From fa28d15c0a7e738587f8cc034438dda133ee722e Mon Sep 17 00:00:00 2001 From: Dmitrii Vasilev Date: Sat, 8 Aug 2026 19:06:12 +0700 Subject: [PATCH] The whole header was dead on six pages, and the homepage sold something else MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Two problems, both reported by the owner and both worse than they looked. **The header did nothing on any routed page.** Its nine links point at sections of the one-page scroll. On /verification, /proof, /ip, /course, /cases and /about those sections do not exist, and the handler called preventDefault and then scrollIntoView on null — so every link was silently inert. Confirmed by asking the page: all nine ids missing, hash unchanged after a click, scrollY still 0. They now go home first and poll for the target until the homepage has mounted, rather than guessing a delay. The footer's section links had a subtler version of the same bug: under HashRouter a bare "#theorems" reads as a route, so clicking one from a routed page dumped the reader on the homepage without scrolling to what they clicked. Same handler, same fix. **The homepage still sold the previous business.** Its position was "GPU-Verified Green AI Economy", its headline metrics were 298K tokens/s on an RTX 3090, and its two calls to action were "Install CLI" — which installed nothing, it scrolled to a section — and "Read Papers". Nothing on the page led to anything that can be bought, and nothing mentioned silicon. The hero now states the position that is actually evidenced: RTL measured on real silicon, bit-exact against a model written from the specification, on a toolchain with no vendor licence in it. The three headline figures are ones I can show the working for — 170,068 cycles with zero mismatches on a third-party design, the SKY130 tape-out, and zero vendor licences. Primary action goes to the service, secondary to the evidence, because that is what a sceptical reader wants next. The φ identity stays as the headline: it is the project's name, and the component special-cases it. Left alone deliberately: the theorems, publications, DePIN and invest sections further down. They belong to a fundraising pitch rather than a services page, and whether to keep them is the owner's call, not mine to make while he is asleep. Co-Authored-By: Claude Opus 5 --- apps/website/messages/en.json | 58 +++++++++---------- apps/website/messages/ru.json | 58 +++++++++---------- apps/website/src/components/Footer.tsx | 25 ++++++-- apps/website/src/components/Navigation.tsx | 22 +++++++ .../src/components/sections/HeroSection.tsx | 18 +++--- 5 files changed, 112 insertions(+), 69 deletions(-) diff --git a/apps/website/messages/en.json b/apps/website/messages/en.json index 5470b7cf10..8e05f81bb1 100644 --- a/apps/website/messages/en.json +++ b/apps/website/messages/en.json @@ -14,57 +14,57 @@ "Invest" ], "hero": { - "tag": "GPU-Verified Green AI Economy", - "badge_publications": "7 PUBLICATIONS", + "tag": "Hardware-verified RTL · measured on real silicon", + "badge_publications": "2 arXiv PAPERS", "badge_doi": "DOI-VERIFIED", "headline": "φ² + 1/φ² = 3", - "subheadline": "298K tokens/s on RTX 3090 | 274K on A100 | VERIFIED", - "quote": "\"Not a claim — a theorem. Not a promise — a proof. Not simulated — GPU verified.\"", - "desc": "First AI platform with 4 mathematical theorems + GPU-verified benchmarks: 298K tokens/s (RTX 3090), 70% noise robustness at 30% corruption", + "subheadline": "Bit-exact against an independent model · measured on a live Artix-7 · no vendor licence anywhere in the flow", + "quote": "\"Not simulated — measured. Not asserted — reproducible by anyone, for free.\"", + "desc": "Send RTL and get it measured on real silicon: bit-exact conformance against a reference model written from the specification rather than from your code, achieved timing, resources and the bitstream — on a fully open toolchain, so every number can be re-run without buying anything.", "metrics": [ { - "value": "298K", - "label": "Tokens/s (RTX 3090)", - "proof": "RunPod GPU verified" + "value": "170,068", + "label": "cycles checked, 0 mismatches", + "proof": "On a third-party design, 8 Aug 2026" }, { - "value": "274K", - "label": "Tokens/s (A100)", - "proof": "RunPod GPU verified" + "value": "SKY130", + "label": "Tape-out completed", + "proof": "Tiny Tapeout — GDS, gate-level, precheck" }, { - "value": "70%", - "label": "Noise @30%", - "proof": "Theorem 1 + GPU test" + "value": "0", + "label": "Vendor licences needed", + "proof": "Yosys · nextpnr-xilinx · iverilog" } ], "newResults": { - "title": "GPU-Verified Results (RunPod)", + "title": "Measured, not estimated", "items": [ { - "value": "298K", - "label": "RTX 3090 tokens/s", - "source": "runpod_full_tests_report.md" + "value": "170,068", + "label": "Cycles bit-exact on a client design", + "source": "0 mismatches, ACC_W = 32 / 24 / 16" }, { - "value": "274K", - "label": "A100 80GB tokens/s", - "source": "e2e_repeat_report.md" + "value": "112 