An unused import edge makes a method receive the wrong class as `self`

Reported by lekkerzeilen-7a 2026-09-20 (530ba53), found while running the attribution that refuted the performance hypothesis on feature-n-register-the-class-shells-of-the-import-closure-before-parsing-any-body.

What happens

Adding from .world import Grid to the top of wind.py — with Grid never referenced anywhere in that module — kills the program with:

AttributeError: 'World' object has no attribute 'flow'

flow is a field of Environment. So a World instance is arriving as self inside an Environment method: the receiver is an object of the wrong class and the field read lands at whatever that offset holds. CPython runs 250 frames on the identical source.

The edges in sim.py and rig.py are fine on their own. The wind.py edge alone is sufficient to break it.

Why the prio is 75 and not lower

It is a WRONG OBJECT, not a wrong value: a method body running with a foreign instance as self. Every field read and write in that body is at a foreign offset. The AttributeError is the lucky outcome, not the characteristic one.
The trigger is adding an import that does nothing. No call site changes, no name is used, nothing in the program references the imported class. There is no edit a reader could look at and predict this from.
It blocks measurement of the hottest dispatch site in the demo (canopy.contains(), 1112 calls per frame, 69% of all dynamic dispatches), because the import edge that would resolve it statically is exactly the edit that triggers this.

THE REASON THIS MATTERS TO THE IMPORT-CLOSURE TICKET, AND IT IS A HAZARD NOT A HINT

feature-n-register-the-class-shells-of-the-import-closure-before-parsing-any-body proposes registering the class shells of the whole import closure before any body is parsed. That is, in effect, giving every module the visibility that an explicit import edge gives it — which is the operation that triggers this bug when done by hand in one module.

So the two tickets are the same family pulling in opposite directions: one is FIXED by adding an edge, the other is CAUSED by adding one. Whoever implements the closure registration must reproduce this first and understand it, or the fix may deliver this failure across a whole program at once, as a wrong object rather than a diagnostic. A green tier would not necessarily catch it: the condition needs several modules and a class-name population that a fixture author has no reason to construct.

NOT REDUCED

7a could not reduce it below the demo and did not locate the failing call site. Per the tree's own rule, two failed reductions in this family mean the next attempt should run candidates against the demo rather than reduce further — mechanism from whoever has the source, verdict from whoever has the failing tree. The demo is the artefact; the trigger is a one-line import.

What would retire this