Memory Persistence, Scope, and Curation for CCAO-F

Durable memory, deliberately kept

The persist response from restart, summarise, persist leans on Memory and Project knowledge bases, and the CCAO-F exam tests, at the apply level, how to use them well. Memory carries durable facts across sessions so you stop re-typing the same recurring context every time. But the exam's emphasis is not just on saving to Memory -- it is on scoping and curating what you save, because Memory that is never reviewed can quietly turn from helpful to harmful.

The skill has three parts: knowing what belongs in Memory, understanding that Memory can be scoped per Project, and treating stored memories as something to maintain rather than set and forget. Together these make Memory a reliable asset rather than a growing pile of possibly-stale facts. This connects back to project anatomy, since the knowledge base is the Project-level companion to Memory.

What to store, and what not to

Memory is for durable, recurring facts -- the things you would otherwise re-type at the start of session after session. Recurring role context, preferences for output format, names of frequent collaborators, and standing constraints that apply across your work are all good candidates. Saving them once means every future session starts with them in place, which is the efficiency Memory exists to provide.

The counterpart is what not to store. Transient, one-time details that will not recur do not belong in Memory; they add clutter without adding future value, and clutter is the enemy of a useful store. This is the same discipline as the persist trap in the three responses: persist durable facts, let one-off specifics go. The judgement is always "will this matter across future sessions," and only a yes earns a Memory entry.

Scope and curation

Memory can be scoped per Project, which matters when you work across separate workstreams. Project-scoped Memory keeps each workstream's facts separate, so context from one client or project does not leak into another. The assumption to avoid is that Memory entries automatically cross Project boundaries -- they do not; they stay scoped to their Project. Setting up separate Projects for separate workstreams keeps the Memory boundaries clean, echoing the conversation isolation principle at the Memory level.

Curation is the part candidates most underrate. Stored memories need periodic review because an entry that was accurate months ago can become actively misleading if left untouched. A fact about a collaborator's role, a client's preference, or a standing constraint can silently go out of date, and an unreviewed Memory will then feed Claude wrong information with full confidence. The remedy is to treat Memory like a working file: review it on a regular cadence -- at least once a month for active users -- update what has changed, and delete what no longer holds. The value of Memory is in the accuracy and relevance of its contents, not in how many entries it has -- a small, current store beats a large, stale one.

There is also a way to keep a session out of Memory entirely. Incognito mode runs a standalone chat that is not written to Memory or chat history, which suits sensitive or exploratory work with confidential inputs you do not want surfacing later. It is worth knowing that this is a Memory-capture control, not a data-retention override: it keeps a session out of your own history, but it does not change any underlying organisational retention that applies to the account.

What the CCAO-F exam trips candidates on

Two errors are tested. The first is treating Memory as a permanent, write-once store that never needs review or deletion. The credited reading treats Memory as a maintained working file, because unreviewed entries can become actively misleading as facts change. Accuracy comes from curation, not from accumulation.

The second is assuming Memory entries automatically cross Project boundaries rather than staying scoped to their Project. Memory can be scoped per Project, and separate workstreams should be kept separate. A question may imply a fact saved in one Project surfaces in another; the credited answer notes that Memory stays scoped and does not leak across Projects on its own.

Common misreadings to avoid

How this shows up on the exam

Questions describe someone relying on Memory that has gone stale, or expecting Memory to cross Project boundaries. Anchor on two points: Memory holds durable, recurring facts and must be curated to stay accurate, and it is scoped per Project rather than global. The reliable answer values accuracy and relevance over volume and keeps workstreams separated by Project.

This knowledge point applies the persist response from restart, summarise, persist, builds on project anatomy, and reinforces conversation isolation at the Memory level.