docs(prp): add forecast intelligence planning docs

[w7-mgfcode]

Summary

Lands the docs-only Forecast Intelligence roadmap — 4 INITIAL planning docs + 3 dependency-chained PRPs (35/36/37). No production code in this PR. Tracked by epic issue #295.

What's included

File	Lines	Purpose
PRPs/INITIAL/INITIAL-forecast-intelligence-index.md	217	Roadmap index linking A/B/C
PRPs/INITIAL/INITIAL-forecast-intelligence-A-feature-frame-v2.md	245	INITIAL for Feature Frame V2
PRPs/INITIAL/INITIAL-forecast-intelligence-B-model-zoo-backtesting.md	233	INITIAL for Model Zoo + Backtesting
PRPs/INITIAL/INITIAL-forecast-intelligence-C-interactive-ui.md	280	INITIAL for Interactive UI
PRPs/PRP-35-forecast-intelligence-A-feature-frame-v2.md	1,103	Backend PRP — V2 contract + sidecar
PRPs/PRP-36-forecast-intelligence-B-model-zoo-backtesting.md	1,356	Backend PRP — model zoo + bucket metrics
PRPs/PRP-37-forecast-intelligence-C-interactive-ui.md	1,221	Frontend PRP — operator workflow surfaces

Total: 7 files, +4,655 lines, all under PRPs/.

PRP shape

• PRP-35 Feature Frame V2 — V1 builders frozen; V2 ships as separate _v2 siblings in rows_v2.py; dispatch lives EXCLUSIVELY at the service layer (forecasting/backtesting/scenarios). Legacy bundles default feature_frame_version=1 via bundle.metadata.get(..., 1). New FeatureGroup enum + sidecar contract. Load-bearing leakage spec at test_leakage_v2.py.
• PRP-36 Model Zoo + Backtesting — hard-depends on PRP-35 merged. New target-only baselines (weighted_moving_average, seasonal_average, optional trend_regression_baseline + random_forest). Backtesting adds rmse + horizon_bucket_metrics (h_1_7 / h_8_14 / h_15_28 / h_29_plus). Registry _find_duplicate + new find_comparable_runs predicate gain feature_frame_version to match key. Ops adds StaleReason.FEATURE_FRAME_VERSION_MISMATCH. Confidence 7/10.
• PRP-37 Interactive UI — hard-depends on PRP-35 + PRP-36 merged. New components under frontend/src/components/forecast-intelligence/. Modifies 5 pages: /visualize/{forecast,backtest,planner,batch}.tsx, /explorer/{run-detail,run-compare}.tsx, /ops.tsx. Uses shadcn@4.7.0 pin + per-component @radix-ui/react-X imports (per .claude/rules/shadcn-ui.md). Partial-execution gates [gate:PRP-35] / [gate:PRP-36] / [gate:always]. Confidence 6.5/10.

Execution gating

Each downstream PRP starts with a contract-verification gate (Task 1):

• PRP-36 Task 1 = Contract Refresh → snapshots PRP-35's shipped contract to PRPs/ai_docs/prp-35-final-contract-snapshot.md, patches PRP-36 in-place if drift surfaces before any code is written
• PRP-37 Task 1 = Contract Probe → reads app/features/**/schemas.py, produces PRPs/ai_docs/contract-probe-report.md with PRESENT/ABSENT + file:line for every probed field; ABSENT field → task DEFERRED, never scaffolded with a placeholder

Excluded (intentionally separate)

The following local-tree items are not part of this PR — they belong to separate fix branches per the prior session HANDOFF:

• app/features/rag/models.py (modified — RAG embedding-dim work)
• app/shared/seeder/generators/facts.py (modified — seeder price-history fix)
• uv.lock (modified — incidental)
• alembic/versions/a2b3c4d5e6f7_rag_embedding_dim_2560_qwen3.py (untracked — pairs with the rag/models.py change)
• scripts/seed_historical_activity.py / seed_phase2_only.py / seed_registry_from_jobs.py (untracked — local demo-DB tooling)

Test plan

Docs-only — no code paths exercised. Validation:

• Files render as markdown (PRPs follow PRPs/templates/prp_base.md shape)
• No production code touched (git diff --cached --stat confirms all 7 paths under PRPs/)
• Commit message follows type(scope): description (#issue) with the umbrella issue #295
• CI green (ruff / mypy / pyright / pytest / migration-check — should pass trivially, no code under their scope)

Next steps (post-merge)

1. PRP-35 execution → branch feat/forecast-feature-frame-v2
2. PRP-36 execution (gated on PRP-35 merged) → branch feat/forecast-model-zoo-and-backtesting
3. PRP-37 execution (gated on PRP-35 merged; PRP-36 may be in-flight) → branch feat/forecast-ui-interactive-workflow

[sourcery-ai]

Sorry @w7-mgfcode, your pull request is larger than the review limit of 150000 diff characters

[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: f0f5fd9b-c460-426b-b9a2-b725021de409

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 docs/forecast-intelligence-prps

---

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.}