Erasure Coding Keeps A Distributed Cloud Online When Homes Die

🪩 The Disco Party is MATA's distributed cloud.

Erasure coding is how a distributed cloud stays online when homes die. Replication copies a whole blob to N machines and hopes enough of them stay plugged in. Erasure coding splits ciphertext into n shards such that any k of them rebuild the object, and the leftover n−k can vanish without a restore ticket. That is the difference between "my friend went on vacation" and "the vault is gone." On MATA the math is Reed-Solomon over GF(2^8). The crate is rusty_erasure. The SpaceDB seam is spacedb-durability. The hosting home cannot read what it holds.

Centralized cloud solves the same problem by buying more racks in one company's region and sending you an invoice when you leave. A distributed cloud has no region. It has kitchens, apartments, and offices. What Is A Distributed Cloud is the definition. This article is the durability layer: how erasure coding, ciphertext shards, and a remade ISA-L keep the Disco Party up when a node unplugs.

Why Erasure Coding Beats Whole-Copy Replication On A Distributed Cloud

Whole-copy replication is easy to explain and expensive to trust. Three full replicas mean three machines that each hold a complete object. If those machines are honest-but-curious, you just handed three strangers the same ciphertext — and if the key ever leaks, you handed them the vault. Erasure coding on a distributed cloud shrinks what each home holds to a shard that is useless alone. k-of-n is the contract: lose a laptop, lose a weekend, lose a whole street, and the object still reconstructs from the survivors.

Security is the Digital Freedom primitive that maps to encryption. Shards are cut after AES-256-GCM. The host sees opaque pieces. Trust is the identity primitive: mID decides who may fetch a shard, not a password on a vendor disk. Incentive is the payout primitive: homes that store shards get paid for capacity and uptime, not for reading your files. The Electronic Frontier Foundation has documented how "we keep a spare copy" becomes "we were compelled to produce it." Erasure coding on a distributed cloud refuses the spare copy. It keeps a spare fraction.

SpaceDB is the local-first CRDT database underneath. The library is open. The network is disco. Seams are Transport (iroh), ShardStore (erasure), KeyDirectory (mID), and Settlement ($MATA / Iron Bank). Read SpaceDB as a Firebase alternative for the database story. This page is the ShardStore story.

How Erasure Coding Puts Ciphertext Shards On A Distributed Cloud

spacedb-durability takes sealed entries and feeds them to rusty_erasure. The output is n shards placed across homes with anti-affinity so a single apartment cannot hold a reconstructable set. Placement is a distributed-cloud job, not a RAID card in one chassis.

rusty_erasure — ISA-L Erasure Coding Remade For A Distributed Cloud

Intel ISA-L is the industry's fast Reed-Solomon library. It is C, and the interesting bits are GPL-adjacent in the way teams quietly avoid. rusty_erasure remakes that work in memory-safe Rust: GF(2^8) Reed-Solomon, SIMD paths up through GFNI, wasm-ready, no C, no GPL. A distributed cloud that wants to compile for wasm32 and for a home PC cannot drag a C toolchain into the replica. The NIST Secure Software Development Framework is blunt about memory-safe implementations at the bottom of the stack. Erasure coding is that bottom for durability.

Source and siblings live under github.com/Remade-With-Rust. The Remade With Rust catalog is the public card wall.

spacedb-durability — k-of-n Ciphertext On The Distributed Cloud

spacedb-durability is the policy layer. It chooses k and n, cuts shards from ciphertext, and talks to ShardStore for placement. The hosting home cannot decrypt. It cannot even see a useful fraction of the plaintext because there is no plaintext on the wire. Peer-to-peer device sync over iroh moves those shards between devices. When a home dies, the mesh notices the missing shard and rebuilds onto a new host from the remaining k. That is erasure coding doing its only job: keep the distributed cloud online.

What The Hosting Home Can See — And Cannot

A shard is a blob addressed by content, stored beside other blobs, paid through Settlement. The host sees size, uptime, and a capability check. It does not see filenames, owners, or bytes that decode to a document. The NIST Zero Trust Architecture says assume the network is hostile. Erasure coding on a distributed cloud assumes the disk is hostile too — including the disk you are renting from a contributor you have never met.

Erasure Coding Is How A Distributed Cloud Survives Real Homes

Homes die. Power cuts. People sell the NAS. Laptops go in bags for a month. A distributed cloud that needs every replica to stay warm is a centralized cloud with worse marketing. Erasure coding is the design that treats churn as weather. The Freedom Guide is how a person joins. Digital Freedom is the era map. What Is Digital Freedom is the why.

Rebuild When The Distributed Cloud Loses A Home

Self-repair is not a weekend script. When a shard disappears, ShardStore schedules a reconstruction from any k survivors and places a fresh shard on a new home. Incentive pays the new host. The owner does not file a ticket. The Federal Trade Commission spends a lot of ink on companies that held the only recoverable copy and then failed. Erasure coding spreads recoverability so no company is that copy.

Erasure Coding Does Not Replace Your Keys

k-of-n is not a backdoor. If you lose the key, the shards are decorative gravel. Trust stays on the keypair. Security stays on AES-256-GCM. Deputy keeps the crate graph from quietly reintroducing a C ISA-L. The Learn cluster and the home page are the rest of the path in.

Remade Primitives, Disco Party, Digital Freedom

rusty_erasure and spacedb-durability are Remade-With-Rust primitives. They become a disco party when ShardStore places ciphertext across homes and Settlement pays those homes. They become Digital Freedom when Trust, Security, and Incentive — identity, encryption, payouts — outlive any single kitchen. Start at the Disco Party. Keep erasure coding on the distributed cloud, or admit you are one power cut away from a restore myth.

Fetch-by-hash instead of a bucket name is content-addressed storage. Why the shards cannot sit on a C heap is why a distributed cloud needs memory-safe Rust. The database those shards protect is SpaceDB.