# Audit Carrier-Data Field Completeness with Clear Denominators

Canonical: https://usdotdata.com/blog/carrier-data-field-completeness-audit
Author: John Hauler
Published: 2026-10-15T07:37:13.532Z
Updated: 2026-10-15T07:37:13.532Z

---

To audit carrier-data field completeness, freeze the input, define which entities need each field, and count values that meet a written rule against that field's eligible denominator. Report missing values, invalid values and unfinished assessments separately. A populated field can still be wrong, outdated or irrelevant to the decision.

This guide answers a narrow research-operations question: **how many eligible carrier records contain usable values for the fields this task requires?** The result is a reproducible field audit with named exceptions. It does not establish that an export contains every intended carrier, that a contact can be reached, or that a company needs your service.

## Define what the percentage will measure

Write the intended task first. For a regional fleet-service research project, a principal-office state might support preliminary account grouping. A business telephone field might support later contact research. Neither proves where trucks receive maintenance or who approves purchases.

Distinguish three checks:

- **Presence:** a value is supplied under the source's missing-value conventions.
- **Structural usability:** the supplied value meets your declared format or category rule.
- **Accuracy and business relevance:** separate evidence supports the value and its use in a real decision.

The [UK Government Data Quality Framework](https://www.gov.uk/government/publications/the-government-data-quality-framework/the-government-data-quality-framework) distinguishes completeness, validity and accuracy. We use those distinctions here, but the audit states and calculations below are our editorial method, not a prescribed government standard.

Name the measure precisely. “Format-usable phone values among eligible accounts” tells a reader more than “verified leads.” If you combine presence with a format check, disclose both rather than presenting the result as pure cell population or established accuracy.

Start with the [carrier-data field dictionary](https://usdotdata.com/blog/carrier-data-field-dictionary) when a column's meaning is uncertain. A principal-office state and a confirmed maintenance-site state are different fields even if both contain `OH`.

## Freeze the input and counting unit

Save the original file, extraction time, selection description, dictionary version and row grain. Record the working-copy version and review cutoff. A colleague should be able to identify exactly which input produced the counts.

For an entity-level audit, reconcile repeated observations before assigning one result per entity and field. Keep the duplicate rule explicit: byte-identical repeated rows may be represented once, while conflicting observations require a documented resolution. Do not keep whichever value makes the completeness percentage higher.

Rows with unresolved identifiers need a separate record count and explanation. Silently dropping them can hide the very identity defect being investigated. If they cannot yet be reconciled into entities, do not add their row count to a distinct-entity denominator.

FMCSA's [Open Data Program](https://www.fmcsa.dot.gov/registration/fmcsa-data-dissemination-program) links dataset-specific dictionaries and describes different record types. Identify the actual file and definitions you used. A familiar USDOT column does not make every row a unique carrier observation.

If the question is whether the expected files or entities arrived, first use the [export delivery reconciliation guide](https://usdotdata.com/blog/reconcile-carrier-export-delivery-boundary). Field completeness concerns values inside a defined input; it cannot prove the delivery's membership.

## Assign eligibility before looking for favorable results

A field's eligible denominator contains the entities for which that field is required by the declared task. Different fields may have different denominators, but each exclusion needs a reason fixed independently of whether a value happens to be present.

For example, a team may need geography for every account while requiring a telephone field only for the subset assigned to contact research. That is a task-specific scope choice. It does not mean excluded businesses have no telephone or that the source is not expected to hold one.

Use “not required for this task” separately from “source says not applicable.” Preserve any source-provided not-applicable code and its documented meaning. Do not turn a blank into an exemption. If applicability is unknown, keep the assessment unresolved until the denominator can be settled; report that uncertainty rather than publishing a definitive eligible percentage.

Write the eligibility rule before inspecting outcomes. Changing it later creates a revised audit, with the earlier result retained. Removing accounts because their phone fields are missing would improve the number by changing the question.

