Skip to content

Axiomatic Incompatibility

Prime #
648
Origin domain
Mathematics Logic
Subdomain
impossibility results → Mathematics Logic

Core Idea

Axiomatic incompatibility names the structural pattern in which a small, individually plausible set of axioms placed on some mapping, aggregation, or design rule turns out to be jointly unsatisfiable: there is no construction over the relevant domain that satisfies all of the axioms at once. The design problem therefore has no solution meeting all the stated properties, and the designer must knowingly sacrifice at least one axiom; the choice of which to sacrifice becomes the load-bearing decision. Crucially, the pattern is given by proof rather than by "no construction found yet": the impossibility is structural, not contingent on cleverness. No further iteration on the design will produce a compliant solution, and the design conversation must move from search to choosing which axiom to relax.

The structural commitment is the existence proof of non-existence. The work of an axiomatic-incompatibility result is to identify a domain, write down a short list of axioms each defensible in isolation, and prove that no element of the domain can satisfy them simultaneously. Once such a result is in hand, a whole class of design searches is closed at once. The pattern's distinctive value is precisely this closure: it converts an open-ended, potentially endless search into a bounded decision over a known trade-off, and it does so with the certainty that only a proof can supply. What looked like a hard engineering problem awaiting a clever solution is revealed to be a settled impossibility awaiting an explicit choice.

How would you explain it like I'm…

Wishes That Fight

Sometimes you want a bunch of nice things all at once, but they secretly fight each other so you can't have them all — no matter how clever you are. It's like wanting a sandwich that is hot AND cold AND only one bite: pick any two and the third won't fit. When that happens, the smart move isn't to try harder; it's to decide which wish to give up.

Rules That Can't All Win

Imagine you write down a short list of fair-sounding rules for a game, and each rule by itself seems totally reasonable. Sometimes you can prove that no possible game can follow ALL of them at the same time. That's not because you haven't found the trick yet — it's been proven that the trick can't exist. So the only choice left is to pick which rule to drop. The hard part stops being 'find a solution' and becomes 'decide what to sacrifice.'

Proven Impossible Trade-off

Axiomatic incompatibility is when a small set of individually sensible requirements turns out to be jointly impossible — there is no design over the relevant domain that satisfies all of them at once. The key word is proven, not 'not-found-yet': a real impossibility result writes down the axioms and demonstrates that no element of the domain can meet them simultaneously. This is much stronger than a problem being hard, because no extra cleverness or iteration will ever produce a compliant solution. Once you have such a proof, a whole family of design searches is closed in one stroke. The conversation has to shift from searching for the perfect answer to deliberately choosing which property to relax — and that choice becomes the real decision.

 

Axiomatic incompatibility names the pattern where a small, individually plausible set of axioms imposed on some mapping, aggregation, or design rule turns out to be jointly unsatisfiable: no construction over the relevant domain satisfies all of them at once. The structural commitment is an existence proof of non-existence — you identify a domain, write down a short list of axioms each defensible in isolation, and prove that no element of the domain can satisfy them simultaneously. This distinguishes it sharply from 'no solution found yet': the impossibility is structural, so no further iteration on the design will ever yield a compliant solution. The designer must knowingly sacrifice at least one axiom, and which one to give up becomes the load-bearing decision. The distinctive value of such a result is closure: it converts an open-ended, potentially endless search into a bounded decision over a known trade-off, with the certainty only a proof supplies. What looked like a hard engineering problem awaiting a clever fix is revealed as a settled impossibility awaiting an explicit choice.

Structural Signature

a domain of candidate constructions — a small set of individually plausible axioms — the joint-unsatisfiability proof — the forced sacrifice of at least one axiom — the induced trade-off frontier — the menu of resolution moves

The pattern is present when each of the following holds:

  • A domain. A well-defined space of candidate mappings, aggregation rules, or designs is fixed — every voting rule, every consensus protocol, every classifier.
  • An axiom set. A short list of properties is imposed, each defensible in isolation and each one a designer would naturally want.
  • The joint-unsatisfiability proof. It is proven, not merely unobserved, that no element of the domain satisfies all the axioms simultaneously; the impossibility is structural, not contingent on cleverness.
  • The forced sacrifice. Because no compliant construction exists, every actual design must violate at least one axiom; the design problem has no all-satisfying solution.
  • The trade-off frontier. The proof induces a Pareto structure over the axiom set: each design occupies a position satisfying some axioms and violating the rest, and the load-bearing decision is which to relax.
  • The resolution menu. A fixed family of escapes — relax an axiom, restrict the domain, approximate, randomise, add information, or reframe — follows from the logical form rather than the substrate.

