ClearWater

Настраиваемый фильтр мусорной информации — оценивает реальное содержание текста и кода.

ClearWater — see through AI slop, manipulation, and filler The web is filling with content that looks substantive but isn't: AI-generated padding, sponsored "research" framed as analysis, statistics with no source, manipulation dressed as authority. ClearWater is a configurable filter that evaluates the real content of text and code on any page — block by block — so you can tell what is real from what is noise. What it does Highlight any text on any page (article, blog, doc, GitHub README, anywhere), or let ClearWater auto-detect the main article. It runs a multi-criteria check that judges each block on: • Factual accuracy — claims cross-checked against live web sources • Logical structure — coherence, conclusions, contradictions • Manipulation — emotional weighting, false dichotomy, dark-pattern framing • Templatedness — AI-generated padding, formulaic filler • Hollow content — words without substance, vapor authority • Goal-relative omissions — what was left out that mattered The result is a 0.0–1.0 Truth value per block plus a one-page summary verdict. Read more confidently, faster. How it works For each block of text, ClearWater calls one or more Gemini models — Flash-Lite for speed and value, Gemini 3.5 Flash for balanced cost and depth, or Gemini 3.1 Pro for maximum quality — runs a custom multi-criteria procedure, and returns a structured judgment with per-criterion flags. You see exactly which parts of the text passed, which raised flags, and why. For maximum-quality checks (Magnum Opus mode), ClearWater runs a deeper two-layer combo: per-block fact-check plus a contradiction graph that catches inconsistencies between blocks. A claim in paragraph 1 contradicting a claim in paragraph 5 only surfaces in a multi-block analysis — and ClearWater does it. Key features • Block-by-block markup — the verdict is painted right on the page; no hidden judgments • Reasoning Mode (Low / Medium / High) — choose how deeply each block is reasoned about • Three Gemini models — Flash-Lite, 3.5 Flash, 3.1 Pro — pick per check • Logic-only mode — skip the web fact-check; judge only the structure of the argument (faster, cheaper) • Auto-detect mode — finds the main article text automatically, no manual selection needed • Saved per-site selectors — set it once for your daily news site; future visits target the right block automatically • Trust Index — a community-shared trust map across the internet, with per-page badges and an Obsidian-style graph view. See which domains carry verified content vs noise • Page-load verdict badge — a transient pill appears top-right of any page that has been previously checked by the community • History — every check is kept locally, browsable per-page or as a flat list • Mode switcher — Text articles vs Code (different criteria sets, optimized for each) Pricing Free tier (no account needed): 3 deep checks on the Flash-Lite model per day, with 1 new check regenerating daily (cap 3). Enough for daily reading; never expires. Paid tier (server-issued account key, no email required): one-time top-ups via PayPal or cryptocurrency — Bitcoin, Solana, Ethereum and 6 other EVM chains, Monero. Privacy-respecting rails: Monero is fully anonymous; the BTC/EVM path uses self-hosted BTCPayServer (no third-party custody of your funds during payment); PayPal is the convenience rail. Paid checks unlock: • Premium models (Gemini 3.5 Flash, 3.1 Pro) • Medium / High Reasoning Mode • Magnum Opus deep two-layer check Privacy Zero retention. The server does not store the text you check after the verdict is computed. The check is computed, returned to you, and the input text is dropped — not logged, not analyzed for training, not retained at all. Only operational metadata (token counts, mode used, cost) is recorded for cost monitoring; no characters of your input are persisted. Your account key is server-issued and account-restorable (no email required). PayPal and crypto payments do not link to your reading history. How to start 1. Install the extension (one click below). 2. Pin it to your toolbar (top-right Chrome icon). 3. Visit any article or code page. 4. Click the ClearWater icon → Check. 5. Read the per-block verdict on the page. The first 3 checks are free, every day. No card needed to start. Do mind. ClearWater is an engine, not an oracle. It evaluates the structural and factual properties of text using current best-available LLM models and surfaces uncertainty explicitly rather than hiding it. Final judgment is yours — ClearWater is a lens that helps you read more clearly, not a verdict you obey. The extension shell is permissioned conservatively (activeTab + scripting + storage); the host-permission grant exists so highlight and markup can paint on whatever page you are reading. Source code paths align with this — what the extension can see is what the extension can do, no hidden network channels. Built with privacy, configurability, and honest reasoning as core values.