Audit Log Service · View 05 of 40 · 2 · People and journeys
The trough, and its answer
- The low point is Search, and it is not about speed. The investigator cannot distinguish 'nothing happened' from 'the index has not caught up', and an audit tool that cannot tell those apart is worse than useless in an incident.
- Answered by a completeness bound on every result: the query states the sealed segment it is complete to, so an empty result is an empty result up to a named point in time rather than a shrug. Views 22 and 28.
- Second pain: the payload is not in the index. Answered by an on-demand segment fetch for the page being read, which keeps the index cost proportional to the fields people filter on rather than to the bytes they occasionally open.
What the platform owes each phase
- Scope: a precomputed RBAC scope so an investigator is not waiting on an authorization round trip per page.
- Search: partition pruning by tenant and time before any scan, and a stated query budget so a 13-month scan cannot silently become a cost incident.
- Hand over: a signed bundle rather than a CSV, because the timeline is going somewhere it will be challenged.
Assumptions
- Investigators work inside a scope granted per matter, not standing tenant-wide read. The justification reference is mandatory and is stored on the read record.