# Tool-call provenance in company records | SourceX

Canonical: https://refer.sourcex.si/guides/concept-tool-call-provenance
Last updated: 2026-11-04

Tool-call provenance is the record of which tool, rule or person changed a business record, with which inputs, under which version, and what happened next. In a help desk it looks like an audit line saying a ticket moved to Solved because a named trigger read a confirmed billing credit. The record is useful when the actor, the inputs, the rule version and the outcome are logged together for each change.

## One audit event, annotated (Illustrative)

Start with a single record rather than a definition. Illustrative: a 260-person B2B software company runs customer support in a help desk and billing in a separate subscription system. A customer reports a duplicate charge. Every value below is fictional.

| Field in the audit event | Illustrative value | What it proves |
|---|---|---|
| Object and change | Ticket 48213, status Pending to Solved | The action itself |
| Actor type | Business rule, not a person | That automation acted, so a rule is accountable |
| Rule name and revision | Trigger "Billing credit confirmed", revision 7 | Which version of the approved workflow ran |
| Input the rule read | Custom field "credit note ID" filled by the billing integration | The fact the rule relied on |
| Who wrote that input | Billing integration, API channel, 14:02 | The upstream tool call that fed the rule |
| Timestamp of the change | 14:03, one minute after the input | The order of events |
| Outcome after the change | No reopen within 14 days; credit visible on the next invoice | Whether the action was right |

Read down the middle column and you have a chain: an integration wrote a credit note ID, a rule read it, the rule changed the status, the customer did not come back. Each step names the tool that acted. That is tool-call provenance in a business setting.

Some help desks expose exactly these fields. Zendesk's developer documentation describes a `via` object on each ticket audit; when a status change is made by a business rule, the event's source is a trigger, with the trigger's ID and title, and a revision ID on Enterprise plans. Other systems record less, so what a given archive holds matters more than any product name.

## What does tool-call provenance mean in an ordinary business workflow?

It means a change in a system of record can be traced to the tool that made it and the inputs that tool used. In a business, the "tool" is often a workflow rule, an integration or a macro, and the provenance is whatever the audit log kept about it.

The term comes from AI agent work. When an AI model uses a tool, the call has a name and arguments. The Model Context Protocol specification, for example, says a `tools/call` request must name a tool exactly as the server listed it and pass arguments defined by that tool's input schema. A record of which tool was called, with what arguments, and what came back is the provenance of the agent's action.

Companies have kept the business equivalent for years without calling it that. The W3C PROV ontology gives a vocabulary that fits both: an entity (the ticket status), an activity that acts on it over time (the update), and an agent that bears responsibility for the activity (the trigger, or the employee who ran a macro). Where an archive records all three, a reviewer can tell an approved workflow from a manual override, which is the distinction AI developers care about when they study how an agent should act.

This page is about one action at a time. For dataset-level origin, consent and rights, see [what data provenance is](/proof/what-is-data-provenance); for the difference between tracing transformations and tracing origin, see [data lineage vs data provenance](/compare/data-lineage-vs-data-provenance).

## What makes a provenance chain unreliable?

A chain is only as strong as its weakest link, and the weak links are predictable.

- **Shared service accounts.** Several integrations write through one API user, so the log says "API" without naming which tool acted.
- **No rule revision.** The log names the trigger but not its version, and the rule was edited twice that quarter. Which logic ran is uncertain.
- **Inputs overwritten later.** The credit note ID field is changed after the status update, so the value the rule read is no longer the value on the ticket.
- **Manual override without a note.** An agent reopens and re-solves the ticket by hand, breaking the clean line from input to rule to outcome.
- **Outcome never recorded.** The ticket closes and nothing links it to the next invoice or a later complaint, so nobody can say whether the action was right.

These gaps do not erase the record's value, but they change what a reviewer can conclude. Vendors who sell agent evaluation services describe the same concern from the buyer side: Appen, for example, markets failure analysis that groups agent breakdowns by cause, which depends on knowing exactly which step and which tool went wrong. Market interest in the concept does not mean every audit log qualifies.

## Who qualifies, and what stays unknown

The introduced company needs 50+ full-time employees at peak (contractors excluded), a documented operating history, rights to license the records, 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.

Establish only metadata: which systems log changes (help desk, ERP, billing), roughly how many years of audit history exist, who owns those systems, and whether customer contracts restrict reuse. The owner is usually support operations or IT; the person who can authorize a licensing discussion is the CEO, often with the COO or CFO as sponsor.

What stays unknown until SourceX reviews the opportunity: whether there is SourceX's current data needs for these records, the outcome of the rights review, the de-identification needed, the company's offer and any reward. SourceX licenses approved business data directly from companies and manages the agreement, rights review, data preparation, delivery and payout; the partner's role ends at a qualified introduction. More record families are listed under [data types](/data-types).

## Go deeper on tool-call provenance

- [Questions an introducer should ask about tool-call provenance](/questions/does-tool-call-provenance-data-qualify): a short question set for checking whether a client's audit history could qualify.
- [Record gaps that make tool-call provenance hard to assess](/questions/tool-call-provenance-evidence-gaps): a guide to the missing actors, versions and outcomes that weaken an audit trail.
- [Tool-call provenance review packet](/resources/tool-call-provenance-review-packet): a checklist of what belongs in an evidence packet for SourceX review.

## 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 a client's audit history names the rule or integration behind each change and links it to an outcome, [register as a partner](/register), check [who qualifies](/who-qualifies), and read the [program terms](/terms) before you introduce the decision-maker.

## Sources

- https://refer.sourcex.si/terms
- https://developer.zendesk.com/documentation/ticketing/reference-guides/via-object-reference/
- https://developer.zendesk.com/api-reference/ticketing/tickets/ticket_audits/
- https://www.w3.org/TR/prov-o/
- https://modelcontextprotocol.io/specification/2025-11-25/server/tools
- https://www.appen.com/agentic-ai
