Fall back to downex instructions during older kid tasks

[zwatson2001]

This PR adds the downward extension instructions to the task if the participant meets the criteria (2 out of the first 4 trials are incorrect) for memory game, matrix reasoning, and mental rotation.

[asengupta3]

LGTM! @zwatson2001 I am wondering if we should also update some of our cypress tests to reflect these new changes?

[zwatson2001]

LGTM! @zwatson2001 I am wondering if we should also update some of our cypress tests to reflect these new changes?

Great, thanks for reviewing @asengupta3! We definitely could - the current tests won't trigger this feature because they only select correct answers, but it might be good to test other paths as well.

[digital-pro]

That’d be cool! I’ve been trying to figure out how to build a “bot” that can deal with CAT, etc. I asked Adam Ritchie for his thoughts and he has a conceptual idea, but implementing it could be someone’s dissertation. Obviously we can take small steps, but it is not that different (maybe harder) than writing alphaGO (because we’re not hoping it learns how to win, but sorting through paths and judging outcomes against “something.” From: Zach Watson ***@***.***> Sent: Friday, February 27, 2026 4:51 PM To: levante-framework/core-tasks ***@***.***> Cc: Subscribed ***@***.***> Subject: Re: [levante-framework/core-tasks] Fall back to downex instructions during older kid tasks (PR #421) [https://avatars.githubusercontent.com/u/156386417?s=20&v=4]zwatson2001 left a comment (levante-framework/core-tasks#421)<https://urldefense.com/v3/__https:/github.com/levante-framework/core-tasks/pull/421*issuecomment-3975909635__;Iw!!G92We9drHetJ8EofZw!btZNmmnMIagg-OeiOrK01l9wCP4YvcKauDoUglAUSDhRRlZ45RGwRzaS8ythRuFrUoQV4Y5Oo-NkV-5U6Dq3XzeSRxrZ$> LGTM! @zwatson2001<https://urldefense.com/v3/__https:/github.com/zwatson2001__;!!G92We9drHetJ8EofZw!btZNmmnMIagg-OeiOrK01l9wCP4YvcKauDoUglAUSDhRRlZ45RGwRzaS8ythRuFrUoQV4Y5Oo-NkV-5U6Dq3XztC4QHq$> I am wondering if we should also update some of our cypress tests to reflect these new changes? Great, thanks for reviewing! We definitely could - the current tests won't trigger this feature because they only select correct answers, but it might be good to test other paths as well. — Reply to this email directly, view it on GitHub<https://urldefense.com/v3/__https:/github.com/levante-framework/core-tasks/pull/421*issuecomment-3975909635__;Iw!!G92We9drHetJ8EofZw!btZNmmnMIagg-OeiOrK01l9wCP4YvcKauDoUglAUSDhRRlZ45RGwRzaS8ythRuFrUoQV4Y5Oo-NkV-5U6Dq3XzeSRxrZ$>, or unsubscribe<https://urldefense.com/v3/__https:/github.com/notifications/unsubscribe-auth/ABJ4N3GJGIIFRDHUJIU6Q2D4ODQ6PAVCNFSM6AAAAACU65RCH6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTSNZVHEYDSNRTGU__;!!G92We9drHetJ8EofZw!btZNmmnMIagg-OeiOrK01l9wCP4YvcKauDoUglAUSDhRRlZ45RGwRzaS8ythRuFrUoQV4Y5Oo-NkV-5U6Dq3X6CGVRfm$>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.******@***.***>>

[zwatson2001]

That’d be cool! I’ve been trying to figure out how to build a “bot” that can deal with CAT, etc. I asked Adam Ritchie for his thoughts and he has a conceptual idea, but implementing it could be someone’s dissertation. Obviously we can take small steps, but it is not that different (maybe harder) than writing alphaGO (because we’re not hoping it learns how to win, but sorting through paths and judging outcomes against “something.”

Oh interesting! I guess eventually we could have the tests read the participant's estimated ability for CATs and check that it changes as expected from correct and incorrect answers, but that's definitely a bigger project.