django-oscar fix shows why code needs 'Laws', ArchSig author says

A familiar problem: human code reading no longer keeps pace with AI-generated code. Its proposed posture is to decide in advance which properties must survive a change, write them as Laws, and define Atoms as the specific facts to read from code. ArchMap records those observations with source locations, so a questionable value can be traced back to the code.

The django-oscar case shows why this matters. A change meant to make stock locking order consistent also reordered order-line creation because both operations shared one loop. Against L1 (stock allocated in stockrecord order) and L2 (order lines created in basket order), the pre-change code failed L1 but passed L2; the reordered code passed L1 but failed L2. Only after the loops were split into PR #4594 and PR #4598 did both Laws pass.

This links to algebraic architecture theory (AAT), framing Laws as equations whose solution space represents software satisfying them, and describing how diagnosis and repair candidates can be explored within allowed change scopes. It acknowledges limits such as Rice's theorem, noting that restricting to finite or linear models makes judgment and repair computable.