01
Offline-first scene capture
A structured mobile workflow that remains usable without a reliable connection and synchronizes when service returns.
What the warehouse shows
The same vehicle can enter the public record many different ways.
Standardization includes legitimate code expansion as well as spelling and formatting repair. A changed label is not automatically an error. It is evidence that unqualified searches and rankings can split one real-world vehicle across multiple labels.
2.17M
vehicle-party labels changed after standardization
2,172,968 of 5,310,041 party records since 2020, or 40.92%.
90,583
distinct raw make-and-model labels
Those labels resolve to 80,450 cleaned labels under the current rules.
76
raw variants associated with Tesla Model 3
A single raw label captures only 7,244 of 33,642 standardized records.
One Toyota field, several source values
TOYTOYOTOYOTTOYOTA /TOYTTOYT - TOYOTATOYTAWhy it changes the answer
Searching only one spelling can undercount a vehicle, alter a ranking, or make two jurisdictions appear less comparable. The repair should happen as close to collection as practical—not years later in every analyst’s private cleanup script.
What we can—and cannot—conclude
The public data exposes the result, not every step that produced it.
CHP’s current Crash Investigation Manual contains the instructions for the CHP 555 report. The form places vehicle year, make, model, and color together, while CHP’s Collision Investigation Unit also says it assists development of the California Automated Reporting System.
That is enough to identify a system-design opportunity. It is not enough to say every record was handwritten, that every agency uses the same software, or that one officer or department caused a particular variant. Those questions require direct workflow research with participating agencies.
The statewide California Crash Reporting System publishes the resulting crash, party, and injury records. Our role is to preserve the source values, show where analysis changes after cleaning, and propose improvements that respect field conditions and officer judgment.
What we would build with a department
Make the correct record easier to create than the fragmented one.
This is a partnership proposal, not a finished specification. The first step would be observing the real workflow with officers, records staff, analysts, and state data stewards.
02
Scan, then confirm
Where policy permits, use VIN or registration information to suggest year, make, and model—while keeping the officer in control of the final record.
03
One shared, versioned vocabulary
Give every participating agency the same maintained vehicle, roadway, injury, and collision definitions with effective dates and change history.
04
Helpful validation at entry
Flag likely misspellings, incompatible values, and missing context before submission without preventing an honest Unknown or Other response.
05
A correction feedback loop
Return recurring data-quality findings to agencies in a form that improves training and software instead of silently repairing the same patterns downstream.
06
Transparent quality measures
Track completeness, revision lag, coordinate coverage, and standardized-label coverage alongside every analytical release.
A friendly critic—and a willing partner
If zero is the goal, better evidence is shared infrastructure.
We would welcome a pilot with a California law-enforcement agency, transportation department, research group, or newsroom that wants to improve collection, validation, publication, or public understanding.
Method note: figures come from the project’s published standardization-impact snapshot generated 2026-08-07T20:26:31.701Z. Source labels are retained for audit. Standardization changes analytical labels; it does not alter the underlying official collision report or establish fault, negligence, or agency performance.