Lab · Case study
Two landing pages and one visual contract
In brief: When adding AI and Engineering offers to 4digital, I wanted them to retain their identity while sharing the site's visual language. Successive corrections led to a contract and browser geometry measurements. The final record covered 1340 cases with no reported violations, but full E2E acceptance remained partial. This case shows what needs checking between writing a rule, implementing an interface and deciding to publish.
01 / 1340 cases and acceptance still partial
After harmonising elements across the 4digital sites, the final record covered 1340 measurement cases with no reported violations. The checks included header geometry, logos, content alignment and document overflow.
Full E2E acceptance still had the status PARTIAL.
Some requirements remained outside the scope of those measurements: a complete set of reviewed reference images and some manual scenarios involving the keyboard, zoom and transitions between domains. The measurement results allowed specific findings to be closed. They did not automatically complete the entire acceptance process.
The work that led to this distinction began much more simply — with two landing pages and elements that ought to line up.
02 / Three cards that were hard to compare
I was expanding the 4digital website with AI and Engineering offers. The AI landing page needed to retain its own value proposition, work in the Polish and English versions of the site, and allow a later move to a subdomain. I did not require a separate repository or deployment.
I also wanted transitions between the new offer, the home page and the Lab to preserve brand consistency.
In the collaboration section, three cards presented diagnosis, delivery and verification. Each had a title, a description and a “WHAT YOU GET” section. The middle title took up more lines, its description was longer, and the separator and outcome started lower than in the neighbouring cards.
Looking across the row meant encountering content with different roles. Comparing the stages required finding the relevant passage separately in each column.
I asked for the descriptions, separators and outcome sections to start at the same heights. A longer title or description needed to affect the layout of the entire row.
Similar observations came up around logos, headings and spacing. Comparing the pages showed that the same visual roles had been given different proportions. Another local correction could tidy up one fragment, but rules for subsequent changes were still needed.
03 / From an observation to a rule
I decided to develop a shared visual contract. The reference points were the existing 4digital.cloud site and the Lab's identity. The distinct content of the offers and the layout suited to publications were to be preserved.
The contract defined relationships that could later be checked. Descriptions in cards within the same row were to start at the same height, even when titles occupied different numbers of lines. Separators and outcome sections were also to share a common height.
A separate rule concerned the project list: the starts of the description column in successive rows were to lie on the same vertical axis. The first case compared Y coordinates; the second compared X coordinates.
The logos required a different criterion again. The specialisation marks included an additional frame, so giving the files the same width did not ensure the same scale for the symbol and wordmark. I chose alignment with the existing Lab. The contract related scale to the height and internal geometry of the marks.
A shared layer for the header, navigation and breadcrumbs was added to the code. Grids took responsibility for alignment. When the layout switched to one column, content could grow naturally, without empty paragraphs or fixed text heights.
It remained to be checked whether the browser actually positioned these elements according to the rules.
04 / What the geometry checks found
File checks could confirm the presence of a component, a CSS declaration or the correct asset. An element's final position also depended on the style cascade, font, text length and screen width.
A measurement panel was prepared that opened pages in frames of specified dimensions. It recorded the coordinates of selected elements, document width, check results and hashes of the artefacts under examination. This tied each result to a specific version of the page.
The table below combines observations recorded in the acceptance report before the corrections with results read from the final measurement record. Dimensions are in CSS pixels.
| Problem | Observation | Cause or finding | Correction | Check result after the changes |
|---|---|---|---|---|
| G05 — description column in the AI project list | Report: a 27.289 px difference in where the column started | Each row determined the widths of its own grid, depending on its content | Shared parent columns inherited through subgrid | AI PL, Chrome, 1440 px: all three X starts = 574.344 px after rounding; spread 0 px, both themes |
| G06 — AI section content relative to its heading | Report: a 10 px offset | The content margin did not match the geometry of the 1:2 heading grid and its gap | Corrected the margin calculation and the breakpoint for switching to one column | AI PL, Chrome, 1440 px: heading and content X = 518.664 px after rounding; difference 0 px in both sections and themes |
| G09 — Polish home page form | Report: 2 px of overflow at a width of 761 px | The columns and form actions did not fit within the available width | Allowed columns to shrink and actions and button text to wrap | At 761 px, the document is 761 px wide; no overflow in Chrome, Firefox or Safari, both themes |
| G07 — Engineering EN card descriptions | Reported misalignment with a title spanning two lines | Not reproduced within the tested scope; the existing subgrid aligned the descriptions | No additional CSS correction for this finding | Engineering EN, Chrome, 1440 px: Y spread of descriptions in the first row 0 px, both themes; the other tested widths meet the tolerance |
The “before” values come from the implementer's report. The old version was not independently reproduced while preparing this article. The “after” results come from the specified samples in the final measurement record.
In G05, the shared grid removed the dependency between the start of the column and the content of an individual row. In G06, the margin calculation had to be matched to the geometry of the adjacent section. G09 showed the importance of checking right at a layout breakpoint: the form pushed the document beyond its width at 761 px.
