SourceScore

Changelog

Every shipped feature, catalog expansion, and methodology update on the SourceScore VERITAS API. Reverse-chronological.

feature

Batch 30 → 346 — search APIs + AI-search infra + 2025 launches

10 hand-verified claims expanding search-infra + AI-grounding tooling + 2025 frontier: Firecrawl (Mendable AI 2024-04-15 — web scraping → LLM-ready markdown), Exa (founded 2021 by Bryk + Wang, formerly Metaphor — AI-native embedding search), Jina AI (founded 2020 by Han Xiao — multimodal infra + Reader API), Brave Search API (Brave Software 2022-03-22 — independent index for AI grounding), Kagi Search (founded 2018 by Prelovac — paid private search + FastGPT API), OpenRouter (founded 2023 by Atallah — unified API gateway 100+ providers), Recraft (founded 2022 by Dorogush — brand-consistent text-to-image + vector), Google Gemini 2.5 Flash (Google DeepMind 2025-04-09 — controllable thinking budget + 1M context), ChatGPT Edu (OpenAI 2024-05-30 — university-tier with GPT-4o), Anthropic Claude Haiku 4.5 (Anthropic 2025-10-16 — fast tier 4.5 family). Coverage strengthens: AI-search infra (Firecrawl + Exa + Jina + Brave Search + Kagi), unified LLM routing (OpenRouter), 2025 hybrid-reasoning (Gemini 2.5 Flash + Haiku 4.5), education-tier (ChatGPT Edu). Bulk 336 → 346 catalog count sync. tags.json now indexes 770 unique tags across 346 claims.

feature

Batch 29 → 336 — enterprise infrastructure + 2024-2025 launches

10 hand-verified claims expanding enterprise infrastructure + AI safety + consumer AI: xAI Colossus (xAI 2024-09-02 — 100k H100 supercomputer Memphis, TN), Hugging Face Inference Endpoints (HF 2022-09-27 — managed inference deployment), Microsoft Copilot (Microsoft 2023-09-21 — consumer AI assistant launched from Bing Chat), Apple Foundation Models (Apple 2024-06-10 — WWDC reveal, on-device + Private Cloud Compute), Meta Movie Gen (Meta AI 2024-10-04 — 30B text-to-video + 13B audio research preview), NVIDIA Cosmos (NVIDIA 2025-01-06 — World Foundation Models platform for physical AI / robotics / AV training), IBM Granite 3 (IBM 2024-10-21 — Apache 2.0 enterprise family), NVIDIA Llama Nemotron (NVIDIA 2025-03-18 — reasoning-tuned Llama for agentic workflows), Anthropic Constitutional Classifiers (Anthropic 2025-02-04 — jailbreak-defense input/output filters), Microsoft AutoGen (Microsoft Research 2023-09-25 — open-source multi-agent framework, used by Magentic-One). Coverage strengthens 2024-2025 enterprise infrastructure (Colossus, Cosmos, Inference Endpoints), AI safety (Constitutional Classifiers), consumer AI (Microsoft Copilot, Apple Foundation Models), multi-agent (AutoGen + Magentic-One). Bulk 326 → 336 catalog count sync. tags.json indexed 741 unique tags across 336 claims (pre-Batch 30).

feature

sameAs entity-coherence — Organization + Person schemas link to GitLab + Dev.to

Aleyda Solis 10-char #3 Recognizable + #7 Credible compound. Added `sameAs` array to: (1) Organization schema in /about/ (the canonical org reference), (2) Organization schema in layout.tsx site-wide JSON-LD, (3) editorialPersonSchema Person entity in /about/. Linked surfaces: gitlab.com/acevault-lab (primary git host post-2026-05-06 per Julian/accounts.md) + dev.to/paulomdevries (active 2026-05-16, canonical cross-post host). Borrowed from holdlens fleet pattern (holdlens/about page has Person+sameAs). LLM citation gravity compound: AI models that verify entity identity across multiple platforms before citing now have explicit sameAs verification path. Verified 3 sameAs JSON-LD blocks rendered in out/about/index.html.

feature

Resource hints in <head> — preconnect + dns-prefetch (fleet pattern from readstacks)

Added 5 resource hints to <head> per readstacks fleet pattern: preconnect + dns-prefetch to pagead2.googlesyndication.com (AdSense, on every page), dns-prefetch to www.clarity.ms (Microsoft Clarity), plausible.io (analytics), googleads.g.doubleclick.net (AdSense fill). Cost: 4 cheap DNS lookups on initial page load. Benefit: ~100-300ms faster first-contentful paint when the script/image actually fires. Direct Core Web Vitals compound (LCP/FCP improvement signals Page Experience to Google + Lighthouse PWA score). Verified: 5 resource hints rendered in /out/index.html post-build. Pure additive; no regression risk.

feature

app/manifest.ts — PWA install + Lighthouse PWA score (fleet pattern from readstacks)

Next.js 15.1+ MetadataRoute manifest.ts adopted from readstacks fleet pattern. Adds installability signal — Chrome desktop install ribbon, iOS standalone mode, Lighthouse PWA score, Android home-screen-add. Modest GEO compound: Google Lighthouse PWA score factors into Page Experience; AI crawlers use manifest.json for some entity recognition. Generates /manifest.webmanifest at build time (static export friendly). Theme: SourceScore brand colors (#0a0a0a) + favicon.svg as icon. Display mode: standalone (true PWA). Build verified; emitted as /manifest.webmanifest in out/.

