Why Full Danksharding Needs Sampling at Scale

If a malicious block producer withholds even a single data chunk, it breaks the sampling guarantee. The core risk isn't a 51% hash power attack, but a coordinated data withholding by a majority of block builders.\n- Impact: Honest validators cannot reconstruct the block, causing chain halt.\n- Mitigation: Requires honest majority assumption among sampled nodes and cryptographic fraud proofs.

Adversarial nodes can isolate honest validators by controlling their peer connections, feeding them only valid samples to hide missing data. This attacks the P2P sampling network layer.\n- Impact: Targeted validators falsely attest to data availability, compromising safety.\n- Mitigation: Requires robust, randomized peer discovery (e.g., Discv5) and out-of-band data sourcing.

Sampling 128 KB chunks across a 128 MB block requires ~1000 samples per validator. At scale, this creates a constant bandwidth and CPU load, risking centralization as smaller nodes drop off.\n- Impact: Validation centralization to professional nodes, reducing censorship resistance.\n- Mitigation: Optimized algorithms (e.g., 2D Reed-Solomon) and proposer-builder separation (PBS) to isolate builder risk.

Pre-Danksharding scaling (e.g., EIP-4844 blobs) often relies on off-chain DACs. This reintroduces trusted third parties, creating a liveness bottleneck and a legal attack surface.\n- Impact: Regulatory takedown of a major DAC could cripple rollup finality.\n- Mitigation: Full Danksharding eliminates DACs via cryptographic sampling, making L1 the trustless DA layer.

Rollups and bridges (e.g., LayerZero, Across) depend on fast DA attestation. If sampling introduces multi-block confirmation delays for safety, it increases cross-chain vulnerability windows.\n- Impact: Bridge exploit risk scales with attestation delay, potentially increasing from minutes to hours.\n- Mitigation: Requires proof of custody games and optimistic fast-finality pathways.

Danksharding uses 2D KZG commitments and Reed-Solomon codes. A bug in the complex encoding/decoding library is a single point of failure.\n- Impact: Catastrophic, irreversible data loss if a widely adopted codec fails.\n- Mitigation: Requires multiple independent implementations (e.g., Geth, Erigon, Reth) and formal verification of core cryptography.

DAS allows light nodes to probabilistically verify the availability of large data blobs by sampling small, random chunks. This is the cryptographic trick that makes scaling beyond a single node's bandwidth possible.\n- Enables Stateless Clients: Clients no longer need to store the entire state.\n- Breaks the Bandwidth Monopoly: Prevents a requirement for nodes with >1 Gbps connections.

Relies on KZG polynomial commitments to create a cryptographic fingerprint of the data. Combined with Reed-Solomon erasure coding, it ensures any 50% of the data can reconstruct the whole. This is the bedrock for reliable sampling.\n- Enables Fast Proofs: Allows for efficient verification of data presence.\n- Guarantees Recoverability: As long as 50%+ of samples are available, the full data can be retrieved.

The system parameters are engineered for statistical security at scale. 64 blobs per slot and 32 sampling committees create the redundancy and parallelization needed to hit the 1.3 MB/slot target while keeping node requirements low.\n- Parallel Sampling: Committees work simultaneously, keeping latency < 2 seconds.\n- Economic Security: Requires a >50% honest majority of samplers, not a supermajority of total stake.

If you don't build sampling into the core consensus, you fragment the ecosystem. Celestia's modular approach separates DA from execution, forcing L2s like Arbitrum Orbit and Optimism Stack to become external DA consumers. This creates a multi-DA security surface and bridging complexity.\n- Architectural Fork: Forces a choice between integrated security and modular flexibility.\n- L2 Vendor Lock-in: Rollups become dependent on an external DA layer's liveness.

This is the raw scaling multiplier. Moving from the current ~83 KB/slot in proto-danksharding (EIP-4844) to the full 1.3 MB/slot target reduces L2 data posting costs by ~100x. This finally makes <$0.01 transactions viable for mass adoption.\n- Cost Driver: DA is the primary cost for rollups like Arbitrum and Optimism.\n- Capacity Ceiling: Sets the theoretical max TPS for the entire L2 ecosystem.

Full Danksharding's endgame is a network where the dominant client type is a light node performing DAS. This radically decentralizes validation compared to today's full-node-heavy networks. It's a shift from resource-based to algorithmic security.\n- Hardware Democratization: Nodes can run on consumer hardware with ~100 Mbps internet.\n- Validator Scaling: Enables >1M validators without proportional bandwidth growth.