BlockRR: Unified Protocols for Privacy & Consensus

• BlockRR is a framework that employs block-structured randomness and scheduling to balance privacy, security, and fairness through deterministic probability partitions.
• It generalizes randomized response protocols by partitioning label spaces into majority and minority subsets using block-wise probability allocations and closed-form normalization.
• In blockchain consensus, BlockRR enables fair, deterministic leader selection by adopting round-robin scheduling and weighted validation to resist adversarial bias.

BlockRR encompasses a class of mechanisms and protocols that leverage block-structured randomness and scheduling, most notably for (1) label differential privacy, (2) distributed storage security, and (3) blockchain consensus. This entry focuses on the principal BlockRR frameworks as introduced in randomized response for label differential privacy (Liu et al., 3 Feb 2026) and consensus selection (Ahmed-Rengers et al., 2018), with pointers to code-theoretic and related usages. The central paradigm is the deterministic (or block-partitioned) allocation of probabilities or roles within a system, trading between privacy, security, and fairness guarantees as a function of system parameters.

1. Unified Randomized Response for Label Differential Privacy

BlockRR, as formulated in (Liu et al., 3 Feb 2026), is a generalization of RR-type mechanisms for satisfying label differential privacy (label-DP). Let $S$ be the label space, partitioned into majority ($S_1$) and minority ($S_2$) subsets based on estimated prior weight. The user specifies a candidate mapping $B: S \to 2^{\tilde S}$, where $\tilde S$ is the set of randomized output labels and $y \in B(y)$. BlockRR further selects a subset $\Delta \subseteq \tilde S_1$ of size $\ell$, allowing the $(y, \tilde y)$ probability plane to be expressed in four blocks:

• $y \in S_2, \tilde y \in \Delta$: uniform probability $S_1$0
• $S_1$1: $S_1$2
• $S_1$3: $S_1$4
• $S_1$5: $S_1$6
• $S_1$7: $S_1$8
• $S_1$9: $S_2$0

Parameters $S_2$1 are computed via a linear system that ensures normalization and differential privacy for any $S_2$2:

$S_2$3

A closed-form solution is derived under $S_2$4 assumptions, as detailed in equations (1) and (2) of (Liu et al., 3 Feb 2026).

BlockRR subsumes prior mechanisms (standard RR, RRWithPrior, RRonBins, RPWithPrior) as special cases by proper selection of $S_2$5, removing the need for individual case analysis.

2. Differential Privacy Guarantees and Weight Matrix Partitioning

BlockRR enforces $S_2$6-label DP (Theorem 3.3, (Liu et al., 3 Feb 2026)) across all label mappings, with the ratio of output probabilities for any pair $S_2$7 tightly controlled. This is realized by setting the highest allowed probability to $S_2$8 times the smallest, as implemented in the block-structured conditional probabilities.

Majority/minority partitioning of $S_2$9 is determined in a data-driven manner. A small fraction of the dataset (or public prior) is processed via the Laplace mechanism to estimate the empirical label prior $B: S \to 2^{\tilde S}$0. A $B: S \to 2^{\tilde S}$1 weight matrix is constructed:

$B: S \to 2^{\tilde S}$2

with $B: S \to 2^{\tilde S}$3 as a tuning parameter. Set $B: S \to 2^{\tilde S}$4, and $B: S \to 2^{\tilde S}$5, yielding a sharply defined criterion for label dominance and block assignment. This partition mitigates "class collapse" by distributing probability mass for minority classes off-diagonal.

Parallel composition is invoked by splitting data into disjoint subsets: one for estimating $B: S \to 2^{\tilde S}$6, one for BlockRR randomization; the overall DP guarantee remains $B: S \to 2^{\tilde S}$7 (not $B: S \to 2^{\tilde S}$8), by the parallel composition theorem.

3. Pseudocode and Implementation Workflow

The operational procedures of BlockRR are formalized in two core algorithms, emphasizing explicit blockwise and labelwise processing. Below is an excerpted version:

$y \in B(y)$6

End-to-end dataset privatization proceeds by splitting the dataset $B: S \to 2^{\tilde S}$9, estimating $\tilde S$0 from $\tilde S$1 under $\tilde S$2-Laplace noise, partitioning $\tilde S$3, and applying BlockRR to $\tilde S$4.

