# Tool-call provenance fit scorecard | SourceX

Canonical: https://refer.sourcex.si/questions/does-tool-call-provenance-data-qualify
Last updated: 2026-11-04

Score six criteria from 0 to 2 points each: company size, operating history, rights to license, record depth (does each automated change name the rule or integration and its revision), linkage (can the input and the later outcome be tied back), and English share where it applies. Ten to twelve points is a strong fit, six to nine is borderline, five or under is weak for now. Every row can be answered by a support or IT lead without sharing a file.

This scorecard belongs to the guide on [tool-call provenance in a business workflow](/guides/concept-tool-call-provenance), which annotates one help desk status change field by field. That guide explains what the record is. This page answers the introducer's question: does this particular client hold enough of it, in a form a buyer could use, to justify an introduction?

## How the 12-point scale works

Each criterion earns 2 points when the answer is clearly yes, 1 point when it is partly true or uncertain, and 0 when it is clearly no. The first three criteria are company-level eligibility; the last three describe the records themselves. Two rules sit on top of the arithmetic:

- **A zero on any eligibility row stops the score.** A company under 50 full-time employees at peak, or one that does not control its own records, is not ready however rich its audit log.
- **A zero on record depth makes the result weak for now, whatever the total.** If the history says only "updated" with no rule, integration or revision, there is no provenance to license, only a status trail.

Why weight record depth this heavily? Provenance is about attribution. NIST's control catalog, written for security auditing rather than AI, makes the same point in its audit-content control (AU-3): a useful audit record establishes what happened, when, where, its source, its outcome and who or what was responsible. An automated change history that drops the source or the responsible rule fails on the very element a buyer studying agent behavior needs most.

## The tool-call provenance scorecard

| Criterion | 2 points (strong) | 1 point (borderline) | 0 points (weak) |
|---|---|---|---|
| Company size | 50+ full-time employees at peak, contractors excluded, clearly | Near 50, or headcount fell after a peak that is hard to document | Under 50, or mostly contractors |
| Operating history | Several years of the same help desk, ERP or billing platform | Two to three years, or one platform migration with history carried over | A recent start, or a rebuild that discarded history |
| Rights to license | The company owns the records and no known customer term restricts reuse | Ownership is clear but customer terms are unknown | A parent, outsourcer or client controls the records |
| Record depth | Each automated change names the trigger or integration and its revision | Names the rule but not the version, or names only "API" | Shows only that a field changed |
| Linkage | The input's source event and the later outcome (reopen, refund, dispute) share a ticket or order key | Linkable by customer and date only | Outcomes live in another team's system with no shared key |
| English share | Company in the U.S., Canada or the EU, or 70%+ of the records in English | Mostly English with a sizable non-English queue | Under 70% English outside the U.S., Canada and the EU |

The English row follows the program's geography rule in full: 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.

## Discovery prompts that fill in each row

Five questions fill in most of the scorecard; each is mapped to the row it scores. Ask the head of support operations, the IT lead or the COO.

| Prompt | Row it scores | A 2-point answer sounds like |
|---|---|---|
| When a ticket or order changes automatically, does the history say which rule or integration did it? | Record depth | "The audit shows the trigger name" |
| Are past versions of your workflow rules kept, with dates? | Record depth | "Rule edits are logged; we can see revision numbers" |
| Does the history keep the old and new value of each field? | Linkage | "Every field change has before and after" |
| Can a reopen, refund or dispute be tied back to the original ticket number? | Linkage | "Same ticket or order number across systems" |
| Who could authorize a conversation about licensing these records? | Rights to license | "The CEO, and counsel knows of no customer term restricting reuse" |

Some platforms make the record-depth questions easy to answer. Zendesk's developer documentation describes a `via` object on each ticket audit: when a trigger makes a change, the source names the trigger's ID and title, and on Enterprise plans a revision ID too. A support lead on that setup can usually say yes to the first prompt and tell you whether revisions are captured. On another platform, the honest answer may be "I would have to check," which scores 1 until someone does.

## Strong, borderline and weak: three scored archives (Illustrative)

All three companies are fictional. The scores show the method, not a forecast.

**Strong: a 310-person subscription software company.** Support runs in one help desk for seven years; billing writes credit note IDs into tickets through an integration with its own API user. Audits name the trigger and revision for every automated status change, and reopens are linked by ticket number. The COO owns the records, though nobody has yet checked what the enterprise customer contracts say about reuse. Scores: size 2, history 2, rights 1 (contracts unchecked), depth 2, linkage 1 (refund disputes sit in the payment portal, linkable only by date), English 2 (U.S. company). Total 10.

**Borderline: a 140-person property management firm.** Maintenance requests close automatically when a vendor marks work complete, and the log says "closed by integration" without naming which of three vendor portals sent it. Rules were rebuilt last year. Scores: 2, 2, 1, 1, 1, 2 for 9. Worth introducing, with the shared integration identity named as a known gap; see [tool-call provenance evidence gaps](/questions/tool-call-provenance-evidence-gaps) for how to describe it.

**Weak: a 60-person logistics broker.** Shipment statuses are updated by staff and a nightly batch job that overwrites the current value. No history of who or what changed a status survives. Scores: 2, 1, 2, 0, 0, 2 for 7 on paper, but the zero on depth makes it weak: there is nothing to trace.

## Who signs off, and what the score cannot tell you

Support operations usually owns the help desk configuration and its rules, IT owns integrations and service accounts, and finance owns the billing records that feed the inputs. The person who can authorize a licensing discussion is the CEO. The company also needs documented operating history and an authorized contact.

A score cannot show how many changes per year were automated, how many years of audit history the platform actually retains, what de-identification customer messages would need, or whether any buyer wants this task type. Those stay unknown until SourceX reviews the opportunity under its own agreement with the company. Partners never request exports, sample tickets or screenshots; discovery stays at metadata level.

Check company basics with the [company qualification checklist](/resources/company-qualification-checklist), see the wider record family under [tool-use training data](/data-types/tool-use-training-data) and [customer support workflows](/data-types/customer-support-workflows), and read [how AI labs and data users evaluate datasets](/proof/how-ai-buyers-evaluate-datasets). After an introduction, the company prepares the [tool-call provenance review packet](/resources/tool-call-provenance-review-packet).

## 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 client scores six or more with record depth above zero, [register as a partner](/register), confirm the basics on [who qualifies](/who-qualifies), and read the [terms](/terms) before you introduce the COO or CEO.

## Sources

- https://developer.zendesk.com/documentation/ticketing/reference-guides/via-object-reference/
- https://developer.zendesk.com/api-reference/ticketing/tickets/ticket_audits/
- https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-53r5.pdf
- https://refer.sourcex.si/terms
