Dummies: distinct-collection exhaustion message over-promises replay for foreign element generators

[Reefact]

Problem

When a distinct collection cannot fill its bounded dedup draw, CollectionState.Exhausted (CollectionState.cs:246-254) reports:

> The arbitrary values were seeded with {seed}; reproduce this run with Any.Reproducibly({seed}, ...).

For a collection over a foreign IAny element generator, this promise is false. AnyCollection.Generate() substitutes the ambient source (SourceOrNull ?? AmbientRandomSource.Instance, AnyCollection.cs:149) because AnyDerivation.SourceOf returns null for a foreign generator. The reported seed is the ambient seed that drove the count and layout — but the foreign generator's own draws ignore it, so replaying with that seed reproduces neither the elements nor the shortfall. (The seed is not null guard at :249 is also dead code: SeededRandom.Seed is a non-nullable int and the source is never null there, so the replay sentence is always appended.)

Impact

Minor deviation from ADR-0013, whose decision states the bounded draw "fails explicitly and reproducibly." For every library-sourced or As/Combine-derived element generator the claim holds; only the genuinely-foreign case is affected — a narrow path. But there, a user replays the reported seed, gets different elements, and concludes reproducibility is broken — the exact trust failure the seed reporting is designed to prevent.

Direction

Thread knowledge of whether the element generator carries an IHasRandomSource into the exhaustion message. When it does not, drop or qualify the replay sentence, e.g.:

> The collection's count and layout were seeded with {seed} (Any.Reproducibly({seed}, ...)), but its elements come from a generator that does not draw from this source, so the element sequence is not reproducible from this seed alone.

Remove the dead seed is not null guard while here.

Acceptance criteria

• The exhaustion message for a foreign element generator no longer claims full Any.Reproducibly replay of the elements.
• The message for a library-sourced / As-derived element generator is unchanged (it is faithful).
• The dead null-guard is removed.
• A test drives a distinct collection to exhaustion with a foreign IAny and asserts the qualified message.

Context

Surfaced by the 2026-07-20 architecture & design audit of the standalone Dummies library (doc/handwritten/for-maintainers/audit/2026-07-20-dummies-architecture-and-design-audit.md, §6 (ADR-0013 compliance); branch claude/dummies-architecture-audit-cru5aq).

---

Filed from the Claude Code session that produced the 2026-07-20 Dummies architecture & design audit: https://claude.ai/code/session_01GvrWFVoG74xiys3oyEft1G

[Reefact]

Status against current main (post-#211) + two AC refinements

Re-checked before implementing. #211 (fcb7ef4, "report a source-appropriate replay hint on failure") merged into main after this issue was filed, which shifts the picture slightly.

1. AC "The dead null-guard is removed" — already done by #211.
fcb7ef4 already removed the dead seed is not null guard from CollectionState.Exhausted and routed the message through the new RandomSource.ReplayHint(seed). That criterion is therefore already satisfied and out of scope for this issue. The message text quoted in Problem and the CollectionState.cs:246-254 / :249 line references are now stale for the same reason.

The core defect is untouched by #211: that change only decides which instruction to name (ambient Any.Reproducibly vs fixed Any.WithSeed), which is orthogonal to whether the element generator draws from the reported source at all. The over-promise for a foreign element generator still stands.

2. Two refinements to the acceptance criteria.

• Reformulate "the message for an As-derived generator is unchanged (faithful)" → faithful iff the propagated source is non-null. An As/Combine-derived generator is not unconditionally faithful: DerivedAny implements IHasRandomSource but carries a null source when derived from a foreign operand (Any.As(foreignGen, …)), and there its elements are exactly as unreproducible as a foreign generator's. So the discriminator must be AnyDerivation.SourceOf(element) is null, not an is IHasRandomSource type test — the type test would misclassify As(foreign, …) as faithful and keep over-promising. (This is also the bit the collection already computes as SourceOrNull.)

• Add a criterion for the twin site: ContainingAny. The same over-promise exists in DrawFresh when a ContainingAny(...) generator is foreign, independently of the main element generator. A complete fix qualifies the message at both exhaustion sites, keyed on the source of the specific generator that exhausted.

Planning the complete version (both sites). To stay correct when the collection's own source is a fixed Any.WithSeed(...) — reachable at the ContainingAny site — the "elements not reproducible from this seed" phrasing will be carried on RandomSource (a twin of ReplayHint) rather than hardcoded, keeping it consistent with the #211 design.

---

Generated by Claude Code