The starting-loadout menu now saves independent quantities for backpack and ground destinations. Older two-field records still mean backpack. Both lists share the existing 32-entry / 64-item budget and only apply on the next custom prison or crash start. Ground stock bypasses inventory insertion and uses the same dry-ground, clearance and wall-access checks as oversized-item fallback.

Native Workbench verification

The final isolated G3 run compiled the current production sources and exercised:

Normal full mission preparation and player admission.
A full-length M16A2 rejected by backpack capacity but retained nearby.
An M16A2/M203 carbine explicitly assigned to ground, despite fitting inside the backpack in the earlier rifle tests. Exactly one was outside, zero inside.
The required field map and one PM pistol retained inside the backpack.
Spawn-scene inclusion and pickup of the original loose rifle into the player inventory.
Opening the real admin menu, selecting the ground destination, and saving two PM pistols through the authenticated catalog command. Both destination entries appeared in the returned snapshot; disk reload preserved one backpack pistol and two ground pistols independently.

native/runner.json reports smoke-completed and confirmed private-test cleanup. The source tree and installed dependencies were unchanged by the runner. The strict reporter still exits nonzero for installed localization resource and world-property errors; it reports no script/runtime failures. This is local scripted Workbench coverage, not a dedicated-server multiplayer certification.

Reproduce with a fresh alphanumeric suffix:

python3 docs/evidence/starting-ground-2026-10-28/isolated-runner.py G4 --full --carbine

The runner uses a frozen /tmp addon copy and private test profile; it does not modify the installed mission or open user Workbench. G1 could not launch under sandbox audio restrictions. G2 verified the physical loadout; G3 added actual menu/save/reload verification. Final evidence is from G3.

Repository verification

tests/test_start_backpack.py executes the production validation and edit methods: legacy compatibility, destination-specific edits/removal, duplicate rejection, shared limits, authority and failed-save retention. tests/test_start_stock_overflow.py executes the preparation loop and verifies explicit ground stock never requests an inventory insertion, remains the same entity and is processed only once. The existing required-stock and asynchronous insertion checks remain intact.

Full python3 tools/check.py passed (exit 0); output is saved in repository-checks.log. Generated-menu freshness and git diff --check also passed.