# elixir-ecto-schema-changeset **Rank**: #5 of 22 **Tier**: A (high-value, good evidence) **Taxonomy path**: `data` / `ecto-orm` / `elixir` **Status**: ✅ Validated by research — `:float` for money is a clinching evidence item ## Specialization From a domain model description, generates Ecto schemas, migrations, and changesets with the right field types, validations, DB-level constraints, and associations (`has_many`, `belongs_to`, `many_to_many`, `embeds_one`, `embeds_many`). Enforces functional Elixir patterns over ActiveRecord/Django carryover. ## Why LLMs struggle Claude treats Ecto schemas like ActiveRecord models — methods on structs, mass assignment without thinking, ORM-style associations. Specific failure modes: - **`:float` for money fields** — silent data corruption; named iron law in `oliver-kriska/claude-elixir-phoenix` - Forgetting `cast/3` + `validate_required/2` on changesets - Mixing up `has_many` vs `belongs_to` direction - Forgetting that `unique_constraint/2` requires a matching DB-level unique index in the migration - Conflating migration code (`alter_table`) and schema code (`field`) - Polymorphic association attempts (Ecto doesn't natively support them) ## Decomposition ### Foundation - **F: `schema-organization`** — Flat schemas vs nested embeds vs cross-context splits. Variants: single-source-of-truth (one schema per concept), context-scoped (separate schemas per bounded context), hybrid. Locks in how every field/association/changeset is shaped. ### Capabilities 1. **C: `field-types-and-decimal`** ⭐ — Picking the right `field` types; **`:decimal` for money, NEVER `:float`**; `:utc_datetime` vs `:naive_datetime` 2. **C: `associations`** — `has_many`, `belongs_to`, `many_to_many`, `has_one`, `through:` associations 3. **C: `embedded-schemas`** — `embeds_one`, `embeds_many`, when to embed vs associate 4. **C: `validations-basic`** — `validate_required`, `validate_length`, `validate_format`, `validate_inclusion`, `validate_number` 5. **C: `validations-custom`** — `validate_change/3`, `validate_confirmation/2`, cross-field validations 6. **C: `unique-constraints-and-indexes`** — DB-level unique index + changeset `unique_constraint/2` (must match!) 7. **C: `cast-and-allowed-fields`** — `cast/3`, `cast_assoc/3`, `cast_embed/3`; mass assignment safety 8. **C: `migrations`** — `create_table`, `alter_table`, `add`, `remove`, indexes, foreign keys 9. **C: `polymorphic-associations`** — The tradeoffs; Ecto doesn't natively support them; common workarounds 10. **C: `soft-deletes-and-timestamps`** — `deleted_at` patterns, `timestamps()` macro, `updated_at` triggers 11. **C: `multi-tenant-schemas`** — `prefix:` option, schema-per-tenant patterns ### Total dimensions **12** = 1 foundation + 11 capabilities ## Evaluation criteria sketch - **Money field test**: build a Product schema with a price field; score.py checks the field type is `:decimal`, not `:float` - **Association test**: build a Post schema with many tags via `many_to_many`; verify join table migration + association declarations match - **Unique constraint test**: build a User schema with unique email; verify both `unique_index` in migration AND `unique_constraint(:email)` in changeset - **Mass assignment test**: build a User schema where role is set by admins only; verify `cast/3` excludes role from public changesets - **Embed vs assoc test**: build an Order with line items; ask the skill to justify embeds_many vs has_many for this case ## Evidence - [Research report Part 1 #5](../../docs/research/elixir-llm-pain-points.md#5-float-for-money-in-ecto) - [oliver-kriska/claude-elixir-phoenix](https://github.com/oliver-kriska/claude-elixir-phoenix) — `:float` iron law ## Notes - The `:decimal`-not-`:float` rule is the single highest-confidence safety fix in this family. Bake it into the foundation's voice. - Closely linked to `elixir-ecto-query-writer` — they share schema definitions. Both families' challenges should reference the same fixture schemas to keep evaluations comparable. - Polymorphic associations capability is interesting because the "correct answer" is "don't" — that's a teachable Elixir vs Rails distinction. ## family.json ```json { "slug": "elixir-ecto-schema-changeset", "name": "Elixir Ecto Schema & Changeset", "language": "elixir", "tier": "A", "curve": "binary", "spec_version": "2.1", "taxonomy": { "domain": "data", "focus": "ecto-orm", "language": "elixir" }, "foundation_dimension": "schema-organization", "capability_dimensions": [ "field-types-and-decimal", "associations", "embedded-schemas", "validations-basic", "validations-custom", "unique-constraints-and-indexes", "cast-and-allowed-fields", "migrations", "polymorphic-associations", "soft-deletes-and-timestamps", "multi-tenant-schemas" ], "challenges": { "total": 100, "by_tier": { "easy": 35, "medium": 35, "hard": 22, "legendary": 8 }, "held_out_ids": [ "elixir-ecto-schema-changeset-easy-07", "elixir-ecto-schema-changeset-easy-13", "elixir-ecto-schema-changeset-easy-18", "elixir-ecto-schema-changeset-easy-20", "elixir-ecto-schema-changeset-easy-25", "elixir-ecto-schema-changeset-easy-29", "elixir-ecto-schema-changeset-easy-34", "elixir-ecto-schema-changeset-medium-02", "elixir-ecto-schema-changeset-medium-09", "elixir-ecto-schema-changeset-medium-15", "elixir-ecto-schema-changeset-medium-22", "elixir-ecto-schema-changeset-medium-29", "elixir-ecto-schema-changeset-medium-31", "elixir-ecto-schema-changeset-medium-34", "elixir-ecto-schema-changeset-hard-01", "elixir-ecto-schema-changeset-hard-09", "elixir-ecto-schema-changeset-hard-17", "elixir-ecto-schema-changeset-hard-22", "elixir-ecto-schema-changeset-legendary-03", "elixir-ecto-schema-changeset-legendary-06" ] }, "evaluation": { "score_script": "evaluation/score.py", "criteria_file": "evaluation/criteria.json", "environment_file": "evaluation/environment.yml" }, "seeded_at": "2026-04-11", "tier_methodology": "heuristic", "tier_methodology_note": "Tiers assigned by drafting agent judgment per SEEDING-PLAN.md item 4. Empirical Haiku+Sonnet calibration is a future workstream." } ```