Title: Browser layout expert: detect content hidden in scroll regions (5 h, hourly)
Budget / SalaryHourly project
TypeFreelance project
LocationRemote
Posted56 minutes ago
About the project
A test harness that renders React components in headless Chromium,
changes their content and screen size, and measures whether the layout
breaks. Breakage is found by geometry, not by comparing screenshots: five
checks run inside the page and report overflow, clipping, content
spilling out of its box, overlap and content cut off by the screen edge.
The checks have been compared against blind human labels.
The problem
The checks treat every scroll container as fine, because content past
its edge can be reached by scrolling. Human labellers disagree in one
kind of case: a navigation row that scrolls with its scroll bar hidden
by CSS, and at 200% text size shrinks until only "Dashb" of its first
link shows. The screenshots show no scroll bars at all, so the rule has
to come from the page's DOM and CSS, not from the picture. The current
checks are frozen, so this is a prototype rule for the next version.
What you'll do
Must have:
1. Define, in plain words first, and then implement a new check for
scroll regions that hide content without any sign that they scroll:
either the scroll bar is hidden by CSS, or the visible area has
shrunk smaller than a single item.
2. Put the check behind a switch, so that with the switch off nothing
in the current results changes.
3. Pin its behaviour with lab cases (minimal HTML with exact expected
counts, both cases it must flag and cases it must not): left-to-right
and right-to-left pages, horizontal and vertical scrolling, nested
scroll containers, scroll bars hidden in each common way, and regions
that are cut off but do show that they scroll.
If time is left:
4. Content pushed past a scroll container's start edge (for example
centred content wider than its container), which cannot be scrolled
to at all.
5. A badge or pill that runs under neighbouring text that sits in no box.
Acceptance (I will run these)
- Switch off: the full test suite passes, and no existing check's
output changes.
- Switch on:
- your lab cases pass;
- none of the 20 hand-written reference components gets a finding in
any of the 19 test cases;
- the real component from the disagreement above fails its case;
- three real cases whose scroll region is already cut in the
unchanged render get no new failure.
- Checks stay plain DOM functions: no timers, no randomness, no network.
Deliverables
1. Commits in the repository I give you (Conventional Commits format, no
trailers), sent back as patch files. You don't push anywhere.
2. A short, factual report of about one page (400 to 600 words), as
plain text or Markdown, covering:
- the rule in plain words, as it would read in documentation, and
the severity you suggest (error or info);
- what it catches and what it misses, naming the lab cases that show
each;
- the acceptance results: each command you ran, pass or fail, and
the key numbers;
- anything you did differently from this brief, and why;
- open questions or risks, and the hours spent.
No screenshots or slides needed. Clarity matters more than polish.
Skills required
- Deep knowledge of how Chromium lays out a page: overflow and scroll
containers, flexbox and grid alignment, right-to-left text and scroll
bar styling. This matters more than any library. The checks are plain
browser code (getComputedStyle, getBoundingClientRect,
Range.getClientRects) with no framework.
- TypeScript, React, Vite and Playwright (Chromium only).
- Tailwind CSS v4: the components being tested are styled with Tailwind
classes, so you need to read what those classes produce.
- Node.js 26 and its built-in test runner (node --test). Node runs the
TypeScript files directly; there is no Jest or Vitest.
- Git, to send your commits back as patch files.
Nothing else is needed: no backend, database, cloud service or API key.
Terms
Hourly at [rate], capped at 5 hours. Must-haves first; if you reach the
cap, stop and report where you are. Scripts must keep working on
Windows, macOS and Linux. I share the code after the award, and all
communication stays on Freelancer.
To apply, answer in your own words
1. Screenshots in this harness show no scroll bars. From inside the
page, how would you tell whether a scroll container's scroll bar is
hidden by CSS, both with `scrollbar-width: none` and with a
`::-webkit-scrollbar` rule?
2. An `overflow-x: auto` container holds a flex row with
`justify-content: center` that is wider than the container. What can
the user scroll to, and what can't they?
