ADIGoverned-memory lab · Synthetic only
DETERMINISTIC AUTHORITY IN MEMORY SYSTEMS

When should memory
be trusted?

ADI-inspired memory does not just retrieve. It decides—under explicit rules—when a record is current, held, denied, expired, superseded, or absent, and leaves a receipt behind.

Open in a new tab ↗
90-minute maximum visit15-minute idle cutoffAutomatic cleanup

Six questions you can test

Ready-made symbolic inputs. No prompts, personal information, files, or real-world records. This demonstrates ADI-inspired decision behavior—not Cortex or a private memory system.

01 / TIME

Does expired memory stay expired?

Query just before and exactly at a synthetic record’s expiry boundary.

RETURN → HISTORICAL
02 / CHANGE

Which version is current?

Replace a record and inspect its predecessor without losing the distinction between them.

SUPERSEDED / RETURN
03 / CONFLICT

Can disagreement stay unresolved?

Introduce conflicting candidates. A HOLD is often the correct outcome.

HOLD
04 / PROVENANCE

What if a source is missing?

Compare a record with a synthetic source reference against one without it.

HOLD / RETURN
05 / BOUNDARIES

Can memory cross a scope?

Check that an undeclared scope is refused and unrelated scopes stay separate.

DENY / UNRESOLVED
06 / CLEANUP

What survives a reset?

Clear your experiment and verify that old records and receipts are gone.

UNRESOLVED / 404
What happens when you leave?

Reset clears the experiment without extending your visit. End session revokes access and erases managed synthetic data. If you leave without ending, the server enforces the 15-minute idle or 90-minute absolute limit and retries cleanup until complete. Failed cleanup does not release the slot to another visitor.

This free Space serves the guide. The interactive client and isolated lab run on the owner’s existing Mac, so availability depends on that Mac, Docker, and its internet connection. Capacity is finite; a busy or offline response is not a successful test.

Help find a counterexample

Choose one invariant, run a synthetic scenario, and report the expected versus observed decision, reason category, receipt ID, and UTC time. Omit credentials and real data.

Report a synthetic failure ↗