PREVIEW · static UI mock · screen 16 / 32 · workflow failures · cluster taxonomy · all screens
‹ Workflows

Demand prediction

● Activev23

Owner: Maria Chen · Supply Chain VP · 3 active clusters · 18 resolved this quarter · open agent UI →

Active clusters
3
1 high · 1 medium · 1 low
Resolved · 30d
7
all gate-passed
Avg time-to-fix
2d 14h
cluster surfaced → deployed
Eval cases promoted
38
from clusters · automatic
Severity Status Source Showing 8 clusters
Active · 3
Seasonal anomaly missed in winter footwear demand
cluster #ce-2284
High14 traces· First seen 2026-04-21· Region: Pacific NW· $340k exposure
Proposal ready
forecast_demand(sku=BOOT-W42, region=PNW, week=2026-W45) → 312 units
actual: 1,847 units · markdown_alert fired W49 (4 weeks late)
Review proposal → Gate: 12 / 12 passing
Promotional uplift underweighted on bundled SKUs
cluster #ce-2278
Medium9 traces· First seen 2026-04-25·$128k exposure
Generating proposal…
forecast for 4-pack bundle ignored 30% promo lift on parent SKU
3 SKU bundles affected · revenue exposure $128k
Supplier delay impact missed for 12-week lead-time SKUs
cluster #ce-2261
Low6 traces· First seen 2026-04-28·monitor signal: low_confidence
Awaiting more signal
forecast did not factor 14-day supplier delay on Northeast restocks · 2 traces show forecast over-estimated post-delay availability
Recently resolved · 4
Supplier price-build missed in Q1 forecasting
cluster #ce-2218
Resolved11 traces· Resolved 9d ago by Maria Chen· Lift: +1.4 pts
Deployed v3
Regional variance underweighted in Southeast
cluster #ce-2196
Resolved8 traces· Resolved 17d ago by Maria Chen· Lift: +1.2 pts
Deployed v2
Promotional uplift handler — initial version
cluster #ce-2174
Resolved17 traces· Resolved 24d ago by Maria Chen· Lift: +1.7 pts
Deployed v2
Aggressive promo discount auto-apply (rejected)
cluster #ce-2241
Reopened4 traces· Rejected 3d ago by Maria Chen· Comment promoted to eval case #128
Reviewer rejected
Failure clusters are semantic groupings of trace failures with the same root cause. Each cluster automatically promotes a representative trace to a new eval case in the regression suite — so once a class of failure is fixed, it stays fixed.