[ChrisRackauckas]

What was stripped

d0061cd — "Remove dependency facade and enforce strict QA (#318)" — deleted 210 lines from src/EasyModelAnalysis.jl: the four bare

using DifferentialEquations
using ModelingToolkit
using Distributions
using Plots

lines and the four explicit export blocks that 713e02c (#317) had added to pin them.

using EasyModelAnalysis went from 1156 names to 26 — the package's own 25 query functions plus the module name. For a package whose entire premise is "one-liner queries over SciML-defined models", that removes the thing the package exists to provide.

Evidence that this was documented use

The stripping commit is its own counter-evidence: it had to rewrite the README quick demo from using EasyModelAnalysis to using DifferentialEquations, EasyModelAnalysis, ModelingToolkit, Plots. And 713e02c before it had already had to add the same imports to every docs page:

page	was	became
getting_started.md	using EasyModelAnalysis, Plots	using DifferentialEquations, EasyModelAnalysis, ModelingToolkit, Plots
tutorials/datafitting.md	using EasyModelAnalysis	+ DifferentialEquations, Distributions, ModelingToolkit
examples/SEIRHD.md	using EasyModelAnalysis	+ DifferentialEquations, Distributions, ModelingToolkit, Plots
examples/ASIR.md, examples/petri.md, examples/Carcione2020.md, scenarios/2–4, tutorials/{ensemble_modeling,sensitivity_analysis,threshold_interventions}.md	same	same

Every one of those pages previously ran on using EasyModelAnalysis alone.

How the list was chosen

Per "we should be reexporting what is normal documented use, but not whole dependencies", this is not a restored blanket @reexport and not the previous 1130-name full-surface restore. It is 393 names, chosen as:

1. Every name used in docs/src/**/*.md code blocks, README.md, and test/ — extracted mechanically by matching each of the 1130 previously-exported names against the code fences (not prose, so fit/probs/alpha/color as local variables and keyword arguments do not count). That yields: ODEProblem, EnsembleProblem, OptimizationProblem, OptimizationFunction, remake, solve, SciMLBase, Tsit5; @independent_variables, @named, @parameters, @variables, @unpack, @mtkbuild, @mtkcompile, Differential, ODESystem, System, Num, structural_simplify, complete, equations, continuous_events, discrete_events, add_accumulations; Normal, Uniform, Truncated, truncated, var; plot, plot!, scatter!.
2. The coherent API cluster around each, because these four are exactly the four capabilities the package advertises — build a model, solve it, sample a parameter distribution, plot the result. A user who can write ODEProblem/solve needs ODESolution, init, remake, the ensemble types, the callbacks, ReturnCode/successful_retcode and the default solver set; a user who can write Normal needs the other distributions and pdf/cdf/quantile/mean/std; a user who can write plot needs scatter, histogram, savefig, xlabel!, @layout.

Deliberately excluded (the ~740-name difference): SciMLBase's internal traits and function wrappers (has_*, is*, *_cache, WOperator, StaticWOperator) and problem/solution types for equation classes this package does not analyze; the SciMLOperators operator algebra; ModelingToolkit's structural-transformation internals (TearingState, BipartiteGraph, StructuralTransformations, Rewriters, RuleSet, pantelides_reassemble, alias_elimination) and the entire sympy_* / sympy_pythoncall_* bridge; the exotic tail of Distributions (WalleniusNoncentralHypergeometric, EdgeworthZ, KSOneSided, the ZeroMean*Canon family); and the ColorTypes color-space types Plots reexports (DIN99d, LCHuvA, AOklch, CIE1931JV_CMF, …) plus the backend-selection functions.

Per-package: DifferentialEquations 74/207, ModelingToolkit 93/327, Distributions 95/243, Plots 131/353.

Documentation

New page docs/src/api/reexports.md, wired into the API section of docs/make.jl. This is the part that matters most here: EasyModelAnalysis reexports across four unrelated packages, so without it a reader has no way to know that plot is documented by Plots, @variables by ModelingToolkit/Symbolics, ODEProblem by SciMLBase and Normal by Distributions.

The page states up front that EasyModelAnalysis owns and documents none of these names, gives an owner/documentation table for the four packages, then lists all 393 names grouped by owning package and by role within it — modeling / solving / distributions / plotting — each group linking to the owner's docs. Every section closes with an explicit boundary line naming what was deliberately excluded and why.

A script check confirms the page's bullet lists and the export blocks name exactly the same 393 symbols.

Drift protection

docs/src/api/reexports.md, the four export blocks in src/EasyModelAnalysis.jl and the REEXPORTS tuple in test/qa/qa.jl are the same list in three places. test/qa/qa.jl declares the same 393 symbols as REEXPORTS, passes them to reexports_allow, reuses them for api_docs_kwargs (ignore/rendered_ignore — these names are documented upstream) and for ei_kwargs.no_implicit_imports.ignore (they are reexported, not used), and adds a testset asserting each is in names(EasyModelAnalysis) and in scope.

Semver

Restoring names that used to be exported is not breaking — a patch/minor bump. No version bump here; a separate release pass handles versions.

Verified

Run locally on this branch (macOS aarch64, Julia 1.11.8):

• Pkg.instantiate() + using EasyModelAnalysis — loads and precompiles clean.
• All 393 approved names are in names(EasyModelAnalysis) and in scope from using EasyModelAnalysis; names(EasyModelAnalysis) is now 419 (26 own + 393 reexported). MISSING: Symbol[].
• Runic v1 --check clean on the changed Julia files (src/EasyModelAnalysis.jl, test/qa/qa.jl, docs/make.jl); typos clean on all changed files.
• All external documentation links in the new page return HTTP 200 (linkcheck = true in docs/make.jl).

Not run: Pkg.test() / the test/qa suite. This package's test environment pulls Turing, GlobalSensitivity, OptimizationBBO/NLopt and SciMLExpectations, and the machine is heavily loaded by other jobs; the full instantiate+precompile alone took ~11 minutes. The change is export-only plus QA declarations, and the reexport-surface property that CI's QA checks is exactly what was verified above.

Note on the docs

The explicit using DifferentialEquations, ModelingToolkit, … lines that #317/#318 added to the README and docs pages are left in place — they are harmless now, and reverting them across 13 files would be churn orthogonal to restoring the public surface. Worth a follow-up if you want the docs to demonstrate the facade again.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Rmh6B9eoW3N8oCbuuVPVxJ