MENSARA.AI · A GOVERNED MIND, IN YOUR ACCOUNT
Sign in
Mensara Create account

Use memory well · GUIDE 06

Ask what is current, then ask why

Two questions often look alike but need different sources: “What applies now?” and “Why did this change?” Ask the first when you need today's guidance. Ask the second when you need history, context, or a review trail.

Use the right view for the job

For a current question, ask the assistant for the rule or decision that should guide work now. Material that is waiting, expired, replaced, or in conflict may stay out of that answer. That does not erase it.

For a historical question, ask for the evidence behind a decision and read the underlying source. Looking at history does not give the assistant access to areas it could not already read.

For example, ask “What is our current refund policy?” when helping a customer today. Ask “Why did we change the refund policy?” when reviewing the decision. The answers may use different notes.

Check the answer

Open the source behind an important answer. A note's status records how it has been handled; it is not a guarantee that every statement in it is true. If the source is missing or access is refused, report that gap instead of filling it with a guess.

If you find a correction, a suggestion still needs its own permission and review path.

Related

Who can read and suggest changes?

NEXT GUIDE · 07 Ask to see the source behind an answer An answer is more useful when you can see what it is based on. Ask the connected assistant to include the notes it used, then open the relevant source before you rely on an important answer. →
Create account

MENSARA.AI · ASK WHAT IS CURRENT, THEN ASK WHY · AGENT VIEW · TEXT/MARKDOWN

# Retrieval rule

Use `memory_view:"current"` for ordinary factual retrieval. Use `memory_view:"evidence"` only when the request explicitly requires history, provenance, or review. Both read and memory-target authorization occur before lifecycle projection; evidence does not widen the principal or zone grants.

**Template only — substitute authorized discovery paths.**

```
Current policy question: read({"path":"<discovered policy memory path>","memory_view":"current"})
Explicit historical question: read({"path":"<authorized historical memory path>","memory_view":"evidence"})
```

For a current exclusion, return the actual condition: `ok=false`, `error=memory_not_current`, plus reported reasons. Do not automatically issue the evidence call. Classifications such as conflicted, superseded, expired, pending, and no-promote keep a classified entry noncurrent; do not claim ordinary files all carry lifecycle metadata.

**Observed synthetic in-process output — not an MCP transport response.** The captured pending current-view fixture returned `ok=false`, `error=memory_not_current`, `reasons=[pending_approval]`. The captured explicit-history fixture read an expired record using evidence view and returned `lifecycle_state=expired` with `current_eligible=false`. Captured denied calls in both views returned `access denied`. Use these as boundary examples, not customer paths or a client setup transcript.

Use `orient` memory-authoring output for possible authorized destinations. If none are discovered, stop rather than inventing a customer taxonomy. A current or approved item is not independent proof of factual truth. For a correction, obtain the current revision: typed replace, edit, append, and delete operations require `base_sha`. Inspect a proposal disposition before saying a change applied.

For memory changes, an authorized memory entry's SHA-256 revision is an accepted base token; the gate also accepts the target Git HEAD blob token. This does not make an arbitrary served-content hash valid for an ordinary file. Obtaining an ordinary file's base token through a supported client remains unverified in this pilot.

## Continue

[Read and proposal permissions](/how-it-works/read-and-propose)

[Generated tool input reference](/how-it-works/tool-inputs.md)