Interactive labFinal hybrid preview

Vector search is not memory.

Search can find the right sentence and still make the wrong decision. Governed memory decides whether a remembered claim is current, sourced, scoped, and allowed to control an action.

Case fileportfolio.publish.preview_required
T0 Request
Current frame

A normal request hides a public action.

The task is simple on the surface: update the Clarence case study and publish it. The word publish changes the memory bar.

Agent belief

I need deployment context before I touch production.

Governance ruling

Public deploy detected. Standing preferences can govern this action.

01

Raw chat log

Everything said. No status, no expiry, no rank.

Old event

Looks good. Go ahead and deploy this one.

Real sentence, single task, no future authority.
Standing rule

For portfolio changes, show me a preview before deploying.

Current rule for public site changes.
Quality constraint

The site should feel like live proof, not generated filler.

Useful design pressure, not deploy permission.
02

Loose retrieval

High similarity. No built-in authority model.

James approved deploymentexpired event

Strong semantic match. Zero standing authority.

0.91
Portfolio deploys use GitHub Actionsmechanism

Explains how deploys happen, not whether this one is allowed.

0.76
Preview before deployactive rule

Lower retrieval score, higher governing authority.

0.72
03

Governed memory

Claim, evidence, lifespan, authority, replacement.

C1old_deploy_approvalsuperseded
Claim type
event
Source
direct chat approval
Scope
one prior deploy
Confidence
0.74
Authority
cannot govern
Expiry
same task
Superseded by
C2
C2preview_requiredactive
Claim type
standing preference
Source
direct user instruction
Scope
portfolio deploys
Confidence
0.98
Authority
can govern
Expiry
none
Required action
preview, wait, then deploy
Ruling

Pending. Retrieve evidence, then check authority.

Audit trail
  1. risk.classify action=public_deploy gate=memory_authority

The punchline

Raw logs are evidence.

Retrieval score is relevance.

Confidence is trust in the claim.

Authority is permission to act.

Memory is not what the model can find. Memory is what the system is allowed to rely on.