The checks
The brief
Your post, line by line.
| Your post says | Where it’s answered | Status |
|---|---|---|
| “A network that grades intelligence.” | Every answer shows the provider’s rank, score and confidence, and routing shows the ranked list it picked from. Routing, answer, thesis. | Done |
| “Where humans, agents, robots plug in and query the network.” | Humans: the ask flow. Agents and robots: API keys with their own limits in the wallet. | Done |
| “An Anjoman… a symposium where poets, writers or intellectuals meet.” | The mark is built on it: one face made of the assembly’s light, with a bilingual wordmark, انجمن. Brand. | Done |
| “A feeling that we are in the age of machines… finding graded intelligence and using it.” | The seal: its ring is Telegraph’s 64 validator slots, lit by who signed; its face is as bright as the answer is sure. Brand, verifying. | Done |
| “Miners provide answers permissionlessly and evaluators score the intelligence.” | “Ranked by measured performance. Nobody pays for placement.” The receipt names the evaluator, and the no-provider state offers “Answer it yourself”. No provider. | Done |
| “Complete redesign of the Anjoman website including all internal pages.” | Built deep, not wide: the ask-to-answer flow in every state, plus the wallet, at desktop and phone width. The other routes are mapped in the audit and priced in the proposal. | Partly |
| “You may use these images.” | Used in the hero and launch kit, plus new images generated from yours and graded to match them on scopes. Imagery. | Done |
| “So intuitive, seamless, and easy to navigate.” | One composer, the answer first, the proof one tap away. Tests with people are built but not yet run (below). | Partly |
| “It has to feel like Claude.” | Calm: one composer, the answer first, details in a side panel, a quiet working state. Not Claude’s cream and clay, which the beta already copies. | Done |
| “Have the features like OpenRouter.” | One balance, spend per day, an activity log, keys with limits, and rankings shown in routing and on the receipt. No model picker. Wallet. | Done |
| “The benefits (end result) are explicitly clearer.” | A verdict in one line, then who answered, their rank, how sure, and that it was checked, under every answer. The first screen states the three benefits in plain words. | Done |
| “Its own viral OpenClaw moment.” | “Share this warning” on every scam answer: copy, post or download a 1200 × 630 card made for that answer. Answer, launch kit. | Done |
| “A logo, and icon.” | One direction, drawn as a vector: tested at 16 and 32 px, in one colour, inverted, in X’s avatar circle and in the app header. Mark. | Done |
| “Social media banners.” | X header 1500 × 500, LinkedIn cover 1512 × 256, a 400 × 400 avatar and per-answer share cards, checked in mocks of the real layouts. Launch kit. | Done |
| “A cool oracle type person, a monk or guru.” | A faceless oracle whose face is light. No one bows to it. Thesis. | Done |
| “Send me a message with your price and timeline.” | In the proposal, as milestones. | Done |
“A network that grades intelligence.”
Every answer shows the provider’s rank, score and confidence, and routing shows the ranked list it picked from. Routing, answer, thesis.
Done
“Where humans, agents, robots plug in and query the network.”
Humans: the ask flow. Agents and robots: API keys with their own limits in the wallet.
Done
“An Anjoman… a symposium where poets, writers or intellectuals meet.”
The mark is built on it: one face made of the assembly’s light, with a bilingual wordmark, انجمن. Brand.
Done
“A feeling that we are in the age of machines… finding graded intelligence and using it.”
The seal: its ring is Telegraph’s 64 validator slots, lit by who signed; its face is as bright as the answer is sure. Brand, verifying.
Done
“Miners provide answers permissionlessly and evaluators score the intelligence.”
“Ranked by measured performance. Nobody pays for placement.” The receipt names the evaluator, and the no-provider state offers “Answer it yourself”. No provider.
Done
“Complete redesign of the Anjoman website including all internal pages.”
Built deep, not wide: the ask-to-answer flow in every state, plus the wallet, at desktop and phone width. The other routes are mapped in the audit and priced in the proposal.
Partly
“You may use these images.”
Used in the hero and launch kit, plus new images generated from yours and graded to match them on scopes. Imagery.
Done
“So intuitive, seamless, and easy to navigate.”
One composer, the answer first, the proof one tap away. Tests with people are built but not yet run (below).
Partly
“It has to feel like Claude.”
Calm: one composer, the answer first, details in a side panel, a quiet working state. Not Claude’s cream and clay, which the beta already copies.