feature

Fleet pattern absorption — GEO compound (llms.txt + .well-known + head alternate links)

Cross-project SEO/GEO/AEO pattern absorption per `rules/cross-project-learning.md`. Three high-leverage patterns lifted from fleet siblings: (1) **llms.txt enriched** with txtfeed-style sections — explicit Permitted block (11 crawlers + rationale per crawler — GPTBot for ChatGPT training, ClaudeBot for Claude citation, etc.), Restricted block (3 paths — /api/v1/verify POST, /embed/*, /og/*), and Preferred Citation block (5 idiomatic citation examples: source-score, methodology, claim-verification, comparison, category). llms.txt grew 142 → 184 lines. Direct GEO compound — AI crawlers see attribution clarity + crawler-by-crawler permission. (2) **/.well-known/llms.txt + /.well-known/ai-sitemap.xml** Cloudflare Pages 200-redirect added (not 301 — serves identical content). IETF emerging-track URL pattern that LLM crawlers probe before /llms.txt fallback. Borrowed from txtfeed fleet GEO pattern. (3) **<head> alternate links** added — RSS feed discovery for blog + claims feed (RSS readers + LLM crawlers auto-detect); JSON twin discoverability via /api/sources.json + /api/v1/claims.json + /api/v1/openapi.json (LLM crawlers follow Aleyda Solis 10-char #4 Extractable; prefer structured JSON). Borrowed from readstacks fleet GEO pattern. All three changes purely additive. Sourcescore (3,840 AI crawls/30d — heaviest bot-traffic site in fleet per LEARNED.md) now signals attribution clarity + provides multiple structured-data access paths per LLM-citation 10-char checklist.

feature

Editor @id chain across all 7 use-cases + Organization @id consistency

Schema entity-coherence pass: added editor @id chain to the 5 use-cases that lacked it (ai-agent-grounding, research-citation, customer-support-bot, rag-pipeline-verification, content-moderation). All 7 use-cases now reference Person #person-editorial-lead from /about/ as editor + Organization @id #organization. Pattern matches the existing chain on /concepts/[X]/ pillars + /topics/[slug]/ hubs + /blog/[slug]/ posts + /comparisons/[slug]/ pages. Entity-coherence is now fleet-wide: every TechArticle/BlogPosting points to the same Person + Organization @id graph. Aleyda Solis 10-char #3 Recognizable + #7 Credible + #8 Differentiated — LLMs see one consistent entity, not fragmented mentions.

feature

Batch 28 → 326 — enterprise open-weight + 2025 frontier expansion

10 new hand-verified claims expanding enterprise open-weight + 2025 frontier: Qwen 3 (Alibaba 2025-04-29 — hybrid thinking mode 0.6B-235B), Magentic-One (Microsoft Research 2024-11-04 — generalist multi-agent on AutoGen), Microsoft Phi-4 Multimodal (Microsoft 2025-02-26 — 5.6B with speech + vision + text), Snowflake Arctic (Snowflake 2024-04-24 — 480B dense-MoE hybrid, 17B active, Apache 2.0), Databricks DBRX (Databricks 2024-03-27 — 132B MoE / 36B active), Reka Core (Reka AI 2024-04-15 — multimodal text/image/audio/video), Liquid AI LFM (Liquid AI 2024-09-30 — non-Transformer foundation models 1B/3B/40B), Mistral Medium 3 (Mistral 2025-05-07 — enterprise-tier 8× cost-efficient), Allen AI Molmo (Allen AI 2024-09-25 — fully-open VLM family 1B/7B/72B Apache 2.0), Replit Ghostwriter (Replit 2022-10-26 — early AI pair-programmer, renamed Replit AI 2023). Coverage strengthens: 2024-2025 enterprise open-weight (Arctic + DBRX + Qwen 3 + Molmo + LFM), multi-agent (Magentic-One), novel architectures (Liquid AI non-Transformer), enterprise multimodal (Phi-4-multimodal + Reka Core). Bulk 316 → 326 catalog count sync. tags.json now indexes 721 unique tags across 346 claims.

feature

/blog/multi-llm-grounding-2026/ — 6th blog post (provider-portable architecture)

6th blog post — multi-LLM grounding architecture pattern. Covers why single-provider lock-in is fragile in 2026 (pricing variance 5-10×, capability gaps shift quarterly, outages routine, regulatory zones, open-weight quality crossed the line), the 3-layer pattern (Router → Adapter → Grounding), Python skeleton code routing across OpenAI + Anthropic + Gemini with provider-agnostic VERITAS grounding, adapter library landscape (Vercel AI SDK, DSPy, LangChain, Instructor, LiteLLM, OpenRouter), 5 production routing rules (task type → model, user tier → cost, latency SLA → streaming, outage → failover, regulatory → compliant provider), why grounding-layer portability matters (provider-locked grounding disappears on failover), when to combine provider-native + portable (user-doc RAG via Citations API + shared facts via VERITAS). BlogPosting + BreadcrumbList schema with editor @id chain. Cross-links to /comparisons/veritas-vs-anthropic-citations + /blog/llm-grounding-strategies-2026 + /blog/llm-framework-comparison-2026 + /concepts/llm-grounding + /topics/llm-releases-2024-2025. RSS feed regenerated (4 posts). Targets queries: 'multi-LLM grounding', 'switch LLMs production', 'LLM provider portability', 'OpenAI Anthropic Gemini router'. Blog count: 5 → 6.

