don't lowercase site names

fix(hosting): retry replica reads whose pooled connection was already closed

fix(firehose): release quarantine fences of sites whose repo is gone from its PDS

feat(observability): chart service memory against uptime so restarts overlay

feat(dashboard): tell owners when the host could not fetch their site's files

feat(hosting): time every response by tier and name the reason behind each 404

perf(tiered-storage): batch access statistics and cap buffered hot promotion

fix(main-app): log the hourly OAuth cleanup at debug when it deletes nothing

fix(hosting): log fail-closed 503s as one summary line per site and reason

fix(dns): run one verification pass fleet-wide and back off failing domains

fix(hosting): log client-aborted requests as info instead of errors

fix(dashboard): keep waiting on the site list for a minute, not 28 s

hosting: bound in-process memory and make the budgets configurable

perf(firehose): scan storage stats only on the leader, and bound the frame queue

perf(tiered-storage): promote streamed reads into warm and hot

migrate firehose and storage ownership to Effect; add Hydrant CID recovery

ops: add scoped Bun canary memory sampling and rollback runbook

ops: preserve log ingestion and alert on stalled operations

webhooks: add scope.backlinksOnly to fire only on other repos' references

type-check komodo actions against real client types

fix csrf, rename cookie as __HOST

sync lastAccessed back to the in-memory metadataIndex

leader election in firehose worker

some cleanup by hand

fix storage-miss revalidation loop and tier reporting

refactor: migrate to maintainable monorepo structure

fix cookie routing

fuck shiki

index db