Phase 4: the public interface and the result type

[adolgert]

Phase 4 of design/20260926_implementation_plan.md, stacked on #54. Review note with the target-screen transcript and open decisions: design/phase4_review.md.

• TestCases{T} <: AbstractVector{T}: typed rows (NamedTuple for named spaces, Tuple for positional), the space, strategy, engine and seed, must-include count, covered and excluded targets with attribution. show prints a summary, an excluded line, and a DataFrame-style table, and performs no search (tested).
• covering, all_values/all_pairs/all_triples, excursions, full_factorial over a TestSpace, a NamedTuple, pairs, or positional vectors; constraints = builds a space from named input and errors on a TestSpace or positional input; must_include accepts partial rows and an existing TestCases; stronger leaves the caller's vector untouched.
• Tables: DataFrame(cases) and CSV.write work for named results; the positional recipe is documented.
• Deprecations via Base.depwarn: all_tuples, n_way, seeds, wayness, values_excursion/pairs_excursion/triples_excursion, GND(M=). Removed: Excursion, Counter, generate_tuples; disallow is a MethodError.
• CI matrix is now Julia 1.10 (LTS), 1.13, and latest.

Full suite: 272,326 pass, 8 pending (Phases 5–6), 0 failures. Aqua and docs build green.

🤖 Generated with Claude Code

[adolgert]

Review round 1 applied (see design/phase4_review.md): an explicit keyword with its deprecated alias is now an error for all five alias pairs (the audit found a sixth silent case, distance vs n_way on the excursion aliases, fixed too); must_include is materialized once, so Iterators.Stateful and generators keep their rows; the excluded line says "impossible under the constraints"; the budget docs describe feasibility_limit (throws) and explanation_limit (certified design, unresolved attribution) separately; the strength = 0 sentinel is documented and the Phase 5 rule for measuring such results is in the plan and contract §1.12.

Full suite: 272,474 pass, 8 pending, 0 failures.

🤖 Generated with Claude Code

[adolgert]

Review round 2 applied (see design/phase4_review.md): the full-factorial limit is checked before any search; every integer keyword is validated before sorting or conversion with a focused message (audit table in the review); field types are computed from the values per contract §2.4 (Any[1, 2] → Int64, Any[1, 1.0] → Union{Float64, Int64}); the Phase 5 report default is min(2, parameter count); the full-factorial and excursion docstrings and contract §7.2 state that supplied duplicates are kept and enumeration adds each remaining valid row once.

Full suite: 273,137 pass, 8 pending, 0 failures.

🤖 Generated with Claude Code