## Use five mutually exclusive audit states

For a field whose eligibility is resolved, give every entity one of these local states:

1. **P — present and passes the declared rule.** A value exists and passes the specified structural test. This does not imply an external truth check.
2. **M — missing observation.** The field was assessed and contains no meaningful value under the documented source conventions. Keep blank, provider-omitted and explicit unknown reasons distinguishable in the detail.
3. **I — present but invalid under the rule.** A meaningful supplied value fails the declared format, range or category check. “Invalid” describes this field result, not the carrier.
4. **U — unassessed.** Eligibility is known, but the check is unfinished or its evidence cannot yet be interpreted. State the reason, such as pending review or inaccessible input.
5. **X — not required for this task.** Excluded from this field's denominator under the saved eligibility rule. Preserve the reason; do not count it as a passing value.

Document ambiguous tokens before classification. A literal `unknown` might be a missing-value marker under one specification and ordinary text in another. A source's explicit zero can be a meaningful observation; do not remove it merely because a spreadsheet treats zero as false. Unknown source semantics belong in U until resolved.

An absent column can be recorded as M with reason “not supplied” if the file is readable, its scope is understood and that field was required. A file you cannot inspect is U. These produce different repair actions even when neither supplies a usable value.

## Report both usable values and completed assessments

With resolved eligibility, the counts must satisfy:

```text
Eligible = P + M + I + U
All entities = Eligible + X
Format-usable share = P / Eligible
Assessment completion = (P + M + I) / Eligible
```

Multiply a share by 100 to express it as a percentage. Keep the underlying counts beside it. If the eligible count is zero, report “not applicable to this task; zero eligible entities,” not 0% or 100%.

You may also report `P / (P + M + I)` as the share passing among completed assessments. Label that denominator prominently. It excludes unfinished work and cannot replace the eligible-denominator result. Neither percentage measures how many companies are safe, commercially suitable or ready to buy.

If you inspected a sample, state the selected input and method. The [carrier-research sampling plan](https://usdotdata.com/blog/carrier-research-sampling-plan) addresses what can be inferred beyond it. A field audit of a convenience sample does not become an industry-wide estimate because its arithmetic is correct.

## Worked example: forty fictional entities, two denominators

Everything in this example is invented for instruction. `C001` through `C040` are local exercise IDs, not USDOT numbers. No USDOTData export, customer dataset or market-quality benchmark is represented.

The frozen exercise contains 43 input rows: one for each of the 40 IDs plus identical repeats of C004, C012 and C026. Removing those three repeated rows under the declared entity rule leaves 40 entities. There are no conflicting or unresolved identities in this exercise.

The saved plan requires a principal-office state for all 40 entities. It requires a business-phone format check only for C001–C032, the preselected contact-research group. C033–C040 are assigned to entity-only account mapping, so their telephone fields are X. Membership was assigned before examining the fields.

The phone rule asks only whether the working value contains ten digits after removing spaces, parentheses and hyphens. It is a deliberately limited test for this fictional input, not a universal international-phone validator or evidence of an assigned, working number. The geography rule accepts a two-letter state code from the saved allowed set for this exercise. It does not validate an address or operating territory.

Here is the complete synthetic classification manifest. Ranges are inclusive; every ID receives exactly one state per field. The manifest lets a second reviewer reproduce the counts; it is not a raw-source validation dataset.

```text
PHONE
C001-C024: P  (24)
C025-C028: M  (4)
C029-C030: I  (2)
C031-C032: U  (2)
C033-C040: X  (8)

PRINCIPAL-OFFICE STATE
C001-C032: P  (32)
C033-C036: M  (4)
C037-C038: I  (2)
C039-C040: U  (2)
X: none
```

The phone results are **24/32, or 75%, format-usable**, with 30/32 assessments completed, or 93.75%. Four missing values and two invalid values are known exceptions. Two checks remain unfinished. Reporting 24/30 = 80% without naming the completed-assessment denominator would conceal those open checks.

