# Environment reproducibility, explained | SourceX

Canonical: https://refer.sourcex.si/guides/concept-environment-reproducibility
Last updated: 2026-11-04

Environment reproducibility means the starting conditions of a task can be rebuilt so that anyone repeating it begins from the same place and sees the same problem. In an ordinary inventory workflow, that means recreating one stock mismatch in a safe test copy of the ERP: the same posted records up to count day, the same software version and extensions, the same expected quantity, and a written note of anything that could not be copied, such as a live integration.

## What does environment reproducibility mean in an ordinary business workflow?

Put simply, it is the difference between replaying a case and re-enacting it from memory. If the starting point drifts, two people working the "same" case are really working different cases, and nobody can say whether a fix worked.

AI developers care because agents are trained and tested by running tasks over and over. Toloka, which builds agent environments, describes versioned environments, deterministic resets and controlled seeds as what allows exact replay of agent runs and human trajectories. That is the builder's side of the problem. The company's side is simpler and older: did it keep enough dated records that a real past situation can be put back together? Our explainer on [why AI agents fail](/questions/why-ai-agents-fail) shows how much of agent failure comes from acting on the wrong state, and [why synthetic data rarely replaces real data](/questions/will-synthetic-data-replace-real-data) covers why real starting states are hard to invent.

Inventory work is a clean place to see it. An ERP records every receipt, shipment and adjustment as a posted entry. A count happens, and the counted number differs from what the system expected. To understand that difference, or to train someone to investigate it, you need the system as it stood on count day, not as it stands today.

## Rebuilding one mismatch (Illustrative)

This case is fictional. A 310-person industrial parts distributor runs Microsoft Dynamics 365 Business Central. On March 4 its warehouse team counts bin A-14 and finds 36 bearing kits of item BK-220. The system expected 48.

The physical inventory journal is where this starts. Microsoft's documentation explains that the Calculate Inventory action creates journal lines in which the counted quantity starts equal to the calculated quantity; the counter enters a different number where the count differs, and posting adjusts inventory to the counted figure. The same page warns users to keep the originally calculated lines and not recalculate, because expected inventory might change. In other words, the calculated 48 is itself a starting-condition record. Recalculate on March 9 and the "expected" number may no longer be 48.

The operations analyst wants to replay the investigation without touching live stock. Business Central lets an administrator copy an environment as a sandbox. According to Microsoft, the copy runs on the same application version as the source and keeps its per-tenant and Marketplace extensions, while integration settings in the base application are cleared, SMTP email details are deleted and agents are copied but turned off. After it is made, the copy does not sync with the source. The analyst makes the copy on the evening of March 4, after the count and before the next day's postings, which is what fixes its starting point. No BK-220 entries posted on March 4 itself, so the item's ledger in the copy runs through March 3.

So the analyst's checklist looks like this:

| Starting condition | Production, March 4 | Sandbox rebuilt to March 4 | Match? |
|---|---|---|---|
| Application version | Version in use that day | Same version, carried over by the copy | Yes |
| Extensions | Warehouse scanning and EDI extensions installed | Same extensions carried over | Yes |
| Posted item ledger entries for BK-220 | All entries through March 3; none posted on March 4 | Same entries, because the copy was made that evening | Yes |
| Calculated quantity in bin A-14 | 48 | 48, from the kept journal line | Yes |
| Counted quantity | 36 | 36 | Yes |
| EDI feed from the main customer | Running, posting shipments | Cleared in the copy | No, written down as a known difference |

Five conditions match and one does not. The EDI difference is not a failure as long as it is recorded: the analyst knows that any shipment the feed would have posted after March 4 will be missing in the copy, and that the replay should stop at the moment of the count. With that note, the sandbox shows the same 12-unit gap. The investigation can then be repeated as often as needed: was it an unposted shipment, a put-away into the wrong bin, or a receipt booked twice?

The workflow behind the gap itself, a cycle count variance investigation, is covered on its own [data type page](/data-types/cycle-count-variance-investigation). This page is about whether the starting state can be rebuilt at all.

## What the rebuilt mismatch shows, and what it cannot

- **A faithful start is not an explanation.** The sandbox reproduces the 12-unit gap. It does not say whether the cause was an unposted shipment, a wrong bin or a double receipt; that is the investigation the replay makes possible.
- **The copy is a snapshot.** Microsoft notes that a copied environment does not sync with its source afterwards. A copy made on March 9 instead of March 4 would carry five days of later postings, and the starting point would be lost.
- **Recorded differences must stay recorded.** The cleared EDI feed is harmless only because the analyst wrote it down. An undocumented difference turns a replay into a different case.
- **One item is not a warehouse.** Rebuilding one bin for one date says little about how many past mismatches the company could rebuild, which depends on how long journals and ledger history are kept.

## Who owns these records, and what stays unknown

The controller or operations finance owns the count journals and item ledger; the warehouse manager owns the counting process; an ERP administrator or IT team controls environment copies and extensions. Authority to discuss licensing usually rests with the CFO, COO or CEO.

The company needs 50 or more full-time employees at peak (contractors excluded), documented operating history, rights to license its 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. For the first questions most owners raise, see [whether a company can license its data to AI companies](/questions/can-a-company-license-its-data-to-ai-companies), [evidence that AI companies pay for data](/proof/ai-companies-pay-for-data) and [questions to ask before licensing company data](/resources/questions-to-ask-before-licensing-company-data).

What stays unknown until SourceX reviews it: how many past situations can actually be rebuilt and pass an acceptance check, how supplier and customer names are de-identified, whether the ERP vendor's or a logistics provider's terms limit use, and whether any buyer wants this task type. Interest in reproducible environments across the industry does not mean every ERP archive qualifies. Partners stay at metadata level throughout; more data types are listed in the [data types hub](/data-types).

## Go deeper on environment reproducibility

- [Questions to ask a client about environment reproducibility](/questions/does-environment-reproducibility-data-qualify): a short discovery script for checking whether a company can rebuild past operating states.
- [Record gaps that make environment reproducibility hard to assess](/questions/environment-reproducibility-evidence-gaps): a guide to the recalculated journals, missing version records and lost history that break a replay.
- [Environment reproducibility review packet](/resources/environment-reproducibility-review-packet): a list of the metadata a company prepares before SourceX reviews its ERP history.

## 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 controller confirms that count journals are kept as calculated and that history can be cut off at a date, [register as a partner](/register), check the company against [who qualifies](/who-qualifies), and read the [terms](/terms) before you introduce the CFO or COO.

## Sources

- https://learn.microsoft.com/en-us/dynamics365/business-central/dev-itpro/administration/tenant-admin-center-environments-copy
- https://learn.microsoft.com/en-us/dynamics365/business-central/inventory-how-count-adjust-reclassify
- https://toloka.ai/virtual-environments-rl-gyms
- https://refer.sourcex.si/terms