The components compose to convert an open-ended search ("find the rule satisfying all desiderata") into a bounded decision ("choose which axiom to sacrifice, transparently"). The closure is what the proof buys: a class of design searches is settled at once, with the certainty only a proof supplies.

What It Is Not

  • Not hierarchical decomposability. hierarchical_decomposability is about whether a system can be broken into nested independent parts; axiomatic incompatibility is about whether a set of desiderata can be jointly satisfied. One concerns structure of a system, the other joint satisfiability of properties.
  • Not commensurability. commensurability asks whether two things can be measured on a common scale; axiomatic incompatibility asks whether several properties can hold at once. Incompatible axioms may be perfectly commensurable yet provably unsatisfiable together.
  • Not a trade-off. trade_offs (and multiobjective_optimization) describe continuous frontiers where more of one good costs some of another; axiomatic incompatibility is the proof that no point satisfies all the stated constraints — a discrete impossibility, not a smooth exchange rate.
  • Not incentive compatibility. incentive_compatibility is a property a mechanism may or may not have; axiomatic incompatibility is the meta-result that a set of such desired properties cannot coexist (as in Holmström's theorem).
  • Not falsifiability. falsifiability concerns whether a claim could be empirically refuted; axiomatic incompatibility is a proven non-existence over a design space, settled by deduction rather than observation.
  • Common misclassification. Treating a hard but open design problem, or a merely conjectured hardness, as a settled impossibility — and abandoning a winnable search. Catch it by locating the actual theorem and reading its assumptions; without a proof of non-existence, the matter is open and search, not trade-off selection, is the right activity.

Broad Use

A growing inventory of named results across domains shares the shape. In social choice, Arrow's theorem shows that unrestricted domain, Pareto, independence of irrelevant alternatives, and non-dictatorship are jointly unsatisfiable for any aggregation rule over three or more alternatives; Gibbard-Satterthwaite shows no non-dictatorial ordinal voting rule is strategy-proof. In apportionment, Balinski-Young shows quota and two monotonicity properties cannot all hold. In mechanism design, Holmström's impossibility rules out budget balance, efficiency, individual rationality, and dominant-strategy truthfulness together in team production. In distributed systems, the CAP theorem rules out consistency, availability, and partition-tolerance jointly under partition, and FLP rules out deterministic asynchronous consensus tolerating even one crash. In physics, the Heisenberg uncertainty principle forbids joint sharp position and momentum, and the no-cloning theorem forbids copying arbitrary unknown quantum states. In computability, Rice's theorem makes any non-trivial semantic property of programs undecidable. In machine learning, the No-Free-Lunch theorems rule out universal superiority across all problem distributions. In fairness, the Chouldechova and Kleinberg-Mullainathan-Raghavan results show calibration and the two balance conditions cannot all hold under unequal base rates. And in logic, Gödel's incompleteness shows consistency, completeness, and sufficient expressiveness are jointly unsatisfiable for a formal system.

Clarity

Naming the pattern clarifies a recurring design conversation that otherwise stalls in "if we just try harder we can find a rule that satisfies all the desiderata." The structural fact — no such rule exists — converts the conversation from search to trade-off. Designers can then make the trade-off explicitly: which axiom is least important to preserve, and what is the substantive cost of relaxing it? The clarification also exposes a common rhetorical move — invoking plausible properties one at a time to criticise an existing rule, without noting that no rule satisfies all the properties the critic appeals to. Recognising axiomatic incompatibility lets an analyst respond that yes, this rule violates property X, but any rule satisfying X must violate Y or Z, so the question is which trade-off the critic proposes. The conversation moves from purity to choice. The clarifying force, in short, is to distinguish a contingent shortfall — this design fails to meet a goal that some other design might meet — from a structural impossibility, in which no design can meet all the goals and the only honest move is to choose among them transparently.

