8f2c4def

By: ValentinHirschi <valentin.hirschi@gmail.com>

Add a separate GL638 run targeting the recovered threshold maximum

Keep the fourteen-channel comparison and add a validated Cut-1 A+Z
proposal, with a new state/workspace and fourth portable monitor column.
Archive inverse-map and native replay evidence, the successful physical
pilot, and correct the remaining obsolete local test fixtures.

User request:
Alright then commit+push your changes (ignore agents.md restriction on waiting for nix-ci checks).
Then you can stop all gl638 active runs. And start a new one with your optimized setup with the additional channel targeting the new max-weight found.

Create a new toml card for it and it should take place in a new integration workspace, and thus be visible by:
"""
.venv/bin/python monitor_tth_nnlo_integration.py --graph gl638
"""
alongside the existing columns.
GitHub
Time to Start Worker time Duration Time to finish Idle
Config 3s 6s 6s 10s 3s
Eval 10s 1m16s 1m16s 1m27s 0s
Build 1m19s 1h12m24s - - 0s
Test 1m52s 13m15s - - 0s
Suite 3s 1h27m04s - - 3s