dbt: verify polymorphic_value_invariant covers all four value columns + value_kind (invariant 8)

[countercheck]

Source: codebase audit, 2026-06-11 (audit finding F6, Tier 2)
Severity: Low-Medium
Category: Invariant surface

Problem

CLAUDE.md invariant 8: "Exactly one of `{option_key, value_numeric, value_text, value_date}` per fact row. dbt test enforces; don't skip."

The existing `polymorphic_value_invariant` test enforces this at mart time. After M5.5 introduced `value_kind` (option / text / numeric / date) as the discriminator computed in `int_survey_questions`, the test may or may not be checking each value column against its `value_kind` — needs verification.

Suggested approach

1. Read dbt/tests/singular/polymorphic_value_invariant.sql and confirm:
   • Counts non-nulls across all four columns (option_key, value_numeric, value_text, value_date) — must equal 1 per row.
   • Cross-checks against `value_kind` if present — e.g. `value_kind='numeric'` ⇒ value_numeric NOT NULL AND others NULL.
2. If the current test only counts non-nulls (correct for invariant 8 but doesn't catch a value_kind/value-column mismatch), either extend it or add a sibling test `value_kind_value_column_alignment` that asserts the discriminator → column mapping.
3. If the test already does both — close as "no change needed" with a note in the PR description.

Verification

```bash
cd dbt && dbt build
```

Should still PASS, with the extended (or unchanged) test catching synthetic violations.

Why this matters

A bug in `int_response_selections` SQL or a future API code path could populate two value slots; today only the next `dbt build` catches it. The check is cheap and the failure mode is silent without it.