Manages Complexity

The pattern compresses a wide range of "why can't we have all of these?" complaints across substrates into a single diagnostic: is there a proof of axiomatic incompatibility for this collection of desiderata, and if so, which axiom does the proposed solution sacrifice? The intervention catalogue is correspondingly uniform. One can relax an axiom — drop independence of irrelevant alternatives, accept some availability loss under partition, allow approximate strategy-proofness. One can restrict the domain, since the result may fail in a narrower setting such as single-peaked preferences or restricted query languages. One can approximate the axioms, accepting bounded failures rather than perfect satisfaction, as in eventual consistency or approximate privacy. One can randomise, since some impossibility results yield to randomised rules. One can add information, introducing stronger primitives or side-channels that change the domain. Or one can reframe the question, recognising the desiderata as irreconcilable rather than the rule as defective. The same six resolution strategies apply whether the impossible quartet is Arrow's, CAP's, or Chouldechova's, because they are properties of the logical form rather than the substrate content. The complexity reduction is that a catalogue of domain-specific "puzzles" and "tensions" becomes one object with one taxonomy of escapes.

Abstract Reasoning

The pattern is naturally formalised as a non-existence theorem over a domain and a set of axioms: for no element of the domain do all the axioms hold. The proof structure is typically a constructive contradiction — assume an element satisfies all axioms, derive a contradiction. Once such a theorem is in hand, it induces a Pareto frontier over the axiom set: every actual design satisfies a subset of axioms and violates the complement. The reasoning supports several cross-substrate moves. For any design proposal claiming to satisfy all of an incompatibility theorem's axioms, the violated axiom can be located in advance, because it is provably present. For any design competition over an axiom-incompatible set, the competition is over which axiom to relax, not over whether to relax any. And for any rhetorical demand for all the axioms at once, the demand is over-determined and can be dismissed as such. The pattern also generates taxonomies of resolution strategies — relax, restrict, approximate, randomise, augment, reframe — that transfer across substrates precisely because they are properties of the logical form, not of the substrate content. This is what makes the abstraction a reasoning tool rather than a mere observation: it tells the reasoner, in any domain, exactly what kinds of escape are available once an impossibility is recognised.

Knowledge Transfer

The structure transfers because the logical form — a small axiom set, joint unsatisfiability proven on a defined domain — is substrate-free, and an analyst who has made the move in one field makes it in another with no retooling. A distributed-systems engineer reading the CAP theorem develops the same intellectual move that a social-choice theorist makes with Arrow: identify the desiderata, find the impossibility result that constrains them, and choose a trade-off explicitly. A fairness-in-ML researcher reading Chouldechova's result recognises the same structure a physicist invokes for Heisenberg, a logician for Gödel, and a mechanism designer for Holmström. Because the resolution taxonomy is a property of the form, the engineer who has learned to relax, restrict, approximate, randomise, augment, or reframe in one setting arrives in the next already holding the menu of escapes. The transfer is especially productive for recognising that an unsolved design problem is in fact a known impossibility. Many design conversations carry a folk conviction that "someone smart enough could solve this," when the underlying issue is a proven impossibility the participants have simply not encountered in their own field's literature. Naming axiomatic incompatibility as a prime gives analysts a standing prompt to check: is there an impossibility theorem in this design space? A team that asks this question early can save months of fruitless iteration — the characteristic failure mode of the pattern is precisely not knowing the impossibility exists and iterating toward an unreachable solution. Because the pattern carries no normative or institutional load — its vocabulary is purely logical and relational, an axiom set and a joint-unsatisfiability proof — it imports into any field without friction, and the recognition that a design problem is closed by proof rather than open to cleverness is the same recognition whether the substrate is voting, consensus, measurement, computation, or allocation. The strongest transfer, then, is the habit of treating a stubbornly unsolvable design problem as a candidate impossibility, searching the relevant literature for the theorem, and, on finding it, redirecting all effort from search to the explicit, transparent selection of which axiom to sacrifice.

