Level.Foundry · How it works
Website work, run in the open
Foundry is how Level runs website engagements: every claim backed by evidence, every step reviewed by a person before it counts, and all of it visible to you as it happens. It ends in something you can act on, never a slide deck about it.
Two ways in
Foundry runs two kinds of engagements on the same discovery, the same standards, and the same human checkpoints. They differ only in destination.
The redesign
Your site, rebuilt on evidence: discovery through structure through visual design, ending in a complete styled specification of the new site.
Ends in: the design specification, ready for the build.
The conversion program
Your current site, made measurably better: the same discovery aimed at a ranked testing program, with split-test variations designed against your live pages.
Ends in: the Test Kit, ready for your testing tool.
Below: the redesign path in full, then how the conversion program differs.
Phase I · Interrogate
Discovery & context
Before anyone sketches anything, we establish what is actually true: about your brand, your market, your site, and your visitors. Twelve investigations, each producing a reviewable document.
Brand profile
your identity transcribed into a testable specification, colors checked against accessibility standards
Data analysis
twelve months of your analytics read line by line, against the prior year
Technical discovery
your platform, integrations, and risks, mapped before they can surprise the build
Industry & competitor analysis
the market's buyers and your rivals' websites, examined first-hand
Accessibility scan
your current site graded against WCAG 2.2 AA, page by page
Questionnaire & interviews
your team and your customers, asked the questions the data can't answer
Heuristics & conversion mapping
every key page judged against named usability principles; every journey to enquiry traced
Content inventory
every URL you own, catalogued and weighed
What you get: A shared, evidenced picture of where you are, and the agenda for what the new site must fix.
Phase II · Ideate
Content, structure & wireframes
Discovery's findings become decisions: what content stays, how the site is organized, what the design stands for, and how the key pages work, before any pixel is styled.
Content strategy
keep, combine, or retire: every page's future decided, with redirects planned
Sitemap & information architecture
navigation your visitors think in, not your org chart
Design point of view
the creative thesis the whole site will argue
Template & component plan
the reusable building blocks the site will be assembled from
Messaging strategy
what each audience hears, in words tested against your positioning
Wireframes with journeys
the key page structures, each one walking a real visitor path
Imagery direction
the photography and art direction the brand will live in
What you get: A blueprint you have approved, so visual design starts from decisions instead of guesses.
Phase III · Implement
Visual design & handoff
The blueprint becomes the site's actual look: explored in competing directions, refined through review rounds, and delivered as working, styled pages.
Homepage concepts
multiple full designs generated and compared, not one take defended
Major & minor templates
every significant page type designed against the approved wireframes
Pattern library
the complete component system, consistent across every template
The handoff artifact
styled, working HTML of the new site: a specification a build team can execute without interpretation
What you get: A complete design specification: real pages, real interactions, ready for the build.
The conversion program
The CRO campaign
When the goal is lifting the site you already have, the same discovery feeds a different second half: instead of designing a new site, Foundry devises and strategizes a testing program.
Discovery & baseline
The relevant discovery investigations, with your analytics elevated to required: tests are undesignable without a real baseline. When a Foundry redesign already exists, its discovery is imported instead of repeated.
Hypotheses & test strategy
Findings become ranked, testable hypotheses, and every one must cite its evidence: an unevidenced hypothesis is a defect, not an idea. The test plan then does the honest math: what your traffic can actually support, in what order, and which measurements must exist first.
Variations & the Test Kit
Each approved test gets variations designed against your live page, changing exactly one variable so a result teaches you why it happened. The Test Kit packages every test for your testing tool: hypothesis, variation, what changed, what to track, and what win, lose, or flat each mean.
The rules: Two rules are absolute: a variation changes one thing or it is not a test, and no experiment ever uses manufactured urgency, hidden costs, or any other dark pattern. Fixes that need no experiment (a broken control, an accessibility defect) ship as fixes: we never hold a repair hostage behind a test.
The detail we go to
In a typical discovery:
Every
URL you own, crawled and catalogued
Sites of a thousand-plus pages are read in full, never sampled.
12months
Of analytics, read line by line
Against the prior year, with bot traffic identified and set aside.
Every
Brand color pair, contrast-tested
So the new design is beautiful and accessible by construction.
2viewports
Desktop and mobile, every key page
Captured, graded, and kept as evidence you can inspect.
Human judgment, built in
Nothing reaches you that a person hasn't signed.
Two checkpoints on every step
Before a step runs, a person confirms its inputs are right. After it runs, a person reviews the result. Work advances on approval, never on momentum.
Standards are checked by machine, taste is judged by people
Objective bars like WCAG 2.2 AA accessibility and Core Web Vitals are verified automatically and must pass. Everything subjective, from tone to layout, passes a named human reviewer.
AI drafts, people decide
Foundry uses AI to go deeper and faster than a manual process can. Its self-assessment is only ever a ranking aid: it cannot approve its own work, and unreviewed drafts never appear in anything you receive.
Every claim carries its evidence
Findings cite what they were observed from. When something is unknown, it says so, and joins a punch list instead of a guess.