Obsidian separates execution from data transport. EVM transactions are computationally intensive but relatively small. The Silica Protocol uses spare network capacity to carry larger, non-executable data payloads that need minimal CPU processing.
Message payloads are propagated as data sidecars. These are gossiped peer-to-peer, erasure-coded, and verified by lane committees. They are part of the blockchain protocol but intentionally kept outside the block body so they do not slow down block propagation or voting.
Where is the data stored? Validators hold erasure-coded chunks during the retention window (~18 days). After that, archive nodes preserve the full history. Anyone can run an archive node and earn from the Archive Pool (50% of PM fees) for serving historical data.
The network is divided into parallel lanes. Each lane has a rotating committee of validators assigned to it. Messages are deterministically routed based on sender address:
lane = Hash("LANE_V2", chain_id, sender) mod lane_countSo all messages from a single sender go to the same lane. Nodes can immediately determine which lane committee should receive a message without knowing the current slot or RANDAO state.
BID > MinPMBid
Include a fee bid and are sorted by bid amount (highest first) within each lane batch. Signed debit authorization means the fee is deducted at inclusion time.
BID == 0 + VDF PROOF
No bid required. A VDF (Verifiable Delay Function) proof provides spam resistance instead. VDFs are sequential computations that cannot be parallelized, creating a natural rate limit.
Most DA solutions use Data Availability Sampling (DAS), where light clients randomly sample chunks to probabilistically verify availability. Obsidian takes a different approach: committee attestation is the acceptance criteria.
| Aspect | DAS-Primary (Ethereum) | Committee-Primary (Obsidian) |
|---|---|---|
| Block acceptance | Requires successful sampling | Requires committee QC (2/3 threshold) |
| Primary verification | Light clients sample randomly | Committee members prove chunk possession |
| Parallelism | Single blob space per block | Multiple lanes with dedicated committees |
| Acceptance | Probabilistic | Deterministic |
A valid certificate proves that a supermajority of the lane committee had the data when they signed. With erasure coding, this means the data can always be reconstructed.
Data doesn't live in blocks. It lives in the Silica data plane. Blocks only contain compact cryptographic commitments (Merkle roots) that reference the actual payloads stored elsewhere.
This separation allows for sharded storage: archive nodes can specialize by storing only specific epoch ranges or lane subsets. A node might store epochs 1-1000, another epochs 1001-2000. Together they preserve the full history without any single node needing all ~84 TB/year.
Each message batch is split into data chunks and expanded with parity chunks using Reed-Solomon encoding. The original data can be reconstructed from any 50% of chunks.
At ~84 TB/year, traditional "store everything" archives become impractical. Obsidian solves this with epoch-range sharding: instead of every archive storing all history, nodes store specific epoch ranges.
Priority Message fees are distributed among protocol participants. Rather than prepaying fees, Priority Messages include a signed debit authorization. The proposer deducts the bid from the sender's balance at inclusion time.
Full technical specification of the Silica Protocol, consensus modifications, and economic model.