What mathematical model and units does Forecourt Prism use?

Forecourt Prism applies b_w = (1 − λ) × mean(V_{d ∈ weekday(w)}) + λ × mean(V_{all}); Backtest MAE = (1/K) Σ_{k=1..K} |V_k − b_{w(k)}|. Units: Fuel volume in litres/day (L/day); observation window in days (35 d); backtest window K = 21 days; forecast horizon H = 7 days.

What are the operational boundaries and limitations of Forecourt Prism?

Synthetic forecourt-volume forecasting with weekday-shrunk baselines and missing-data diagnostics. Not EV telemetry, live sales data or streaming analytics. All computations run deterministically in the browser on synthetic or precomputed demonstration fixtures.

TECHNICAL METHODOLOGY & VERIFICATION SPECIFICATION

Forecourt Prism: Mathematical Model, Units & Reproducible Benchmarks

Q: How can I reproduce the benchmark results in Forecourt Prism?

Open the interactive browser workspace at ./run/, select Case 1 to inspect the worked result (Case 1 (35 observed days, synthetic baseline 600 L/day, zero missing days, 7-day forecast horizon): The weekday-shrunk trailing estimator computes a Daily Baseline of 603.4 L/day with a 21-point held-out Backtest MAE of 62.2 L/day. For 2026-09-05, the point forecast is 557.7 L with heuristic band [448.3 L, 667.0 L].), and export the JSON or CSV evidence ledger.

Role & Domain Intent: Forecourt demand analytics (Synthetic volume forecasting; not EV telemetry or real-time ingestion.)

1. GOVERNING FORMULATION

Equations, Variables & Dimensional Rules

Synthetic forecourt-volume forecasting with weekday-shrunk baselines and missing-data diagnostics.

b_w = (1 − λ) × mean(V_{d ∈ weekday(w)}) + λ × mean(V_{all}); Backtest MAE = (1/K) Σ_{k=1..K} |V_k − b_{w(k)}|

Declared Units: Fuel volume in litres/day (L/day); observation window in days (35 d); backtest window K = 21 days; forecast horizon H = 7 days.

2. REPRODUCIBLE WORKED EXAMPLE

Step-by-Step Numerical Verification

Case 1 (35 observed days, synthetic baseline 600 L/day, zero missing days, 7-day forecast horizon): The weekday-shrunk trailing estimator computes a Daily Baseline of 603.4 L/day with a 21-point held-out Backtest MAE of 62.2 L/day. For 2026-09-05, the point forecast is 557.7 L with heuristic band [448.3 L, 667.0 L].

Verified benchmark cases from the Forecourt Prism browser edition
Case ID	Observed Days	Missing Pattern	Daily Baseline (L)	Backtest Points	Backtest MAE (L)	Horizon (d)
Case 1 (Complete Series)	35	None (0%)	603.4	21	62.2	7
Case 2 (Every 5th Day Missing)	28	20% periodic gaps	598.1	17	74.8	7
Case 3 (High Volume Site)	35	None (0%)	1,206.8	21	124.4	14

3. DATA PROVENANCE & OPERATIONAL BOUNDARIES

Validation Status, Scope Exclusions & Disclaimers

Data Status: Synthetic / precomputed browser demonstration fixtures generated deterministically from the project engine.
Operational Limitation: Not EV telemetry, live sales data or streaming analytics.
Client-Side Execution: All interactive views, filters, and JSON/CSV exports execute locally in the user's browser with zero external API calls, zero telemetry, and no server-side persistence.
Author & Principal Investigator: Mohammad Rezwan Khan, PhD (Aalborg University · BESS QA/QC & Power Systems Specialist).