# Error localization evidence gap analysis | SourceX

Canonical: https://refer.sourcex.si/questions/error-localization-evidence-gaps
Last updated: 2026-11-04

Four groups of gaps make error localization hard to assess in clinical testing records: missing context, such as interface logs purged after a few weeks; missing permissions, such as corrections made under shared logins; versions that moved, such as a mapping fixed without a dated change record; and unrecorded final outcomes, such as a second corrected report never linked to the first. Each hides at least one stage of the value's journey, so the first wrong step cannot be pinned down. All four can be spotted from metadata.

This analysis is part of [error localization](/guides/concept-error-localization), which follows a glucose result whose unit went wrong at the interface engine. The task here is to set what a stage-by-stage review would expect beside what testing sites usually keep, and to describe the shortfall honestly.

## What a stage-by-stage review needs to see

Localizing an error means comparing the value at each system it passed through with the instrument reading, and finding the first one that differs. That works only if a version survives in every system and the versions can be joined.

| Evidence | Stage it covers | Why the review needs it |
|---|---|---|
| Instrument file with value and unit | Analyzer | The reference every later version is compared with |
| Original and corrected report | Lab information system | Shows the change itself |
| Correction reason, signer and time | Lab information system | Shows who acted and why, and whether the change was a fix |
| Outbound message content | Interface engine | The only record of what left the lab |
| Mapping and template change dates | Interface configuration | Separates cases before and after a fix |
| Final report linked to earlier versions | Receiving record and lab system | Records the outcome, so the case has an answer |

Units deserve a word. UCUM exists so that machines exchange quantities and their units without ambiguity; when a unit is filled from a template default instead of the source, that is exactly the kind of error only the outbound message can reveal.

## Expected against typically kept

The pattern below reflects ordinary testing operations; it is not a statistic.

| Gap group | What labs usually keep | What tends to be missing | Metadata question |
|---|---|---|---|
| Missing context | Reports, because regulation requires them | Outbound message content; logs roll over after days or weeks | "How long are outbound messages kept, and where?" |
| Missing permissions | Named logins in the lab system | Shared bench or night-shift logins; no role recorded | "Are corrections signed by named users with roles?" |
| Versions that moved | Current interface mapping | Dated history of unit, range and template changes | "Are mapping changes dated and tied to a ticket?" |
| Unrecorded final outcome | The last corrected report | A link from the final report back to earlier versions | "Is each corrected report linked to the ones before it?" |

The report side is usually strong for a reason. 42 CFR 493.1291 requires a U.S. laboratory that finds an error to issue a corrected report and keep duplicates of the original and the corrected version, and 42 CFR 493.1105 requires original, preliminary and corrected reports to be retrievable for at least 2 years. Nothing comparable fixes the interface engine's message log, so it is the first place to look for a gap.

## Illustrative gap profile: a 260-person outpatient laboratory network

The network is fictional and the figures are invented. Each percentage is the share of corrected results from the last two years for which that evidence exists, as the quality manager and integration lead estimated it.

| Evidence | Share covered | What is missing | What it hides |
|---|---|---|---|
| Instrument files | 80% | Files from two older analyzers were not archived | The reference value for those cases |
| Original and corrected reports | 100% | Nothing; both are kept as required | Nothing |
| Correction reason and signer | 75% | Night-shift corrections under a shared login | Who acted at the lab-system step |
| Outbound message content | 30% | Logs kept for 60 days until an archive was added last spring | Whether the interface changed the value |
| Mapping change dates | 40% | Changes before the ticketing system went live | Whether a case predates a fix |
| Final report link | 60% | Second corrections issued as new reports, not linked | Whether the case ever reached a right answer |

The outbound message row is the one that matters most here. With only 30% coverage, most older cases look as if the error jumped from a correct lab-system entry straight to a wrong receiving record, and the interface is blamed or cleared by guesswork. The cases worth describing are the recent ones where all six rows exist together.

## How each gap distorts the answer

A missing stage does not just leave a blank. It moves the apparent location of the error.

- **Missing context** removes a stage, so the error can only be placed somewhere between the surviving versions. In the glucose example, purged interface logs leave a correct lab-system entry next to a wrong receiving record, with no way to say whether the interface or the receiving system changed the unit.
- **Missing permissions** keep the stage but lose the agent. FHIR's Provenance resource, for comparison, records the person, software or device responsible for each version; without that, a reviewer cannot tell a human slip from a template default.
- **Versions that moved** mix cases that behave differently. A template fixed a week after the incident means cases before and after the fix answer differently, and an undated fix makes them indistinguishable.
- **An unrecorded outcome** removes the answer key. If the right unit eventually went out on a second report that nobody linked, the case looks unresolved.

Where interface logs usually live, and how long, is covered on the [integration error records](/data-types/integration-error-handling-records) page.

## Describing the shortfall in an introduction

Write one sentence per gap, with the window that is complete. For the fictional network:

- "Corrected and original reports are complete for two years; outbound message content is complete only since last spring."
- "About a quarter of corrections were made on a shared night-shift login; the quality manager can say which shifts."
- "Mapping changes are ticketed since the ticketing system went live; earlier changes are undated."

Phrases to avoid: "we keep everything" (the interface log usually says otherwise), "every error is traceable" (untested) and "the records are compliant" (compliance covers reports, not the stage you need). Eligibility comes first in any case: 50 or more full-time employees at peak, documented operating history, rights to license, and an authorized contact. Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English.

The [qualification scorecard](/questions/does-error-localization-data-qualify) turns these answers into a fit rating, and the [review packet](/resources/error-localization-review-packet) lists what the lab prepares next. Non-clinical laboratories are covered on the [testing laboratory industry page](/industries/testing-laboratory-data-licensing); more record families are under [data types](/data-types).

## Next step

> **Refer a company. No cap on rewards.** Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.

If the laboratory has a recent window where every stage survives, [register as a partner](/register), check [who qualifies](/who-qualifies) and read the [terms](/terms), then introduce the owner with that window and the gaps stated plainly.

## Sources

- https://www.ecfr.gov/current/title-42/chapter-IV/subchapter-G/part-493/subpart-K/subject-group-ECFR9482366886d579f/section-493.1291
- https://www.law.cornell.edu/cfr/text/42/493.1105
- https://hl7.org/fhir/provenance.html
- https://ucum.org/about
- https://refer.sourcex.si/terms
