PREDICT · W45 · 7 ENGINES

Real signals. Real math. Zero ML hype.

Predictions on Akekart come from published statistics — rolling medians · exponential smoothing · velocity gradients · linear depletion. No black-box ML · no fake confidence scores · every forecast cites its method.

7 W45 engines Published methods Real confidence bands Audit-logged
78%
Avg Confidence · 7d
0
Predictions Today
78.0%
Avg Confidence
0
Refills Due 7d
0
Inventory Alerts

4 prediction tracks Each track surfaces a W45 engine · pure stats · no hidden ML

Refill Predictions
Per-user per-SKU next-purchase prediction · from real interval history
RefillPredictionEngine
Demand Forecast
7/30/90-day demand by category · seller capacity planning
DemandForecastEngine
Trend Momentum
Catalog-wide trending products · velocity-weighted, anti-recency-bias
TrendMomentumEngine
Inventory Depletion
Per-seller stock-out projection · pre-warns 7-14 days before run-out
InventoryDepletionPredictor
Method transparency · how each prediction is computed
Refill: rolling-interval median
last 6 purchases → median gap days → next = last_purchase + median. Confidence drops as gap variance grows.
Demand: exponential smoothing
90-day weighted moving avg with seasonal multiplier (Diwali/Eid/wedding-season known months). No ML black box.
Trend: velocity gradient
7-day order-velocity slope vs 30-day baseline. Filters one-day spikes. Outputs momentum score 0-100.
Inventory: linear depletion + buffer
current_stock / 7-day-velocity = days-of-stock. Alert at 14d-buffer threshold.
Sample forecasts · live from `predictive_forecasts` (last 3)
Forecast engine warming up — sample predictions appear here as data flows in.

What makes this honest Pure stats · confidence bands · no hidden ML · audit log

Pure stats
Median · smoothing · gradient · linear · documented math
Confidence bands
Every forecast ships with ± band · variance honestly shown
No hidden ML
No black-box LLMs · no "AI predicted" without method shown
Audit log
Every forecast → predictive_audit_log · replayable input + method