Examples

Formal/abstract

Arrow's impossibility theorem is the pattern's foundational worked instance. The domain is the space of social welfare functions — rules that aggregate the ranked preferences of three or more individuals over three or more alternatives into a single collective ranking. The axiom set is short and each member is individually compelling: unrestricted domain (the rule must handle any profile of individual rankings), Pareto efficiency (if everyone prefers A to B, the collective ranks A above B), independence of irrelevant alternatives (the collective ordering of A and B depends only on individuals' orderings of A and B, not of some third option C), and non-dictatorship (no single individual's preference always determines the outcome). The joint-unsatisfiability proof shows that no aggregation rule satisfies all four at once: assume one does, and a constructive argument identifies a "pivotal" voter whose switch flips the social outcome for one pair, then extends this pivotality across all pairs until that voter is shown to be a dictator — contradicting non-dictatorship. The forced sacrifice follows: every actual voting rule must violate at least one axiom. The trade-off frontier is then visible — ranked-pairs and Borda methods relax independence of irrelevant alternatives, restricted-domain methods relax unrestricted domain, and so on. The resolution menu is exactly the form-level catalogue: restrict the domain (single-peaked preferences escape the theorem entirely, as the median-voter rule then satisfies the rest), relax an axiom (accept IIA violations and the attendant vulnerability to agenda-setting), or add information (cardinal utility, as in range voting, changes the domain and sidesteps the ordinal-only premise). The proof converts an endless search for the "fair voting system" into the bounded, honest decision of which property to give up.

Mapped back: Arrow instantiates every role of the signature — a fixed domain of constructions, a small set of plausible axioms, a constructive joint-unsatisfiability proof, a forced sacrifice, an induced trade-off frontier, and a form-level resolution menu — and shows the prime's central closure: the search for an all-satisfying rule is settled at once by proof, leaving only the choice of which axiom to relax.

Applied/industry

The CAP theorem performs the identical structural work in distributed-systems engineering, and the ML-fairness impossibility result performs it in algorithmic decision-making — three genuine domains sharing one logical form. In distributed databases the domain is the space of replicated data stores; the axioms are consistency (every read returns the most recent write), availability (every request receives a non-error response), and partition tolerance (the system keeps operating despite dropped messages between nodes). The joint-unsatisfiability proof shows that when a network partition occurs, a system cannot remain both consistent and available: a node on one side of the partition either answers (risking staleness, violating consistency) or refuses (violating availability). The forced sacrifice is real and operational — under partition, an architect must choose CP (refuse writes to stay consistent, as in a strongly consistent quorum store) or AP (accept writes and reconcile later, as in eventual-consistency stores). The resolution menu is the prime's: approximate the axioms via eventual consistency, restrict the domain to partition-free single-datacenter deployments, or reframe by tuning per-operation consistency levels. The fairness result is the same object again: under unequal base rates between groups, a risk score cannot simultaneously be calibrated and satisfy both balance-for-the-positive-class and balance-for-the-negative-class — a proof that converts heated "make the algorithm fair on every metric" debates into the disciplined recognition that the metrics are jointly unsatisfiable and a transparent choice among them is the only honest move. In each case, the practitioner who recognises the impossibility stops iterating toward an unreachable solution — the pattern's characteristic failure mode — and moves directly to selecting the trade-off.

Mapped back: CAP and the fairness impossibility are the same logical object as Arrow under different substrates — a small axiom set proven jointly unsatisfiable on a defined domain — and in each the engineering payoff is identical: redirect effort from a doomed search to the explicit, defensible choice of which axiom to sacrifice.

Structural Tensions

T1 — Domain as Stated versus Domain as Restrictable (Scopal). The impossibility holds only over the domain the axioms quantify across; narrow the domain and the theorem can simply evaporate, as single-peaked preferences dissolve Arrow. The failure mode is treating the impossibility as a law of nature that forbids the goal everywhere, when in fact the real-world domain is a tractable subset where all axioms can hold. Diagnostic: ask what range of inputs the proof actually requires, then ask whether your instances live there; a team paralysed by a theorem whose worst-case profiles never occur in their setting has mis-scoped the result. The competing move is domain restriction, not trade-off.

T2 — Sacrifice Whole Axiom versus Relax Continuously (Sign/Degree). The clean statement forces sacrificing an axiom, framing the choice as discrete: keep four, drop one. But many results admit graded relaxation — approximate strategy-proofness, epsilon-consistency, bounded unfairness — where every axiom holds to a degree. The failure mode is binary thinking that throws an axiom out entirely when a small, quantified violation of two would have served better. Diagnostic: ask whether the axioms are all-or-nothing predicates or admit a tolerance parameter; if the latter, the real decision space is a continuous frontier, and discrete "which one to drop" framing hides the best designs in the interior.

