docs: close the cahier — §12 walk, F4, retire docs/reports/ (E6-C)

[ocots]

Final phase of the documentation-rewrite campaign. Walks the cahier's §12 global acceptance criteria, reconciles the stale symbol count (F4), documents the last shim-coverage gap, and retires the now-spent specification set (decisions.md §6).

§12 global acceptance criteria — the walk

#	Criterion	Evidence	Verdict
1	docs/Project.toml resolves against root Project.toml	make.jl LOAD_PATH-stacks the root env; the full E6-C build resolves and executes	✅ met
2	make.jl runs with draft = false, log clean of resolution errors	Mechanism changed in phase E5: make.jl ships draft = true globally; all 43 content pages opt back in with Draft = false in their @meta; only solve/gpu.md stays drafted (issue #885), api/ecosystem.md has no executable code. Build log: exactly the 6 pre-existing upstream @extref (Plots.plot(::CTModels.Solutions.Solution) ×4, CTBase.Strategies.parameter ×2) in the tolerated warnonly classes; 0 internal @ref breakage; 0 failed to run.	✅ met on substance — the criterion's literal wording (draft = false) predates the E5 per-page opt-in and should be reworded to "every content page executes except the deliberately-drafted gpu.md"
3	No user-facing page contains Lie(, ⋅, HamiltonianLift, OptimalControl.VectorField, bare autonomous=, Flow(ocp, u, g, — except the Migration page	grep: off migration.md, the only occurrences are the rename tables on geometry/ad.md and geometry/overview.md, labelled stale-form demos on geometry/lie-macro.md and flows/from-hamiltonians.md, and comments on functional-api.md explaining CTModels' real time_dependence!(…; autonomous=…) signature. No live-code violation.	✅ met on substance — the rename tables on the two geometry hub pages are deliberate (attic audit §1); the criterion should say "except migration/rename tables and labelled stale-form demos", not "except the Migration page"
4	Every capability in §6 has a page	walked bullet-by-bullet: Modelling (5), Direct resolution (6), Results (3), Flows (10), Geometry (6) — each resolves to a live page. ∂ₜ shares geometry/ad.md (as the §7 sitemap and api-coverage.md §9 both intend).	✅ met
5	docs/api-coverage.md has no uncovered re-exported symbol	enforced by the build: docs/api_reference.jl errors the build if any of the 203 names(OptimalControl) is absent from API_THEMES, or if a theme lists a non-exported, non-qualified name. Passes in this build.	✅ met (command: the docs build)
6	docs/attic/ is gone	test ! -d docs/attic	✅ met (#940)
7	Every §8.2 deprecated spelling throws a PreconditionError naming its replacement, or is documented in the Migration page as unshimmable with the reason	runtime-checked: 9 shims throw PreconditionError (Lie, ⋅, HamiltonianLift, Flow(f::Function), the 5-positional Hamiltonian-flow call, the 4-positional state-flow call, time(ocp), time(sol), success(sol)); Flow(ocp, u, g, μ) throws upstream in CTFlows naming the keyword form (CTFlows#401 closed); autonomous=/variable=/inplace= and augment=true are unshimmable and documented (migration.md "What could not be shimmed"); the OptimalControl.VectorField prefix is not breaking. CTSolvers.Modelers.ADNLP() / CTDirect.Collocation() — now added to that table by this PR.	✅ met (after this PR's migration.md row)

Changes

• F4 — docs/reports/99-api-coverage.md → docs/api-coverage.md. names(OptimalControl) is 203, not the stale 193: top-line and the §4/§7/§8 subtotals corrected (47→58, 22→25, 24→26), the "flagging for whoever next" note resolved (the 3-symbol gap is the §13 deprecated shims Lie/⋅/HamiltonianLift, now real exported bindings). Header reframed — the theme half of coverage is enforced by docs/api_reference.jl's build-time assertion; this file is the guide-side record.
• docs/src/migration.md — the Flow(ocp, u, g, μ) row's "suggestion text is currently wrong" note is stale (CTFlows#401 fixed; verified it now names Flow(ocp, u; constraint=g, multiplier=μ)). New "What could not be shimmed" row for CTSolvers.Modelers.ADNLP() / CTDirect.Collocation(): UndefVarError on the module name — CTSolvers/CTDirect/CTParser are deliberately not re-exported; the constructors themselves are, so write OptimalControl.ADNLP() / OptimalControl.Collocation().
• Retire docs/reports/ — the spec set was tracked only to be a stable contract for the campaign's ~13 PRs; that job is done. Keep only the coverage matrix; delete 00–10 and the Work board. Not built, not consumed programmatically. Four comment paths in docs/api_reference.jl updated.

Verification

Docs build (julia --project=. docs/make.jl): exit 0, VitePress site builds, the api_reference.jl completeness assertion passes (all 203 symbols), only the 6 pre-existing upstream @extref "cannot resolve", 0 internal @ref breakage.

Follow-up (not in this PR)

Investigating criterion 7 surfaced that the ecosystem deprecates removed v2.0 spellings four different ways — hard PreconditionError (OptimalControl deprecated.jl, CTFlows guards), soft Base.depwarn still working (CTBase is_parameter_type), silent alias (CTParser euler_b→euler_implicit, trapezoidal→trapeze), and nothing at all (CTSolvers, CTDirect, CTModels, CTLie). A coherent deprecation policy across the repos is a separate work item.

🤖 Generated with Claude Code