# Environment reproducibility review packet | SourceX

Canonical: https://refer.sourcex.si/resources/environment-reproducibility-review-packet
Last updated: 2026-11-04

An environment reproducibility review packet holds nine metadata-only documents, assembled in this order: a scope note, the task specification, the starting-state field list, the cut-off method, the version and extension history, the integration register, the acceptance check, a volume and retention summary, and a rights note. Together they let SourceX judge whether past stock mismatches can be rebuilt and scored. The company prepares them from what it already holds; no ledger exports go in before an agreement.

This list is part of [environment reproducibility](/guides/concept-environment-reproducibility). It comes after the conversation an introducer has with the owner: the [qualification scorecard](/questions/does-environment-reproducibility-data-qualify) decides whether to introduce, and the [evidence gaps analysis](/questions/environment-reproducibility-evidence-gaps) explains the weak spots this packet should state openly.

## Before assembling: what the packet is and is not

The packet describes a class of past situations, not individual records. Its model is the dataset datasheet proposed by Gebru and colleagues, which documents a dataset's motivation, composition, collection process, preprocessing, recommended uses, distribution and maintenance. Applied to an ERP inventory archive, that becomes: which situations, rebuilt how, from which records, checked against what.

| The packet is | The packet is not |
|---|---|
| A description of count-variance cases and how their starting states can be fixed | A ledger export, backup file or set of count journals |
| Dates, counts, system names and owners | Supplier, customer or employee names |
| A statement of known differences between original and rebuild | A promise that every case will rebuild |
| Prepared by the company | Prepared or handled by the introducer |

## The packet in order: nine documents

Order matters because each document depends on the one before it. The acceptance check, for example, cannot be fixed until the starting-state fields are known.

| # | Document | What it shows | Who usually prepares it |
|---|---|---|---|
| 1 | Scope note | Which ERP, sites and years the packet covers, and which are excluded | Controller |
| 2 | Task specification | The one task the cases support, stated in a sentence | Operations director with controller |
| 3 | Starting-state fields | The fields that define each case on count day | Controller |
| 4 | Cut-off method | How history can be stopped at a date | ERP administrator |
| 5 | Version and extension history | Application versions and extensions in force over the period | ERP administrator or IT |
| 6 | Integration register | Feeds posting into the ERP on each date, and which a copy would not carry | ERP administrator |
| 7 | Acceptance check | The test a rebuilt case must pass, fixed in advance | Controller with operations director |
| 8 | Volume and retention summary | Investigations per month, months retained, migration dates | Controller |
| 9 | Rights note | Who owns the records, vendor or 3PL terms, identifiers present | CFO or counsel |

**1. Scope note.** One paragraph. Name the ERP, the warehouses and the date range, and list anything left out, such as a site run by a logistics provider.

**2. Task specification.** For example: starting from the system state at the moment of a physical count, determine why the counted quantity differs from the calculated quantity, and identify the entry or missing entry that explains the difference.

**3. Starting-state fields.** Count date and location as recorded on the journal; calculated and counted quantities as first recorded; item and bin identifiers. Microsoft's documentation shows why "as first recorded" matters: Calculate Inventory creates journal lines with a calculated quantity, and users are told not to recalculate because expected inventory might change.

**4. Cut-off method.** Dated backups, dated test copies or posted entries that can be filtered by date, with how far back each reaches. Note the timing rule: a copied Business Central environment does not sync with its source after it is made, so a copy fixes the state of the moment it was taken, not of an earlier count.

**5. Version and extension history.** The versions and extensions live on each date, or the log that shows them. Microsoft says a copy keeps the source's version and extensions; the history shows whether a past case ran under the same ones.

**6. Integration register.** Each feed, its start date and whether a copy would carry it. Base-application integration settings are cleared in a Business Central copy, so these feeds are the usual known differences.

**7. Acceptance check.** See the next section.

**8. Volume and retention summary.** Variance investigations per month, months of retained history, any ERP migration dates and any years with purged journals.

**9. Rights note.** Whether customer, supplier or employee identifiers appear, whether a logistics provider or ERP vendor contract limits use, and who can authorize a licensing discussion. Whether a company can license such records at all is covered in [can a company license its data to AI companies](/questions/can-a-company-license-its-data-to-ai-companies).

## The acceptance check, fixed before anyone replays

Without a fixed check, "the rebuild matches" is an opinion. The packet states four conditions in advance:

- [ ] The rebuilt state shows the same calculated and counted quantities as the original journal.
- [ ] Every known difference between the original and the rebuild, such as a cleared EDI feed, is listed before the task starts.
- [ ] The recorded cause, a correcting entry or reason code, explains the full difference, not part of it.
- [ ] The person who posted the correction is identifiable by role.

A case that fails any condition stays out of the set. Counting how many fail is itself useful: it tells the reviewer how much of the archive is genuinely rebuildable.

## Illustrative packet cover sheet

This cover sheet is fictional, for a 270-person HVAC parts distributor. The figures show the level of detail expected, not a real company.

| Field | Entry |
|---|---|
| ERP and sites | Business Central, two warehouses; a third site run by a logistics provider excluded |
| Period | 18 months of retained history |
| Investigations | About 40 count variances investigated per month |
| Cut-off method | Dated backups for 12 months; posted entries filterable for the full period |
| Known differences | EDI feed from the largest customer; scanner app added mid-period |
| Acceptance check | Four conditions, as above |
| Identifiers present | Customer and supplier names on postings; employee user IDs |
| Authorizer | CEO |

Agent builders want tasks like this because they can be replayed and scored; the live guide to [RL environments in enterprise workflows](/guides/rl-environments-enterprise-workflows) covers the demand side, and the evidence that [AI companies pay for data](/proof/ai-companies-pay-for-data) explains why operational records attract interest. None of that means a given packet will be accepted. How SourceX tracks SourceX's current data needs over time is outlined in its [quarterly SourceX's current data needs review](/resources/quarterly-buyer-program-review).

## Who owns the records, and what the packet leaves open

The controller usually owns count journals and variance investigations, the ERP administrator or IT owns versions, copies and integrations, the CFO or COO may sponsor the introduction, and the CEO can authorize a licensing discussion. The company itself must have 50 or more full-time employees at peak (contractors excluded), 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 packet cannot show how many cases pass the acceptance check, how supplier, customer and employee identifiers would be de-identified, whether a logistics provider's or ERP vendor's terms limit use, or whether any buyer wants this task. Those are settled in SourceX's review with the company.

## 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 owner can see how its controller would fill most of the nine slots, [register as a partner](/register), confirm the company against [who qualifies](/who-qualifies) and read the [terms](/terms), then make the introduction and let the company prepare the packet with SourceX.

## Sources

- https://learn.microsoft.com/en-us/dynamics365/business-central/inventory-how-count-adjust-reclassify
- https://learn.microsoft.com/en-us/dynamics365/business-central/dev-itpro/administration/tenant-admin-center-environments-copy
- https://arxiv.org/abs/1803.09010
- https://refer.sourcex.si/terms