T3 — Worst-Case Proof versus Typical-Case Reality (Measurement). An impossibility is usually a worst-case existence claim — some profile breaks every rule — yet practice runs on typical inputs where the offending case is vanishingly rare. The failure mode is over-conservatism: refusing a design that violates an axiom only on adversarial inputs that the deployment never sees, or conversely assuming the rare break is harmless when an adversary can force it. Diagnostic: ask how often, and at whose instigation, the impossibility-triggering case actually arises; the answer decides whether to engineer for the worst case or accept the typical-case-safe design that the bare theorem appears to forbid.

T4 — Proven Impossibility versus Unproven Hardness (Epistemic). The prime's force rests on a proof of non-existence, which closes search with certainty. The characteristic confusion is mistaking an unsolved-but-open problem, or a merely conjectured hardness, for a settled impossibility — and abandoning a search that was actually winnable. The inverse failure is iterating forever on a problem that a known theorem already closed. Both are misreadings of epistemic status. Diagnostic: locate the theorem and read its assumptions; if there is no proof, the matter is open and problem_space search reasoning applies; if there is, search is the wrong activity and only trade-off selection remains.

T5 — Axioms as Fixed versus Axioms as Negotiable (Reframing). The result takes the axiom set as given and proves them jointly unsatisfiable — but the axioms are chosen, often inheriting a framing no longer load-bearing. Independence of irrelevant alternatives looks essential until cardinal information is admitted and it dissolves. The failure mode is accepting the incompatibility as fate while never auditing whether one axiom is a disguised modelling convenience rather than a genuine requirement. Diagnostic: for each axiom ask "who needs this, and would a weaker primitive satisfy them?"; an impossibility built on an over-strong axiom is reframed away by abstraction of the requirement, not resolved by sacrifice.

T6 — Single Result versus Composed Constraints (Coupling). Real designs sit under several impossibility theorems at once, and the escapes from one can collide with another — relaxing consistency to dodge CAP may reopen a problem another theorem governs. The failure mode is treating each impossibility in isolation, picking a locally optimal sacrifice that violates an axiom a second theorem makes binding. Diagnostic: inventory all the incompatibility results live in the design space and check whether the resolution menus intersect; when two frontiers couple, the trade-off is joint, and choosing on one axis without the other produces a design that is Pareto-honest locally but globally infeasible.

Structural–Framed Character

Axiomatic Incompatibility sits firmly at the structural end of the structural–framed spectrum. It is a pure logical form — a small axiom set proven jointly unsatisfiable over a defined domain — and nothing about its meaning depends on a particular field's vocabulary or assumptions. Every diagnostic points one way, consistent with its aggregate of 0.0.

The pattern carries no home vocabulary that must travel with it: the identical "axiom set, joint-unsatisfiability proof, forced sacrifice" shape is told as Arrow's theorem in social choice, the CAP theorem in distributed systems, Heisenberg's principle in physics, Gödel's incompleteness in logic, and the Chouldechova fairness result in machine learning, each in its own field's words, with the logical skeleton imported nowhere. It carries no inherent approval or disapproval — an impossibility result is neither good nor bad, merely a fact about what cannot coexist; the only normativity is the designer's downstream choice of which axiom to relax, which is external to the prime. Its origin is formal: the signature is stated as a non-existence theorem over a domain, with no appeal to any institution or human norm. It runs indifferently across logical, physical, and computational substrates — quantum no-cloning and program-property undecidability instantiate it as readily as voting rules — so it requires no human practice to exist; the impossibilities in physics and computability hold whether or not anyone is designing anything. And to invoke it is to recognise a proven incompatibility already present in the design space, not to import an interpretive frame: the diagnostic is simply to locate the theorem and read its assumptions. On every criterion the prime reads structural.

