# QA scoring disagreements: a system map | SourceX

Canonical: https://refer.sourcex.si/guides/quality-scoring-disagreements-records-across-systems
Last updated: 2026-10-19

Rebuilding a quality scoring disagreement takes four joins: the contact center's interaction ID stored on each evaluation, the snippet timestamps that place reviewer comments on the recording, reviewer and adjudicator IDs that survive staff turnover, and the form version in force on the session date. A fifth link, from the interaction to the customer account, brings in the context a ruling may depend on. With those keys, one split decision can be followed from the call to the ruling to the changed form.

This map is part of the [quality scoring disagreements guide](/guides/workflow-quality-scoring-disagreements), which covers the inventory before a QA vendor change. It does a narrower job: show which system holds what and which key gets you from one to the next.

## Four systems and what each one knows

No single platform holds a full disagreement. Each system sees one slice.

- **Contact center suite.** The interaction itself: recording, transcript where enabled, agent, queue, timestamps. It knows nothing about scores.
- **QA platform.** Evaluations, calibration sessions, comments, adjudication notes, corrected scores and usually the forms. In Genesys Cloud, for instance, a calibration session bundles evaluations of a single interaction by several evaluators and flags one of them as the scoring index. The QA platform knows the judgment, not the context.
- **CRM or account system.** Account notes, approvals recorded against the customer, case history. It often holds the fact a ruling turned on.
- **Form library or settings history.** The wording of each scoring item on each date. Sometimes inside the QA platform, sometimes only in a document someone kept.

## The linking-key table

The evidence asset for this page: each record, the system that holds it, the key that joins it to the next, and who usually owns it.

| System | Record | Linking key | Typical owner |
|---|---|---|---|
| Contact center suite | Recording and interaction metadata | Interaction ID | IT or contact center operations |
| QA platform | Evaluation with item scores and comments | Interaction ID; evaluation ID | Quality director |
| QA platform | Snippet citation in a comment | Start and end timestamps against the recording | Quality director |
| QA platform | Calibration session | Session ID; evaluation IDs; scoring index | QA calibration lead |
| QA platform or email | Adjudication note | Session ID or evaluation ID | QA calibration lead |
| Form library | Form version | Form ID and version date | Quality director |
| CRM | Account note or approval | Account ID reached from the interaction | Customer service operations |
| Directory | Reviewer and agent identities | Employee or user ID | HR and IT |

## One fully joined case (Illustrative)

A fictional 420-person home warranty contact center; every ID and date is invented. This case is different from the over-limit credit in the worked example: here the agent, not a second reviewer, raised the dispute.

1. **Evaluation E-60318** in the QA platform marks interaction 88-51904 as a fail on the "hold etiquette" item: the reviewer's comment says the customer sat silent for almost a minute without a check-back.
2. **Agent dispute.** The agent contests the score through the platform's dispute queue, saying the hold was a warm transfer to the claims team.
3. **Interaction ID to the suite.** 88-51904 opens the interaction record in the contact center suite. Its event log shows a consult at 04:10 and a transfer at 05:05, not a hold.
4. **Calibration session CS-2307.** Two more reviewers score the call. Both mark the item as not applicable, citing the same 04:10 to 05:05 segment.
5. **Ruling.** The QA lead overturns the fail: consult time before a warm transfer is not hold time. The ruling is written on the session.
6. **Form version 8**, dated two weeks later, adds a line to the item's guidance separating consults from holds.

Here the explaining record is not a CRM note but the suite's own event log. Without the interaction ID on the evaluation, the dispute would read as an agent's word against a reviewer's. The [worked exception example](/guides/quality-scoring-disagreements-exception-examples) follows a split between two reviewers instead; together the two cases show why both the suite and the CRM links matter.

## Where joins usually break

Each break below is named by the key that fails and what the joined case from the previous section would lose if it did.

| Failing key | How it happens | What case E-60318 would lose |
|---|---|---|
| Interaction ID | Evaluations were imported into a QA tool that kept only its own evaluation number | Step 3: the event log showing a consult and transfer, so the dispute becomes one person's word against another's |
| Snippet timestamps | The suite purges audio on a shorter clock than the QA platform purges evaluations | Step 4: the 04:10 to 05:05 segment both reviewers cited no longer plays |
| Reviewer and adjudicator IDs | User accounts are deleted when staff leave | Steps 1 and 5: who failed the call and who overturned it |
| Form ID and version date | The item guidance is edited in place with no dated copy | Step 6: proof that version 8 changed the rule, rather than the reviewers changing their minds |
| Session ID on the ruling | The overturn was agreed in a mailbox thread | Step 5: the session shows a split with no recorded outcome |
| Account ID | Account notes live in a CRM the QA team never reaches | Nothing in this case, but the over-limit credit in the [worked example](/guides/quality-scoring-disagreements-exception-examples) would lose its explanation |

The form row deserves a second look. McHugh's review of the kappa statistic treats interrater reliability as different raters scoring the same variables; when the wording of an item shifts between two dates, an apparent disagreement across those dates may only reflect the edit. The recordings row is covered from the retention side in the [guide to recordings at platform retirement](/systems/contact-center-migration-call-recordings-retention).

## Questions for the quality director

Score each join pass, partial or unknown, from the director's own knowledge.

| Join | Question | Answer format |
|---|---|---|
| Evaluation to recording | Does each evaluation store the suite's interaction ID? | Pass, partial, unknown |
| Comment to snippet | Do comments cite timestamps, and do recordings still exist for those years? | Years with both |
| Session to ruling | Is the adjudication written on the session, or elsewhere? | Platform, email, none |
| Score to rulebook | Are form versions kept with dates? | Yes, partly, no |
| Interaction to account | Can the account be reached from the interaction? | Pass, partial, unknown |

The [data readiness survey](/resources/company-data-readiness-survey) gives the company a wider format for the same self-assessment, and the [AI use case prioritization framework](/resources/ai-use-case-prioritization-framework) helps a client place this record set among other candidates.

The quality director knows the joins; the CEO decides whether to discuss licensing, often with the COO, CFO or general counsel advising. The introduced company needs 50+ full-time employees at peak (contractors excluded), a 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 90% of the data is in English. Buyer interest, the rights position on recordings and any offer stay unknown until SourceX's review. Partners sign the Channel Partner Agreement before submitting; the first valid introduction within the 90-day window is credited when the company submits a verified application. Records of this shape are what make [exception handling data](/guides/exception-handling-ai-agents-data) useful.

## Next step

> **Refer a company. Earn up to $100,000.** Default reward: 25% of Actual Supplier Payout, capped at $100,000 cumulatively per referred company. Eligibility and payment conditions apply, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.

If the quality director can mark most joins as pass for several years of calibration, [register as a partner](/register), check the company on [who qualifies](/who-qualifies) and read the [program terms](/terms), then fill in the [referral brief template](/resources/quality-scoring-disagreements-referral-brief-template).

## Sources

- https://help.mypurecloud.com/?p=380404
- https://pmc.ncbi.nlm.nih.gov/articles/PMC3900052
- https://refer.sourcex.si/terms