Error Correction Zoo: ECC Catalog

• Error Correction Zoo is a structured catalogue that organizes classical, classical–quantum, and quantum error-correcting codes using a detailed taxonomy and unique identifiers.
• It cross-references code parameters, decoding algorithms, and mathematical objects like sphere packings, lattices, and combinatorial designs to aid practical code selection.
• The platform facilitates research in communication, storage, and computation by linking theoretical constructs with actionable insights and benchmarking performance.

The Error Correction Zoo (EC Zoo) is a structured, extensible catalogue of error-correcting codes (ECCs) and their relationships to mathematical objects such as sphere packings, lattices, combinatorial designs, groups, and classical and quantum phases of matter. It provides a unified taxonomy and cross-referencing for classical, classical-quantum, and quantum codes, supporting a rigorous and machine-accessible knowledge base for code selection, comparison, and discovery in communication, storage, and computation contexts (Albert et al., 9 Jun 2026).

1. Taxonomy and Data Model

The EC Zoo introduces a taxonomic hierarchy for codes based on communication domain and alphabet. At the highest level, the three principal Domains are: Classical, Classical–Quantum (c-q), and Quantum. Within each Domain, codes are grouped into Kingdoms according to their alphabet or physical substrate—Binary, $q$-ary, real, group-based, qudit, bosonic, etc. Each entry is defined by a unique identifier (“code_id”) and annotated with parameters $(n,k,d)$ (or generalizations), key properties (distance, locality, thresholds), mathematical parentage and cousin relations, decoding algorithms, and implementation references. The data format (YAML+LaTeX) ensures modularity and supports automated aggregation into interactive web, static HTML, and print-ready handbook formats (Albert et al., 9 Jun 2026).

2. Principal Code Families and Constructions

The EC Zoo catalogs all major families of classical and quantum codes, detailing their constructions, properties, and interrelations.

Classical Code Families

• Linear Block Codes: Hamming $[2^r-1,2^r-r-1,3]_2$, extended Hamming, Simplex, Reed–Muller $$\left[2^m,\sum_{i=0}^r \binom{m}{i},2^{m-r}\right]_2$$, BCH $[n=2^m-1,k\ge n-mt, d\ge2t+1]_2$, Reed–Solomon $[n\le q,k,d=n-k+1]_q$, Generalized Reed–Solomon, alternant/Goppa.
• Cyclic/Quasi-Cyclic Codes: Codes generated as cyclic shifts, including BCH, QR, and QC-LDPC.
• LDPC Codes: Sparse parity-check matrices; includes protograph, spatially coupled, and algebraic LDPC.
• Polar Codes: Binary $(N,K,d)$ codes with $N=2^m$, achieving channel capacity by polarization.
• Convolutional and Turbo Codes: Streaming codes generated by shift registers; concatenation and iterative decoding.
• LRC/LDC/LTC Codes: Locally recoverable, decodable, and testable codes with sublinear query and recovery properties.
• Geometric/Spherical Codes: Lattice-based, constant-energy, and design-based code structures (Albert et al., 9 Jun 2026).

Quantum Code Families

• Stabilizer and CSS Codes: Qubit $\llbracket n,k,d\rrbracket$ codes fixed by Abelian subgroups; CSS constructions from pairs of classical codes.
• Topological Codes: Surface/toric, color, fracton, and hyperbolic codes exploiting geometric/topological structures.
• Bosonic Codes: GKP (oscillator lattice), cat, binomial, and related continuous-variable encodings.
• Quantum LDPC Codes: Hypergraph-product and related product constructions with constant or growing rate and distance.
• Quantum Polar and Concatenated Codes: Codes leveraging channel polarization and concatenated architectures for fault tolerance (Albert et al., 9 Jun 2026).

3. Key Algorithms and Notable Zoo Entries

The EC Zoo compiles both canonical and novel ECC methods, highlighting unique algorithms for encoding, decoding, and error synthesis.

Compositional Multi-hop Factual Error Correction (CECoR)

CECoR is a reasoning-aware framework developed for multi-hop Factual Error Correction (FEC), in which input claims are decomposed into compositional reasoning steps. Controlled perturbations are injected at the step level to synthesize realistic and diverse error patterns. Correction models are trained via a two-stage pipeline: supervised fine-tuning (SFT) on synthetic pairs, followed by reinforcement learning (RL) with rewards for factuality, fluency, and semantic fidelity. No architectural changes are made to the transformer core; all reasoning and error synthesis occurs upstream via planning and injection modules. CECoR demonstrates superior benchmark performance and robust generalization to both single- and multi-hop correction under noisy evidence (Zhu et al., 4 May 2026).

