# State repair data in business workflows | SourceX

Canonical: https://refer.sourcex.si/guides/state-repair-training-data-enterprise
Last updated: 2026-11-04

State repair is the act of putting a business record back into the correct state after something broke it: a sync overwrote a field, an order stuck on hold, an invoice posted to the wrong account. The records that show it are a before value, the repair action with who and why, and an after value someone checked. A company that keeps all three, linked, holds a rare record of how its workflows get fixed.

## What is state repair in a business workflow?

State repair is a correction of the system state, not of a document. The workflow's records said one thing, reality said another, and a person changed the records until they matched again.

Most workflow data shows the happy path: a request comes in, it is processed, it closes. State repair shows the other branch. Somebody noticed a wrong value, worked out which records were affected, changed them in the right order, and confirmed the result. That chain carries judgment that rarely gets written down anywhere else: which field to trust when two systems disagree, what to fix first, and how to tell the repair worked.

The useful unit is not the error. It is the pair of states with the action between them.

## One repair, field by field (Illustrative)

The table follows a fictional 240-person industrial parts distributor. A nightly integration between its CRM and ERP mapped the wrong source field, and a long-standing customer's payment terms flipped to prepay. The customer's next order went straight to credit hold.

| Record | Broken state | Repair action | Corrected state | Where the evidence lives |
|---|---|---|---|---|
| Customer payment terms | Prepay, written by the nightly sync | Credit analyst reverts the field and notes the reason | Net 30 | CRM field history: old value, new value, user, time |
| Sales order status | Credit hold | Hold released with reason code "terms error" | Released to warehouse | ERP order audit trail |
| Integration field mapping | Terms mapped from a retired field | Systems admin corrects the mapping under a change ticket | Mapping fixed, next sync clean | IT ticket and integration change log |
| Customer communication | Automated delay notice sent | Account manager emails a new ship date | Order shipped two days late | Helpdesk or email thread |
| Verification | Not checked | Analyst reruns the open-orders report next morning | No other accounts affected | Saved report and ticket closing note |

Read across any row and you get a complete before, action and after. Read down the rows and you get the order of the repair: fix the data, release the work, fix the cause, tell the customer, check for other damage. The last row is the one most companies skip. Without it, nobody knows whether the repair was complete.

## Why does the end state matter so much?

Because it is checkable. A before value and an after value can be compared without anyone having to interpret free text.

Agent researchers use the same idea. The τ-bench benchmark judges whether an agent completed a customer-service task by comparing the database state at the end of the conversation with an annotated goal state. A company's own repair history supplies real pairs of that kind: a wrong state, the steps taken, and a goal state a person confirmed. Data services firms describe related work as well; Appen's trajectory analysis service, for example, has experts write the correct continuation from the point where an agent failed.

That is a reason the records might interest an AI buyer, not a promise that they will. SourceX reviews whether a specific company's records are licensable and wanted before anything is agreed.

## Which records show how a workflow was corrected?

Four kinds of record do the work. A company rarely has all four for every repair, and that is fine; the question is which ones it has consistently.

- **Field-level change history.** The before and after values. Salesforce, for example, documents field history tracking that stores the old value, new value, user and date for each tracked field. Two catches matter: tracking only covers fields someone switched on, and it starts from that day, not retroactively. Salesforce also documents exporting field history for up to 24 months, so older repairs may only survive in an archive.
- **Workflow or order audit trails.** Status changes with reason codes, such as a hold placed and released. These place the repair inside the process.
- **Tickets and change requests.** The diagnosis: why the state was wrong and what fixed the cause. These connect a one-off correction to a lasting fix.
- **Verification evidence.** A rerun report, a reconciliation, a closing note that says what was checked. This is what separates a repair from a guess.

If the company runs several systems of record, ask which one is authoritative for each field. Our explainer on [what a system of record is](/questions/what-is-a-system-of-record) covers why that decides which history counts. Related exception records, such as [reconciliation exceptions](/data-types/reconciliation-exceptions-ai-agent-training-data), often sit next to repair history and are worth asking about in the same conversation.

## Questions to settle before an introduction

Keep this to metadata. You are finding out what exists and who controls it, never collecting records.

| Question | Why it matters | Who usually knows |
|---|---|---|
| Which systems keep field-level history, and for which fields? | Untracked fields leave no before value | CRM or ERP administrator |
| How many months or years of history survive? | Retention limits decide how many repairs remain | IT lead or systems owner |
| Do tickets reference the affected records? | Links turn separate logs into one repair chain | Service desk or operations manager |
| Is there a habit of verifying repairs? | Unverified fixes are weaker evidence | Finance or operations lead |
| Whose data is in the records: the company's, or its clients'? | Rights to license depend on ownership | Owner, COO or general counsel |
| Who can authorize a licensing discussion? | SourceX needs an authorized contact | Owner or executive sponsor |

The company itself must meet the program basics: 50 or more full-time employees at peak (contractors excluded), a documented operating history, rights to license the data, 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 full checklist is on [who qualifies](/who-qualifies), and our list of [questions to ask before licensing company data](/resources/questions-to-ask-before-licensing-company-data) helps the owner prepare.

## Limits and what stays unknown

Several things cannot be judged from the outside, and you should not try:

- **Personal data.** Repair records often name customers and employees. Whether and how they can be de-identified is decided during SourceX's rights and data review, and state privacy rules can apply; see our note on [state AI laws and training data](/guides/state-ai-laws-2026-training-data).
- **Volume and consistency.** Ten well-documented repairs are an anecdote. Whether a company has enough, recorded the same way, is only clear once SourceX reviews metadata with the owner.
- **Value.** No one can price repair history from a conversation. Any company compensation is set in the company's own agreement with SourceX, separately from your reward.
- **Client-owned records.** If the repairs happened in a client's system, the client owns that history, not the service provider who did the work.

For the broader family of exception records, our guide to [exception handling data for AI agents](/guides/exception-handling-ai-agents-data) is a useful companion. More guides are in the [guides hub](/guides).

## 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 an owner tells you their team keeps field history and links repairs to tickets, [register as a partner](/register), check the company against [who qualifies](/who-qualifies), and read the [terms](/terms) before you make the introduction.

## Sources

- https://help.salesforce.com/s/articleView?id=xcloud.tracking_field_history.htm&language=en_US&type=5
- https://help.salesforce.com/s/articleView?id=000383373&language=en_US&type=1
- https://arxiv.org/abs/2406.12045
- https://www.appen.com/agentic-ai-trajectory-analysis
- https://refer.sourcex.si/terms
