consider
Do you need it to live in the EU — really?
Not "in an EU region of a US hyperscaler", but in an EU-domiciled operator with EU staff and EU jurisdiction. Epher Cloud — or your own hardware on Epher Enterprise.
Compare
Our neighbours,
at a glance.
Most peers occupy a single point on the time axis or a single pillar. Epher Continuity Computer covers the whole storage shelf — milliseconds to forever — and adds deterministic compute on the same wire, with one signature chain and one EU-domiciled operator. Below is the buyer's-view summary. The architectural deep-dive lives on the protocol site.
When you'd pick us
Three deciding questions.
consider
Do you need to prove it — to a regulator?
DORA, the AI Act, ICT-incident audit records: Epher CC is the durable end of the shelf, with PQC and inclusion proofs by default — and deterministic-compute receipts when the workflow itself needs auditing.
consider
Do you need the same primitive at both ends of the time axis?
Session state for a real-time app and a six-year audit log of the same workload, with one wire? That is the design.
Side by side
Where each system shines — and where it stops.
| System | Best for | Time horizon | EU sovereign | HW attestation | PQC | Token |
|---|---|---|---|---|---|---|
| Epher CC | EU compliance + agent state + audit + verifiable workflows | seconds → eternal | yes (operator) | HATP + KVM | Falcon (T5·T6·T7) | none |
| MagicBlock (Solana ER) | Web3 gaming session state | seconds | — | TDX (announced) | — | none |
| Celestia / EigenDA / Avail | Rollup data availability | ~7 days | — | — | — | native |
| Filecoin | Bulk leased storage | months → years | — | SGX in places | — | FIL |
| Arweave | Permanent public storage | eternal | — | — | — | AR |
| Algorand state proofs | PQC chain anchoring | chain-eternal | — | — | Falcon | ALGO |
| EBSI (EU) | EU public-sector notarisation | years → eternal | yes | research | roadmap | none |
| QANplatform | EU-hosted PQC L1 | chain-eternal | hostable | — | Dilithium | CryptoQNT |
| In-house log + HSM | DIY audit trail | whatever you build | depends | depends | unlikely | n/a |
"—" means the system does not attempt that property; not a criticism, a scope statement. For the full architectural treatment of each peer, see ephernity.org/compare.
Per-call commit · measured 2026-05-31
AMD EPYC 7313P · nested KVMThousands of TPS, per call, committed.
Most chains hide their TPS in batched-block scheduling: throughput is high, but every call waits for a block. Epher CC is built for the per-call commit regime — every contract call returns after the canonical record has landed on a Raft majority, with no batching and no block scheduling. Below is a side-by-side measurement against the field, every number taken on the same host inside the same nested-KVM rig.
| System | Topology | Per-call commit TPS | Ratio to Epher 3-node |
|---|---|---|---|
| Epher CC — 3-node | Raft over TCP loopback | 1,547 | — |
| Epher CC — 5-node | Raft over TCP loopback | 1,216 | −21% |
| Epher CC — 7-node | Raft over TCP loopback | 995 | −36% |
| Anvil (Foundry single-node EVM) | instant-mine | 43 | 36× slower |
| Solana base layer | CLI per-call confirm | 39.89 | 39× slower |
| MagicBlock Ephemeral Rollups | per-call, undelegated | 36.06 | 43× slower |
| CometBFT — 3-node sequential | block-time-bound | 1 | 1,547× slower |
| Geth POA Clique — 3-signer | mine + in-turn rotation | 0.33 | 4,688× slower |
All Epher numbers are multi-process Raft over TCP — no in-process shortcut. Every measurement is reproducible from the reference implementation's benchmark scripts. We have not yet locally measured MagicBlock's delegated mode (vendor: ~50,000 TPS), Aptos Block-STM (vendor: ~6,000 TPS), or Hyperledger Fabric 2.x BFT (vendor: ~3,000 TPS) — those are pending local replication.
Batched regime · for completeness
When chains amortise commit cost across a block.
Batched chains group transactions into blocks and report throughput per block, not per call. Below is the same hardware, measured at the batched regime each system was designed for.
| System | Batched TPS | Source |
|---|---|---|
Solana base — solana-bench-tps | 3,322 avg · 4,920 peak | measured |
| CometBFT — batched, 100 ms blocks | 171 – 188 | measured |
| Hyperledger Fabric 2.x BFT | ~3,000 | vendor |
| Aptos Block-STM | ~6,000 | vendor |
| MagicBlock ER — delegated accounts | ~50,000 | vendor |
Read this table column by column with the previous one in mind: at per-call commit, Epher CC is in a band of its own. At the batched regime, Epher's per-mesh throughput sits in the same band as Solana base; Aptos and Fabric publish higher (locally unverified) numbers; MagicBlock's delegated path is the only credible 50,000+ TPS claim in the field — and we have not yet replicated it.
The architectural distinction
Per-mesh per-contract throughput in the 1,000–4,000 TPS band — with per-call commit, no batched-block scheduling.
Global L1s divide their headline TPS across the planet's workload; an Epher mesh delivers that throughput on a per-contract basis to a tenant, with every call returning carrying a tamper-evident, EU-attested audit record.
Validation · 2026-05-31
Six fault categories. Invariants hold.
Performance numbers without correctness numbers are marketing. Epher CC has been exercised through a chaos and stress harness across four real contracts — escrow, DAO voting, supply-chain attestation, insurance claim — under the fault categories on the right. Ledger, storage, and idempotency invariants hold in every category.
Network delay
Replication path injected with bounded latency; commits still return ordered and signed.
Contract trap
Borz meter-budget exhaustion; the call returns a typed error; ledger remains consistent.
Storage failure
Underlying object backend faulted mid-write; payload-CID write retries idempotently.
Replay attack
Replayed inbound message rejected by the idempotency store; no duplicate commit.
Tampered arguments
Caller mutates the inbound payload after signing; the contract surface rejects the call.
Mixed-fault stress
All five above injected concurrently; invariants still hold; throughput degrades gracefully.
vs. roll-your-own
The build-vs-buy
that's worth doing the math on.
An append-only Postgres table + a hash chain.
A weekend's work — and a year of liabilities. No per-entry signatures, no offline verification, no PQC, no public-chain anchor, and a question mark over what "tamper-evident" means when the DBA can edit the table.
A self-hosted HSM + a Merkle library.
Closer — but the HATP-style delegation, the key rotation, the offline verifier, and the regulator-facing paperwork are all on you. Auditors will ask who measured the boot of the signing host.
Epher Continuity Computer
A POST request, an inclusion proof, an offline-verifiable export, and a DPA you can sign on day one. The compute pillar adds deterministic Borz contracts with the same receipt format. Built once, audited once, billed per entry or per host depending on the edition.