3. Your plan for the 5 hours, in a few lines.
A test harness that renders React components in headless Chromium,
changes their content and screen size, and measures whether the layout
breaks. Breakage is found by geometry, not by comparing screenshots: five
checks run inside the page and report overflow, clipping, content
spilling out of its box, overlap and content cut off by the screen edge.
The checks have been compared against blind human labels.
The problem
The checks treat every scroll container as fine, because content past
its edge can be reached by scrolling. Human labellers disagree in one
kind of case: a navigation row that scrolls with its scroll bar hidden
by CSS, and at 200% text size shrinks until only "Dashb" of its first
link shows. The screenshots show no scroll bars at all, so the rule has
to come from the page's DOM and CSS, not from the picture. The current
checks are frozen, so this is a prototype rule for the next version.
What you'll do
Must have:
1. Define, in plain words first, and then implement a new check for
scroll regions that hide content without any sign that they scroll:
either the scroll bar is hidden by CSS, or the visible area has
shrunk smaller than a single item.
2. Put the check behind a switch, so that with the switch off nothing
in the current results changes.
3. Pin its behaviour with lab cases (minimal HTML with exact expected
counts, both cases it must flag and cases it must not): left-to-right
and right-to-left pages, horizontal and vertical scrolling, nested
scroll containers, scroll bars hidden in each common way, and regions
that are cut off but do show that they scroll.
If time is left:
4. Content pushed past a scroll container's start edge (for example
centred content wider than its container), which cannot be scrolled
to at all.
5. A badge or pill that runs under neighbouring text that sits in no box.
Acceptance (I will run these)
- Switch off: the full test suite passes, and no existing check's
output changes.
- Switch on:
- your lab cases pass;
- none of the 20 hand-written reference components gets a finding in
any of the 19 test cases;
- the real component from the disagreement above fails its case;
- three real cases whose scroll region is already cut in the
unchanged render get no new failure.
- Checks stay plain DOM functions: no timers, no randomness, no network.
Deliverables
1. Commits in the repository I give you (Conventional Commits format, no
trailers), sent back as patch files. You don't push anywhere.
2. A short, factual report of about one page (400 to 600 words), as
plain text or Markdown, covering:
- the rule in plain words, as it would read in documentation, and
the severity you suggest (error or info);
- what it catches and what it misses, naming the lab cases that show
each;
- the acceptance results: each command you ran, pass or fail, and
the key numbers;
- anything you did differently from this brief, and why;
- open questions or risks, and the hours spent.
No screenshots or slides needed. Clarity matters more than polish.
Skills required
- Deep knowledge of how Chromium lays out a page: overflow and scroll
containers, flexbox and grid alignment, right-to-left text and scroll
bar styling. This matters more than any library. The checks are plain
browser code (getComputedStyle, getBoundingClientRect,
Range.getClientRects) with no framework.
- TypeScript, React, Vite and Playwright (Chromium only).
- Tailwind CSS v4: the components being tested are styled with Tailwind
classes, so you need to read what those classes produce.
- Node.js 26 and its built-in test runner (node --test). Node runs the
TypeScript files directly; there is no Jest or Vitest.
- Git, to send your commits back as patch files.
Nothing else is needed: no backend, database, cloud service or API key.
Terms
Hourly at [rate], capped at 5 hours. Must-haves first; if you reach the
cap, stop and report where you are. Scripts must keep working on
Windows, macOS and Linux. I share the code after the award, and all
communication stays on Freelancer.
To apply, answer in your own words
1. Screenshots in this harness show no scroll bars. From inside the
page, how would you tell whether a scroll container's scroll bar is
hidden by CSS, both with `scrollbar-width: none` and with a
`::-webkit-scrollbar` rule?
2. An `overflow-x: auto` container holds a flex row with
`justify-content: center` that is wider than the container. What can
the user scroll to, and what can't they?
3. Your plan for the 5 hours, in a few lines.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.