feature

/comparisons/veritas-vs-anthropic-citations/ — 4th head-to-head (direct competitor buyer intent)

4th /comparisons/[X]/ — direct head-to-head with Anthropic Citations API (launched 2025-01-23). At-a-glance comparison table (released date · source of truth · cite-format · provider lock · verifiability · external citability · pricing · latency · scope), explainer of what each does, decision matrix (when to use Anthropic Citations vs VERITAS vs both), explicit honest 'what VERITAS does NOT do' + 'what Anthropic Citations does NOT do' sections. Verdict: complementary not competitive — Anthropic Citations API solves user-doc-RAG citation problem within Claude API; VERITAS solves shared-knowledge-base + cryptographic-provenance + multi-LLM problem. Most production AI products use both. TechArticle + BreadcrumbList schema with editor @id chain. Targets queries: 'VERITAS vs Anthropic Citations API', 'alternative to Anthropic Citations', 'multi-LLM grounding API', 'externally citable LLM citations'. /comparisons/ index 3 → 4 head-to-heads.

feature

/topics/voice-and-audio-ai/ — 14th topic hub (Whisper + ElevenLabs + Suno + Stable Audio)

14th /topics/[X]/ hub — voice and audio AI catalog. CollectionPage schema groups catalog claims covering speech recognition (Whisper, Whisper large-v3), text-to-speech (ElevenLabs), music generation (Suno v3/v4, Stable Audio 2.0), voice-emotion (Hume AI), and end-to-end voice agents (ElevenLabs Conversational AI). 4 editorial sections (voice + audio is the next-most-important modality after text, the 3-layer audio stack ASR/TTS/audio-generation, why voice agents are the 2025 frontier, why this catalog matters for voice-agent verification) + 6 DefinedTerms (ASR · TTS · Voice agent · Whisper · Suno · ElevenLabs). Cross-links to multimodal-ai + llm-releases-2024-2025 + /concepts/{multimodal, llm-grounding} + openai-tools + anthropic-sdk integrations. Targets queries: 'voice AI', 'AI voice agent', 'best text-to-speech 2025', 'Suno vs Udio', 'Whisper alternatives', 'ElevenLabs Conversational AI'. 13 → 14 topic hubs.

feature

Batch 27 → 316 — 2025 model releases + emerging architectures

10 new hand-verified claims expanding 2025 frontier + emerging architectures: xAI Grok 4 (xAI 2025-07-09 — Grok 4 + Grok 4 Heavy multi-agent), Anthropic Claude Memory (Anthropic 2025-04-15 — Claude.ai Memory beta), Mistral OCR (Mistral 2025-03-06 — document-understanding OCR with table + math extraction), OpenAI Sora 2 (OpenAI 2025-09-30 — text-to-video with synchronized audio), Google Imagen 4 (Google DeepMind 2025-05-20 — improved typography rendering), Inception Labs Mercury (Inception Labs 2025-02-26 — first commercial diffusion LLM, 10× faster than autoregressive), Anthropic Files API (Anthropic 2025-03-25 — file upload + reference API for Claude), Stability AI Stable Audio 2.0 (Stability AI 2024-04-03 — long-form text-to-music 3min tracks), Cohere Aya Vision (Cohere For AI 2025-03-04 — open-weight multilingual VLM 23 languages), Glean (founded 2019 by Jain + Vishwanath + Gentilcore + Prahladka — enterprise AI work assistant). Coverage strengthens: 2025 frontier reasoning + multi-agent (Grok 4 + Magistral + o3/o4-mini), emerging architectures (Mercury diffusion LLM), video gen (Sora 2 audio-sync), document AI (Mistral OCR), enterprise platforms (Glean), multilingual VLM (Aya Vision). Bulk 306 → 316 catalog count sync across app/**/*.tsx + content/**/*.md + openapi.json. tags.json now indexes 703 unique tags across 346 claims.

feature

Batch 26 → 306 — 2025 frontier + video gen + new modalities

10 new hand-verified claims expanding 2025 frontier + video-gen + emerging modalities: Anthropic Skills (Anthropic 2025-10-16 — curated capability packs for Claude), ChatGPT Atlas (OpenAI 2025-10-21 — AI-native web browser), Mistral Magistral (Mistral 2025-06-10 — first Mistral reasoning model, Magistral Small open-weight + Medium API), Black Forest Labs Flux.1 Kontext (BFL 2025-05-29 — in-context image editing with text + image input), Cohere Command A (Cohere 2025-03-13 — flagship enterprise model with 256k context + agentic focus), Stability AI Stable Video 4D (Stability AI 2024-07-24 — first 4D = 3D+time novel-view synthesis from single video), Tencent Hunyuan Video (Tencent 2024-12-04 — 13B open-weight text-to-video), ByteDance Doubao 1.5 Pro (ByteDance 2025-01-22 — MoE matching GPT-4o on Chinese benchmarks), Krea AI (founded 2023 by Rodriguez + Perez — real-time AI image generation tools), ElevenLabs Conversational AI (ElevenLabs 2024-11-19 — voice-to-voice agent platform combining ASR + LLM + TTS). Coverage strengthens: 2025 frontier (Anthropic Skills + ChatGPT Atlas + Magistral + Command A + Doubao) + video generation (Stable Video 4D + Hunyuan Video) + new modality (ElevenLabs Conversational AI voice-agent). Bulk 296 → 306 catalog count sync across app/**/*.tsx + content/**/*.md + openapi.json.

