<aside> 🧠

A memory your AI agents can trust - because every fact has proof and an expiry date.

Agents that only collect notes get worse over time. This register keeps one fact per row, shows which facts are still safe to act on, and makes old facts retire instead of quietly going stale.

</aside>

Start here

Look at the eight sample facts first. They show a current fact, a rule that never expires, one due for a re-check, one expired, one retired, and one with no proof.

  1. Open Memory Entries and add one fact your agent keeps getting wrong. Fill Why It Matters, paste an Evidence Link, and set Re-verify By (90 days is a good start for anything that can change).
  2. Read Trust Status on that row. Your first useful output is one fact showing 🟢 Current.
  3. Once a week, open the rows marked 🟡 or 🔴, check each against the real thing, and log the result in Review Log.

QualityGate

A fact is trusted only while Trust Status is 🟢 Current. That needs an evidence link, a reason it matters, and either a re-verify date that has not passed or the Never Expires box ticked by a person. A fact without proof is a rumour.

A review counts only when Review Gate is green: an entry, a date, an outcome and an evidence link.

Working databases

Memory Entries

Review Log

Decision rules

Trust Status

Status Meaning What to do
🟢 Current Has proof and is inside its re-verify date, or is a rule marked Never Expires Agents may act on it
🟡 Re-verify this week The re-verify date is seven days away or less Check it against the real thing and log a review
🔴 Expired - advisory only The re-verify date has passed Treat as a hint, not a fact, until someone re-checks it
⛔ Incomplete - not trusted No evidence link, or no reason it matters Add the proof or delete the row
⛔ No re-verify date Nothing says when to check it again Set a date, or have a person tick Never Expires
⚫ Retired Found to be wrong, and kept for the record Do not apply. Do not delete.

After a review

Before writing a new fact