G07 ended differently. The reported misalignment of Engineering EN descriptions was not reproduced within the tested scope. The existing subgrid met the tolerance. The finding retained its “not reproduced” status, with no additional CSS correction.
Failing to reproduce a reported issue is a different result from confirming that a correction worked.
05 / What the positive result covered
The number 1340 refers to measurement cases from a defined matrix of pages, themes and widths.
Chrome covered the full matrix of page addresses, two themes and the basic desktop/mobile sizes. Measurements were also taken in Chrome, Firefox and Safari for representatives of page families at width boundaries. The browser scopes were not identical.
The measurements allowed coordinates to be compared and overflow to be detected. They did not replace every test of using the interface. Full acceptance still lacked, among other things, a complete set of manual keyboard and zoom scenarios and reviewed reference images.
That is why a positive result for this matrix and the status PARTIAL could apply at the same time. They referred to different scopes.
A check result only becomes meaningful alongside an account of what the check covered.
06 / From a one-off measurement to a reusable tool
I decided to extract the measurement tooling into Web Factory so it could be used in subsequent projects.
The measurement mechanism, profile configuration and a synthetic example were extracted. A profile defined page families and check criteria. The example allowed the tool to be checked without copying the entire site.
The extraction received its own validation: tool tests and 604 browser cases in Chromium. The earlier Firefox and Safari results applied to the version embedded in the project. They were not treated as tests of the code after its move.
This made it possible to check the same relationships again in subsequent deployments. The tool's effect on the time needed for that work has not yet been measured.
The question for a future experiment is: will using the existing measurement tooling in another, independent project shorten qualification time and reduce the number of manual corrections compared with the previous process?
07 / A separate boundary before production
A separate requirement emerged while preparing the release. An experimental Discovery assistant was running on DEV. I wanted to continue testing it and retain a separate decision on its public launch.
Geometry measurements did not settle this question. The publication path needed its own control.
Checks of the finished artefact were added to the supported builds and deployment scripts, including when restoring a historical version. Detecting assistant files or known markers stops this path. The widget and API also received host restrictions, and DEV continued to require authorisation.
In a negative test, the script was given a historical artefact containing the assistant. It stopped before connecting over SFTP. A separate readback of production confirmed the absence of the feature's markers on 31 checked pages and 404 responses at the three assistant addresses tested.
I then decided to freeze the tested version with a DEV tag. This preserved a reference point for subsequent changes, together with evidence and outstanding acceptance criteria.
That decision allowed testing to continue. It did not constitute approval for the public launch of Discovery.
08 / What was resolved
The initial problem concerned the consistency of the new offers with the existing site. The work produced shared rules, an implementation and a way to measure whether those rules were met.
Specific geometry differences were closed. One reported issue was not reproduced. Full acceptance remained partial, and the measurement tool received separate validation after being extracted for further use.
Each result had its own scope. Preserving that scope in the report made it possible to determine which work could be closed, what needed further checking and which decision remained to be made.
In this case, a useful DEV state could be frozen without declaring full production readiness. The acceptance limitations remained part of the result.
09 / Limitations
This account concerns work on the 4digital site and the state frozen on 3 November 2026. It is the history of an engineering effort, not a controlled experiment comparing the efficiency of two website development processes.
The measurements cover specified pages, widths and checks. They are not a full accessibility audit or proof that every interaction is correct. Changes in conversion, reading time and maintenance costs were not measured.
The Discovery block concerns the code examined and the supported release paths. It does not replace hosting permissions or controls over deliberate changes to the block itself. The production readback describes the state at the time of the check, not continuous monitoring.
10 / Material provenance
The goal of the work and the decisions described are supported by the preserved conversation from the period when the site was being expanded. An AI coding agent was used for implementation, preparation of the measurement tooling and execution of the checks.
While editing this article, those records were compared with the code, change history and saved reports. Site measurements were not rerun. Figures from before the corrections were attributed to the implementer's report, and the specified final results were checked against the preserved measurement record.
Implementation materials and full reports remain in the internal archive. This article does not provide a complete package for publicly reproducing the work.
11 / From the author
I just wanted three cards on a page to line up.
Really. I had not planned to develop a visual contract, build measurement tooling or extract a tool into a separate project. I wanted the titles, descriptions and separators to start at the same heights.
A few iterations later, I had 1340 measurement cases, my own tool for subsequent projects and a report that, despite the positive results, still said PARTIAL.
Along the way, we also found two pixels of overflow that I apparently could not leave unattended.
The funniest part is that it all makes technical sense. Each decision followed from the previous one, and the resulting tool may still prove useful.
Was it worth it? We will see on the next project.
Could we simply have aligned the three cards and moved on?
Of course.
But then this article would not exist.
How to cite
Bartłomiej Pietrzyk, “Two landing pages and one visual contract”
4digital Lab, 3 November 2026, https://lab.4digital.cloud/en/two-landing-pages-and-one-visual-contract/
Terms for using the write-ups and data are agreed individually.