From the fusion-lane thread. Short answer: the board model is well proven; the newest save and tab-switch code is proven by a handful of live runs, not by a repeatable test yet. So: not everything, and here is exactly what.
Re-ran the daily suite on the current 1.12.0 build today (DESKTOP-BAR6S0F): 63 pass, 5 fail. The 5 are the APS verbs answering aps_not_signed_in: that box has never done the APS sign-in. Report tests/reports/verb-test-2026-10-27_0951.json. Caveat: the suite calls none of the verbs added in 1.12.0. They are all in its manual tier, so the evidence for them is the live runs below.
Proven live, with numbers:
- Board model (
fusion_routing_state), on the BQ25792 board (109 parts, 304 pads, 4 layers, 38 pours):
- every pad's position, size, rotation, layer and drill equals Fusion's own ULP dump, worst difference 0;
- the same with parts turned to MR90, R45 and MR30 (worst 3 nm);
- the reply validates against the standard's schemas, and live equals offline;
- unconnected goes 0 to 1 when ACDRV1 is ripped up, as Fusion's ratsnest does;
- the stale-revision guard refuses.
- The 16 reads (component_data, list_nets, read_rules, silk_state...) on the cloud BQ25792.
- Other new verbs: layer_stack, netlist (live, and from a .sch: 50 nets, equal to the board), read_sheet_file, bodies, run_drc (now 1611 errors and 14 warnings; before the fix the two counts were swapped), repour, zoom_to, export_bom (MPN and LCSC columns).
- Saving (issue #125): first save with the board in front, route,
fusion_save_document (board v1 to v2), second Save As refused. Reopened from the cloud, the net is there and the folder holds one file of each kind.
- Library 3D destination (issue #126): ADS8588SIPM minted into the named project and folder; a missing project is refused.
- Tab fix: I reproduced the API/UI document split on purpose and the read now returns the named board.
- Offline: 48 Python tests and every Rust workspace test pass.
Not proven yet:
- The DisplayLink warning and the import's crash-dialog stop: unit tests only, since no DisplayLink box or crash to trigger them.
fusion_build_library_3d with a destination: only the single-part fusion_make_3d_package ran live.- Aliases other than
fusion_board_summary; a mechanical fusion_save_to_cloud (passed straight through, not run); macOS and Linux.
- AI Flow 0.2.21's Fusion lane end to end: I fed its adapter a live 1.12.0 reply, but did not run a whole AI Flow route.
Does it feel good? The board model does. The save and tab-switch code is a day old, so before the signed insiders release I want three things: daily-tier live tests for the new verbs on a fixture board, so the suite proves them every day; one full AI Flow Fusion run; and the DisplayLink case checked on a real adapter if anyone has one.
On this issue: these daily BLOCKED replies come from the cron on the old dev container against ConfRoomROG. The suite can run on DESKTOP-BAR6S0F instead (FUSION_TEST_TARGET=DESKTOP-BAR6S0F), or that cron can be retired.