Private 10,000-vehicle development sample
The next inspection, in sample context
The research question is intentionally narrow: predict whether a returning vehicle's next inspection episode has a first-attempt non-pass. Non-pass combines fail, reject, and abort.
Data through
Unavailable
Waiting for manifest
Dashboard data is unavailable
The approved aggregate files could not be validated. No estimates are being shown.
Sample association over time
Observed first-attempt non-pass trend
Non-pass (fail + reject + abort)
Partial periods and source changes are called out when present.
Feed availability
County coverage
Availability only—this is not a county outcome ranking. Teal feeds are represented in the sample; gray counties are unavailable, not zero.
Sample association by age
Observed non-pass rate by vehicle age
Sample aggregates only, without confidence intervals. The pattern is descriptive and not a population estimate or causal effect.
Private 10,000-vehicle development sample
Sample cohort explorer
Explore supported prior make-and-model cohorts and their observed next-episode first-attempt non-pass rates. These are sample associations—not reliability grades or individual predictions.
Explorer unavailable
Approved scorecards could not be validated.
Descriptive sample comparison
Observed cohort associations
Ordering is a viewing aid within this sample, not a reliability ranking or population comparison.
No supported cohorts match
Try clearing one or more filters. Suppressed cohorts are never included in the browser bundle.
Private 10,000-vehicle development sample
Model & benchmark
Calibrated logistic regression is the final development-prototype model for the next-episode first-attempt non-pass target. The boosted tree is retained only as a benchmark.
Model diagnostics unavailable
The approved diagnostic summary could not be validated.
Final prototype2024 calibration-fit check
Calibrated logistic regression
The selected model for the research conclusion and any prototype-level discussion. Platt scaling calibrates its probabilities.
- PR-AUC
- —
- ROC-AUC
- —
- Brier ↓
- —
Benchmark only2024 calibration-fit check
Calibrated boosted tree
Retained for development comparison only. It does not replace the logistic model or define dashboard outputs.
- PR-AUC
- —
- ROC-AUC
- —
- Brier ↓
- —
Finished publication boundary
No vehicle-level prediction service
This static dashboard serves supported aggregate sample rates and diagnostic summaries only. It contains no vehicle lookup, personal inputs, operational connection, or row-level prediction output.
- Returning-vehicle next-episode target only
- First attempt; pass versus non-pass
- Final model: calibrated logistic regression
- Boosted tree: benchmark only
Private 10,000-vehicle development sample
Data & methods
How episodes, labels, temporal evaluation, coverage limits, and privacy controls shape every published result.
Live diagnostics unavailable
Static methodology remains below; no performance values are being shown.
Prediction unit
The first attempt of the next inspection episode. Attempts no more than 30 days apart form one episode.
Binary target
Pass versus non-pass. Non-pass combines fail, reject, and abort; blanks and unrecognized outcomes are not labels.
Point-in-time features
Every history window ends before the target episode. Current-test results, diagnostics, stations, and later attempts are excluded.
Availability by source era
Coverage timeline
The slc and slco labels are source eras for Salt Lake County, not separate physical counties.
No random row split
Chronological evaluation
Historical events establish prior context. The 2025 development-sample holdout was evaluated once after model choices were frozen, so it is no longer a pristine final test. A later complete-data run is confirmatory; the partial 2026 feed is monitoring only.
- 2010–15ContextHistory only
- 2016–22TrainFit preprocessing + model
- 2023TuneSelect regularization
- 2024CalibrateScale probabilities
- 2025One-time holdoutAlready evaluated
- 2026ShadowPartial-period drift
Model governance
Evidence status
- Calibrated logistic regression is the declared final prototype model.
- The boosted tree remains a benchmark only.
- Published diagnostics cover train, tune, and calibration partitions.
- No locked-test result or population performance claim is published.
Interpret with care
Known limitations
- Participating inspection feeds do not cover all 29 counties.
- Feed and program changes can resemble real-world trends.
- The 10,000-vehicle draw retained 9,996 vehicles; page sampling can over-represent vehicles with more inspection records.
- Reject and abort can reflect process or readiness issues.
- Association is not causation or a mechanical diagnosis.
Publication boundary
Only approved aggregates reach this site
Private source
Read-only local pipeline Suppression
Minimum support + complements Public assets
Suppressed summaries only
Read-only local pipeline Suppression
Minimum support + complements Public assets
Suppressed summaries only
VINs, plates, ZIPs, stations, technician identifiers, private vehicle tokens, raw source JSON, credentials, operational records, and row-level predictions are outside this site and its public data contract.