A quiet product shell makes automated advertising tangible.
Cherrie presents an AI advertising manager through a stable shell, large claims, and one explanatory object per promise. The pricing estimator concentrates the system in a direct manipulation with deterministic values, motion, and gesture-gated sound.
What this is
This report studies the bounded six-route public marketing site captured on August 12, 2026. The reconstruction preserves route structure, content, layout logic, and measured interaction behavior. The exact mirror lives under /preview/cherrie/.
Unaffiliated archive study. This private specimen is for visual and implementation analysis. It does not represent Cherrie, does not change the source site, and stays noindex.
Mirror
/preview/cherrie/
Loading
Browse the archived mirror in place. The pricing estimator remains interactive, with gesture-gated generated sound and mobile haptics. Forms, sign-up controls, booking, social links, analytics, and external network behavior are inert.
Archive state
Mirror coverage
The archive includes landing, about, pricing, demo, privacy, and terms. It was built from static extraction, rendered desktop and mobile captures, accessibility trees, runtime summaries, scroll states, pricing interaction captures, temporal evidence, Web Audio probes, hover states, and reduced-motion observations. Pricing received the deepest treatment.
Safeguards
The deployed mirror uses local HTML, CSS, JavaScript, and SVG only. All forms and external affordances are inert. The pricing AudioContext is created only after a trusted gesture. A scoped content security policy blocks form actions and third-party connections. The mirror and report remain noindex.
A fixed floating navigation pill sits over one internal scroll region. Landing moves from a centered, oversized hero into alternating feature bands. About uses a large statement and loose polaroid collage. Pricing compresses the claim into one centered estimator and a regular card grid. Demo narrows into a practical form card, while the legal routes use a restrained reading measure.
Typography
A heavy grotesque display style carries the large claims. Body copy is lighter, gray, and spacious. Scale creates most of the hierarchy, while coral marks categories, actions, and changing values. The source custom font binaries were not frozen, so the mirror uses a system grotesque approximation.
Color and material
The palette is a warm near-white ground, white cards, near-black type, muted gray explanation, hairline borders, and a coral-to-cherry accent family. Rounded rectangles, soft shadows, and faint dotted textures make the product feel approachable. The pricing result card concentrates the richer material treatment in a one-pixel animated border beam and restrained glow.
Rhythm
Wide whitespace isolates each major claim. Short declarations alternate with interface demonstrations, so the long landing page reads as claim, object, pause, then the next claim. Pricing uses the reverse strategy: a compact instrument first, followed by a calm regular grid.
Motion and feedback
Landing uses ambient movement and animated accents. Pricing turns motion into feedback: odometer digits spring to new values, slider bars reshape over 150 ms, card content reveals over 300 ms, and a border beam rotates every 1.96 seconds while its hue shifts over 12 seconds. Each crossed slider line produces a 15 ms sine click after a trusted gesture. Pitch rises from 1000 to 2000 Hz with value.
Responsive behavior
Display headlines shrink but remain dominant, content becomes one column, and the landing email form stacks. The pricing slider changes from 80 bars to 40 below 768px, with matching feedback quantization and a 3 ms vibration request per crossed line. The local mirror also adds keyboard controls, ARIA range semantics, and a reduced-motion mode.
Implementation mechanics
Static reconstruction from a Next.js source
The public source is a Next.js application. The mirror uses seven generated static HTML files, one shared stylesheet, two small browser scripts, and local assets. This preserves the visible and measured behavior without recreating the source framework.
architecture
Internal scroll ownership
The body stays fixed while #info-scroll-container owns vertical scrolling beneath the floating navigation. This shell behavior is shared across all six routes and is essential to the source framing.
layout
Deterministic pricing engine
Slider state is an integer from 0 to 100. Daily spend follows a cubic curve from $100 to $500,000 and snaps to $100 increments. A 43-row rate table lowers the displayed fee from 5.0% to 0.8%. Views, clicks, conversions, and revenue derive from fixed formulas.
calculation
Gesture-gated Web Audio
The first trusted pointer or keyboard gesture creates or resumes the AudioContext. A crossed slider line generates a short sine click with a 2 ms attack and 15 ms decay. Resting animation has no sound, and sound never loops.
interaction
Responsible parity additions
The source custom slider has no keyboard or assistive-technology contract and continues moving under reduced-motion preference. The mirror adds range semantics and keyboard input, then stops continuous motion and snaps value changes for reduced motion without changing the commercial state model.
accessibility
Reviewable safety boundary
Forms, booking, sign-up, social affordances, and external requests remain inert. Final revision 02 passed 46 Chromium Playwright tests and a static safety scan, and all 71 frozen deployment-source artifacts matched their accepted hashes before packaging.
verification
How it makes a complex corporate claim feel legible
Construction reading
The page asks the reader to believe that Cherrie can take over a fragmented, technically demanding advertising operation and make it simple. "The AI Ads Manager" names a category but leaves unanswered what the system manages, which channels it spans, how automation affects performance, and what paying for it might mean. The site answers through increasingly concrete artifacts. A dashboard gives the promise an interface and visible metrics. Brand marks and campaign creative imply customer, channel, and asset breadth. Feature bands divide the platform into familiar jobs. The pricing estimator carries the most explanatory weight because it turns a broad commercial promise into a model the reader can manipulate. Spend, fee tier, projected revenue, views, clicks, and conversions move together inside one field. Each route assigns one main object to its job: dashboard and feature sequence on landing, estimator on pricing, photo collage on about, and booking form on demo.
