Accuracy & back-test
Two honest views of the engine's own accuracy. Reconstruction proves the pricing mechanics reproduce six months of invoices to the cent (in-sample). The expanding-window forecast is the out-of-sample test — each month's rate index is calibrated only from prior months, then applied to that month's actual activity (no circularity). The head-to-head against Denise's trailing average lives on the vs. Denise tab.
How to read this — bias matters more than per-month error
What the forecast does and doesn't claim.
Each carrier applies a volatile monthly rate index (±30%), but monthly spend is stabilized (rate and volume move opposite ways), so per-month percentage error on a forecast is inherently noisy — we don't chase it. What an accrual must not do is miss in the same direction every period: bias compounds in the P&L; per-month noise averages out. The engine is near-unbiased (-0.2%), decisively prices the one structurally predictable error — the Heartland quarter-to-date reset — and prices the shipments that actually moved, so April reflects April's mix with shipment-level backup and a confidence band instead of a bare point estimate. The reconstruction below proves the mechanics reproduce every historical invoice to within ±$0.
Forecast vs actual — by carrier-month
Headline = expanding-window forecast: each month's rate index is calibrated only from prior months, then applied to that month's actual shipment activity. The first month is cold-start (printed card) and excluded from headline metrics. Reconstruction below proves the pricing mechanics reproduce every invoice to the cent.
| Month | Carrier | Engine estimate | Actual | Engine error |
|---|---|---|---|---|
| March 2026 | Coastal Express | $25,593 | $26,300 | -2.7% |
| March 2026 | Heartland Freight | $37,366 | $33,200 | +12.5% |
| March 2026 | Peak Logistics | $40,346 | $34,217 | +17.9% |
| February 2026 | Coastal Express | $24,557 | $24,900 | -1.4% |
| February 2026 | Heartland Freight | $35,279 | $34,600 | +2.0% |
| February 2026 | Peak Logistics | $25,128 | $34,606 | -27.4% |
| January 2026 | Coastal Express | $18,877 | $25,400 | -25.7% |
| January 2026 | Heartland Freight | $32,185 | $38,200 | -15.7% |
| January 2026 | Peak Logistics | $32,843 | $34,396 | -4.5% |
| December 2025 | Coastal Express | $26,875 | $21,200 | +26.8% |
| December 2025 | Heartland Freight | $25,475 | $28,500 | -10.6% |
| December 2025 | Peak Logistics | $27,376 | $27,844 | -1.7% |
| November 2025 | Coastal Express | $19,491 | $25,800 | -24.5% |
| November 2025 | Heartland Freight | $32,727 | $31,900 | +2.6% |
| November 2025 | Peak Logistics | $46,254 | $31,045 | +49.0% |
| October 2025 | Coastal Express | $27,958 | $24,100 | +16.0% |
| October 2025 | Heartland Freight | $31,741 | $32,800 | -3.2% |
| October 2025 | Peak Logistics | $34,872 | $31,738 | +9.9% |
Listed newest first. The earliest month (cold-start — no prior invoices to calibrate from) is excluded from the headline metrics.
By carrier — engine MAPE & bias (forecast)
| Carrier | Engine MAPE | Engine bias | Months |
|---|---|---|---|
| Peak Logistics | 20.1% | +6.7% | 5 |
| Heartland Freight | 8.7% | -1.8% | 5 |
| Coastal Express | 16.2% | -5.5% | 5 |
| Overall | 15.0% | -0.2% | 15 |
Peak (20.1%): the most volatile index in the data (0.6113–1.0705) plus an incomplete printed mileage table — the noisiest per-month estimates, which is exactly why Peak's printed card can't be trusted at face value. The misses are flagged (MAPE alarm) and carried in the confidence band, not hidden.
Heartland (8.7%): the structurally predictable carrier — the quarter-to-date tier reset is contractual, and pricing it exactly is where a bottoms-up engine adds the most value.
Coastal (16.2%): monthly spend is the most stabilized in the data, so per-month error is dominated by noise; the engine adds shipment-level backup that ties to the JE, exception flags, and a band that says how much the accrual could move.