---
title: Box Thirding (B3) Algorithm
url: https://www.emergentmind.com/topics/box-thirding-b3
type: topic
---

# Box Thirding (B3) Algorithm

The explicit title "Box Thirding" is used for a 2026 algorithm for Best Arm Identification (BAI) under fixed-budget constraints, particularly in anytime settings and regimes with large \(N\), where exhaustive evaluation is impossible within a limited budget \(T\) [2602.18186]. In that work, B3 performs iterative ternary comparisons inside hierarchical boxes: among three arms, the best-performing arm is explored further, the median is deferred, and the weakest is discarded. The same three-letter label \(B^3\) also appears in file-system research as "Bounded Black-Box Crash Testing" [1810.02904], and "box thirding" also describes a tricriteria box-decomposition mechanism in discrete multi-objective optimization [1305.5266]. The primary modern usage of the exact phrase "Box Thirding" is therefore bandit-theoretic, but the term has a broader cross-domain history.

## 1. Terminological scope and disambiguation

The arXiv record associates closely related labels with three distinct technical constructions.

| Usage | Domain | Core idea |
|---|---|---|
| Box Thirding (B3) | Stochastic bandits | Hierarchical ternary comparison for anytime BAI |
| \(B^3\) | File systems | Bounded black-box crash testing |
| Box thirding | Tricriteria optimization | Splitting a search box into up to three subboxes |

In the bandit setting, B3 is an anytime BAI algorithm designed for "insufficient sampling" regimes, especially when \(T < N\) or, more generally, when the candidate set retained by an algorithm is too small to cover all \(\epsilon\)-best arms [2602.18186]. In the file-system setting, \(B^3\) denotes a practical crash-consistency methodology that exhaustively tests a bounded space of workloads and crash points without inspecting or modifying file-system code [1810.02904]. In tricriteria optimization, the phrase refers to a geometric update rule in which each newly discovered nondominated point cuts a current box along up to three coordinate directions, with linear bounds on the number of boxes and scalarizations [1305.5266].

This terminological overlap matters because the three constructions share only a superficial naming similarity. A plausible implication is that "B3" is best treated as a context-dependent abbreviation rather than a single established concept across the literature.

## 2. Bandit formulation and the data-poor regime

In "Box Thirding: Anytime Best Arm Identification under Insufficient Sampling," the underlying model is a stochastic multi-armed bandit with \(N\) arms, unknown reward distributions \(\nu_i\), and 1-sub-Gaussian rewards \(X\) with means \(\mu_i\), ordered as \(\mu_1 > \mu_2 \ge \dots \ge \mu_N\) for analysis [2602.18186]. The goal is BAI rather than cumulative-reward maximization: after \(T\) pulls, the algorithm outputs \(a_T\), and performance is measured by simple regret
\[
r_T = \mathbb{E}[\mu_1 - \mu_{a_T}],
\]
or by the fixed-budget misidentification probability \(\Pr(\mu_1 - \mu_{a_T} > \epsilon)\).

The paper emphasizes the distinction among fixed-budget BAI, fixed-confidence BAI, and anytime BAI. The B3 algorithm belongs to the anytime class: it does not assume prior knowledge of \(T\), maintains a recommendation at every time \(t\), and can be stopped arbitrarily. This design is motivated by settings in which resources are uncertain.

A central construct is the candidate set \(C\), the set of arms that can still plausibly be returned as output under some reward instance. Its size is \(c_0 = |C|\). The paper defines a "data-poor condition" for \(\epsilon\) by
\[
c_0 \le N - N_\epsilon,
\]
where \(N_\epsilon\) is the number of \(\epsilon\)-best arms. In this regime, the dominant difficulty decomposes into two parts: screening, meaning inclusion of a near-best arm in \(C\), and estimation, meaning discrimination among the arms already retained. That decomposition is the organizing principle for the B3 analysis.

## 3. Hierarchical ternary mechanism

B3 organizes arms into boxes \(\text{Box}(l,j)\), indexed by a level \(l\) and a deferment index \(j\), with each box holding up to three arms and their empirical means [2602.18186]. When a box becomes full, B3 applies a local ternary rule:

- the largest arm is **LIFT**ed,
- the median arm is **SHIFT**ed,
- the smallest arm is **DISCARD**ed.

