Symmetry Based Fairness¶
Essence¶
Symmetry-Based Fairness is the intervention pattern of treating equivalent cases equivalently unless a relevant asymmetry justifies different treatment. It turns a broad fairness claim into a practical comparison structure: define which cases count as the same for the decision, identify what treatment should remain invariant, and require reasons when that treatment changes.
This archetype is not a generic command to make all outcomes equal. It is a disciplined way to ask, "Would this case be treated the same if an irrelevant label, location, reviewer, ordering, or presentation changed?" When the answer is no, the system must either identify a relevant difference or repair the decision rule.
Compression statement¶
When decisions vary across similar cases, define the equivalence relation that matters, apply an equal-treatment rule to cases inside it, and require documented relevant differences before asymmetric treatment is allowed.
Canonical formula: same_relevant_case + same_rule -> same_treatment; asymmetric_treatment -> relevant_difference + recorded_reason + review_path
When to Use This Archetype¶
Use this archetype when similar cases receive different benefits, burdens, access, penalties, rankings, standards, or review intensity, and the system cannot clearly explain which differences matter. It is especially useful when decisions recur across locations, reviewers, departments, time periods, or categories.
It fits policy administration, adjudication, eligibility, resource allocation, moderation, grading, hiring, promotion, procurement, and other cases where consistency and legitimate exceptions both matter. It is weak when cases cannot be scoped, when equal treatment would erase relevant need, or when the real problem is procedural voice rather than treatment symmetry.
Structural Problem¶
The structural problem is unexplained asymmetry. A system claims to use a rule, but treatment shifts across cases that appear equivalent. Sometimes the variation comes from hidden bias, habit, convenience, reviewer preference, historical accident, or local interpretation. Sometimes the variation is justified, but the justification is not recorded.
The result is a legitimacy gap. People cannot tell whether different treatment reflects a relevant difference or an arbitrary one. Decision-makers may also lose track of their own standards, creating drift over time.
Intervention Logic¶
The intervention starts by naming the decision and the treatment dimension: access, benefit, burden, penalty, ranking, eligibility, timing, or review intensity. Next, it defines a case set and an equivalence rule. The rule says which facts make cases the same for this decision.
Once equivalence is defined, the equal-treatment rule becomes an invariant: equivalent cases should receive equivalent treatment. When treatment differs, the system must run a relevant-difference test and document the exception rationale. Repeated exceptions are then audited to distinguish justified policy refinement from creeping inconsistency.
Key Components¶
Symmetry-Based Fairness operates as a disciplined comparison structure that turns a broad fairness claim into a checkable rule: equivalent cases should receive equivalent treatment unless a relevant difference is shown. The Case Set Scope defines the population of decisions, claims, or applications being compared, preventing actors from selecting convenient comparison groups after the outcome is already preferred. The Equivalence Rule is the heart of the archetype — it names the facts that make two cases relevantly the same for this decision and excludes facts that should not affect treatment, such as reviewer identity, label, location, or presentation order. The Equal-Treatment Rule then specifies what must stay invariant across equivalent cases — eligibility, priority, sanction, access, or burden — tied to a concrete treatment dimension rather than left as a vague value.
The remaining components handle the legitimate exceptions and the audit trail that distinguishes principled departures from creeping inconsistency. The Relevant Difference Test protects the archetype from rigid sameness by asking whether a difference between cases is strong enough to justify divergent treatment, while also preventing exception-making from becoming disguised favoritism. The Exception Rationale records why an otherwise equivalent case was treated differently, naming the relevant difference, the rule it activates, and who accepted the departure — without that record, justified asymmetry and arbitrary inconsistency are indistinguishable. The Audit Trail preserves case facts, equivalence judgments, treatment outcomes, and exception rationales so later reviewers can test whether the symmetry rule is real in practice and whether exceptions are clustering in suspicious or policy-relevant ways. Together these six components let the system answer not only whether treatment was equal, but whether unequal treatment was explainable, scoped, and reviewable.