MHz", + "label": "Post-route on XC7A200T", + "source": "nextpnr-xilinx, sequential MAC" }, { - "value": "70.2%", - "label": "Noise Robustness @30%", - "source": "Both GPUs identical" + "value": "100%", + "label": "Held-out accuracy, training on-chip", + "source": "Backward pass in RTL, no host" }, { - "value": "293W", - "label": "A100 Power (vs 348W RTX)", - "source": "nvidia-smi verified" + "value": "0", + "label": "DSP48 blocks in the GF16 matmul", + "source": "32,252 LUTs, fabric-only mapping" } ] }, - "cta": "See the Proofs", - "ctaSecondary": "Run a Node" + "cta": "Verify my RTL", + "ctaSecondary": "See the evidence" }, "navExtra": { "dashboard": "Dashboard", diff --git a/apps/website/messages/ru.json b/apps/website/messages/ru.json index 554e5271dc..f4e1058ed2 100644 --- a/apps/website/messages/ru.json +++ b/apps/website/messages/ru.json @@ -14,57 +14,57 @@ "Инвестиции" ], "hero": { - "tag": "Доказанная Экономика Green AI", - "badge_publications": "7 ПУБЛИКАЦИИ", - "badge_doi": "DOI-ПОДТВЕРЖДЕНО", + "tag": "Верификация RTL на живом кремнии", + "badge_publications": "2 СТАТЬИ arXiv", + "badge_doi": "ПОДТВЕРЖДЕНО DOI", "headline": "φ² + 1/φ² = 3", - "subheadline": "4 Теоремы + Верифицированные Результаты = Доказанная Ценность", - "quote": "\"Не заявление — теорема. Не обещание — доказательство. Не симуляция — проверено на GPU.\"", - "desc": "Первая AI платформа с 4 математическими теоремами + верифицированными бенчмарками: 20× сжатие памяти (проверено на CPU), до 3000× энергоэффективность (проекция FPGA)", + "subheadline": "Побитово против независимой модели · измерено на живом Artix-7 · ни одной вендорской лицензии во всём флоу", + "quote": "«Не симуляция — измерение. Не утверждение — воспроизводимость, бесплатно и любым желающим.»", + "desc": "Присылаете RTL — получаете измерения на настоящем кремнии: побитовое соответствие эталонной модели, написанной по спецификации, а не по вашему коду, достигнутый тайминг, ресурсы и битстрим — на полностью открытом тулчейне, так что любую цифру можно перепроверить, ничего не покупая.", "metrics": [ { - "value": "15.7×", - "label": "Сжатие", - "proof": "Теорема 3 + ternary_quant_report" + "value": "170 068", + "label": "циклов проверено, 0 расхождений", + "proof": "На чужом дизайне, 8 авг 2026" }, { - "value": "8.1×", - "label": "Ускорение GPU", - "proof": "Теорема 2 + GPU benchmarks" + "value": "SKY130", + "label": "Тейпаут пройден", + "proof": "Tiny Tapeout — GDS, вентильный тест, precheck" }, { - "value": "100%", - "label": "Устойчивость к шуму", - "proof": "Теорема 1 + hdc_double_q_report" + "value": "0", + "label": "вендорских лицензий нужно", + "proof": "Yosys · nextpnr-xilinx · iverilog" } ], "newResults": { - "title": "Последние верифицированные результаты", + "title": "Измерено, а не оценено", "items": [ { - "value": "15.7×", - "label": "HDC сжатие", - "source": "ternary_quant_report.md" + "value": "170 068", + "label": "циклов побитово на клиентском дизайне", + "source": "0 расхождений, ACC_W = 32 / 24 / 16" }, { - "value": "43%", - "label": "Бонус разреженности", - "source": "ternary_quant_report.md" + "value": "112 МГц", + "label": "post-route на XC7A200T", + "source": "nextpnr-xilinx, последовательный MAC" }, { "value": "100%", - "label": "Устойчивость при 20% ошибок", - "source": "hdc_double_q_report.md" + "label": "на отложенной выборке, обучение на кристалле", + "source": "Обратный проход в RTL, без хоста" }, { - "value": "8.1×", - "label": "SIMD ускорение", - "source": "GPU benchmarks" + "value": "0", + "label": "блоков DSP48 в матмуле GF16", + "source": "32 252 LUT, только логика" } ] }, - "cta": "Смотреть доказательства", - "ctaSecondary": "Запустить ноду" + "cta": "Проверить мой RTL", + "ctaSecondary": "Посмотреть доказательства" }, "navExtra": { "dashboard": "Панель", diff --git a/apps/website/src/components/Footer.tsx b/apps/website/src/components/Footer.tsx index ded3f8cc2c..fd2a1102c2 100644 --- a/apps/website/src/components/Footer.tsx +++ b/apps/website/src/components/Footer.tsx @@ -3,6 +3,23 @@ import { motion } from 'framer-motion' import { Link } from 'react-router-dom' import { useI18n } from '../i18n/context' +// Under HashRouter a bare `#section` is read as a route, so these links used to +// dump the reader on the homepage without scrolling to what they clicked. This +// sends them home when needed and then finds the section once it exists. +function goToSection(e: React.MouseEvent, id: string) { + e.preventDefault() + const hash = window.location.hash + const onHome = hash === '' || hash === '#' || hash === '#/' + if (!onHome) window.location.hash = '#/' + let tries = 0 + const findIt = () => { + const el = document.getElementById(id) + if (el) { el.scrollIntoView({ behavior: 'smooth' }); return } + if (++tries < 40) requestAnimationFrame(findIt) + } + requestAnimationFrame(findIt) +} + export default function Footer() { const { t } = useI18n() @@ -59,10 +76,10 @@ export default function Footer() {