Version Lab log
How to use it: Record what changed between v1 and v2 without deleting the recoverable baseline.
v1 identifier
Targeted failures
Changed examples
v2 identifier
Release decision
Mission 10 kit · screen and print
Targeted retraining may fix failures but also break earlier successes, so versions need regression tests and rollback.
How to use it: Record what changed between v1 and v2 without deleting the recoverable baseline.
v1 identifier
Targeted failures
Changed examples
v2 identifier
Release decision
How to use it: Lock four earlier successes and test the same anchors against both versions.
Anchor A
Anchor B
Anchor C
Anchor D
Same conditions for v1 and v2
How to use it: Compare both target failures and all four anchors side by side.
Six locked cases
v1 result
v2 result
Improved, same, or regressed
How to use it: Keep v1 when any locked anchor is lost, then plan a smaller targeted change.
ROLL BACK
Lost anchor
Keep v1
Investigate before another release
Safe mission inputs
Retrain a copy of the prop classifier and use AI Studio only for the adult-led version demonstration.