feature

/use-cases/developer-copilot/ — 7th use-case (Cursor/Windsurf/Bolt/Lovable/v0 grounding)

7th /use-cases/[X]/ — buyer-intent use-case for the 2024+ wave of AI coding tools (Cursor, Windsurf, Continue, Bolt.new, Lovable, Vercel v0, Microsoft Copilot Studio, Claude Code, GitHub Copilot, Codeium). Covers 5 common hallucination failures (invented package names / slopsquatting, hallucinated API signatures, fabricated config flags, wrong release dates, mis-attributed paper citations), 3 integration patterns (post-generation IDE verification, server-side pre-suggestion gate, sidebar reference panel with citation badges), what VERITAS catches (model spec claims + paper authorship + org facts + framework release dates + license facts) vs what to handle separately (function-signature hallucinations need type-checker integration; slopsquatting needs registry-side checks; code bugs need SAST). TypeScript code snippet for Continue/Cursor extension integration. Economics fitted to coding-tool scale (Free / Startup €99 / Scale €499). TechArticle + BreadcrumbList schema with editor @id chain. /use-cases/ index now 6 → 7 patterns.

feature

Batch 25 → 296 — AI coding assistants + 2024-2025 developer tools

10 new hand-verified claims expanding AI coding tools + developer-platform ecosystem: Bolt.new (StackBlitz 2024-09-16 — AI web app builder), Lovable (founded 2023 by Anton Osika — AI app builder evolved from GPT-Engineer, Stockholm), Windsurf (Codeium 2024-11-13 — AI-native IDE forked from VSCode), Anthropic Claude Sonnet 4.5 (Anthropic 2025-09-29 — production model with extended-thinking), NotebookLM (Google Labs 2023-07-12 — AI research assistant grounding in source material), Suno v3 (Suno 2024-03-21 — full 2-minute music generation), Microsoft Copilot Studio (Microsoft 2023-11-15 — low-code enterprise AI copilot builder), Continue.dev (Continue Dev Inc 2023-07-26 — open-source AI coding assistant for VS Code + JetBrains), Vercel v0 (Vercel 2023-10-31 — generative UI tool for React + Tailwind), Google AI Studio (Google 2023-12-13 — developer playground for Gemini API). Coverage strengthens: AI coding tools (Bolt + Lovable + Windsurf + Continue + v0 + Copilot Studio + Google AI Studio) + grounding tools (NotebookLM) + music gen (Suno v3) + frontier model (Claude Sonnet 4.5). Bulk 286 → 296 catalog count sync across app/**/*.tsx + content/**/*.md + openapi.json. tags.json indexed 662 unique tags across 296 claims.

feature

/concepts/context-window/ — 12th concept pillar (RoPE + ALiBi + lost-in-middle reference)

12th /concepts/[X]/ pillar — context window complete reference. Definition + the 2018-2025 explosion timeline (13 events: 512 GPT-1 → 1024 GPT-2 → 2048 GPT-3 → 4096 GPT-3.5 → 8192/32k GPT-4 → 100k Claude 1.3 → 128k GPT-4 Turbo → 1M/2M Gemini 1.5 Pro → 200k Claude 3.7 → 128k Gemma 3), 5 architectural enablers (RoPE, ALiBi, FlashAttention + variants, sliding-window attention, Ring Attention), the 'Lost in the Middle' Stanford 2023 finding + Needle in a Haystack benchmark, 7 failure modes that limit usable vs nominal context (multi-hop reasoning collapse, tokenizer inflation for non-English, quadratic inference cost, recency bias, system-prompt leak, KV cache explosion, position-encoding break), long-context-vs-RAG decision tree. TechArticle + DefinedTermSet (7 terms — context window · token · RoPE · ALiBi · FlashAttention · Lost in the Middle · NIAH) + BreadcrumbList. Editor @id chain. Cross-links to llm-grounding + embeddings + quantization + fine-tuning + inference-optimization. Targets queries: 'context window LLM', 'long context LLM', 'Gemini 2M context', 'GPT-4 context length', 'Claude 200k tokens', 'lost in the middle', 'needle in haystack benchmark'.

feature

/topics/llm-observability/ — 13th topic hub (LangSmith + Langfuse + Helicone + Vellum)