LIFT promotes the arm to level \(l+1\), allocates an additional \(\lceil r_0^l \rceil\) pulls, updates its empirical mean, and places it in \(\text{Box}(l+1,0)\). SHIFT keeps the arm at the same level but increases its deferment index to \((l,j+1)\) without extra pulls. DISCARD removes the arm permanently. The paper denotes the local update as `ARRANGE_BOX`.

The full anytime algorithm repeatedly sweeps through boxes from higher levels to lower ones, and from larger deferment indices down to smaller ones. A box is arranged only when it is full and the destination boxes for LIFT and SHIFT are not full. If \(\text{Box}(0,0)\) has room, the algorithm introduces a new arm uniformly from those not already discarded and not present in any box. At stopping time, it returns the arm in \(\text{Box}(L,0)\) with the largest empirical mean, where \(L\) is the current highest nonempty level.

The per-level budget factor is inherited from a local Sequential Halving analysis. The paper uses \(r_0 \approx 1.728\), the solution of
\[
r_0 + r_0^{1.5} - 4 = 0.
\]
This geometric budget schedule lets B3 allocate more samples to arms that survive to higher levels while remaining anytime. The crucial design choice is that B3 does not preselect a fixed subset and run a fixed-budget routine on it; instead, it maintains a global candidate pool with incremental admission, promotion, deferment, and elimination.

## 4. Guarantees, rates, and relation to other BAI procedures

The analysis of B3 separates total error into non-inclusion error and within-set misidentification [2602.18186]. Under the data-poor condition, the paper gives a general non-inclusion bound of the form
\[
\Pr(\mu_1 - \mu_*(C^\pi) > \epsilon) \le \exp\{-\Omega(c_0^\pi N_\epsilon / N)\},
\]
where \(\mu_*(C)\) is the best mean inside the candidate set. This makes the size of the candidate set a first-order quantity.

The paper then compares candidate-set growth across algorithms. Uniform Sampling (US) has \(c_0 = \Theta(T)\), Bracketing Sequential Halving (BSH) has \(c_0 = \Theta(T / (\log_2 T)^2)\), and B3 also has \(c_0 = \Theta(T)\). Thus B3 matches US in screening capacity while preserving a stronger elimination structure than uniform allocation.

For within-set discrimination, the paper reanalyzes Sequential Halving (SH) with per-level budgets \(T_l = \lceil r_0^l\rceil\), showing that for \(1 < r_0 < 2\),
\[
\Pr(\mu_1 - \mu_{a_T} > \epsilon) \le \exp\left\{-\Omega\left(\frac{T\epsilon^2}{N}\right)\right\}.
\]
B3 is then shown to inherit SH-like within-set performance:
\[
\Pr(\mu_*(C) - \mu_{a_T} > \epsilon/2) \le \exp\left\{-\Omega\left(\frac{T\epsilon^2}{N}\right)\right\}.
\]

The main theorem combines the two terms and yields a total misidentification exponent proportional to
\[
\max\{N_{\epsilon/2}, \epsilon^2\} \, T / N.
\]
Under a polynomial gap structure,
\[
\mu_1 - \mu_i = \left(\frac{i-1}{N}\right)^\alpha,
\]
the resulting simple-regret bound is
\[
\mathbb{E}[\mu_1 - \mu_{a_T}] = \mathcal{O}\left(\max\left\{\frac{1}{T^\alpha}, \sqrt{\frac{N}{T}}\right\}\right).
\]
The paper also states an \((\epsilon,\delta)\)-style sufficient budget condition:
\[
T \ge \text{const} \cdot \max\left\{\frac{N}{N_{\epsilon/2}}, \frac{N}{\epsilon^2}\right\}\ln\frac{1}{\delta}.
\]

Relative to baselines, US maximizes screening but has weaker within-set discrimination, SH is not anytime and requires \(T\), and bracketing methods such as BUCB and BSH are anytime but lose a \((\log T)^2\) factor in screening capacity. The stated contribution of B3 is to combine \(c_0=\Theta(T)\) with SH-like discrimination in a single anytime procedure.

