Orphan-prune: pinned fixed (PR #17), but should isolated neurons get the same age/access grace as dead ones?

[RobertSigmundsson]

Hi Toni! Quick one, and like #15 this is genuinely your call - we are happy either way.

Running consolidation on a live brain we lost a batch of valuable isolated neurons to the orphan path of _prune. It short-circuits above the pinned/age/access guards that the dead-neuron path right below honors. We made two calls and shipped both to our fork:

1. Pinned guard - we treat this as an outright bug and have sent it as a small PR (fix(consolidation): never auto-prune pinned isolated neurons #17): a pinned neuron was permanently deleted just for being isolated, even though the dead path already honors pinned. The PR hoists the pinned guard to cover both paths, with a regression test, and does not change pruning of unpinned orphans, so your existing orphan invariants stay green.

2. Age/access parity for orphans - this is the actual design question. We decided orphans should get the same prune_dead_neuron_days grace and access_frequency protection that dead neurons get, because a fresh isolated neuron is usually a young single observation that has not had time to form connections, not noise. On our fork we run this parity version and it costs nothing: genuinely old, never-accessed, unpinned orphans still get pruned, just after the grace period.

We are not asking permission - just full transparency. The pinned fix we would recommend taking (clean bug). The age/access parity is an opinion; entirely your call whether to adopt it upstream or keep eager orphan pruning. If you want the parity version as a PR too, say the word and we will send it with tests. Happy either way.

[acidkill]

Hi Robert — thanks for the transparency, and for catching this on a live brain rather than letting it sit.

Both calls are the right ones, so let's take both.

1. Pinned guard (fix(consolidation): never auto-prune pinned isolated neurons #17): agreed, this is an outright bug. Pinning has to mean "never auto-delete" regardless of connectivity — a pinned neuron getting deleted just for being isolated, while a pinned-but-dead one survives, is indefensible. Taking fix(consolidation): never auto-prune pinned isolated neurons #17.

2. Age/access parity for orphans: yes, adopt it. Your reasoning matches mine exactly — an isolated neuron is usually a young, single observation that simply hasn't had time to form connections yet. The clearest case is the start of a new project: the first memories land with nothing to connect to, and eager orphan-pruning would punish them for the engine's own timing. Synapses form over time; we shouldn't expect them instantly. Orphans should get the same prune_dead_neuron_days grace and access_frequency protection as dead neurons — genuinely old, never-accessed, unpinned orphans still get pruned, just after the grace period, which is the correct behavior.

So please do send the parity version as a PR with tests — I'll take it. Thanks again for handling this so cleanly.