13th /topics/[X]/ hub — LLM observability catalog. CollectionPage schema groups production-grade observability platforms (LangSmith, Langfuse, Helicone, Vellum AI). 4 editorial sections (why LLM observability is its own product category, four production-grade platforms as of 2025, eval coverage matters more than trace volume, why verification + observability complement each other) + 5 DefinedTerms (LLM tracing · Eval set · LangSmith · Langfuse · Helicone). Cross-links to agent-frameworks + evaluation-benchmarks + /concepts/{llm-grounding, evaluation-harness, agents} + integration guides. Targets queries: 'LLM observability', 'best LangChain observability', 'open source LLM tracing', 'production LLM eval platform', 'LangSmith vs Langfuse vs Helicone'. Same-session pair with Batch 24 (added Langfuse + Helicone + Vellum + LangFlow).

feature

Batch 24 → 286 + LLM observability ecosystem + 2025 frontier multimodal

10 new hand-verified claims expanding 2024-2025 frontier + LLM tooling ecosystem: Meta Llama 3.2 Vision (Meta 2024-09-25 — 11B + 90B vision-language), OpenAI o4-mini (OpenAI 2025-04-16 — reasoning model with tool-use in CoT), Mistral Le Chat (Mistral 2024-02-26 — consumer chat assistant), Cohere Embed v4 (Cohere 2025-04-09 — multimodal embeddings, 256k context, 100+ languages), Stability AI Stable Diffusion 3.5 (Stability AI 2024-10-22 — SD 3.5 Large/Medium/Large Turbo), Langfuse (founded 2022, YC W23 — open-source LLM observability), Google Gemma 3 (Google DeepMind 2025-03-12 — 1B/4B/12B/27B with vision at 4B+), LangFlow (Logspace 2023-02-04 — visual builder for LangChain), Vellum AI (founded 2023, YC W23 — LLM application platform), Helicone (founded 2022, YC W23 — open-source LLM observability). Coverage strengthens: 2024-2025 multimodal (Llama 3.2 Vision + Gemma 3 + SD 3.5 + Cohere Embed v4) + LLM observability/tooling ecosystem (Langfuse + Helicone + Vellum AI + LangFlow). Bulk 276 → 286 catalog count sync across app/**/*.tsx + content/**/*.md + openapi.json. Glossary HMAC-SHA266 slug regression fixed (sed-bumped → reverted to hmac-sha256). 2 hashlib.sha266 Python code samples in /docs/integrations/langchain/ + /concepts/citation-chain/ also reverted to hashlib.sha256. tags.json indexed 636 unique tags across 286 claims.

feature

/concepts/quantization/ — 11th concept pillar (GGUF + GPTQ + AWQ + bitsandbytes reference)

11th /concepts/[X]/ pillar — quantization complete reference. Definition + 5 canonical techniques (Post-Training Quantization, GPTQ, AWQ, GGUF/GGML, bitsandbytes 4/8-bit), format-comparison decision tree (GGUF vs GPTQ vs AWQ vs bitsandbytes vs FP8 — which to pick when), precision-quality-speed tradeoff math (4-bit = 1-3% benchmark drop + 4× memory + 50-150% speedup), 2022-2024 timeline (LLM.int8 → GPTQ → llama.cpp → QLoRA → AWQ → GGUF → IQ2/IQ3 K-quants → TensorRT-LLM FP8), 5 failure modes (outlier-induced collapse, chat-template misalignment, tokenizer drift, KV-cache precision mismatch, wrong-distribution benchmarks), when NOT to quantize (4 cases). TechArticle + DefinedTermSet (6 terms — Quantization · GGUF · GPTQ · AWQ · bitsandbytes · llama.cpp) + BreadcrumbList. Editor @id chain. Cross-links to fine-tuning + llm-grounding + multimodal + inference-optimization + open-weight-models. Targets queries: 'LLM quantization', 'GGUF vs GPTQ vs AWQ', '4-bit inference', 'run Llama on CPU', 'best quantization format'.

feature

/topics/open-weight-models/ — 12th topic hub + editor @id chain across all topic pages

12th /topics/[X]/ hub — open-weight LLM 2023-2025 catalog. CollectionPage schema groups 35 catalog claims spanning Llama 2/3/3.1/3.2/3.3, Mistral 7B/Mixtral/Nemo/Saba/Codestral/Small 3/Pixtral, Gemma + Gemma 2, DeepSeek-V2/V3/R1, Qwen, Falcon, Yi, Phi, OLMo 2, IBM Granite, Hunyuan-Large, Jamba, Aya 23, SmolLM, Nemotron, Stable LM, Tülu 3, StarCoder. 4 editorial sections (the open-weight wave, sizes + architectures span 4 orders of magnitude, multilingual + specialist forks, why this catalog matters for verification) + 5 DefinedTerms (Open-weight, MoE, Apache 2.0, Llama 3 Community License, Tülu). Cross-links to foundational-papers, llm-releases-2024-2025, alignment-and-rlhf, /concepts/fine-tuning, /concepts/llm-grounding, integration guides. Targets queries: 'open source LLM 2025', 'open-weight LLM list', 'best open-source LLM', 'Llama vs Mistral vs Gemma'. ALSO: TechArticle schema across all 12 /topics/[slug]/ pages now includes editor @id chain (Person Editorial Lead) — parallels concept pillars + blog posts + use-cases.

