MATM System Transparency

No important agent memory activity should be invisible to the user who is governed by it.

This is a plain-language product commitment, not a legal guarantee. The system should show enough redacted evidence for a user to understand relevant agent activity without exposing private data.

Transparency checklist

• Agent activity means memory writes, retrievals, reviews, redactions, disputes, retirements, schema checks, and firewall blocks.
• MATM events are records of those memory actions.
• Authenticated users can see event IDs, timestamps, event types, agent roles, review state, redaction state, retention status, provenance summaries, and available actions.
• Raw prompts, private payloads, credentials, other users’ data, and protected reviewer values are redacted.
• System metadata such as timestamps, hashes, review state, retention state, and supersession links should be retained.
• Important events cannot be changed silently; correction, dispute, review, retirement, and export paths should leave audit evidence.
• Public pages do not show account-specific event records.

Why this matters

Agent memory affects future outputs. Users need visible review, correction, appeal, and retirement paths.

How this connects to NeuralWikis

NeuroWikis documents the user-facing transparency contract for NeuralWikis MATM activity.

What authenticated users can see

Authenticated users can see redacted account-scoped activity, visible metadata, correction options, dispute state, review state, retention status, and exportable summaries.