Non-Obvious Manipulability in Mechanism Design

• Non-Obvious Manipulability (NOM) is a framework that relaxes strict strategy-proofness by only ruling out deviations that obviously improve an agent’s best-case or worst-case outcomes.
• It employs formal benchmarks—such as BNOM, WNOM, and option sets—to assess whether manipulations yield uniformly superior results compared to truthful reporting.
• NOM is applied in diverse settings including voting, fair division, allocation, and procurement, offering a flexible and incentive-compatible alternative to dominant strategy truthfulness.

Non-Obvious Manipulability (NOM), also described in much of the literature as a mechanism being “not obviously manipulable,” is an incentive notion that weakens strategy-proofness by excluding only those manipulations that are transparently advantageous under coarse reasoning about attainable outcomes. Rather than requiring truthful reporting to dominate every misreport state by state, NOM asks whether a deviation improves an agent’s most salient benchmarks—typically best-case and worst-case outcomes, or, in infinite-alternative environments, an entire option set. The concept has become a unifying tool across voting, allocation of divisible and indivisible goods, cake-cutting, assignment, hedonic coalition formation, bilateral trade, temporal elections, and budget-feasible procurement, where it often enlarges the feasible design space relative to dominant-strategy truthfulness while still ruling out strategically simple deviations (Aziz et al., 2021, Arribillaga et al., 2023, Archbold et al., 2022, Keijzer et al., 17 Feb 2025).

1. Formal concept and principal definitions

In finite-outcome voting models, NOM is usually formulated through best and worst outcomes under a report. For a voting rule $f$, true preference $\succ_i$, and report $\succ_i'$, the literature defines $B_{\succ_i}(\succ_i',f)$ as the best possible outcome for $i$ and $W_{\succ_i}(\succ_i',f)$ as the worst possible outcome for $i$, both evaluated over all reports of the other voters. A rule is NOM if every profitable manipulation $\succ_i'$ satisfies

$$W_{\succ_i}(\succ_i,f)\ \succeq_i\ W_{\succ_i}(\succ_i',f)$$

and

$$B_{\succ_i}(\succ_i,f)\ \succeq_i\ B_{\succ_i}(\succ_i',f).$$

A deviation is therefore “obvious” if it improves either the worst-case or the best-case consequence of misreporting relative to truth-telling (Aziz et al., 2021).

A parallel formulation appears in direct mechanisms with transfers, where the literature decomposes NOM into best-case and worst-case components. For type $\succ_i$0, misreport $\succ_i$1, and direct mechanism $\succ_i$2, BNOM requires

$\succ_i$3

while WNOM requires

$\succ_i$4

A mechanism is NOM iff it is both BNOM and WNOM. This formulation is central in single-parameter mechanism design, bilateral trade, and procurement settings (Archbold et al., 2022, Keijzer et al., 17 Feb 2025).

In environments with infinitely many alternatives, the best/worst formulation may be ill behaved. The allotment literature therefore replaces extrema by option sets. Given a rule $\succ_i$5, a report $\succ_i$6, and endowment $\succ_i$7, the attainable set for agent $\succ_i$8 is

$\succ_i$9

A profitable misreport $\succ_i'$0 is obvious if

$\succ_i'$1

This says that every possible outcome under the manipulation is strictly better than some truthful possibility. The option-set formulation is used precisely because in infinite spaces worst outcomes may fail to exist, and in efficient single-peaked allotment models the truthful best case is often already the peak (Arribillaga et al., 2023, Arribillaga et al., 2024).

2. Relation to strategy-proofness and adjacent incentive concepts

All strands of the literature treat NOM as strictly weaker than strategy-proofness. Strategy-proofness forbids any profitable misreport whatsoever, whereas NOM allows manipulations to exist provided they are not obvious in the relevant best/worst or option-set sense. Accordingly, strategy-proofness implies NOM in voting, allotment, division, and direct-mechanism settings, but not conversely (Arribillaga et al., 2023, Arribillaga et al., 2024, Arribillaga et al., 17 Dec 2025).

