[ChrisRackauckas-Claude]

Closes #556.

What changed and why

wronskian built the param_values / initial_conditions / input_values arguments of
power_series_solution with untyped dict comprehensions. TimerOutputs v1's @timeit emits
the wrapped body twice (inside if enabled … else … end), and that duplication makes Julia's
closure conversion box the locals those comprehensions capture — so their element types stop
inferring. For any model with no inputs the empty comprehension then produces
Dict{Any, Any} and power_series_solution has no matching method.

Spelling the element types out fixes the dispatch and is independent of whatever @timeit
does to the surrounding scope. Only src/wronskian.jl is touched; no behavior change for
models that do have inputs.

Full analysis, standalone TimerOutputs MWE and the registry timeline are in #556.

Verification

Julia 1.12.7, environment resolved with Pkg.update() (picks TimerOutputs v1.2.1,
RationalFunctionFields v0.3.5), ModelingToolkit master Pkg.developed in exactly as
ModelingToolkit.jl/.github/workflows/Downstream.yml does.

Failing before the fix

GROUP=Core julia +1.12 --project=StructuralIdentifiability.jl -e 'using Pkg; Pkg.test()'
on unmodified master (90f8452a61109fe29e434d604bd22ad80d474fdd):

Assessing identifiability: Error During Test at test/bodies/identifiability.jl:3
  Got exception outside of a @test
  MethodError: no method matching power_series_solution(::ODE{Nemo.fpMPolyRingElem}, ::Dict{Nemo.fpMPolyRingElem, Int64}, ::Dict{Nemo.fpMPolyRingElem, Int64}, ::Dict{Any, Any}, ::Int64)

  Closest candidates are:
    power_series_solution(::ODE{P}, ::Dict{P, Int64}, ::Dict{P, Int64}, !Matched::Dict{P, Vector{Int64}}, ::Int64) where P<:(MPolyRingElem{<:FieldElem})
     @ StructuralIdentifiability src/ODE.jl:202
    ...
  Stacktrace:
    [1] wronskian(io_equations::Dict{QQMPolyRingElem, QQMPolyRingElem}, ode::ODE{QQMPolyRingElem})
        @ StructuralIdentifiability src/wronskian.jl:218

Test Summary:                 | Error  Total   Time
Core/identifiability.jl       |     1      1  35.5s
ERROR: Package StructuralIdentifiability errored during testing

The group aborts there, so nothing after identifiability.jl runs at all.

Passing after the fix

Same command, same environment, with this branch:

Core/identifiability.jl |   11     11  2m13.7s
Core/identifiable_functions.jl |  800    800  37m14.5s
Core/io_cases.jl |    6      6  0.4s
...
Core/y_saturation.jl |    9      9  4m14.2s
     Testing StructuralIdentifiability tests passed

All 39 Core files pass; every one of them is new coverage relative to the current master
run, which never got past identifiability.jl.

QA

GROUP=QA julia +1.12 --project=StructuralIdentifiability.jl -e 'using Pkg; Pkg.test()'

Test Summary: | Pass  Total     Time
QA/qa.jl      |   21     21  1m36.9s
     Testing StructuralIdentifiability tests passed

Formatting / spelling

Runic.main(["--check", "src/wronskian.jl"]) → exit 0. typos src/wronskian.jl → clean.

What I did not verify

• The Downgrade and 32-bit (Core 32-bit) lanes — not run locally.
• The ModelingToolkitSIExt group — not run locally (it precompiled cleanly during the QA run).
• The docs build — this PR touches no docstring, docs/, or public API.

Note for the reviewer

The underlying defect is in TimerOutputs v1's @timeit, not here; this is a targeted
workaround at the one call site whose dispatch depends on the dict type. The Core.Box
that @timeit introduces is present in every @timeit-annotated function in src/
and is silently costing inference elsewhere, so an upstream fix is still worth chasing —
see #556 for the standalone MWE to file with TimerOutputs.jl.

The alternative fix would be to cap TimerOutputs = "0.5.10" again, but #548 relaxed that
deliberately to resolve a Groebner conflict, so re-capping would reintroduce it.

Please ignore until reviewed by @ChrisRackauckas.

🤖 Generated with Claude Code (model: claude-opus-5[1m])

https://claude.ai/code/session_01GdSpCLd7NBZuuePJmcDzU7

[ChrisRackauckas-Claude]

Downstream mirror: ModelingToolkit's failing job, reproduced and fixed locally

Exactly the configuration of ModelingToolkit.jl/.github/workflows/Downstream.yml
(StructuralIdentifiability.jl/Core/1/): MTK master Pkg.developed into this project plus
its lib/* subpackages, Pkg.update(), GROUP=Core Pkg.test(), Julia 1.12.7.

Before (master 90f8452a61109fe29e434d604bd22ad80d474fdd):

Test Summary:                 | Error  Total   Time
Core/identifiability.jl       |     1      1  35.5s
ERROR: Package StructuralIdentifiability errored during testing

After (this branch, same environment):

Core/y_saturation.jl |    9      9  4m15.2s
     Testing StructuralIdentifiability tests passed

45 test files, 0 errors. So merging this turns
https://github.com/SciML/ModelingToolkit.jl/actions/runs/34395760914/job/102614954099 green
without any change to ModelingToolkit.

Docs build

Master's docs build is broken by this same bug — julia +1.12 --project=docs docs/make.jl
on unmodified master gives three doctest failures, all
MethodError: no method matching power_series_solution(…, ::Dict{Any, Any}, ::Int64):

┌ Error: doctest failure in src/parametrizations.jl:165-183
┌ Error: doctest failure in src/observation_field.jl:22-37
┌ Error: doctest failure in src/identifiable_functions.jl:33-46

The same command on this branch, same package versions, exits 0 with no doctest failures.

The two red CI checks are both pre-existing and unrelated

tests / Core (julia 1, ubuntu-latest, x86) — fails during precompilation, not in any
test: MethodError: no method matching crt_precompute!(::BigInt, ::BigInt, ::BigInt, ::Vector{BigInt}, ::Vector{UInt64})
from Groebner on 32-bit. It has failed the same way since the lane was added, and #554 was
merged with it red:
https://github.com/SciML/StructuralIdentifiability.jl/actions/runs/34163517295/job/101771799884

Documentation — fails on src/lincomp.jl:35-44, purely equation ordering
(x4' x3' x1' x2' vs the documented x1' x3' x2' x4'). That job runs Julia 1.13.0,
which changed Dict iteration order. On unmodified master:

julia +1.12 → ORDER: x1' x3' x2' x4' y1     (matches the doctest)
julia +1.13 → ORDER: x4' x3' x1' x2' y1     (matches what CI printed)

No source from this PR is involved — linear_compartment_model never calls wronskian.
Master's docs build simply dies earlier, on the power_series_solution failures above, so
this ordering failure only becomes reachable once those are fixed. Filed separately as #559.

Green checks

Core (julia 1), Core (julia lts), Core (julia pre), QA (julia 1),
ModelingToolkitSIExt (julia 1), Downgrade (Core), Runic Format Check,
Runic Suggestions, Spell Check.

🤖 Generated with Claude Code (model: claude-opus-5[1m])

https://claude.ai/code/session_01GdSpCLd7NBZuuePJmcDzU7