| Component | Description |
|---|---|
| Case Set Scope ↗ | The case set scope defines which cases, claims, applications, users, events, or decisions are being compared. Without it, the symmetry claim can become either too broad or too narrow. The scope prevents decision-makers from choosing convenient comparison groups after the outcome is already preferred. |
| Equivalence Rule ↗ | The equivalence rule states when two cases count as relevantly the same. It is the heart of the archetype. A good equivalence rule names the facts that matter and excludes facts that should not affect treatment, such as irrelevant labels, reviewer identity, presentation order, or location. |
| Relevant Difference Test ↗ | The relevant difference test asks whether a difference between cases is strong enough to justify different treatment. This component protects the archetype from rigid sameness. It also prevents exception-making from becoming disguised favoritism. |
| Equal-Treatment Rule ↗ | The equal-treatment rule specifies what should remain the same across equivalent cases. It may concern eligibility, priority, sanction, access, review depth, service standard, or burden. Equal treatment must be tied to a specific treatment dimension, not left as a vague value. |
| Exception Rationale ↗ | An exception rationale records why an otherwise equivalent case was treated differently. It should name the relevant difference, the rule or value it activates, and who accepted the departure. Without this record, justified asymmetry and arbitrary inconsistency look the same. |
| Audit Trail ↗ | The audit trail preserves the case facts, equivalence judgment, treatment outcome, and exception rationale. It lets later reviewers test whether the symmetry rule is real in practice and whether exceptions are clustering in suspicious or policy-relevant ways. |
Common Mechanisms¶
A policy symmetry test checks whether a written rule or implementation practice treats equivalent cases the same. It is a test, not the archetype itself, because it does not by itself define the whole fairness structure.
A consistency audit looks for unexplained variation across reviewers, units, time periods, or populations. It implements the archetype when the audit uses a clear equivalence rule and follows up on deviations.
Precedent analysis compares current cases with prior analogous cases. It is useful when legitimacy depends on explaining why this case is like or unlike earlier cases.
An exception register records departures from equal treatment and the reasons for them. It is an implementation artifact that supports accountability and later rule revision.
An equal-treatment checklist helps reviewers ask the same questions each time: What is the equivalence rule? What treatment should remain invariant? What relevant difference justifies a departure? What record is needed?
A classification fairness review tests whether labels or categories are driving unequal treatment without decision-relevant differences. An anti-discrimination check is a narrower mechanism for sensitive or prohibited markers; it should not be confused with the full archetype.
Parameter / Tuning Dimensions¶
The first tuning dimension is equivalence strictness. A strict rule treats cases as equivalent only when many facts match; a looser rule groups cases by fewer decision-relevant facts. Too strict a rule excuses inconsistency. Too loose a rule creates false equivalence.
The second dimension is exception burden. Low-stakes decisions may need lightweight reasons. High-stakes decisions need stronger rationales, review paths, and records.
The third dimension is comparison scope. The case set may cover one team, one jurisdiction, one time period, or an entire organization. Larger scopes improve consistency but may miss local context.
The fourth dimension is audit cadence. Continuous auditing may be necessary for automated or high-volume systems, while periodic review may be enough for slower decisions.
The fifth dimension is equity sensitivity. The archetype must allow relevant asymmetries when need, vulnerability, risk, or repair obligations justify different treatment.
Invariants to Preserve¶
The main invariant is that equivalent cases receive equivalent treatment. If irrelevant labels, order, reviewer identity, or presentation format changes the outcome, the symmetry rule has failed.
A second invariant is that relevant differences remain explicit. The system may treat cases differently, but the basis for doing so must be visible and reviewable.
A third invariant is that exception reasons stay attached to outcomes. Without records, later actors cannot distinguish justified departures from arbitrary ones.
A fourth invariant is that the equivalence rule itself remains open to revision. Repeated justified exceptions may reveal that the rule is wrong, not that fairness has been achieved.
Target Outcomes¶
The desired outcomes are consistent treatment, legitimate exceptions, reduced arbitrary variation, clearer decision reasoning, and better auditability. The archetype should make it easier to answer: "What made these cases the same? What made them different? Why did treatment follow or depart from the rule?"
A successful use does not eliminate all unequal outcomes. It makes unequal treatment explainable, scoped, and reviewable.