This weakening can be substantial. In voting, many familiar rules remain manipulable in the Gibbard–Satterthwaite sense but are nevertheless NOM under zero-information reasoning about best and worst outcomes. At the same time, later work shows that even NOM-worst may be too permissive as a behavioral benchmark for electoral settings. “Worst-case strategy-proofness” (WCSP) strengthens NOM-worst by requiring truth-telling to remain optimal at the same profile of others’ reports that makes truthful reporting worst; formally, strategy-proofness implies WCSP and WCSP implies NOM-worst. In that framework, plurality, Borda, and Dowdall satisfy NOM-worst but violate WCSP, which isolates a narrower class of profile-contingent but still salient manipulations (Ikeda, 20 Jun 2026).

The decomposition into BNOM and WNOM is not merely technical. In bilateral trade, it produces a sharp asymmetry: under efficiency and individual rationality, no finite subsidy suffices for BNOM, whereas WNOM can be achieved together with weak budget balance. This suggests that best-case and worst-case “obviousness” may have very different design consequences once transfers are present (Archbold et al., 2022).

3. Allotment and division of a single commodity

A major NOM literature concerns the allocation of a single non-disposable divisible commodity. On the single-peaked domain, the central characterization identifies a large class of simple rules. These are own-peak-only rules under which, in excess demand, agents with peaks below equal division get exactly their peaks and all remaining agents receive amounts between equal division and their peaks; the excess-supply case is symmetric. The characterization theorem states that a rule on the single-peaked domain satisfies own-peak-onliness, efficiency, the equal division guarantee, and NOM iff it is a simple rule. The same paper shows that the single-plateaued domain is maximal for these properties, so replacing strategy-proofness by NOM expands the class of admissible rules but not the maximal domain in any essential way (Arribillaga et al., 2023).

A later result restores uniqueness of the classical uniform rule by adding consistency. In the standard allotment problem with single-peaked preferences, the uniform rule

$\succ_i'$2

with $\succ_i'$3, is shown to be the only rule satisfying efficiency, the equal division guarantee, consistency, and NOM. The proof’s distinctive NOM step uses reports whose peak equals equal division, because the equal division guarantee then creates a singleton option set at $\succ_i'$4, making any profitable deviation to that report automatically obvious (Arribillaga et al., 2024).

On more general domains requiring only a unique global maximum, the positive picture changes. In that setting there is no rule satisfying efficiency, NOM, own-peak-onliness, and the equal division guarantee; for two agents, efficiency, NOM, and own-peak-onliness already force dictatorship. If efficiency is weakened to unanimity, however, NOM admits a full characterization: the rules satisfying unanimity, NOM, own-peak-onliness, and the equal division guarantee are exactly the agreeable rules, under which each agent receives either her peak or equal division and the set of peak-receivers must form an agreeable coalition. A key step in the proof is that NOM collapses each agent’s option set to

$\succ_i'$5

This suggests that under general preferences NOM retains significant bite, but only after efficiency is relaxed (Arribillaga et al., 17 Dec 2025).

4. Voting and electoral environments

In voting, NOM has been used both as a permissive alternative to strategy-proofness and as a structural lens on when manipulation is salient. One broad result shows that many prominent rules are NOM: Condorcet extensions, plurality, STV, plurality with runoff, Borda, and strict scoring rules with weakly diminishing differences. For positional scoring rules, the paper also gives the threshold

$\succ_i'$6

as a sufficient condition for NOM via almost-unanimity, and it fully characterizes $\succ_i'$7-approval: $\succ_i'$8 Accordingly, obvious manipulability tends to arise when the number of alternatives is large relative to the number of voters, especially under deterministic tie-breaking (Aziz et al., 2021).