Two-Faced Process Error Correction

This method increases symbol interdependence in binary sequences via a two-faced Markov process. No parity bits are appended; instead, an invertible transformation $\phi_{l,w}$ crafts output with locally coin-like but globally correlated statistics. Linear-time decoding exploits this structure to correct erasures or flips by leveraging high mutual dependence in $(n,k,d)$0 windows. BER bounds indicate that for high bias ($(n,k,d)$1) and moderate noise, residual error rates can be driven two orders of magnitude below raw channel error without complex iterative decoding (Ryabko, 10 Jan 2026).

Number-Theoretic Error-Correcting Code (NT-ECC)

NT-ECC encodes a $(n,k,d)$2-bit message by appending a product of small primes (representing bits) modulo a large prime, leveraging unique prime factorization. Correction uses modular division and 2-dimensional lattice reduction to identify error locations. NT-ECC is systematic and concatenates naturally with other block codes. Although less efficient than high-rate codes like Turbo, NT-ECC achieves better rates than some traditional codes for specific parameters and application domains. Decoding complexity is polynomial but dominated by lattice reduction for large error-correcting capacity (Brier et al., 2015).

Code/Method	Key Feature	Complexity
CECoR	Compositional, SFT+RL	SFT+RL, modular
Two-Faced Processes	No redundancy, Markov corr.	$(n,k,d)$3
NT-ECC	Number-theoretic, modular	poly. in $(n,k,d)$4

4. Mathematical Structures and Cross-Relations

The EC Zoo emphasizes mathematical connections between codes and related combinatorial or geometric objects:

• Sphere Packings and Lattices: Via Construction A, linear codes map to high-dimensional packings, e.g., $(n,k,d)$5, Leech lattice $(n,k,d)$6, and analogous quantum structures (Albert et al., 9 Jun 2026).
• Designs and Groups: Combinatorial designs (Steiner systems) correspond to constant-weight codes; group codes exploit algebraic and homogeneous space symmetries.
• Topological Field Theory and Category Theory: Quantum codes connect to fusion categories, topological phases, and TQFT, aligning CSS and homological codes with boundary operator complexes.

Code duality, chain complexes, expander graphs, and polynomial/algebraic geometry principles underlie many class families and their generalizations.

5. Fundamental Bounds, Evaluation, and Benchmarking

Canonical bounds such as the Hamming, Singleton, and Gilbert–Varshamov bounds (and their quantum analogues) are catalogued in canonical EC Zoo entries (Albert et al., 9 Jun 2026). Evaluation metrics and benchmarks—SARI, ROUGE-2, LLM-judged factuality/fluency—are standardized for modern methods like CECoR (Zhu et al., 4 May 2026). Ablation studies quantify the impact of architectural and methodology variants on SARI and LLM-based metrics.

6. Navigation, Usage, and Applications

The EC Zoo provides applied guidance for practitioners:

• Code selection: Entries are cross-referenced for given channel models, noise rates, and block lengths. For moderate data sizes and low bit-flip rates, extended Hamming or BCH codes are recommended. For high-speed, near-capacity regimes, LDPC and polar codes are prominent (Albert et al., 9 Jun 2026).
• Quantum/Fault Tolerance: Surface/toric and color codes facilitate 2D qubit arrays and transversal logical gate measurement.
• Continuous Variable/Bosonic Storage: Codes such as GKP and binomial offer logical error protection against photon loss and amplitude damping.

Classification charts enumerate both classical and quantum code hierarchies, facilitating systematic exploration.

7. Implications, Limitations, and Future Directions

The EC Zoo demonstrates that there is no universal code; optimality is determined by contextual constraints, channel models, and desired physical implementations. The taxonomy and relational graph structure enable identification of equivalences, descendant structures, and convergence of classical and quantum techniques.

Among current limitations: reliance on external modules (e.g., in CECoR, the quality of the decomposition planner), and, for some constructions (e.g., NT-ECC), decoding bottlenecks inherent to lattice algorithms. Open directions include automated end-to-end reasoning module learning, robust handling of open-domain and adversarial errors, and expansions of the Zoo to new alphabets and quantum phases (Zhu et al., 4 May 2026, Albert et al., 9 Jun 2026).

The EC Zoo's formal organization and reference model are intended to support both foundational theory and practical applications, providing a map for both established and newly emerging error-correcting methods.