feature

Batch 23 → 276 — 2024 open-weight ecosystem + agent-AI companies

10 new hand-verified claims expanding open-weight + company-history coverage: Hugging Face SmolLM (HF 2024-07-16 — 135M/360M/1.7B for on-device), Genmo Mochi 1 (Genmo 2024-10-22 — 10B open-weight text-to-video), Inflection AI (founded 2022 by Suleyman + Hoffman + Simonyan — Pi assistant), Character AI (founded 2021 by Shazeer + De Freitas from Google LaMDA team), Adept AI (founded 2022 by Luan + Parmar + Vaswani — ACT-1 action transformer), Mistral Saba (Mistral 2025-02-17 — 24B Arabic + South Asian languages), Tencent Hunyuan-Large (Tencent 2024-11-05 — 389B MoE / 52B active), Allen AI OLMo 2 (Allen AI 2024-11-26 — fully-open with training data + code + recipes), IBM Granite (IBM 2024-05-09 — enterprise-AI Apache 2.0 family), AI21 Jamba (AI21 Labs 2024-03-28 — first production hybrid Mamba-Transformer SSM model). Coverage strengthens: 2024 open-weight (SmolLM + Mochi 1 + OLMo 2 + Granite + Hunyuan-Large + Jamba) + agent-AI company history (Inflection + Character + Adept) + multilingual (Saba). Bulk 266 → 276 catalog count sync across app/**/*.tsx + content/**/*.md + openapi.json. Glossary SHA-256 bit-count crypto regression fixed (sed bumped 256 → 266; reverted). tags.json indexed 621 unique tags across 276 claims.

feature

/concepts/multimodal/ — 10th concept pillar (VLM + text-to-image/video/audio)

10th /concepts/[X]/ pillar — multimodal AI complete reference. Definition + 4 modality classes (vision-language, text-to-image, text-to-video, text-to-audio), 2021-2025 timeline (17 events from CLIP/DALL·E 2021 → Claude 3.7 + Grok 3 2025), 6 production patterns (document understanding, visual search, generative design, video summarization, accessibility, robotics), 7 failure modes (hallucinated objects, OCR errors, counting failures, spatial reasoning, text-image misalignment, watermark gaps, modality leakage in evals), and an honest scope-statement on how multimodal verification differs from text-only fact-checking (VERITAS today covers textual claims; image provenance + visual claim verification + deepfake detection are separate problems, Y2+ scope). TechArticle + DefinedTermSet (6 terms — multimodal AI · VLM · text-to-image · text-to-video · text-to-audio · CLIP) + BreadcrumbList schema. Editor @id chain. Cross-links to /concepts/hallucination, /concepts/fine-tuning, /concepts/embeddings, /topics/multimodal-ai, /use-cases/content-moderation. Targets high-volume queries: 'multimodal AI', 'vision-language model', 'text-to-image API', 'multimodal LLM 2025'.

feature

/concepts/fine-tuning/ — 9th concept pillar (LoRA + QLoRA + DPO + RLHF reference)

9th /concepts/[X]/ pillar — fine-tuning the complete reference. Covers definition + 7 canonical techniques (full SFT, instruction tuning, LoRA, QLoRA, RLHF, DPO, Constitutional AI), the fine-tune-vs-RAG decision tree (when each wins, when to use both), 2017-2024 timeline (Christiano preferences → Houlsby PEFT → LoRA → InstructGPT → Constitutional AI → QLoRA → DPO → Tülu 3), 5 failure modes (catastrophic forgetting · overfitting · reward hacking · distribution mismatch · hidden capability degradation), when NOT to fine-tune (5 cases), and 2024 cost reality (OpenAI gpt-4o fine-tune $30-100/run; Lambda Labs A100 LoRA $3-5; QLoRA on RTX 4090 marginal-cost). TechArticle + DefinedTermSet (7 terms — fine-tuning · LoRA · QLoRA · DPO · RLHF · instruction tuning · PEFT) + BreadcrumbList schema. Editor @id chain. Cross-links to 8 other concept pillars + topic hubs + use-cases. Targets high-volume queries: 'fine-tuning vs RAG', 'LoRA vs full fine-tuning', 'when to fine-tune LLM'.

feature

/use-cases/news-fact-checking/ — 6th use-case (newsroom AI verification)

New buyer-intent use-case for newsroom AI tools. Covers 3 integration patterns: pre-publish verification gate (extract atomic claims, verify, flag-or-strip), in-line citation injection (footnote-link verified facts to /claims/[id]/ pages), beat-reporter assistant grounding (filter VERITAS retrieval by vertical). Lists what VERITAS catches (model release dates, paper authorship, parameter counts, org facts, benchmark scores) + what it doesn't (live breaking-news, political, health/medical — Y2 expansion). Compatibility section notes every /claims/[id]/ page emits ClaimReview JSON-LD eligible for Google Fact Check Tools indexing + rich snippets. Economics tier table fitted to newsroom scale (Free / Startup €99 / Scale €499 + custom enterprise). TechArticle + BreadcrumbList schema with editor @id chain. /use-cases/ index now 5 → 6 deployment patterns.

