Every design direction, side by side

What this is

v1.0 is a faithful, one-to-one rebuild of www.cloudmaven.com — all 286 addresses, the same words in the same order on every page. It changes nothing on purpose. It exists so that any later change can be proved to be only that change, and so that everybody in this review is looking at the same "before".

One thing has to be said plainly about it, and the baseline's own record says it first:

“visual — is this page verified to look like the live page. It is unverified on every row, for one reason, recorded once in REPORT.md: the live design was never captured and cloudmaven.com is not reachable from the environment this was built in.”

review/true-baseline/pages.md, on origin/claude/true-baseline

So the text of these pages is verified against the captured live site; the look of them is a rebuild that has never been placed next to a photograph of the real page. That is a limit on the baseline, not on the directions below — but it is worth knowing before you say "that isn't what our site looks like".

Current verified status

• Addresses286 pages rebuilt
• Photographed17 sample pages, at 1440 and 390
• CopyMatches the captured live body text, in order
• Live comparisonNot possible — cloudmaven.com was not reachable from the build environment
• Broken imagesNone recorded in the capture index
• DeployedNo. Branch only

What to judge

• 1 secondLook at the homepage below and stop after one second. What did the graphic tell you? Whatever that is, it is the bar every direction after this one has to beat — and the answer to this question is the reason the other six exist.
• 3 secondsGive it three. Does a core message land, or does the page simply say "we do business software", which is what every competitor's homepage says on the same screen?
• 30 secondsScroll the interior page (/digital-solutions/netsuite) end to end. Note where your attention drops. Those are the places the later directions claim to fix.

What this is

Attempt A takes v1.0 and swaps out images. Nothing else — not a word of copy, not an element, not a page. Its own report opens with exactly that claim, and the branch carries a script that proves it rather than asserting it:

“v1.1 is v1.0 with images swapped. Nothing else. Not one word of copy changed, not one element added or removed, not one page byte different. That is checked by a script, not asserted.”

review/v1-1/REPORT.md, on origin/claude/v1-1-graphics-r8wvri

It went looking for bad artwork and mostly did not find it. The one picture the owner had called out — the computer in the yellow "create an individual offer" band, which sits on 204 of the 286 pages — was redrawn: its monitor, stand and foot rebuilt from flat primitives, while the envelope and the photograph inside it were carried through as the original pixels. Everything else in the drawn set turned out to need colour corrections rather than redrawing:

“The drawn-illustration set is in far better shape than the brief assumed. The one genuinely bad drawn asset was the computer — which is what the owner said.”

review/v1-1/REPORT.md

It also made one deliberate, arguable trade: the word OFFER was full-strength brand yellow on white, which the report measures at a contrast ratio of 1.75:1 and calls not legible at the size the page renders. It is now the darker readable step named by the corporate design. That reads more muted than the original yellow, and the report says so, and says it is reversible in one line.

Current verified status

• Files in build812 in v1.0, 812 in v1.1
• Files changed22, all images, all named in the manifest
• Pages changed0 of 287 — no markup was touched at all
• Verdict printed“PASS — v1.1 is v1.0 with only the manifest images swapped”
• Robot mascot48 assets carried through byte for byte, enforced in three places
• DeployedNo. Branch only

What this attempt says it deliberately left alone. The five header_software_* images — the large pictures at the top of every solution page. Its reason: they wrap real Oracle NetSuite and EPM screens, and “redrawing them means fabricating a vendor's interface and putting it on a consultancy's website as if it were the product. That is worse than dated, and a different kind of wrong from dated.” It then records the cost of that ruling honestly: “header_software_* remain dated. They are the largest images on the interior hero of every solution page and they are the weakest thing left on the site.”

What to judge

• 1 secondThe redrawn computer is the one picture this attempt actually replaced, and it appears on 204 of the 286 pages. In the before-and-after below, at the size the page really uses it: does it now read as a computer in one second, or is it still a shape you have to work out?
• 3 secondsThe world map and the card plates were colour corrections, not redraws. Does correcting colour without changing the drawing move the message at all at three seconds — or is this a change only the person who made it can see?
• 30 secondsThis attempt keeps the dated hero images and says so. Read a solution page end to end with that in mind and decide the real question: is "leave them dated" acceptable, or does the thirty-second read fail on the largest picture on the page?

What this is

Attempt B is the same idea with a different appetite. It also changes only images and no words, and it proves it the same way — but it replaces twenty pictures rather than the handful attempt A judged necessary.

“What v1.1 is: v1.0 — the 1:1 copy of www.cloudmaven.com — with 20 drawn images replaced and nothing else changed. What it is not: a redesign, a rebuild, or the Astro lane. No copy is edited. No photograph of a person is touched. The robot is not redrawn.”

review/v1-1/REPORT.md, on origin/claude/v1-1-graphics-x9kcgx

The big move is the five large images at the top of every solution page — the ones attempt A left alone. B's diagnosis of them:

“at hero size it is four panes of near-equal size with 12px type in them, so the eye finds no way in; the palette is Oracle's and perk's, not Cloudmaven's; and it puts another company's trade dress under a Cloudmaven headline.”

review/v1-1/REPORT.md

All five were redrawn as flat, abstract dashboards — “one dominant screen, one medium, one small, so depth comes from scale”. The second visible move is the card artwork: where attempt A tested removing the coloured plates and rejected it, attempt B removed them, on the reading that this is what "without backgrounds" asked for.

