Walrus runs availability checks on stored blobs regardless of access.
No reads are required. No application activity is needed. Proofs run anyway.
That behavior defines how Walrus treats storage. Availability is enforced continuously, not inferred from usage. A blob does not stay “alive” because someone accesses it. It stays alive because nodes keep proving they hold their assigned fragments.
Each epoch, Walrus issues availability challenges to storage nodes. These challenges require cryptographic responses derived from actual Red Stuff fragments. A node cannot respond correctly unless the fragment exists on disk. Metadata is not enough. Past proofs are not reusable. The response must be computed fresh.
This is where Walrus diverges from replication-heavy storage models. In those systems, unused data often drifts toward risk. Operators prioritize hot replicas. Cold data survives until someone notices it is missing. Walrus does not wait for access to test integrity. It tests absence directly.
Red Stuff encoding makes this enforceable at scale. Fragments are interchangeable within thresholds, so Walrus does not challenge every fragment every time. It challenges enough of them to maintain statistical certainty. Availability is measured probabilistically but enforced deterministically. If too many proofs fail, the system reacts.
Nodes know when challenges occur. They cannot hide behind low traffic or idle periods. A quiet network does not reduce responsibility. WAL rewards continue only if availability proofs pass. Storage work does not pause because demand pauses.
This changes node behavior in practice. Operators optimize for consistency, not popularity. Disk pruning, lazy replication, or partial corruption surfaces as failed proofs long before users complain. Walrus detects risk upstream, not at the read path.
There is a cost to this model. Availability challenges consume bandwidth and compute even when no one is using the data. Small operators feel this overhead immediately. Walrus accepts that cost explicitly. It chooses predictable enforcement over opportunistic efficiency.
For applications, the effect is subtle but important. Data does not decay silently. If availability drops below threshold, Walrus knows before reads fail. Recovery is not instant, but loss is detected while options still exist.
Reads remain simple. Proofs remain mandatory. These two paths never merge.
Walrus does not assume data exists because nobody complains.
It requires nodes to prove it exists, continuously, whether anyone is watching or not.
