Storage that publishes what it measured
Erasure-coded object storage on independently operated nodes. Priced in USD, settled in USDC, no network token.
What we measured
A durability claim is unfalsifiable to a customer until someone says which numbers came from a measurement and which came from a model. These came from a measurement.
Random sampling reported this code as tolerating one more loss than it does. Exhaustive search found 1,464 fatal patterns the sample never drew. We publish the exhaustive number.
Live, not asserted
The claims above are checkable while you read them. The contract runs against real buckets at two providers every thirty minutes, and the results are URLs, not screenshots.
Conformance
The shard-store contract against Cloudflare R2 and Backblaze B2. Returns 500 if anything fails, so it works as a check.
https://kura-conformance.04-feasts-minded.workers.dev/conformance
Status
The probe series: availability per backend, median read latency, and an explicit refusal to quote a rate until the window is long enough.
https://kura-conformance.04-feasts-minded.workers.dev/status
Durability, demonstrated
Stores an object across both providers, destroys shards on purpose, repairs, and compares the recovered bytes. One shard lost costs four reads. Eight lost is refused rather than guessed at.
https://kura-conformance.04-feasts-minded.workers.dev/durability
Fleet audit
How many genuinely independent failure domains the fleet has. It currently says the fleet is NOT survivable — two providers, and the code needs at least three.
https://kura-conformance.04-feasts-minded.workers.dev/audit
The audit answering no, and the durability run including a case that must fail, are the reasons to trust the rest. A status page that only ever reports success is a status page nobody wired to anything.
Pricing
| Configuration | Storage multiplier | Storage / TB-month | Egress / TB |
|---|---|---|---|
| Launch (today) | 2.00× | $4.62 | $3.08 |
| Target (after Phase 0) | 1.625× | $3.75 | $3.08 |
Launch is deliberately conservative. At launch the node-loss rate is unmeasured, and the costs are asymmetric: launching optimistic and finding the rate is worse loses data irreversibly, while launching conservative and finding it is better is a price cut. The 18.7% reduction goes to customers when the measurement earns it — the data that produces the measurement is theirs.
Egress is not multiplied by the code. A systematic read reconstructs nothing, so serving one logical TB moves one physical TB. Charging egress at the storage multiplier would bill for work we do not do.
Run a node
Onboarding is permissionless. The bond is the barrier, not a gatekeeper — anyone may join, and what filters node quality is collateral, not an invitation.
The bond is priced against the repair bill, not against what a cheat would gain. A strong audit makes deterrence cheap; rebuilding what a departed node held is the expense that remains, and it is the larger of the two.
Get your own numbers — the endpoint runs the same function this page and the tests do, so it cannot drift from them:
https://kura-coordinator.04-feasts-minded.workers.dev/operator/quote?tb=50
Registering records intent. It takes no bond — the escrow contract is not deployed — and places no data, because Phase 0 accepts none. The registry is public, because a permissionless network whose participant list is private is not one anybody can check the failure-domain spread of.
https://kura-coordinator.04-feasts-minded.workers.dev/operator/registry
What this is not
Not a proof-of-storage chain
Audits are unpredictable sampling against a committed Merkle sum tree, with collateral behind them. That is detection with an economic consequence, not the continuous on-chain proof Filecoin publishes. If public, chain-verifiable proof is your requirement, Filecoin is the correct choice and we will say so.
Not trustless placement
A coordinator decides placement, issues audits and computes payouts. It cannot lie about a payout total or a node's share without contradicting a root it already published — but it is a trusted point, the same one Storj's satellites are. Quorum coordination is planned, not shipped.
Not a measured durability number yet
The code's minimum distance is measured. The node-loss rate that turns it into a nines figure is not, and will not be until Phase 0 has run. Anyone quoting you eleven nines for a network that has not measured its own node population is quoting a model, including us — so we quote the model and label it.
Not independent by default
Twenty-six shards on one provider's account are one failure domain, whatever the code says. Every backend must declare its independence, and the audit reports when a placement collapses into fewer real domains than the code needs.
Built in the open
Five libraries, each with the reasoning in its README and tests that assert the claims.
erasure
GF(2^8) locally recoverable codes. Minimum distance measured exhaustively, not cited.
kura
Placement, repair, audit and payment-order decisions. Moves no bytes.
kura-node
Shard backends and the fast multiply-accumulate, held byte-equal to the reference.
kura-bugyo
Pricing, collateral and epoch payout roots anyone can verify.
kotobase-storage-kura
Drop-in backend for anything already speaking kotobase.storage.
Status
Phase 0: the coding, placement, audit and settlement planes run against rented backends as pseudo-nodes, to measure the node-loss rate that sets the storage multiplier. Not yet accepting customer data. The libraries are public and the numbers above are reproducible from them.