On the offer computer the two attempts land in an interesting place: they reached the same fix for the unreadable word OFFER independently, but they disagree about the drawing around it. Attempt A redrew the monitor. Attempt B built the same redraw, compared it against the original, and threw it away — “The redraw was built, rendered and compared, and it lost… the picture got the one surgical fix it actually needed.”

Current verified status

• Files in build812 in each build
• Files differing46 — “Every one of them is an image. No HTML byte differs.”
• Page reflowNone — all ten full-page shots identical in height across both builds
• Palette gatePasses — “every colour emitted is in the palette”
• Robot mascot53 files identical by checksum, enforced twice
• Image weightThe replaced set went from 6.83 MB to 2.59 MB
• DeployedNo. Branch only

Two things this attempt flagged as still open. Four partner diagrams (netsuite-datev, netsuite-hubspot) are off-palette and were left alone, because fixing them properly means re-inking the artwork — “a design decision about four partner diagrams rather than a background removal, and I would rather flag it than take it”. And Bildschirmfoto-2025-12-20-um-21.41.41.png carries four other companies' wordmarks, so it was not touched either.

What to judge

• 1 secondOpen the EPM hero pair below — v1.0 on the left, v1.1 on the right, labelled in the picture itself. Attempt B calls this “the clearest before/after on the site”. Which side tells you what the page is about in one second? It is also the exact change Carsten has to rule on.
• 3 secondsThe redrawn dashboards are invented shapes, not real product screens. At three seconds, do they read as "this is how your finance systems look when they work" — or do they read as generic stock-illustration filler that says nothing a competitor could not also say?
• 30 secondsLook at the /service band with the yellow plates removed. Read it end to end. Does the page hold together with the icons standing free — or were the plates doing structural work that the grid now misses?

What this is

v1.2 changes exactly one thing on the whole site:

“v1.2 is v1.0 plus one thing. The hero's right-hand column — a screenshot collage of four product UIs on a yellow circle — is replaced by the live flow diagram… Nothing else on the site changes, and a build gate proves it byte for byte.”

review/v1-2/REPORT.md, on origin/claude/v1-2-hero-n5fk7m

The diagram shows the same nine things twice. In state A — as it runs today, everything routes through a spreadsheet in the middle, and the lines that reach it are visibly broken where a person has to re-key something by hand. In state B — after, the gaps close and one architecture carries the work. The centre of state A is a box labelled recon_final_v7.xlsx, marked THE ACTUAL SYSTEM OF RECORD.

Two design choices in it are worth knowing about before you look. A manual gate carries no colour at all — it is drawn as a genuine break in the line with a pause mark in the gap, so that the drawing needs no key. And there is no legend, on purpose: “The acceptance test says a graphic that needs a legend has already failed tier one.”

Current verified status

• GatesSix, all passing — copy, figures, spelling, palette, no-JavaScript, German homepage
• Negative testedThree gates broken on purpose, watched to fail, restored
• CopyHeadline, both paragraphs and the button provably untouched
• Without JavaScriptA still image stands in, in two shapes; caption stays real text
• Reduced motionDraws a still, both states stay switchable
• German homepageUnchanged — keeps the v1.0 hero, held there by a gate
• Branch baseRepaired. It had been pointing at unrelated work; reset onto v1.0
• DeployedNo. Branch only

The one measured cost, stated by the lane itself. On the phone the diagram is taller than the picture it replaces, so the headline moves down: “the h1 moves from y=442 to y=752, and the cover grows from 963px to 1273px. At an 844px viewport the first line of the headline is just on screen and the rest needs a scroll.” Two fixes are named — a cheap one worth about 90px, and an expensive one that would reorder the mobile column. Neither was applied.

What to judge

• 1 secondThis is the direction that stakes everything on tier one. Look at state A for one second: does "a spreadsheet is running this company" arrive without reading anything? If it does, this is the only graphic in the review that makes an argument rather than naming a category.
• 3 secondsNow compare A against B. The message is meant to land in the switch between them. Does it — or do you need the caption to understand what changed?
• 30 secondsOn the phone, the headline now begins at y=752 and needs a scroll. Is a graphic that wins the first second worth a headline that needs a scroll? That trade is the whole decision on this lane.

The second decision on this lane — how a manual gate should be marked

The brief said that if a second accent colour turned out to be unavoidable, it should be proposed, rendered, and placed beside the no-colour version rather than chosen silently. Both are built from the same code and both are photographed. The lane's own recommendation is quoted underneath — it is a recommendation, and the decision is open.