For tops-only rules, NOM admits an exact veto-based characterization. A tops-only rule is NOM iff every veto is a strong veto, meaning that if an agent can veto alternative $\succ_i'$9 at all, then any report not putting $B_{\succ_i}(\succ_i',f)$0 on top must also veto $B_{\succ_i}(\succ_i',f)$1. Formally, a tops-only rule is NOM iff $B_{\succ_i}(\succ_i',f)$2 for every agent $B_{\succ_i}(\succ_i',f)$3. This yields sharp subclasses: a median voter scheme is NOM iff its extreme fixed ballots satisfy

$B_{\succ_i}(\succ_i',f)$4

while a voting-by-quota rule is NOM iff

$B_{\succ_i}(\succ_i',f)$5

The result shows that, in tops-only environments, NOM is fundamentally a condition on the robustness of veto power rather than on the absence of manipulation altogether (Arribillaga et al., 2022).

The recent literature also studies stronger refinements motivated by partial strategic information. In scoring-rule voting with fixed-order tie-breaking, WCSP is much more selective than NOM-worst: plurality, Borda, and Dowdall fail WCSP, while anti-plurality satisfies WCSP exactly when

$B_{\succ_i}(\succ_i',f)$6

This does not displace NOM, but it clarifies that the latter is a comparatively permissive benchmark once voters can reason about feared truthful worst-case scenarios (Ikeda, 20 Jun 2026).

A related application appears in temporal elections. In an offline sequential-choice model with additive approval utilities over time, the utilitarian mechanism GreedyUtil is strategyproof, while any deterministic mechanism that always outputs an egalitarian-welfare-maximizing outcome is not NOM in the general domain. Under the complete-preference restriction $B_{\succ_i}(\succ_i',f)$7 for every agent and timestep, however, a mechanism $B_{\succ_i}(\succ_i',f)$8 that selects an egalitarian optimum and breaks ties lexicographically in agents’ utility vectors satisfies NOM. The contrast isolates empty approval sets as the main source of obvious manipulation in that model (Elkind et al., 2024).

5. Fair division, assignment, and coalition formation

NOM has also been used to revisit fair division without money. In deterministic cake-cutting, strategy-proofness is known to be prohibitively costly for fairness, but the leftmost leaves mechanism shows that proportionality and NOM are compatible. Leftmost leaves is a Robertson–Webb implementation that, at each stage, asks every remaining agent to cut a $B_{\succ_i}(\succ_i',f)$9-share of the current remainder and allocates the leftmost such piece. The mechanism is proportional and NOM, whereas cut-and-choose, cut-middle, Banach–Knaster last diminisher, and the original Dubins–Spanier moving knife are all obviously manipulable. The comparison between moving knife and leftmost leaves is especially notable because the two procedures coincide under truthful play, yet only the latter avoids obvious manipulation (Ortega et al., 2019).

For indivisible goods with additive valuations, NOM bypasses classical impossibilities in a similarly strong way. Round-Robin is deterministic, EF1, and NOM; utilitarian-welfare maximization is NOM for $i$0 agents but obviously manipulable for $i$1; and every mechanism that exactly maximizes egalitarian welfare or Nash welfare is obviously manipulable. The paper’s main technical result is an approximation-preserving black-box reduction from designing EF1 and NOM mechanisms to designing clean, non-wasteful EF1 algorithms. This yields deterministic fair-and-efficient NOM mechanisms in settings where truthful mechanisms are known not to exist (Psomas et al., 2022).

In assignment problems, the rank-minimizing objective offers another separation between strategy-proofness and NOM. A random assignment mechanism $i$2 is NOM if, for every agent $i$3, true ranking $i$4, and misreport $i$5,

$i$6

and

$i$7

Every full-support rank-minimizing mechanism is NOM, so the uniform rank-minimizing mechanism is NOM. Without full support, NOM depends on the selection rule: rank-minimizing serial dictatorship is NOM in unit-capacity markets, but becomes obviously manipulable when some object has capacity greater than $i$8, and even unit-capacity markets admit non-full-support rank-minimizing mechanisms that are obviously manipulable (Troyan, 2022).