The geography results are **32/40, or 80%, format-usable**, with 38/40 assessments completed, or 95%. The fact that its denominator is larger does not mean it is a better dataset: it is a different field with different task eligibility.

For work requiring both fields, inspect their intersection at the entity level. Here C001–C024 pass both, giving 24 of the 32 entities that require both fields. Do not average 75% and 80%, multiply the percentages, or infer an intersection from totals alone. The saved ID assignments establish this result.

The appropriate handoff is narrow: 24 contact-research accounts have both fields passing the defined format checks; their contactability, operating location and buying need remain unverified. Observed audit results support a next research step. Possible service fit is a hypothesis. Confirmed buying need requires separate business evidence.

## Turn exceptions into a bounded repair plan

Assign each exception an owner and next action. Missing supplied fields may require clarifying the delivery specification. Invalid values may require fixing an import transformation. Unassessed cases may require finishing review. Do not ask for new data until you know which problem you are addressing.

Keep a versioned before-and-after result. If C031 and C032 later pass the phone check, P becomes 26, U becomes zero and the eligible denominator stays 32. The usable share becomes 81.25%; assessment completion becomes 100%. That is a documented resolution of unfinished work, not proof that the underlying carriers improved or that a provider's entire database became more accurate.

For subsequent batches, keep field definitions, eligibility and assessment coverage comparable. A higher rate after excluding a difficult segment is a scope change unless you separately show results on unchanged membership. Use the [export-evaluation checklist](https://usdotdata.com/blog/evaluate-trucking-lead-list-export) to decide which research uses the remaining limitations permit.

## Copyable audit worksheet and release checklist

Keep one batch block and one field block in your existing authorized system:

```text
BATCH
Purpose / reviewer / cutoff and time zone:
Original file / extraction time / selection version:
Dictionary version / row grain / entity key:
Input rows / duplicate rule / distinct entities:
Unresolved identity rows and treatment:

FIELD
Exact field / intended use / structural test:
Eligibility rule / excluded IDs and reasons:
Source missing-value and not-applicable rules:
Raw observation reference / assessment state:
P / M / I / U / X counts:
Eligible denominator / usable share:
Assessment completion / unresolved applicability:
Exceptions / owner / next action:
Revision and evidence of changed classifications:
```

Before releasing the result, confirm that counts reconcile, exclusions have independent reasons, unknowns remain visible and another reviewer can reproduce the totals. Keep raw values and transformation evidence available under your existing access rules. Publish only the bounded conclusion the audit supports.

## Frequently asked questions

### Does a filled phone column mean the contacts are usable?

Only under the exact rule you tested. A format pass does not establish an assigned number, successful connection, correct business, decision-maker or permission to contact. Describe those as separate checks with their own evidence.

### Should an optional field be excluded from the denominator?

It can be excluded when the saved task genuinely does not require it. State which entities were excluded and why. Do not make it optional after discovering blanks merely to improve the rate.

### Is a missing field the same as an unfinished check?

No. Missing means the relevant input was assessed and supplied no meaningful value under a known convention. Unassessed means the work or interpretation remains incomplete. Report both because their resolution differs.

### Can we combine all fields into one completeness score?

An unexplained average hides required-field gaps and different denominators. Report field results first. If a task requires several fields together, count entities meeting all those requirements using their actual memberships, and retain each field's exceptions.

### What is a good completeness percentage?

There is no universal threshold in this guide. Define what the next task requires and what unresolved work its owner can accept. A high optional-field rate cannot compensate for an unresolved identity or a required field that is missing.

## Start with one decision and its required fields

Choose one existing research input, freeze the entity set and audit the few fields needed for its next use. Deliver the counts with the exception list and allowed conclusion. That gives the next researcher a clear starting point without turning a neat spreadsheet into an unsupported quality promise.

*By John Hauler, USDOTData's editorial persona. Sources reviewed October 15, 2026. USDOTData is an independent information service, not FMCSA. Audit states, formulas and fictional examples are editorial methods, not official classifications, product-quality findings or promised product features.*