Return to presentation: page 1 · page 25 · page 32 · page 35 · page 40
Numerical Verification
These summaries connect the presentation to the underlying records and calculations. Choose Predictions, State Characterization, or QC Exclusions.
Predictions
The nine historical outcome-redacted predictions below follow the chart order. Each is the mean of ten qualifying Table executions. Prices are in dollars; the presentation chart displays thousands of dollars.
| Property | Ensemble mean prediction (F) | Realized sale price (Y) | Absolute percentage error |
|---|---|---|---|
| P1A | $639,000 | $650,000 | 1.69% |
| P2A | $673,000 | $670,000 | 0.45% |
| P3A | $781,000 | $790,000 | 1.14% |
| P4A | $748,750 | $749,900 | 0.15% |
| P5A | $641,700 | $635,000 | 1.06% |
| P6A | $647,500 | $650,000 | 0.38% |
| P7A | $607,900 | $580,000 | 4.81% |
| P8A | $725,500 | $725,000 | 0.07% |
| P9A | $607,500 | $600,000 | 1.25% |
The genuinely prospective case is shown separately and is not included in historical OPVAL aggregates:
| Property | Ensemble mean prediction (F) | Realized sale price (Y) | Absolute percentage error |
|---|---|---|---|
| P1P — genuinely prospective | $714,250 | $720,000 | 0.80% |
Verify these values: Open or download the Qualified Records workbook. Select S0 Qualified Verification, then find the OPVAL Table property rows and the separate PROSP row. Mean F corresponds to the prediction and Actual Y to the realized outcome. S0 Qualified Executions contains the constituent records.
These are the results referenced by presentation pages 1 and 35. They do not establish a Table-versus-No-Table accuracy or precision comparison.
State Characterization
The same configured DF produced fresh held-out LOOCV estimates to characterize each state. The summaries below concern Table executions and use ten qualifying runs per held-out property.
| Metric | State 0 · LOOCV0 Table | State 1 · LOOCV1 Table |
|---|---|---|
| Held-out properties | 9 | 9 |
| Qualifying executions | 90 | 90 |
| Mean absolute percentage error | 3.23% | 3.72% |
| Median absolute percentage error | 2.70% | 2.74% |
| Maximum absolute percentage error | 6.94% | 7.05% |
| Held-out estimates within 5% of Y | 6/9 | 5/9 |
| Mean coefficient of variation | 0.93% | 0.69% |
Accuracy and precision change differently: State 1 has a higher mean absolute percentage error and a lower mean coefficient of variation. Standards A and B, and their implications for qualification, are developed in the presentation rather than redefined here.
Verify these values: Open or download the Qualified Records workbook. Select S0 Qualified Verification and locate the LOOCV0 Table and LOOCV1 Table summaries and property-level detail. The state summaries use held-out estimates, not carried-forward operational predictions.
These results support presentation pages 32 and 40. The laboratory examined these two states, not every state encountered in operational evaluation. LOOCV and evidence-state characterization explains the method and its role in continuing qualification.
QC Exclusions
The OPVAL comparison concerns detected execution-validity failures, distinct from prediction accuracy. Total attempts include both qualifying and excluded executions.
| OPVAL population | Qualifying executions | Excluded executions | Total attempts | Detected exclusion rate |
|---|---|---|---|---|
| No Table | 90 | 32 | 122 | 26.2% |
| Table | 90 | 2 | 92 | 2.2% |
Verify these counts: Open or download the Registry workbook. Select S0 Registry Verification for the adjacent OPVAL No Table and Table rows, or S0 Registry Executions for record-level dispositions. Excluded records remain inspectable while staying outside qualifying ensemble calculations.
This is the comparison on presentation page 25. LOOCV0 rates have a different-process limitation: table-assisted ER review was used for Table records but was not retrospectively applied to the original No Table records. Populations without a No Table counterpart present a separate absence of a direct comparison.
The P4A Supporting Artifacts provides an inspectable exclusion and replacement sequence. Quality Control and Execution Integrity explains the validity requirements and detection limitations.
Working review note: summaries are generated from the current workbooks. Dollar amounts are shown as whole dollars and percentages are rounded for reading; source calculations retain greater precision. The Qualified workbook opens on its verification sheet: Table-only summary first, then OPVAL, separate PROSP, LOOCV0, and LOOCV1 property detail; No Table detail and summaries are retained below. Opening a workbook may require downloading it and using Excel; links do not guarantee a particular selected row.
Return to presentation: page 1 · page 25 · page 32 · page 35 · page 40