4. Empirical Evaluation and Regime Characterization

Evaluation is performed on imbalanced CIFAR-10 variants, contrasting BlockRR, RR (classical), and RRWithPrior. For privacy budgets $\tilde S$5 ("high"- or "moderate"-privacy regimes), BlockRR yields a more favorable balance between test accuracy and average per-class accuracy. In particular, it avoids the minority-class suppression characteristic of prior-based mechanisms and achieves lowest variance in per-class accuracy for intermediate $\tilde S$6.

Table: Representative Results on CIFAR-10$\tilde S$7

ε	Overall Acc (RR)	RRWithPrior	BlockRR	Avg Per-class Acc (RR)	RRWithPrior	BlockRR
0.6	39.75 ±1.44	37.74 ±4.06	40.31±2.08	29.14	26.17	28.52
0.8	46.18±1.51	47.54±4.32	47.11±1.43	37.25	33.07	35.40
1.0	53.66±2.18	54.86±4.19	55.12±2.69	46.01	38.42	43.30
2.0	78.74±0.48	73.20±1.53	78.86±0.56	-	-	-
3.0	87.67±0.50	83.35±1.93	87.60±0.51	-	-	-
4.0+	≈91–92	≈91–92	≈91–92	-	-	-

For $\tilde S$8, all approaches converge to standard RR behavior and no relative advantage persists. This regime separation precisely characterizes where data-driven blockwise randomization is most impactful.

5. Instantiation, Specialization, and Theoretical Connections

BlockRR encompasses, as proper specializations, all existing major RR-type label-DP mechanisms:

• Standard RR: all labels majority ($\tilde S$9, $y \in B(y)$0, identity mapping), uniform output probabilities.
• RRWithPrior: designate top-k majority classes as $y \in B(y)$1, others as $y \in B(y)$2; minority classes omitted from outputs.
• RRonBins: bin-mapped output labels, $y \in B(y)$3, $y \in B(y)$4 as binning map.
• RPWithPrior: candidate sets are intervals $y \in B(y)$5 per label, with appropriate block definitions.

BlockRR's unification relies on its explicit four-block probabilistic structure and linear solution of normalization/equity constraints, permitting arbitrary prior-driven or structural partitionings.

6. BlockRR in Blockchain Consensus Protocols

A distinct usage of the term BlockRR ("Robust Round Robin") appears in blockchain consensus, denoting a deterministic leader selection protocol for Proof-of-Stake systems (Ahmed-Rengers et al., 2018). Here, the deterministic schedule ensures fairness (block production proportional to stake), and the protocol incorporates:

• Long-lived identities based on attested SGX quotes or PoW-mined commitments.
• Candidate selection by round-robin over identities, ranked by age.
• Endorsement by a sampled committee among active identities, with a block proposed only after a quorum confirm.
• Formal analysis of fork probabilities, adversarial bias resistance, and realized transaction finality.

This protocol eliminates bias inherent in Verifiable Random Function leader selection and achieves throughput and fairness guarantees supported by empirical and probabilistic analysis.

7. Extensions and Related Block-Structured Mechanisms

BlockRR embodies a block-structured philosophy in privacy, security, and consensus. In distributed storage, "block security" refers to security guarantees not for the entire dataset but for blocks (subsets) of data, offering a spectrum between no secrecy and perfect secrecy by tuning the block size parameter (see (Dau et al., 2013)). In Block Failure Resilient codes (Calis et al., 2014), block structure governs resilience and repair dynamics in distributed storage.

A plausible implication is that the block-structural paradigm, as instantiated in BlockRR-type schemes, flexibly mediates resource tradeoffs—accuracy, privacy, repair bandwidth, and fairness—by algorithmically partitioning the system's operational space, and solving the resulting system of constraints globally rather than casewise. Empirical evidence suggests that such approaches yield optimal or near-optimal performance in regimes of strong privacy or high security demand.

---

References:

• "BlockRR: A Unified Framework of RR-type Algorithms for Label Differential Privacy" (Liu et al., 3 Feb 2026)
• "Don't Mine, Wait in Line: Fair and Efficient Blockchain Consensus with Robust Round Robin" (Ahmed-Rengers et al., 2018)
• "On Block Security of Regenerating Codes at the MBR Point for Distributed Storage Systems" (Dau et al., 2013)
• "Repairable Block Failure Resilient Codes" (Calis et al., 2014)