Thank you.
Each lamp is one recording that was live at 15:56 CET.
On Friday you joined the first UI testing stress test of the Int4 Suite. You recorded, you pressed Play, and you kept going when it got hard.
This page is what that afternoon achieved. Every number on it was measured on the day.
The afternoon in numbers
54
recordings
47 became scripts. That is 87%.
63
playbacks
Play in the workbench, on 25 scripts. 8 scripts reached PASSED.
296
played steps passed
Out of 353 steps played back.
40
UI test script versions
Saved in the Suite on the day.
5,411
SAP dialog steps
On 2 systems, ED1 and S42.
32
real SAP documents
Sales orders, purchase orders and material documents.
The peak
By 15:56, 30 were live at once, each in its own browser on the runner.
15:56 CET
30
live recordings
17 playbacks ran through to PASSED on the day. The others stopped with a finding, and each finding is on file.
| Minute, CET | Most live recordings | Recordings started | Playbacks passed | Playbacks with a finding | Node CPU, cores | Runner memory, GB |
|---|
What reached SAP
Your recordings and playbacks sent 5,411 dialog steps to 2 SAP systems and created 32 documents.
2,818
dialog steps
VA01279ME21N214MIGO97SE16N502,593
dialog steps
VA01101MIGO69ME21N24SE16N13How the platform held
That is the API median, measured while ~30 recordings ran on one node with 16 vCPU and 64 GB.
Peak 11.7 of 16 cores at 15:55 CET
Peak 14 GB at 15:56 CET
0
pod restarts
0
out-of-memory kills
One live recording costs≈0.4 CPU · ≈450 MBA measured number, and the basis of the capacity plan for Copenhagen.
The hard moments
Some recordings stopped. Some playbacks met a surprise. Each one left evidence, and each piece of evidence becomes a concrete improvement before Copenhagen.
SIGTRAP 5×
5 Chromium renderer crashes between 15:56 and 16:09. 3 recordings could not stop and were ended by the runner's safety timeout after 60 s.
The gain
Core dumps and crash reports captured. We know exactly where, and the recorder will learn to survive a crashed tab.
EXPECTATION_FAILED 13×KEY_NOT_ACTIVE 6×UNEXPECTED_POPUP 3×
Playback checks met popups that appear only sometimes, such as ME21N personalisation and VA01 value help.
The gain
Concrete cases for smarter popup rules recorded with the script.
WRONG_SCREEN 8×
SAP opened on another menu or screen than the one recorded.
The gain
Start-screen normalisation before replay.
AMBIGUOUS_TARGET 5×RECORDING_UNRESOLVED
Seen on toolbar overflow buttons. In 2 recordings a picked suggestion was not the field's own proposal.
The gain
Sharper UI5 targeting and suggestion handling in the recorder.
≈0.4 CPU≈450 MB
That is the cost of one live recording. At the peak of ~30 live recordings the machine used up to 11.7 of 16 CPUs while the Suite's API answered in 27 ms (median).
The gain
A measured capacity plan for Copenhagen instead of a guess.
Assignment numbers and passwords, GUI vs Fiori choice, copy-paste into the recorder screen, unpublished S42 Fiori services.
The gain
A clearer guide, clipboard support on the list, and a system checklist.
The evidence
Nothing from Friday was thrown away. The whole afternoon is on file, ready to reproduce each hard moment and to check each fix.
Each dot is one raw recorder event.
The people
38 people at 37 seats. Seat 34 was shared.
You gave the recorder its first real crowd. You found its limits in one afternoon, and you left us the evidence to move them.
The crash dumps, the error codes and the sizing numbers from Friday are turning into fixes now. You will see them at the workshop.
See you in Copenhagen.
Marcin GalczynskiInt4
Next
Int4 Suite UI Testing · first stress test · Friday 13 November 2026 · 15:29–16:40 CET
All times in CET. All figures measured on the day.
(Confidential) This document is the property of Int4 AG, and without the owner's permission, it may not be copied or distributed.