feat(ui): add forecast champion selector foundation

[w7-mgfcode]

Closes #356 · Slice A of 3 (A → B → C) of the Forecast Champion Selector.
PRP: PRPs/forecast-champion-selector-slice-a-selection-capability.md.

First interactive Champion Selector surface — configuration + availability
triage. It deliberately stops before running the comparison (that is Slice B)
and before any training/prediction/promotion (Slice C).

Backend

• GET /model-selection/models → ModelCatalogResponse — a backend-owned
  capability catalog (one entry per forecasting ModelType, 11 total), declared
  before GET /{selection_id} so Starlette matches the literal path (not a
  selection_id capture).
• New pure module capabilities.py (build_model_catalog(), no DB/IO); each
  entry carries model_type, label, family (derived from
  forecasting.feature_metadata.model_family_for), feature_aware,
  requires_extra, default_params, supports_auto_predict, description,
  plus a default_candidate_model_types list.
  • feature_aware == the set whose forecasters set requires_features=True
    (random_forest/regression/lightgbm/xgboost/prophet_like) — i.e. exactly the
    set ForecastingService.predict() rejects.
  • requires_extra == {lightgbm, xgboost}; supports_auto_predict == not feature_aware.
  • default_params pinned from the live ModelConfig defaults.
• ModelSelectionService.get_model_catalog() thin delegate; response schemas
  CandidateModelInfo + ModelCatalogResponse added to schemas.py.
• No migration, no new mutation surface, no agent tool.

Frontend

• /visualize/champion page shell (pages/visualize/champion.tsx) — lazy
  route + Visualize nav entry. Store → Product → Time period → Horizon → Models
  → Backtest settings, plus a live availability panel.
• components/champion-selector/: searchable-entity-select (Popover +
  Input + filtered list — no new dependency), availability-panel
  (ready/limited/unusable + metrics + recommended split; not-enough-data empty
  state), backtest-settings-form (simple/advanced split, SplitConfig bounds,
  WAPE-default ranking metric with the locked tie-break + bias copy),
  candidate-model-picker (catalog-fed, family-grouped, opt-in-extra badges,
  cap 10), plus copy.ts / split-config.ts / run-request.ts helpers.
• hooks/use-model-selection.ts: useModelCatalog + usePairAvailability
  (availability gated on a valid pair).
• types/api.ts: full Model-Selection workflow contract added (Slice A
  consumes catalog/availability/SplitConfig only; the run/ranking/chart/forecast
  types are declared-for-later so Slices B/C inherit, not redefine).

Scope boundaries

• "Run comparison" CTA is intentionally disabled with explanatory copy —
  Slice B wires the POST mutation. The page assembles a typed
  ModelSelectionRunRequest (with auto_train_winner:false / auto_predict:false
  pinned, both no-ops in the async path) but never sends it.
• No Slice B/C behavior: no run/progress/cancel/results, no ranking charts,
  no winner explanation/business summary, no train/predict/override/promotion.
  Exactly one new backend route; no calls to any /{selection_id}* endpoint.

Validation

• ruff check . + ruff format --check ✅
• pytest -m "not integration" ✅ 1753 passed, 12 skipped (incl. 8 new
  test_capabilities + 2 new /models route/ordering tests)
• pnpm tsc --noEmit ✅ · pnpm lint ✅ (exit 0; only a pre-existing TanStack
  Table warning remains) · pnpm test --run ✅ 282 passed (28 new Champion tests)
• pyright app/features/model_selection ✅ 0 errors

Known pre-existing (NOT introduced here)

• mypy app/ / pyright app/ report lightgbm/xgboost optional-dep
  import-resolution errors in forecasting/ + registry/ (CI installs the
  extras). None are in model_selection; documented in the PRP's Level-4 notes.

[sourcery-ai]

Sorry @w7-mgfcode, you have reached your weekly rate limit of 500000 diff characters.

Please try again later or upgrade to continue using Sourcery

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 1c5709f3-8095-47bf-be96-578218909aa1

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/champion-selector-slice-a

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}