User Story

As Maya, in order to choose which LLM drives form shaping and see how different models perform on that task, I want shaping variants to be selectable from Settings → Variants, backed by a quantitative benchmark.

Preconditions

• #58 (Story 10 variant picker) merged to main

Acceptance Criteria

• New evaluation kind shaping-commands scores a variant against scripted intents with expected Command[] outputs (precision/recall on command-kind + args)
• Variants registered: shaping/haiku, shaping/sonnet (promoted baseline), shaping/opus
• Shaping tab in Settings → Variants renders all three variants with descriptions and Learn more links
• Provenance recorded in shaping-log.json entries (extend existing entry schema with variantId + modelId)
• <VariantBadge task="shaping" ...> rendered wherever shaping output is shown (review/compare views)
• New catalog suite catalog/experiments/shaping-model-comparison/ with _suite.md, haiku.md, sonnet.md, opus.md, each containing metrics, approach, and findings
• catalog/experiments/_roadmap.md updated: row marked shipped, one-line finding added
• Existing catalog/experiments/shaping-architecture/ entries remain intact (architecture story is separate from model comparison)

Success Metrics

• Meaningful recall/precision separation between variants across ~6 scripted intents (same set as the shaping-architecture qualitative comparison)
• Picker UI renders cleanly on every screen that renders shaping output

Notes

• Seeded intents already exist in catalog/experiments/shaping-architecture/_suite.md — reuse them as the benchmark corpus
• src/services/forms/shaping/registry.ts already exists with a Sonnet-only entry — extend it, don’t replace
• The picker’s filling/mapping tabs stay empty until their respective stories land

Definition of Done

• Acceptance criteria met
• Tests pass (bun run check)
• Type checking passes
• Threat model updated if security-relevant
• CI pipeline green
• Deployed and demoable