AI03
AI Explainability
What the model predicted, why, and how it has done.
Schema v01 has no model registry, no prediction log, no feature attribution and no forecast. Its two model-shaped tables are extraction_log (field, proposed, confidence, outcome, model_version, prompt_version) and remnant_match (score, explanation, decision), and both hold no rows.
AI Features
—
no model or feature registry; extraction_log carries a model_version string per row and nothing names the model itself
Predictions Logged
askingrows
extraction_log: one row per field a model proposed, with its confidence and what the person did with it
Matches Explained
askingrows
remnant_match: score, explanation and decision for a lot against an enquiry line
Model Accuracy
—
accuracy would be confirmed over proposed on extraction_log, and there are no rows to divide
Avg Confidence
—
extraction_log.confidence has no rows
Last Prediction
—
no extraction_log row has been written
Prediction Log
DateFeatureSubjectPredictedConfidenceOutcomeModel
extraction_log holds 0 rows: no RFQ has been read by a model (rfq holds 0 rows) and no mill certificate has been parsed (certificate_file holds 0 rows). The workbook carried neither.
What the database does hold
extraction_log rowsasking
remnant_match rowsasking
rfq rowsasking
certificate_file rowsasking
Model Performance
accuracy over time
Nothing to show yet
No prediction rows and no dated outcome, so there is no series.
Fairness & Bias
Nothing to show yet
No prediction rows and no protected attribute anywhere in schema v01.
Feature Insights
global feature importance
Nothing to show yet
No feature-attribution table; extraction_log keeps a confidence, not the factors behind it.
Key Factors
contribution per factor for one prediction
Nothing to show yet
No prediction to explain, and no factor column.
Explanation
the natural-language summary
Nothing to show yet
remnant_match.explanation is the one explanation column in the schema, and it holds no rows.
Data Used
time range, sources, records
Nothing to show yet
No provenance table for a prediction; rfq_line.extraction would carry it per line and rfq_line holds 0 rows.
Related Predictions
Nothing to show yet
No prediction rows to relate.
Prediction vs Actual
Nothing to show yet
No forecast in schema v01 and no predicted series to compare against the ledger.