Taste reading
The visual and temporal choices make a potentially intimidating system feel calm, premium, and approachable. Large type gives confidence to short claims. Wide whitespace keeps a dense product category from feeling dense. Muted neutrals lower the visual temperature, while coral gives actions and changing values one continuous pulse. Most movement stays faint or contained. The pricing card's glow and sound turn an ordinary calculator into the focal instrument without making the whole page restless. The strongest transferable choice is the alternation between plain claims and proof-like objects. A short statement appears, then a dashboard, creative transformation, people network, support queue, or calculator shows what the statement might mean.
Evidence boundary
Illustrated: the page depicts a cross-channel dashboard, brand learning, creative production, pooled rewards, support access, and projected advertising outcomes. Observed: the public surface contains the six documented routes, section sequence, interactive pricing formulas, generated audio behavior, responsive breakpoint, motion timings, and fixed scrolling described here. Proven: within this archive, route structure, pricing source code, deterministic calculations, runtime behavior, and mirror safety are supported by extracted source, browser probes, hashed manifests, review, and automated tests. Gap: the product's advertising performance, AI quality, customer adoption, $500M training claim, projected outcomes, uptime, and business results are not independently proven by this website study. Product access, customer evidence, technical documentation, original design tokens, source image binaries, tablet source rendering, dark theme, and several decorative-motion details remain outside the packet.
What makes it work
Major claims are paired with one explanatory object rather than another paragraph.
The pricing model lets a reader manipulate spend, rate, and projected outcomes in one visual field.
The floating pill, pale ground, rounded cards, and coral accent hold together pages with different jobs.
Large display type and small explanation create hierarchy with few typographic styles.
Pricing feedback uses shape, number motion, a brief generated sound, and mobile haptics, all tied to direct interaction.
Mobile keeps the product claim, primary action, and estimator prominent after single-column reflow.
Deterministic formulas, exposed states, network boundaries, and route-wide tests make parity claims inspectable.
Reusable patterns
Claim-object pairing
Give each consequential promise a distinct artifact that clarifies a mechanism or outcome. The object should carry inspectable information rather than decoration.
Instrumented pricing
Turn a nonlinear price or value model into a direct manipulation. Keep the input, fee, assumptions, and outputs together, and label estimates honestly.
Quiet field, active focal object
Keep most of the page neutral and still, then concentrate motion and chromatic detail around the object that deserves attention.
Fixed shell, route-specific density
Use a stable product shell to connect marketing, company, pricing, form, and legal routes while letting each route adopt the density its job requires.
Responsive interaction density
Reduce visual marks on narrow screens while preserving the range and state model. Cherrie changes from 80 slider bars to 40 and keeps the same 0 to 100 value contract.
Responsible parity additions
A faithful study can preserve source appearance while improving keyboard semantics and reduced-motion behavior. Name the difference instead of presenting it as source behavior.
Do not copy blindly
The coral palette, oversized grotesque claims, floating pill, glow, and rounded cards form a specific brand surface. Copying the bundle would produce imitation.
The estimator is mechanically coherent, but the page does not prove that spend alone produces the displayed revenue, views, clicks, or conversions.
Continuous motion needs an accessible alternative. Preserve the mirror's reduced-motion correction rather than the source behavior.
Generated sound should remain brief and gesture-gated. Autoplay or sustained audio would change the experience and safety profile.
A custom slider needs keyboard and assistive-technology semantics.
A fixed inner scroll region complicates browser scrolling, screenshots, and initial keyboard movement. Use it only when the framing benefit outweighs native document behavior.
Dashboard mockups, brand marks, and creative examples illustrate scope. They do not establish product performance or customer results.
The crowded mobile header, transparent menu at page top, low-contrast coral uses, malformed calendar semantics, and small footer targets are source quirks rather than reusable patterns.
Open questions
What do the production design tokens and custom font files specify beyond the screenshot-sampled approximations?
How does the source render at exactly 768px, especially the header, footer, and feature layouts?
What is the intended open mobile-menu background and focus behavior at the top of a page?
What do the landing canvases render, and which decorative animations carry meaning rather than ambience?
How does the uncaptured dark theme change hierarchy, glow, contrast, and imagery?
What assumptions or historical data support the estimator's CPM, click-through, conversion, and revenue multipliers?
What product or independent evidence supports the $500M training claim, 99.9% uptime claim, and projected campaign outcomes?
Which original images and logos can be archived lawfully and durably for a future higher-fidelity specimen?
How does the experience behave in Safari and Firefox? Accepted automated QA used Chromium.