memory_edit: a long range removal is quoted by both ends and names the defs it took

[dp-web4]

What happened

At 2026-10-07 19:03Z cbp-being was told to delete a 13-line duplicate decoder block in scratch/train-autoencoder-latent5.py. It sent start_line 75, end_line 107 with new "". That removed 33 lines: the block, plus AutoEncoder.forward and the start of train_model (its def, optimizer, criterion, scheduler, and for epoch loop).

The receipt quoted removed[:400] + "...". Those 400 chars were exactly the block the being meant to delete. The lines it should not have removed were all past the cut, so the being had nothing to tell it the delete went too far. It then "fixed" the orphaned return by replacing it with pass. The file parsed again, and the next run (seat seq 7178) raised NameError: train_loader inside __init__.

Change

• _head_and_tail: a removal longer than 400 chars is now quoted as its first lines, then [... N more removed lines not shown ...], then its last lines. When a range overreaches, the extra lines are at the end, so the end is what has to be shown.
• _defs_removed: the receipt now names every def or class line the removal took, e.g. This removed 'def forward', 'def train_model'. A list of names is short enough that it doesn't get cut.

Test

• Added test_memory_edit_long_range_receipt_shows_the_end_and_names_removed_defs. It fails on main and passes here.
• All 64 tests in test_reference_f1a.py pass.

🤖 Generated with Claude Code