Done
“Have the features like OpenRouter.”
One balance, spend per day, an activity log, keys with limits, and rankings shown in routing and on the receipt. No model picker. Wallet.
Done
“The benefits (end result) are explicitly clearer.”
A verdict in one line, then who answered, their rank, how sure, and that it was checked, under every answer. The first screen states the three benefits in plain words.
Done
“Its own viral OpenClaw moment.”
“Share this warning” on every scam answer: copy, post or download a 1200 × 630 card made for that answer. Answer, launch kit.
Done
“A logo, and icon.”
One direction, drawn as a vector: tested at 16 and 32 px, in one colour, inverted, in X’s avatar circle and in the app header. Mark.
Done
“Social media banners.”
X header 1500 × 500, LinkedIn cover 1512 × 256, a 400 × 400 avatar and per-answer share cards, checked in mocks of the real layouts. Launch kit.
Done
“A cool oracle type person, a monk or guru.”
A faceless oracle whose face is light. No one bows to it. Thesis.
Done
“Send me a message with your price and timeline.”
In the proposal, as milestones.
Done
Instruments
Run on the built site, the same files that are deployed.
- Impeccable detector, in a real browser
- 10 routes at 1440 and 390 px: 0 findings on 8 of 10. On the not-sure and no-provider states it flags the seal (64 points and a hood) as an illustration built from shapes. It is the logo glyph, so it stays.
- Taste-skill lint, on the source
- 0 failures. 19 lines carry a written exception: real table rows, floating overlays (menus, sheets, tooltips), the scrim under the hero headline, and monospace kept for hashes and keys.
- Taste-skill inspector, presentation pages
- 0 failures on brand, thesis, audit and before/after at 1440, 768 and 390 px. Warnings left: Bone text reads as “cream” to the meter, there is no video, the brand board shows 12 type sizes because it reproduces other products’ layouts, and three typefaces, with mono only for machine strings.
- axe-core, WCAG 2.2 AA
- 11 routes, night and day themes, 1440 and 390 px: 0 violations, after fixing one. The wallet’s sideways-scrolling tables weren’t keyboard-focusable on a phone.
- Link previews
- Every answer has its own page, so a shared link unfurls with that answer’s card. Checked on the live site in a preview renderer (opengraph.xyz): 0 errors. It caught a real problem: X lays the page title over the bottom left of the card, right on top of the proof line. The facts moved up and the bottom 160 px stays clear now.
- Lighthouse
- Accessibility 100 on all four pages measured. Best practices 100 on mobile and 96 on desktop. SEO 60 to 63, because the spec is set to noindex on purpose. Performance 98 to 100 on desktop (LCP 0.8 to 1.1 s). On mobile, Lighthouse’s simulated slow 4G gives 71 to 86 with LCP 3.7 to 6.8 s; with applied throttling the home page scores 93 (LCP 2.5 s) and the brand board 77 (TBT 0.75 s). The gap is JavaScript: about 218 KB unused, which server components would cut in the real build.
Tests with people
Built for at least five people outside our studio, comparing the live beta with the redesign. Each test runs on a phone or a laptop and gives back a block of answers to paste to us.
- Five-second test →Both home screens, in random order, five seconds each. Then: what is it for, what would you do first, one word for how it felt.
- First-click test →Two tasks on both versions: check a scam text, and find who answered and how sure they are. Records the first click and the time.
- Blind image test →Your six Midjourney images mixed with four new ones. Mark the pictures that don’t belong.
Results so far: none. Not one of the five people per test has taken them yet.
Not done yet
Reported as plainly as what passed.
- The five-second and first-click tests, and the blind image test, need at least five people outside our studio. The kits below are live; none has been run yet, so there are no results to report.
- Mobile LCP is above 2.5 s in Lighthouse’s simulated profile on every page measured.
- Two scope metrics land just outside your references’ range: shadow warmth (kept a little cool for night scenes) and midtone level (0.193 against a floor of 0.194).
- Your Midjourney profile wasn’t available, so new images were generated in FLORA from your images and graded to match, not made in your exact house style.
- Second opinion, changing the job type, scanning the link, keeping a job and registering a provider say what they would do; this spec replays only the four real jobs.
- The link cards haven’t been posted on X, LinkedIn or Slack themselves yet; they were checked in a preview renderer (above).
- Only the core flow and the wallet are designed. The other beta routes are mapped, not redrawn.