacit knowledge that everyone has to acquire privately.
3. Why Exemptions Multiply
A single waiver always looks like a stopgap. In practice, systems rarely stop at the first one. Three mechanisms push it forward.
First, precedent changes where the argument starts. After the first exception, a later applicant no longer has to show that the thing ought to be permitted; pointing out that it was permitted last time is enough. "It was allowed before" is the cheapest of all arguments, because it reverses the burden of proof: now the refuser must explain why this case is different.
Second, once an exemption can be obtained, people invest in obtaining it. Lobbying, connections, restructuring an activity so that it happens to meet the exemption conditions, hiring someone to study the approver's phrasing — none of this is incidental corruption. It is the normal response to an incentive. At that point the rule itself becomes a cost to be paid to get around, and the price of getting around it is set by the approver. The stricter the rule, the higher the rent. Severity and waiver are therefore not opposites; they are often the two halves of one arrangement.
Third, an authority that can grant an exception at any time cannot credibly promise not to grant the next one. Because the promise is not credible, everyone plans on "it might be granted," and the expectation-stabilising function of the rule is systematically discounted. The discount is irreversible: even if no further waiver is ever issued, people have already learned not to rely on the rule.
The Dictator's Handbook uses the idea of a winning coalition to explain why rulers concentrate resources on a small set of essential supporters instead of providing public goods universally. Transposed here, it yields an uncomfortable but genuinely explanatory inference: an exception is a more precise distributive instrument than a general rule. It reaches only those who need to be reached and need not be given to everyone, which makes it cheaper for the distributor.
Hobbes places the sovereign outside the contract and outside civil law in Leviathan, on the grounds that an unconstrained arbiter is needed to end disputes. The strength of that argument is its honesty. If rules have edges, someone must decide where the edges are, and that person cannot be fully bound by the same rule. Exemption is not a malfunction; it is designed in.
4. The Counter-Case: Abolishing Exemptions Is Neither Possible Nor Obviously Better
Everything so far invites the conclusion that exemptions should be driven to zero. That conclusion does not hold, for three independent reasons.
One, rules cannot exhaust circumstances. Aristotle retained equity precisely because universal wording must sometimes go wrong in a particular case; mechanical application can do more damage than a careful exception. Mill's defence of experiments in living in On Liberty points the same way: if every departure had to await an amendment to the general rule, no experiment could occur, and society would lose its chance to discover that the rule itself was wrong.
Two, writing every exception into the rule produces a rule nobody can read. The list always lags behind new circumstances, and the denser it gets, the more it favours those who can find the applicable clause. More commonly, an attempt to eliminate discretion simply moves it from the visible to the invisible, from a level where it can be appealed to one where it cannot. Formally there are no exceptions left; in practice everyone has to guess how a front-line official will read the rule that morning.
Three, a waiver is sometimes an error signal. When one class of application keeps being exempted, that is often evidence the rule should change. A persistent exception may be the only visible indicator that the rule and reality have come apart.
There is a deeper objection still. Treating every exception as an injustice smuggles in an unargued premise: that the existing rule is itself just and was agreed to. When the rule is the problem, a waiver may be the only cushion left working. A bad rule that permits no exceptions is harder to live with than a bad rule that permits traceable ones. The target therefore cannot be zero exceptions. It can only be that exceptions must not happen in a certain way.
5. Three Tests, and the Limits of the Tests
Counting exceptions tells us little about whether a system has lost control of them. Three attributes tell us more.
- A trace. Does the waiver enter a record that can be examined later, so that someone can count how often it happened and where it fell?
- Replicability. Is the same situation handled the same way, or does the outcome depend mainly on who asked and whom they know?
- A duty to justify. Must the beneficiary or the approver give reasons that can be questioned, rather than a phrase like "special circumstances"?
Where all three hold, an exception resembles equity with a review clause attached. Where none hold, the rule retains only its appearance and what actually operates is an unpredictable licensing system.
But the limits of these tests must be stated too, or this essay has merely substituted one patent medicine for another.
- A trace can be formalised. A fully documented "compliant waiver" with hollow reasons is harder to challenge than no record at all, because it now wears the costume of procedure.
- Replicability has no "same case" in genuinely novel situations. Forcing an analogy manufactures a bad precedent and freezes a stopgap into a long-term rule.
- A duty to justify rewards whoever writes the best reasons, transferring discretion from power to paperwork without any guarantee of greater fairness.
- There is a privacy cost. For waivers involving medical hardship or relief, disclosure itself may injure the person concerned; a trace is not the same thing as publicity.
These three are therefore not sufficient conditions, nor a checklist to be ticked. Their use is narrower and more practical: they turn "is this exception acceptable" from a moral argument in which every side talks past the others into a design question that can actually be inspected.
6. Judgement: Put a Price on Exemption
Return to the opening distinction. The enemy of a rule is not the violator: when a violation is punished, the rule is reaffirmed. The enemy is the case in which someone is permitted not to comply and the permission leaves no trace, cannot be replicated, and needs no defence. It preserves the entire appearance of the rule while removing its single function.
The problem can then be restated as a piece of distribution. What a waiver saves is one open negotiation, and that saving goes to the approver. What it costs is an extra burden of guessing borne by everyone, above all by those who depend most on the rule and have least access to an exception. The saving is privatised and the cost externalised — a structure so standard as to be dull, yet rarely recognised as a distribution at all, because it produces no event and leaves no record.
One enforceable judgement follows: every waiver should automatically trigger a review of the rule itself. The reasons given at the moment of exemption should not serve only to establish that this case is special; they should be used to test whether the rule ought to change. When the same class of exemption recurs, write it into the rule. Where it genuinely cannot be written in — because privacy is at stake or the situation resists classification — publish at least how often it occurs and how it is distributed, so that outsiders can judge the rule's real scope.
The maturity of a rule is therefore measured neither by how strict it is nor by how many exceptions it allows, but by whether it is willing to book each exception to its own account. In a system that keeps accounts, an exception is an entrance through which the rule gets corrected. In one that does not, the rule slowly degenerates into a sentence everyone cites and nobody relies on.
