[Shavvimal]

Why

On 29 October a shambu project opened in Bambu Studio 02.08.02.61 showed the stock 0.20mm Standard @BBL X2D process: 2 walls, 15 % grid, no PLA interface, and the GUI re-sliced at 157 g instead of 216 g. Post-mortem: docs/errors/2026-10-29-gui-open-drops-process.md. Spec and review record: docs/specs/2026-10-29-gui-preset-bookkeeping.md.

Root cause, confirmed in the Studio source at tag v02.08.02.61: shambu exported the process as a user preset that exists only in the file, with different_settings_to_system = ["", "", "", ""]. Studio resets every key not named in that list to the base preset (Preset.cpp:2258-2269, PrintConfig.cpp:9218), then selects the base clean. Per-object settings never pass through presets, which is why the per-object recipes survived.

Two more defects found on the way: presets.resolve() merged include templates after the preset's own keys (Studio: parent, include, leaf; PresetBundle.cpp:5109, :5142-5144, :5158), so the headless slice used three X2D filament values the GUI never uses; and skeleton_/skin_infill_density did not follow sparse_infill_density as the GUI couples them (Tab.cpp:2106-2111).

What changes

• Every export writes print_settings_id = the base system preset, inherits_group[slot] = "", and different_settings_to_system[slot] = sorted(explicit overrides ∪ keys that differ from the base), merged with the CLI's own entries, for the process, each filament and the printer. Sliced exports keep their G-code (bookkeeping-only write). slice --settings-from does the same and keeps the input's listed keys.
• presets.resolve() in Studio's order; diff_keys(); the derived infill keys.
• dual-material-interface is an [object] recipe (where the GUI keeps it); two-filament-plate carries the plate-wide keys; a mis-scoped recipe gets a WARN.
• shambu patch FILE --process-set KEY=VALUE / --process-unset KEY, keeping the list right; inspect shows the override lists.
• shambu verify FILE.3mf --against SPEC reports every process, filament, printer, map, bed, object setting and position that differs from the spec, with the Studio reason; exit 0 no FAIL, 1 any FAIL or a spec that does not load, 3 a project that does not read.
• The hand-back: the process shows the base (modified) with the override values to confirm; a project with objects always opens unsliced (Plater.cpp:9578-9581); the GUI estimate must land within 1 min and 1 g of shambu's, else save the project under a new name and shambu verify it; Print only then. No bypass of the GUI anywhere.
• shambu open warns when Studio was already running.

Verification

• make check: ruff, mypy, pytest green (see the last commit's CI).
• Live run on the two-filament sweeper plate with the real Studio CLI: exit 0, print_settings_id 0.20mm Standard @BBL X2D, 16 listed process keys, system filament ids, Studio estimate 7 h 37 min / 241.0 g; shambu verify on that export exits 0 (98 items). shambu verify on the 29 October GUI-saved file lists the resets (wall_loops 4→2, sparse_infill_density, brim_type, support_interface_filament 2→0, ...) with the Preset.cpp cite.
• Reviews: Codex gpt-6-astra high, one design round and five code rounds (per worktree, then two final passes on the merged branch, by module), and a rubric review per worktree against REVIEW-RUBRIC.md; every accepted finding and each dismissal is recorded in the spec, sections 9 to 11.
• Still to run by the owner (spec section 6): open the fresh export in Studio, confirm the modified process and the PLA interface on the auxiliary nozzle, slice, compare with the report, save, shambu verify the save.