You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Scope. Slice B of the Forecast Intelligence roadmap. Land a disciplined model zoo with fair, leakage-safe backtesting comparison on top of the PRP-35 Feature Frame V2 contract. PRP at PRPs/PRP-36-forecast-intelligence-B-model-zoo-backtesting.md; final contract snapshot at PRPs/ai_docs/prp-35-final-contract-snapshot.md.
Deliverables (18 tasks, V1 only per #299 deferral).
New target-only baselines: weighted_moving_average, seasonal_average, trend_regression_baseline (Ridge).
New feature-aware optional model: random_forest (sklearn, gated by forecast_enable_random_forest).
No new Alembic migration; all new state rides in existing JSONB columns.
No agent tool added; agent_require_approval unchanged.
Validation gates. All five must be green: ruff check, ruff format --check, mypy --strict, pyright --strict, pytest -m "not integration" (+ pytest -m integration for slice-touching tests).
Confidence. 7/10 per the PRP's confidence section. Library claims (sklearn HGBR/RF, np.average) verified at PRP-create time.
Epic: #295
Depends on: PRP-35 (merged in #300)
Blocks: PRP-37
Defers V2 backtesting: #299
PRP-36 execution issue under epic #295.
Scope. Slice B of the Forecast Intelligence roadmap. Land a disciplined model zoo with fair, leakage-safe backtesting comparison on top of the PRP-35 Feature Frame V2 contract. PRP at
PRPs/PRP-36-forecast-intelligence-B-model-zoo-backtesting.md; final contract snapshot atPRPs/ai_docs/prp-35-final-contract-snapshot.md.Deliverables (18 tasks, V1 only per #299 deferral).
weighted_moving_average,seasonal_average,trend_regression_baseline(Ridge).random_forest(sklearn, gated byforecast_enable_random_forest).h_1_7,h_8_14,h_15_28,h_29_plus), aggregate across folds._find_duplicatedistinguishes V1/V2;find_comparable_runs(grain, overlapping window, same V).RunResponse/RunDetailResponse: Optionalfeature_frame_version+feature_groupsfromruntime_info.FEATURE_FRAME_VERSION_MISMATCHstale-reason; comparable-run predicate honours V.examples/forecasting/model_zoo_compare.py.Explicitly out of scope.
agent_require_approvalunchanged.Validation gates. All five must be green:
ruff check,ruff format --check,mypy --strict,pyright --strict,pytest -m "not integration"(+pytest -m integrationfor slice-touching tests).Confidence. 7/10 per the PRP's confidence section. Library claims (sklearn HGBR/RF, np.average) verified at PRP-create time.
Epic: #295
Depends on: PRP-35 (merged in #300)
Blocks: PRP-37
Defers V2 backtesting: #299