I was not trying to write about Walrus. I was tracing where availability assumptions crack.
Walrus kept showing up there.
As of mid-Jan (last check, dashboard snapshot), capacity sits in the petabyte range and utilization is still around a quarter. Not empty. Not "future potential" either. Data is getting written, renewed and... yeah—allowed to die.
Blobs on Walrus don't "persist"They live inside a paid window, backed by erasure-coded redundancy, and the obligation ends on schedule. That one rule forces earlier choices. Retention gets shorter. Renewals get intentional. Storage turns over instead of piling up.
On the operator side: 100+ nodes, independent operators. So availability doesn’t fail like an outage most of the time. It fails like variance. Proof timing. Retrieval that technically works… after the meeting. After the release. After the audit question you can’t stall.
That is the pressure beat people miss. Not “down.”
“Late, but valid.”
The identity dataset is the one I can not ignore though... millions of records stored as blobs. 'Mostly available" doesn’t count there. Miss the window and the problem is not UI. It’s ops. Sometimes legal.
And the cost shows up the first time a renewal gets missed during a busy week.