The synthetic forecast-override audit
The final number is evaluable only when the baseline, intervention, information cutoff, later actual, loss function, and disposition remain visible.
| ID | Cutoff and horizon | Baseline | Override and reason | Final | Later actual | Loss disposition |
|---|---|---|---|---|---|---|
| F-01 | 2026-09-01, 30 days | 100 | +5, signed customer commitment | 105 | 108 | Improved: 8 to 3 |
| F-02 | 2026-09-01, 30 days | 80 | -15, capacity constraint | 65 | 60 | Improved: 20 to 5 |
| F-03 | 2026-09-01, 30 days | 120 | +25, unverified optimism | 145 | 115 | Worsened: 5 to 30 |
| F-04 | 2026-09-01, 30 days | 50 | 0, no intervention | 50 | 44 | Unchanged: 6 to 6 |
| F-05 | 2026-09-01, 90 days | 200 | +10, horizon changed after cutoff | 210 | 205 | Hold: horizon not comparable |
| F-06 | 2026-09-01, 30 days | 90 | -10, reason recorded after actual | 80 | 82 | Hold: information timing invalid |
Swipe or scroll horizontally if the table is wider than your screen.
Reference & Evidence
Source: Author's synthetic override audit grounded in Fildes, Goodwin, Lawrence and Nikolopoulos (2009) and Fildes, Goodwin and De Baets (2025). Values, reasons, actuals, losses, and dispositions are illustrative.
Each line is a claim from the register this journal publishes against, resolved from the register at build time.
- A The dataset is counted in the paper: "we collected data on more than 60 000 forecasts and outcomes from four supply chain companies", 68,984 complete triples in Table 1, and "in three of the companies on average judgmental adjustments increased accuracy" Fildes, Goodwin, Lawrence & Nikolopoulos (2009), abstract and Table 1 ·
FGLN09-C1 - A Upward adjustments were more likely to damage accuracy than downward adjustments Fildes, Goodwin, Lawrence & Nikolopoulos (2009), section 4.1 ·
FGLN09-C3 - A A "combined dataset of 147,131 forecasts" and actuals, "obtained from 10 organizations with 22 business units" Fildes, Goodwin & De Baets. (2025) ·
FGBD25-C1 - A "Six datasets were used in this study", and the unit is a triple: "weekly or monthly statistical systems forecasts; the final forecasts, which may be the same as the system forecast or adjusted; and the corresponding actual outcomes for each SKU", with "we consider only one-step-ahead forecasts" Fildes, Goodwin & De Baets. (2025) ·
FGBD25-C2 - B A forecast override is a recorded change from a baseline or system forecast to a final forecast at a declared cutoff Author framework grounded in FGLN09-C1 and FGBD25-C1 ·
V05-OWN-C1 - B The override is a data intervention that must preserve baseline, final value, signed delta, reason, information available, and later actual Author operating framework ·
V05-OWN-C2 - B Override magnitude is final forecast minus baseline forecast under the declared unit and scale Author formula set ·
V05-OWN-C3 - B The six-row override audit is synthetic and contains no company forecast, seller judgment, customer demand, or observed accuracy result Author synthetic object ·
V05-OWN-C7
Grades: A, verified against the printed page of the primary source · B, primary source, text layer only · C, authoritative secondary · D, reported.
Related exhibits
-
Directional impact of judgmental forecast adjustments
From the essay The number you call
-
What does forecast override mean?
From the essay What is a forecast override? Judgment is an intervention in the data
-
The AI forecast control map
From the essay What is AI demand forecasting? A forecast is a system, not an oracle