fix(forecast): model_exogenous re-forecast is inert to price assumptions (0.0 delta)

[w7-mgfcode]

Summary

The model_exogenous scenario re-forecast path is now reachable from the What-If Planner (#229, PR #236), but the re-forecast appears inert to price assumptions — it returns a 0.0 units/revenue delta regardless of the price change magnitude.

Observed behaviour

Verified during Level-3 validation of PR #236 against a freshly seeded DB (store 332 / product 456, regression train job run_id=dc6dc4aaea18, n_observations=356):

POST /scenarios/simulate
{"run_id":"dc6dc4aaea18","horizon":14,
 "assumptions":{"price":{"change_pct":-0.15,"start_date":"2026-05-16","end_date":"2026-05-22"}}}
→ method="model_exogenous"  units_delta=0.0  revenue_delta=0.0

# same baseline, stronger change:
{"...":"...","price":{"change_pct":-0.40,"start_date":"2026-05-16","end_date":"2026-05-29"}}
→ method="model_exogenous"  units_delta=0.0  revenue_delta=0.0

Every per-day point comes back with delta=0.0, applied_factor=1.0, baseline == scenario. The heuristic path on a naive baseline reacts correctly to the same assumption (units_delta≈203.5).

Expected behaviour

A genuine model-driven re-forecast through a regression baseline should produce a non-zero demand delta when a price assumption is applied — otherwise method="model_exogenous" is indistinguishable from a no-op.

Likely root cause (to confirm)

Pre-existing PRP-27 code, not introduced by PR #236. Two hypotheses:

1. The price assumption's change_pct is not flowing into the future feature frame — the exogenous price column built by app/features/scenarios/feature_frame.py is not adjusted before _simulate_model_exogenous re-runs the model.
2. The seeded regression model genuinely learned ~zero price elasticity (the model leans entirely on autoregressive lag features), so a future-window price change has no effect.

Hypothesis 1 should be checked first: confirm whether the price-adjusted column reaches X_future, and whether the model has any non-lag price feature to respond to.

Scope / references

• Surfaced by: PR fix(jobs,ui): reach model_exogenous + block empty assumption dates in the planner #236 (fix(ui): model_exogenous scenario path is unreachable from the What-If Planner #229 + fix(ui): What-If Planner submits empty assumption dates → backend 422 #228), PRPs/planner-model-exogenous-and-empty-dates.md
• Code: app/features/scenarios/feature_frame.py, app/features/scenarios/service.py (_simulate_model_exogenous)
• Out of scope for fix(ui): model_exogenous scenario path is unreachable from the What-If Planner #229/fix(ui): What-If Planner submits empty assumption dates → backend 422 #228 — those only made the path reachable.