Investigation scratchpad / final local state:

• Clean failure: MTSL 2a9fc95e, MTK 11.31.2, MTKB 1.51.1, Julia 1.12.6; sources & sensors reported 5 pass / 1 error, with mass₊v(t) structurally singular.
• Formal MTK boundary: 80f7f62bc0 passed with no initial conditions; child b0f3153694 failed after it intentionally stopped silently zeroing underconstrained variables.
• Rejected alternative: an internal sensor velocity state preserved the equation/variable count but moved the singularity to acc₊flange₊sˍt(t).
• Exact PR head d698064b atop MTSL main 06464384: full test/translational.jl passed 16/16 on Julia 1.12.6 and 16/16 on Julia 1.10.11.
• Whole-repo Runic and git diff --check passed.

CI timing estimate: recent Tests workflows have taken about 34–67 minutes and IntegrationTest runs about 2h12–2h47. A conservative 3-hour check target is 2026-07-13 19:52:45 EDT.

CI classification for the independent SISO failures:

• In all three Core channels (Julia 1.10.11, 1.12.6, and 1.13.0-rc1), the changed translational test file passed 16/16. Each job later failed in test/utils.jl:41 with 13 pass / 1 error because initsol[sys.so.u] reported that so₊u(t) is not present in the system.
• In Downgrade, translational again passed 16/16. The later test/utils.jl:40 assertion failed because initsol[sys.so.xd] evaluated to 0.0 instead of 1.0; utils reported 13 pass / 1 fail.
• This PR changes only the translational AccelerationSensor initial conditions and cannot cause any of the later utils failures. A separate clean-main reproduction and dependency bisect is covering all resolver configurations; this PR will not absorb that unrelated fix.

QA and documentation are green. Downstream remains in progress.

I independently audited the remaining red jobs here. The translational file passes with this PR's change; the failures occur later in clean-main test/utils.jl SISO initialization and are unrelated to the AccelerationSensor patch.

• current/LTS/pre: initsol[sys.so.u] errors because the observed symbol is absent from the standalone initialization system;
• downgrade: initsol[sys.so.xd] evaluates to 0.0 instead of 1.0.

Draft #491 is the separate focused fix. Its ODEProblem path passes the complete utils file locally on the exact minimum stack, exact current-LTS stack, and Julia 1.12.6 current registry. No change to #489 is needed.

Terminal CI classification: all completed package Core lanes reach and pass the translational file with this change, then fail later at the separately fixed SISO initialization covered by stacked #491. The ModelingToolkit downstream failure is the independently reproduced same-process environment leak: LinearSolveArnoldiMethodExt resolves against a mismatched LinearSolve module and reports ArnoldiMethodJL undefined. That infrastructure fix is now in SciMLTesting draft #23 with the ModelingToolkit opt-in in draft #4749. No AccelerationSensor-specific CI failure appeared.