Hedonic coalition-formation models supply two further contrasts. For friends-appreciation hedonic games, there exists an optimal NOM mechanism, although computing an optimal partition is NP-hard; the same paper gives a polynomial-time NOM mechanism with a $i$9-approximation and shows that no optimum mechanism can be NOM under enemies-aversion preferences (Flammini et al., 1 Jan 2025). In additively separable and fractional hedonic games with arbitrary weights, every optimal mechanism is NOM; for arbitrary, non-negative, and bounded continuous domains, scale-independence implies NOM, and for any approximation ratio $W_{\succ_i}(\succ_i',f)$0, the existence of a NOM $W_{\succ_i}(\succ_i',f)$1-approximation is equivalent to the existence of a scale-independent one. The resulting matching-based NOM mechanism gives an $W_{\succ_i}(\succ_i',f)$2-approximation for additively separable hedonic games and a $W_{\succ_i}(\succ_i',f)$3-approximation for fractional hedonic games (Ferraioli et al., 19 May 2025).

6. Single-parameter design, procurement, and broader frontier

A major mechanism-design development places NOM in the single-parameter domain. Using a cycle-monotonicity framework adapted to best-case and worst-case labels, the literature shows that an allocation rule for single-parameter agents is BNOM- or WNOM-implementable iff it is overlapping: one can choose, from each report’s attainable allocations, a monotone sequence across reports. This is far weaker than the pointwise monotonicity required for dominant-strategy implementability. For single-line labelings, the corresponding payments satisfy an envelope-style formula

$W_{\succ_i}(\succ_i',f)$4

In bilateral trade, the same framework yields a stark dichotomy: no efficient, individually rational BNOM mechanism admits any finite subsidy bound, whereas there exists an efficient, individually rational, weakly budget-balanced WNOM mechanism (Archbold et al., 2022).

Budget-feasible procurement pushes the decomposition further. In that setting, NOM is the conjunction of BNOM and WNOM, and each component admits a threshold characterization. Under individual rationality and normalized payments, BNOM is equivalent to threshold golden tickets, while WNOM is equivalent to threshold wooden spoons. These characterizations imply a tight deterministic frontier: no deterministic budget-feasible NOM mechanism can beat approximation ratio $W_{\succ_i}(\succ_i',f)$5, and the mechanism $W_{\succ_i}(\succ_i',f)$6 achieves exactly that bound for monotone subadditive valuations. The same work shows that WNOM alone has frontier $W_{\succ_i}(\succ_i',f)$7, and that randomized universally NOM mechanisms can achieve an expected approximation ratio arbitrarily close to $W_{\succ_i}(\succ_i',f)$8 (Keijzer et al., 17 Feb 2025).

A broader implication is that NOM often enlarges the design space without removing all structure. In some domains it yields sharp characterizations—the simple rules of single-peaked allotment, agreeable rules under unanimity, overlapping allocation rules for single-parameter agents, strong-veto tops-only voting rules, or threshold golden-ticket procurement mechanisms. In other domains it remains selective enough to rule out exact welfare maximization, as with egalitarian and Nash welfare for indivisible goods or deterministic egalitarian maximization in temporal elections. A plausible computational implication is suggested by the partial-information manipulation literature: although that work does not formalize NOM, it shows that under incomplete information weak and opportunistic manipulation often become computationally hard even when complete-information manipulation is easy, which is consistent with viewing “obvious” manipulation as a structurally smaller target than manipulation simpliciter (Dey et al., 2016).

Taken together, the NOM literature identifies a persistent middle ground between unrestricted manipulability and dominant-strategy truthfulness. The central pattern is that once strategic reasoning is restricted to salient best-case, worst-case, or option-set comparisons, fairness and efficiency objectives that are impossible under strategy-proofness frequently become attainable, but only through mechanisms whose attainable-outcome structure is tightly constrained.