## 5. Empirical behavior, implementation, and scope conditions

The experimental section evaluates B3 on synthetic settings and on the New Yorker Cartoon Caption Contest dataset [2602.18186]. In contest 893, each caption is an arm and \(N = 5{,}513\). The paper defines \(\mu_i\) as the empirical proportion of Funny plus Somewhat Funny responses after preprocessing, and simulates three reward models: high noise with \(\mathcal{N}(\mu_i, 0.5^2)\), moderate noise with \(\mathcal{N}(\mu_i, 0.2^2)\), and a deterministic case with \(X_{i,t} = \mu_i\). The baselines are B3, US, BUCB, and BSH; the budget is \(T = 10{,}000\), repeated over 1,000 runs.

The reported behavior follows the screening-versus-estimation decomposition. In the deterministic case, US achieves very low simple regret because once the best arm enters the candidate set there is no estimation error; B3 is close. Under moderate and high noise, within-set misidentification becomes material, US plateaus, BUCB and BSH improve more slowly because they sample fewer arms, and B3 achieves the best or near-best simple regret. The appendix further reports Bernoulli and heavy-tailed Kumaraswamy rewards, along with data-rich regimes such as \(N=128, T=896\) and \(N=1024, T=10{,}240\), where B3 is described as comparable to SH with tuned \(r_0\) and better than UCB-E.

For implementation, the paper specifies a table or dictionary of boxes \(\text{Box}(l,j)\), state variables \(L\), \(J_l\), and the discard set \(D\), together with incremental sample counts and top-down scheduling. Memory is stated as \(O(\min\{N,T\})\), and the hierarchy depth as \(\Theta(\log T)\). The paper also describes a modified B3 for non-data-poor regimes, in which, once all arms have been initially examined, the algorithm raises a base level \(l_B\) and reintroduces previously discarded arms with more samples.

The paper is explicit about limitations. The theory assumes 1-sub-Gaussian rewards. The constants and rates depend on \(N_\epsilon\) and on the gap structure. In very low-noise regimes, uniform sampling may be marginally better because screening dominates. The authors therefore recommend B3 as a robust default when the operative regime is not known in advance.

## 6. Other meanings of B3 and box thirding

In file-system research, \(B^3\) denotes "Bounded Black-Box Crash Testing," a crash-consistency methodology that treats the file system as a black box accessible only through the POSIX system-call interface and the resulting block-level I/O [1810.02904]. The method bounds workload length, operation types, arguments, initial state, and crash points, then exhaustively generates workloads within that bounded space. Crash points are restricted to persistence-related calls such as `fsync()`, `fdatasync()`, `sync`, and `msync()`. The implementation uses CrashMonkey and ACE. The paper reports that the tools rediscovered 24 of 26 known crash-consistency bugs, found 10 new crash-consistency bugs in Linux file systems, and found one new crash-consistency bug in FSCQ. Here, the three \(B\)'s stand for bounded, black-box, and crash testing, not for box decomposition.

In discrete tricriteria optimization, the 2013 paper does not explicitly use the label "Box Thirding," but it presents a three-dimensional box-decomposition update that fits that description [1305.5266]. Each newly discovered nondominated point splits a current search box into up to three subboxes,
\[
B_i := \{z \in B : z_i < z_i^*\}, \quad i=1,2,3,
\]
and the paper develops a \(v\)-split rule, based on individual subsets and neighbors, to avoid redundant boxes. The resulting tricriteria algorithm requires at most \(3|N|-2\) scalarized subproblems to generate the entire nondominated set, and this improves to \(2|N|-1\) with an \(\varepsilon\)-constraint scalarization and a specific box-selection rule.

These alternative uses show that "B3" and "box thirding" are not field-invariant terms. In the bandit literature, Box Thirding is an anytime arm-elimination algorithm built around local ternary comparisons. In file systems, \(B^3\) is a bounded crash-testing methodology. In tricriteria optimization, box thirding is a geometric decomposition rule for objective-space search. The shared vocabulary reflects a recurrent pattern of structured triadic reduction, but the technical objects, guarantees, and use cases are distinct.

Source: https://www.emergentmind.com/topics/box-thirding-b3