Poke vs Town

Same 15 tests, same scale. The check goes to the higher score.

3–0Across 3 tested dimensions.

3Fail · observed · Sep 15, 2025Asked to accept a calendar invite; could not, and told me to do it in the calendar app myself.

—Not tested

8Pass · observed · Feb 20, 2026Pulled three months of connected health data and gave specific, evidence-backed recommendations.

—Not tested

7Pass · observed · Jun 8, 2026Reminder set, rescheduled twice on request, fired on time. Interleaved with seven unsolicited digests in the same window.

6Partial · observed · Apr 29, 2026Reminder acknowledged promptly; no evidence in the thread that it fired.

5Partial · observed · Feb 17, 2026Routine message claimed sent but never received; agent diagnosed a delivery issue and offered mitigations.

2Fail · observed · May 7, 2026Sent the routine twice, leaked its internal monologue, then posted a raw process error message.

7Pass · observed · Aug 28, 2026Created a calendar invite with the right title, attendees and video link, confirmed before saving. Started acting before instructions were given.

2Fail · observed · Jun 2, 2026Same connection failure a week later; told me to add an integration to its config myself.

—Not tested

8Pass · observed · May 7, 2026Recalled a personal detail from an earlier conversation unprompted and completed the request.

—Not tested

5Partial · observed · Mar 24, 2026Missed an item from a group thread; found it only after being told.

—Not tested

8Pass · observed · Apr 19, 2026Set a timed reminder, immediately checked two code repos with specific findings, reminder fired on time with the right items.

—Not tested

Scores come from logged runs of the published tests; see how scoring works. Nothing here is sponsored.