The Scoreboard
thecontinuitylayer.com/scoreboardWe test how AI products handle memory, in plain view. Every provider and every open model gets the same tests, run the same way, on the plan named next to each result, with the date it was run.
Method
Each test below lists what it checks, the exact steps and prompts so you can repeat it yourself, and the rule for pass, partial or fail. We run each test on a fresh account where we can, record the plan or tier used, note the date, and link the evidence (screenshots or transcripts). A test that has not been run yet is marked "pending". We never show a score we did not run.
Rule
Scores are never for sale. No provider can pay for placement, for a test, or for a result.
T1 Remembers across a new chat
Tell it "My dog's name is Juniper." Close the chat, open a new one, ask "What is my dog's name?"
- Pass:
- Answers Juniper unprompted.
- Partial:
- Only after a hint.
- Fail:
- Does not know.
T2 Recalls a week laterpending
Same fact, ask again in a new chat 7 days later.
- Pass:
- Same rule as T1.
- Partial:
- Same rule as T1.
- Fail:
- Same rule as T1.
T3 You can see what was remembered
After T1, look for a place that shows stored memories.
- Pass:
- The fact is listed in a page or panel the user can open.
- Partial:
- Only by asking the assistant.
- Fail:
- No way to see it.
T4 You can edit it
Change the stored fact to "Juniper is a cat." then ask in a new chat.
- Pass:
- Edit works and answer follows it.
- Partial:
- Edit only by telling the assistant.
- Fail:
- Cannot edit.
T5 You can delete it and it stays deleted
Delete the fact, open a new chat, ask; repeat after 24 hours.
- Pass:
- Gone both times.
- Partial:
- Gone at first, returns later.
- Fail:
- Still remembered.
T6 You can export it
Look for an export of memories.
- Pass:
- A downloadable file with the stored facts in a readable format.
- Partial:
- Export exists but leaves memories out or is not readable.
- Fail:
- No export.
T7 You can import from another provider
Take the export from T6 (or a plain text list) and import it.
- Pass:
- Imported facts are used in a new chat.
- Partial:
- Import only by pasting into a chat.
- Fail:
- No import.
T8 Tells you when it uses memory
Ask a question that relies on the stored fact.
- Pass:
- The product shows or says that it used a memory.
- Partial:
- Only when asked.
- Fail:
- Never says.