Reconciling Probabilistic Fairness with Realized Outcomes

This direction is merged into another direction: ex-ante fairness. Its last version stays readable and citable here.

Version 1 of 132 papers43 conceptsfrozen 2026-10-04

When resources are indivisible—like a single kidney for transplant, a seat in a prestigious school, or a specific plot of public housing—deterministic fairness is often mathematically impossible. You cannot give half a kidney to two people and expect a good result. To solve this, researchers turn to randomization. By using a lottery, we can ensure that everyone has an equal chance of receiving the resource.

However, this creates a fundamental tension between two stages of the decision-making process: ex-ante fairness (fairness in expectation, before the dice are rolled) and ex-post fairness (fairness in the actual outcome, after the winner is chosen).

The Gap Between Expectation and Reality

The core problem is that a process can be perfectly fair on average while being perceived as deeply unfair in practice. As Andreas Krause and Hoda Heidari point out in Preventing Disparate Treatment in Sequential Decision Making, even if a model’s internal probability scores are consistent, the realized binary decisions can vary wildly.

Consider a lottery where two agents, Alice and Bob, both want a single item. A 50/50 coin flip is perfectly ex-ante envy-free: both agents are satisfied with their 0.5 expected utility. But once the coin lands, the ex-post reality is that one person has everything and the other has nothing. This discrepancy is a major hurdle for the public acceptance of algorithmic systems. As noted by Richard Cole and Yixin Tao in On the Existence of Pareto Efficient and Envy Free Allocations, a mechanism that is fair in expectation may still be rejected by participants once the final, lopsided assignment is revealed.

The Best of Both Worlds Paradigm

To bridge this gap, a growing body of research pursues the Best of Both Worlds (BoBW) framework. The goal is to design randomized mechanisms that provide strong guarantees at both stages: the lottery itself must be fair, and every possible outcome that could be sampled from that lottery must also meet a minimum standard of fairness.

In Best of Both Worlds: Agents with Entitlements, Giovanna Varricchio and Martin Hoefer extend this to settings where agents have different "weights" or entitlements, proving that it is possible to satisfy both ex-ante and ex-post guarantees simultaneously even with unequal players. Similarly, Haris Aziz and Jeremy Vollen explore this in Best-of-Both-Worlds Fairness in Committee Voting, seeking algorithms that ensure a fair process for all voters while guaranteeing that any selected committee is representative.

Mathematical Limits and Incompatibilities

The search for BoBW solutions is difficult because ex-ante and ex-post properties often conflict. Jannik Peters demonstrates in Candidate Monotonicity and Proportionality for Lotteries and Non-Resolute Rules that certain ex-ante proportionality definitions are mathematically incompatible with the strongest ex-post axioms.

This "no free lunch" reality is further echoed by Telikepalli Kavitha and Rohit Vaish in Best-of-Both-Worlds Guarantees with Fairer Endings. They prove that exact ex-ante envy-freeness is fundamentally incompatible with the strong ex-post guarantee of EFX (envy-freeness up to any good). Because of these limits, the research direction has shifted toward finding the optimal approximation ratios—how close can we get to perfect fairness in expectation without ruining the fairness of the final result?

Finding the Middle Ground: Interim Fairness

If ex-ante is too weak and ex-post is too strict, is there a middle ground? Ioannis Caragiannis and colleagues propose a "Goldilocks" solution in Interim envy-freeness: A new fairness concept for random allocations.

Their concept, Interim Envy-Freeness (iEF), evaluates fairness at the moment an agent learns their own assignment but before they know what everyone else received. This captures the intuition of a participant who thinks, "I'm happy with what I got, and I don't think anyone else is likely to have received something better than my fair share."

Why It Works: From Theory to Practice

This research isn't just theoretical; it has direct implications for high-stakes resource allocation:

  1. Kidney Exchanges: Sharmila Duppala and Aravind Srinivasan show in Group Fairness in Set Packing Problems that randomized algorithms can ensure different patient groups are represented fairly in transplant chains, even when a deterministic "perfect" packing is impossible.
  2. Participatory Budgeting: In Fair Lotteries for Participatory Budgeting, Toby Walsh and co-authors argue that lotteries can prevent "winner-take-all" outcomes where a slight majority of voters gets all the funding, ensuring minority groups have a fair expected share of the budget.
  3. Social Media: Stefania Ionescu and Aniko Hannak illustrate the stakes in The role of luck in the success of social media influencers, showing how systems that are fair in expectation can still lead to extreme ex-post inequality due to early random advantages (luck).

By formalizing the relationship between the "dice roll" and the "landing," this field provides the tools to build systems that are not just mathematically equitable on paper, but demonstrably fair in the lives of the people they affect.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_reconciling_probabilistic_fairness_with__99646682,
  title = {Reconciling Probabilistic Fairness with Realized Outcomes},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/reconciling-probabilistic-fairness-with-realized-outcomes/rdr_99646682da10410dace06f2fe2a2b6e8},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_reconciling_probabilistic_fairness_with__99646682_v1,
  title = {Reconciling Probabilistic Fairness with Realized Outcomes},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/reconciling-probabilistic-fairness-with-realized-outcomes/rdr_99646682da10410dace06f2fe2a2b6e8/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}