Tradeoffs¶
The core tradeoff is consistency versus contextual responsiveness. A stable rule improves predictability and legitimacy, but it can become unfair if it ignores relevant differences.
There is also a tradeoff between transparency and discretion. Written reasons improve accountability but may slow decisions or expose contested value judgments.
A third tradeoff is auditability versus burden. Strong records make fairness testable, but they require administrative capacity and may feel bureaucratic if the stakes are low.
Finally, there is a tradeoff between precedent and learning. Consistency with prior cases is valuable, but unfair precedent should not be preserved merely because it is consistent.
Failure Modes¶
False equivalence occurs when cases are treated as the same despite relevant differences in need, risk, entitlement, vulnerability, or harm. Mitigate it with a serious relevant-difference test.
Rigid sameness occurs when equal treatment is applied mechanically and suppresses justified equity adjustments. Mitigate it by making exception paths legitimate rather than shameful.
Reference-class manipulation occurs when actors choose comparison groups to make preferred treatment appear consistent. Mitigate it by publishing case-set scope and checking alternative comparisons.
Exception creep occurs when departures accumulate until the equal-treatment rule is nominal only. Mitigate it with an exception register and periodic rule revision.
Hidden value smuggling occurs when a supposedly technical equivalence rule embeds contested values. Mitigate it by naming the value priority behind the rule.
Neighbor Distinctions¶
The closest first-wave neighbor is Symmetry Breaking for Differentiation. That archetype intentionally breaks equivalence so roles, order, or structure can emerge. Symmetry-Based Fairness does the opposite: it preserves equivalence in treatment unless relevant asymmetry is shown.
Procedural Fairness Design is broader. It designs voice, notice, impartiality, reasons, and review. Symmetry-Based Fairness is narrower: it asks whether equivalent cases receive equivalent treatment.
Equivalence Class Consolidation groups or merges items that are equivalent. Symmetry-Based Fairness uses equivalence to govern treatment decisions.
Equity Adjustment is a neighbor because it justifies asymmetry when relevant context differs. Symmetry-Based Fairness must preserve that possibility rather than collapse fairness into sameness.
Canonical Classification creates stable categories. Symmetry-Based Fairness audits whether categories are causing unjustified treatment differences.
Variants and Near Names¶
Policy Symmetry Review is the governance variant for checking whether a policy is applied consistently across units, locations, populations, or time periods.
Precedent Consistency Review applies the same logic to historical cases. It asks whether a current departure from precedent is justified by a relevant difference.
Anti-Discrimination Symmetry Check is a sensitive variant where identity, status, location, or group markers may be functioning as hidden asymmetry triggers.
Reciprocal Burden Symmetry checks whether parties in reciprocal positions bear equivalent obligations unless a relevant difference justifies asymmetry.
Near names include "like cases alike," "equal treatment principle," "policy symmetry test," and "consistency fairness check." These should generally point back to this archetype or to a recognized variant rather than becoming separate drafts.
Cross-Domain Examples¶
In public administration, applicants with equivalent eligibility facts should receive equivalent benefit decisions, and statutory exceptions should be recorded.
In organizational promotion, candidates with equivalent rubric evidence should be evaluated under the same standard, while job-relevant differences must be documented.
In platform moderation, similar content cases should not receive different enforcement merely because of reviewer, speaker identity, topic label, or presentation order when those factors are irrelevant.
In healthcare operations, patients with equivalent acuity and waiting-time criteria should be scheduled equivalently, while clinical exceptions should be explicit.
In education, equivalent work should receive equivalent marks, while accommodations should be documented as justified asymmetries rather than arbitrary departures.
Non-Examples¶
Randomly assigning one of several tied applicants to a slot is not this archetype. It breaks symmetry to resolve a coordination problem.
Holding a public hearing is not this archetype by itself. It may support procedural fairness, but it does not define equivalent cases or equal treatment.
Merging duplicate IDs or categories is not this archetype unless a fairness-relevant treatment decision depends on the equivalence relation.
Giving extra support to a group with clearly different needs is not a failure of this archetype. It may be a justified equity adjustment when the relevant difference is explicit.