The uploaded connectome is split into shards, each stored as k independent replicas on separate physical media. Every replica has an annual failure probability λ, so a single replica survives t years with:
P_replica(t) = exp(-λ·t)
P_shard_lost(t) = (1 - exp(-λ·t))^k (no maintenance)
Each simulated year, every surviving replica independently rolls against λ. A shard is only truly gone once all k of its replicas have failed — that's why redundancy matters so much more than raw storage. A shard with at least one surviving replica can be repaired: with probability equal to the maintenance rate, its lost replicas are re-copied from the survivor back up to k, restoring full redundancy. Once every replica of a shard is gone, there is nothing left to copy from — that fraction of the mind is permanently unrecoverable.
- Scan resolution — finer sampling of the connectome means more shards and a larger archive, at higher fidelity per shard.
- Redundancy — more replicas per shard exponentially cuts the odds of total shard loss, at k× the storage cost.
- Failure rate — how quickly the storage medium itself degrades (bit rot, disk failure, decay of the physical substrate).
- Maintenance — how often a repair cycle re-replicates degraded-but-not-lost shards back to full redundancy.
This is the real bottleneck the article's "long-term data storage" section gestures at: mind uploading is not just a scanning problem, it's an indefinite, unbroken maintenance obligation — skip enough repair cycles and the archive erodes to nothing no matter how good the original scan was. This 2D companion renders the same shard population as a flat grid instead of a rotating 3D lattice, with a live fidelity-over-time strip along the bottom of the canvas so the survive-or-collapse trend is visible at a glance.