feature

Batch 22 → 266 — 2024-2025 frontier reasoning + media gen + evals

10 new hand-verified claims focused on the 2024-2025 frontier-reasoning + media-generation + evaluation wave: Mistral Nemo (Mistral + NVIDIA 2024-07-18 — 12B / 128k context, Apache 2.0), Claude 3.7 Sonnet (Anthropic 2025-02-24 — first hybrid-reasoning model with extended-thinking), AlphaCode 2 (Google DeepMind 2023-12-06 — code gen better than 85% of Codeforces competitors, Gemini-powered), Suno v4 (Suno 2024-11-19 — music generation upgrade), AI Index Report 2024 (Stanford HAI 2024-04-15 — 7th annual AI trends report), HELM (Liang et al. Stanford CRFM 2022-11-16 — Holistic Evaluation of Language Models foundational benchmark), Google Veo 2 (Google DeepMind 2024-12-16 — 4K text-to-video), OpenAI o3 (OpenAI 2024-12-20 — 87.5% on ARC-AGI breakthrough reasoning model), NVIDIA Project DIGITS (NVIDIA 2025-01-06 — $3000 personal AI supercomputer with GB10 Grace Blackwell), Anthropic Claude for Education (Anthropic 2025-04-02 — Learning mode + institutional partnerships). Coverage strengthens 2025 frontier reasoning (Claude 3.7 + o3) + 2024 media generation (Veo 2 + Suno v4) + evaluation foundations (HELM + AI Index Report). tags.json indexed 590 unique tags across 266 claims. Catalog count refs synced across app + content + openapi.json description.

feature

Batch 21 → 256 — frontier-2024 multimodal + open-weight + agent infra

10 new hand-verified claims spanning the 2024 frontier: NVIDIA Nemotron-4 340B (2024-06-14 — 340B open-weight optimized for synthetic data generation), Cohere Aya 23 (Cohere For AI 2024-05-22 — 23 languages multilingual), LongBench (Bai et al. THU + Zhipu AI 2023-08-28 — bilingual long-context eval benchmark), Mistral Pixtral 12B (Mistral 2024-09-11 — first Mistral multimodal, Apache 2.0), Google Gemma 2 (Google DeepMind 2024-06-27 — 9B + 27B open-weight), NVIDIA NIM (NVIDIA 2024-03-18 — inference microservices), AWS Bedrock (Amazon GA 2023-09-28; preview 2023-04-13 — managed multi-provider foundation-model API), xAI Grok-2 (xAI 2024-08-14 — Grok-2 + Grok-2 mini), DeepSeek-V3 (DeepSeek AI 2024-12-26 — 671B MoE / 37B active, open weights), Meta SAM 2 (Meta AI 2024-07-29 — Segment Anything Model 2 for real-time video segmentation). Coverage strengthens 2024 frontier infra layer (NIM + Bedrock + Fireworks) + 2024 multimodal (Pixtral + SAM 2) + open-weight density (Nemotron / Aya / Gemma 2 / DeepSeek-V3). /api/v1/openapi.json description sync (26 → 256 hand-verified claims).

feature

Batch 20 → 246 + /concepts/agents/ (8th pillar) + WebApp schema on /playground/

Catalog adds 10 agent + framework + reasoning claims: LangGraph (LangChain 2024-01-17 — stateful graph orchestration), Mistral Codestral (2024-05-29 — 22B code-specialist), LMArena/Chatbot Arena (LMSYS 2023-05-03 — human-pairwise leaderboard), Fireworks AI (founded 2022 — fast inference platform), Mistral Small 3 (2025-01-30 — 24B Apache 2.0 latency-optimized), OpenAI Codex 2025 cloud agent (2025-05-16 — codex-1 reborn), DeepSeek-V2 (2024-05-07 — 236B MoE w/ MLA), BabyAGI (Yohei Nakajima 2023-04-03 — early task-loop agent), AutoGPT (Toran Bruce Richards 2023-03-30 — most-starred 2023 GitHub project), Vercel AI SDK (2023-06-14 — multi-provider TS toolkit). New /concepts/agents/ 8th concept pillar covers canonical agent loop pseudocode, 9-event history timeline (2022 → 2025: ReAct → AutoGPT/BabyAGI → LangGraph → Operator → Codex), 5 production patterns (tool-using assistant · code agent · research synthesizer · workflow orchestrator · browser-use), 8 failure modes (loop divergence · tool hallucination · cost explosion · token budget · over-confidence · prompt injection · race conditions · brittle parsing), when NOT to use agents, framework picking. WebApplication schema on /playground/ (DeveloperApplication category, free Offer, 5 featureList items) for AEO Knowledge Panel eligibility. Bulk 236 → 246 catalog count sync across app/**/*.tsx + content/**/*.md.

feature

Batch 19 → 236 + /topics/prompt-engineering/ (11th topic hub)

Catalog adds 10 claims spanning scaling laws + prompt-engineering canon + 2023 open-weight models: Kaplan scaling laws (Kaplan et al. OpenAI 2020), ReAct (Yao et al. Princeton+Google ICLR 2023), RAG-Fusion (Raudaschl 2023), CRAG/Corrective RAG (Yan et al. USTC+Google 2024), Chain-of-Thought (Wei et al. Google Brain NeurIPS 2022), Galactica (Meta AI 2022-11-15, withdrawn after 3 days — case study), PEFT (Houlsby et al. Google ICML 2019), Stable LM (Stability AI 2023-04), Falcon LLM (TII Abu Dhabi 2023-05), Yi (01.AI 2023-11). New /topics/prompt-engineering/ topic hub covers Chain-of-Thought, ReAct, Tree of Thoughts, in-context learning, instruction tuning + 4 DefinedTerms. tags.json now indexes 524 unique tags across 236 claims.