Substrate Independence

Axiomatic Incompatibility is a maximally substrate-independent prime — composite 5 / 5 on the substrate-independence scale. Its domain breadth is total: the joint-unsatisfiability-by-proof shape is recognised, not translated, as Arrow's and Gibbard-Satterthwaite's theorems in social choice, the CAP and FLP theorems in distributed systems, the Heisenberg uncertainty and no-cloning theorems in physics, Gödel's incompleteness and Rice's theorem in logic and computability, Holmström's impossibility in mechanism design, the No-Free-Lunch theorems in machine learning, and the Chouldechova/Kleinberg fairness results in algorithmic decision-making. Its structural abstraction is complete because the signature is a pure logical form — a small axiom set proven jointly unsatisfiable over a defined domain — carrying no field vocabulary, no normative load (an impossibility is neither good nor bad), and no human-practice presupposition; the physics and computability impossibilities hold whether or not anyone is designing anything. Its transfer evidence is concrete and unusually strong: the same six-move resolution menu (relax, restrict, approximate, randomise, augment, reframe) is a property of the logical form rather than the substrate, so an analyst who has chosen a trade-off under CAP arrives at Arrow or Chouldechova already holding the menu, and a string of named theorems across utterly different fields demonstrates the carry directly. Nothing caps this prime; every component reads at ceiling.

  • Composite substrate independence — 5 / 5
  • Domain breadth — 5 / 5
  • Structural abstraction — 5 / 5
  • Transfer evidence — 5 / 5

Neighborhood in Abstraction Space

Axiomatic Incompatibility sits among the more crowded primes in the catalog (31st percentile for distinctiveness): several abstractions describe nearly the same structure, so a description that fits it will tend to fit its neighbors too — transporting it usually means disambiguating within this family rather than landing on it exactly.

Family — Formal Methods & Idealized Models (31 primes)

Nearest neighbors

Computed from structural-signature embeddings · 2026-06-14

Not to Be Confused With

The most consequential confusion is with trade_offs (and its optimisation cousin multiobjective_optimization). A trade-off describes a continuous frontier: you can have more of one objective by accepting less of another, and the design task is to choose a point on a smooth Pareto curve where every axis is satisfied to a degree. Axiomatic incompatibility is a categorically different claim — a proof that no construction satisfies all the stated properties at all, so at least one must be sacrificed entirely (or, where graded relaxation exists, violated by a definite amount). The trade-off frame presupposes the desiderata are jointly achievable in degrees; the incompatibility frame establishes they are not jointly achievable in full. The two interact — an incompatibility theorem often induces a trade-off frontier among the achievable subsets — but conflating them is a live error in both directions: reasoning about Arrow's theorem as if it were a tunable trade-off invites the futile search for a voting rule that "balances" all four axioms, when the theorem proves no rule satisfies all four; conversely, reading a genuine continuous trade-off as a hard impossibility prematurely forecloses designs that the smooth frontier actually permits. The discriminating question is whether there is a proof of non-existence over the design space, or merely a cost curve.

Axiomatic incompatibility must also be separated from commensurability. Commensurability asks whether two values or quantities can be placed on a shared scale so they can be compared or summed; incommensurability is a property of the measurement relation between things. Axiomatic incompatibility asks whether a set of properties can be jointly satisfied by some construction. These are independent: a set of axioms can be perfectly commensurable (all expressible in the same currency) yet provably jointly unsatisfiable, and incommensurable values need not be axiomatically incompatible at all. A practitioner who diagnoses a stuck design as an incommensurability problem will look for a common scale, when the real situation is that even on a shared scale no construction meets all the constraints — a different problem with a different resolution menu (relax, restrict, approximate, randomise, augment, reframe).

A third confusion, sharpened by tension T4, is with the difference between a proven impossibility and falsifiability or mere conjectured hardness. Falsifiability concerns whether a claim is empirically refutable; a problem's being "open" or "believed hard" concerns its epistemic status as unsolved. Axiomatic incompatibility rests specifically on a deductive proof of non-existence, which closes search with certainty no empirical or conjectural status can supply. The characteristic mistake is treating an unsolved-but-open design problem as a settled impossibility (abandoning a winnable search) or, inversely, iterating forever on a problem a known theorem already closed. The fix in both cases is the same: locate the theorem, read its assumptions, and check whether a proof exists.

For practitioners these distinctions decide the activity. If the situation is a trade-off, the work is choosing a frontier point. If it is incommensurability, the work is constructing or abandoning a common scale. If it is an open problem, the work is search. Only when there is an actual joint-unsatisfiability proof does the work become the disciplined, transparent selection of which axiom to sacrifice — and mistaking any of the neighbours for axiomatic incompatibility (or vice versa) sends effort in exactly the wrong direction.

Solution Archetypes

No catalogued solution archetypes reference this prime yet.