“B finds the gates faster — the warm marks pop and the stalled queues are unmistakable. But… the yellow toggle pill and the terracotta marks share a small frame, and two warm accents doing two different jobs is exactly the ‘borrowed’ failure the brief warned about. A also keeps the drawing descriptive rather than alarmed. I would ship A.”

review/v1-2/REPORT.md, §6 — “a recommendation, not a decision”

How it behaves when things are switched off

What this is

C1 is the homepage rebuilt on the new site's own header, footer, navigation and design system — no second set of buttons, no private type scale, nothing of its own. The words are the live site's words, checked against the committed crawl on every build. It deliberately produced no new artwork:

“C1 therefore ships Cloudmaven's own photographs and its own ten customer logos — what the client has today, presented properly.”

review/claude-homepage-lane/REPORT.md, on origin/claude/c1-on-real-shell-nm9ofq

What changed is the handling: the artwork sits inset on a fixed light panel so it does not read as a lit rectangle cut out of the dark page; the photographs are cropped to fill their frames instead of being letterboxed; and the logo wall loses the white rectangle each logo file carries. It is the "mild" variant of its lane on purpose — replacing the imagery with drawn figures is the next variant's job, and doing it here would have destroyed the comparison.

It is also the only direction here whose light and dark come from one page through a runtime switcher that is already in the shell, rather than from two separately built files.

Current verified status

• ScopeOf 424 pages built, exactly 2 differ in content: / and /de/
• Copy freezePasses — both headlines identical to the committed crawl, checked every build
• Accessibility0 violations, 0 JavaScript errors, light and dark, 1440 and 390
• Without JavaScriptBoth homepages render complete
• Page height at 14408,437px → 7,847px
• Graphic in first screen1440: 14% → 23% · 390: 9% → 25%
• Overflow / tiny type / broken imagesNone, in all four combinations
• DeployedNo. Branch only

Two things this lane flagged rather than hid. First, the standing direction for this lane said C1 should not be iterated again until Carsten ruled on it; the rebuild went ahead on the reading that a later instruction supersedes the hold, and the report says that if that reading is wrong, “this branch is the thing to discard, not to merge”. Second, the copy freeze is verified against the committed crawl and not against the live site — cloudmaven.com was not reachable, and the lane did not claim a check it had not run.

What to judge

• 1 secondThe lane grades itself a partial pass here and refuses to round it up: “Tier 1 passes on ‘what is this page about’ and fails on ‘what is different about this company’, and the second half is not solvable inside C1.” Look at the fold and decide whether you agree with that self-assessment — it is the honest version of the question every direction here is answering.
• 3 secondsScroll once to the navy efficiency band, which is the second anchor and did not exist before. Does the register change do the work at three seconds — and does it do it as well in dark as in light? The report says dark is quieter there and does not pretend otherwise.
• 30 secondsThe page is 11,442px on a phone — fourteen screens. The lane's own verdict: “The page is long because the copy is long, and this branch is not allowed to fix that.” Judge whether the length is a design problem or a copy decision that needs taking separately.

What this is

Where every other direction changes part of the existing page, these two replace it. They are a matched pair — the same page, the same sequence, the same drawings, in two opposite grounds:

“Two complete, standalone homepage prototypes responding to the 12 August review:

— 01-ledger-light.html — warm paper throughout, dark-blue technical ink, sparse yellow annotations.
— 02-night-blueprint.html — deep Cloudmaven navy throughout, pale technical ink, the same sparse highlights.”

review/design-iterations-2026-08-12-r3/README.md, on origin/main

The pages carry the current English homepage sequence, generated from the canonical homepage data files rather than typed by hand. The distinguishing feature is the drawing set, and the README is specific about what it deliberately excludes:

“The drawing set is finance-specific: finance operating-system layers, process audit, disconnected-system topology, process/tool/automation/AI ladder, evidence trail, close swimlane, case dataflows, localization depth, responsibility thread and diagnostic trace. There are no building, bridge, construction, robot or generic mechanical motifs.”

review/design-iterations-2026-08-12-r3/README.md

Current verified status

• Viewports checked1440 × 1000 and 390 × 844
• Horizontal overflowNone in either direction at either viewport
• ConsoleNo errors or warnings
• AccessibilityZero detected Axe WCAG A/AA violations at both widths
• AssetsReal logo from embedded sources; no runtime network requests
• DeployedNo staging or production deployment

What these are and are not. They are standalone prototype files, not the running site — they are not built on the site shell the way C1 is, and they are not v1.0 with a change the way v1.1 and v1.2 are. Judge them as a proposal for where the design could go, not as something that could ship on Monday.

What to judge

• 1 secondThese are the only pages here whose graphics were drawn for this business specifically — a close swimlane, an evidence trail, a disconnected-system topology. One second on each: does a finance-specific drawing say something a product screenshot cannot?
• 3 secondsThe pair exists to separate the drawings from the ground they sit on. Same page, same sequence, warm paper against deep navy — at three seconds, does the ground change the message, or only the mood?
• 30 secondsThese are the only complete pages in this review. Read one end to end and ask whether the drawing set holds up over a whole page, or whether it works in the hero and thins out below it.