Business Central's Record-and-Replay Test Tool Reached GA This Week. It Is Not UAT.

What shipped with version 29

Business Central 29 — 2026 release wave 2, build 55407 — was released on 29 September 2026, and its what's-new page lists "Page Scripting enters General Availability" among the AL tooling items. The matching message centre post, RM573384, was published on 30 September 2026 and is marked Launched, GA target October 2026. It names two additions alongside the status change: multi-row grid selection, and validation of dialog text.

The tool itself is not new. The 2024 release wave 1 plan entry for "Use in-client page scripting tool for acceptance testing" records a public preview date of 1 April 2024 and, in the general availability column, a dash. So GA is not a feature arriving after thirty months. It is a caveat being removed: the documentation called it a production-ready preview subject to supplemental terms of use, the phrasing that stops a cautious partner putting a tool on a client's critical path. Note how the news arrived, too — release plans stopped in September 2026, so a change like this surfaces through the message centre rather than a document you can diff, as we covered in scoping the BC29 regression pass without a release plan.

The reference documentation has not caught up

At the time of writing, the Microsoft Learn article for the tool is still titled "Use page scripting tool for acceptance testing (preview)", still states "This is a production-ready preview feature", and carries a last-updated date of 1 October 2025 — a year before the GA announcement.

More usefully, that page still documents the behaviour the GA note says has been improved. On multi-row selection it states, verbatim, that multiple line selections aren't recorded as expected, and that if you select multiple records in a list during a recording, only the last line selection is recorded and played back. On dialog text it says nothing; the validation it describes is asserting that a control has a specific value.

The honest reading is narrow: the capabilities shipped and the documentation has not been revised to describe them. But it leaves both unspecified — no Microsoft page says how a dialog-text assertion is written or what it matches on. Test them in a sandbox before you build a pack around them.

What a recording actually proves

Recordings save as YAML, which is what makes the tool interesting to a partner rather than to a lone consultant: a YAML file goes into source control, gets reviewed, parameterised and reused across tenants. Microsoft also ships an npm package, @microsoft/bc-replay, so the same files run headless from a command line or a build pipeline rather than only in a browser session.

The documented boundary is equally clear, and gets glossed over in summaries: it is not a generic HTML automation tool, and cannot automate control add-ins, embedded Power BI reports, or anything outside the Business Central web client experience. On a mid-market implementation that exclusion list is not academic: control add-ins are where much bespoke UI lives, and an embedded report is often the thing a finance user actually looks at.

A recording therefore proves that a sequence still completes and that nominated controls still hold nominated values. A good regression check; not the claim that the business accepts the system, which is the distinction we drew in UAT versus QA.

Where it belongs in your process

The failure mode to avoid is treating a GA automation tool as a replacement for the acceptance cycle. What sinks an ERP go-live is rarely a step that stopped working. It is scripts nobody executed, a module nobody could test, a defect accepted by nobody in particular, and a user who finished the journey and said it would not survive a Monday. None of that is a replay problem.

Where it earns its place is the layer underneath. Record the stable, high-volume journeys — order to cash, purchase to pay, the posting routines — and replay them against every monthly update and extension deployment, so the human UAT window goes on judgement instead of re-proving that sales order entry still saves. That monthly pass is the work most partners quietly under-do, and exactly the work a machine should do. Regression testing and acceptance testing are different jobs, and this tool is squarely in the first.

What it does not give you is the record. A pipeline tells you which replays passed last night, not which agreed scripts were never run, which failures were accepted and by whom, or what the client signed. That lives in test management, and the sign-off process is what turns one into the other.

What to do in this upgrade cycle

Record five journeys per client, now. Not a full suite — the five highest-volume ones. The value appears at the second replay, so what matters is a first pass in source control before the next monthly update.

Verify the two new capabilities yourself. Multi-row selection and dialog-text validation are announced but undocumented. Confirm them on build 55407 before either reaches a client-facing test plan.

Map the exclusions against your extensions. Every control add-in and embedded report stays manual, and knowing that list is what stops an automated coverage figure misleading you. Our guide to UAT on Business Central covers scoping the rest.

Keep acceptance attributable. Automation changes what gets re-proved. It does not change who has to accept what, by name, before a date.