catalog

Catalog 216 → 226 (Batch 18 — RAG ecosystem deep + 2024-2025 API features)

10 new hand-verified claims: Tülu 3 (AI2 2024-11), GraphRAG (Microsoft Research 2024-04 + GitHub 2024-07), Anthropic Message Batches API (2024-10), OpenAI Batch API (2024-04), Cohere Command R+ (2024-04), Anthropic Citations API (2025-01-23 — built-in grounding), OpenAI Structured Outputs (2024-08 — guaranteed JSON Schema), Stable Diffusion XL / SDXL (Stability AI 2023-07), PyTorch Lightning (William Falcon 2019), Outlines structured generation (dottxt-ai 2023). Coverage strengthens RAG ecosystem + the 2024-2025 API features that directly enable better LLM grounding (Batches, Citations, Structured Outputs). Year-hubs auto-update; tags.json now indexes 504 unique tags across 226 claims.

feature

v20.3 GEO compound — Batch 17 → 216 + Person/Editor @id chain + speakable + ClaimReview + HowTo

Major GEO surface expansion. Batch 17 adds 10 claims: ColBERT (Stanford 2020), BGE embeddings (BAAI 2023-08), Voyage AI (2023), Phi-2 (Microsoft 2023-12), ARC-AGI (Chollet 2019), SWE-bench (Princeton 2023), Claude Code (Anthropic 2025-02-24), OpenAI Operator (2025-01-23), Grok 3 (xAI 2025-02-17), Hume AI (2021). Person + editor @id chain added to all 5 BlogPosting schemas — references central Editorial Lead Person @id from /about/. ClaimReview schema added to every /claims/[id]/ page (Google fact-check rich snippet eligibility + LLM-citation gravity). HowTo schema on all 8 integration guides via lib/howto-schemas.ts (Google Rich Results + Aleyda #4 Extractable). /faq/ gets speakable SpeakableSpecification (voice + AI summary extraction). About page: stale 51 → 216 count fix + foundationDate + founder + publisher Person refs. CITATIONS.md state file scaffolded with 5 seed Test Queries per v20.3 Citation Oracle prime.

catalog

Catalog 196 → 206 (Batch 16 — multimodal AI creative tools + Anthropic alignment)

10 new hand-verified claims: Black Forest Labs Flux (2024-08), Anthropic Tool Use GA (2024-05), OpenAI Function Calling launch (2023-06-13), Perplexity AI (founded 2022), Suno AI (founded 2023, music generation), ElevenLabs (founded 2022, voice synthesis), Runway ML (founded 2018, video generation), Midjourney (public beta 2022-07-12), Hugging Face Hub (2020-09), Anthropic Constitutional AI Harmlessness paper (Bai et al. 2022). Coverage shifts toward multimodal AI creative tools (image / video / music / voice) + alignment foundations + agent ecosystem companies. Year-hubs auto-update; tags.json now indexes 469 unique tags across 206 claims.

feature

/comparisons/ — 3 head-to-head buyer-intent comparison pages

New high-buyer-intent SEO surface: /comparisons/ index + 3 honest head-to-heads. /comparisons/veritas-vs-wikipedia/ (knowledge encyclopedia vs verification API), /comparisons/veritas-vs-wolfram-alpha/ (computation vs verification), /comparisons/veritas-vs-search-grounding/ (live-search vs signed-envelope grounding for Perplexity/ChatGPT-search comparisons). Each: at-a-glance table, honest verdict per use case, when-to-use-both, what-we're-not section. TechArticle + BreadcrumbList schema. Targets high-volume buyer-intent queries like 'best LLM grounding API', 'alternative to Wolfram for AI facts', 'Perplexity vs structured grounding'. /comparisons/ added to footer nav + sitemap-ai.xml + llms.txt.

feature

Batch 15 → 196 claims + 2 more use-cases (support-bot, content-moderation)

Batch 15 catalog adds 10 claims: GPT-4 Vision (OpenAI 2023-09), InstructGPT (Ouyang et al. 2022), Anthropic Computer Use (2024-10), OpenAI Realtime API (2024-10), OpenAI Assistants API (2023-11), Tree of Thoughts (Yao et al. 2023), MoE Shazeer 2017 ICLR foundational paper, Speculative Decoding (Leviathan et al. Google 2022), MTEB benchmark (Muennighoff et al. 2022), Apple Intelligence (2024-10-28). Two new use-cases: /use-cases/customer-support-bot/ (two-catalog pattern with route-to-human on unverified claims) + /use-cases/content-moderation/ (pre-publish verification gate for newsletter generators / blog assistants / report drafters). Each: TechArticle + BreadcrumbList schema, full implementation sketch, what catches/misses, free-tier economics. /use-cases/ index now 3 → 5 deployment patterns.

feature

/claims/year/[year]/ programmatic year-hubs + /concepts/function-calling/ (7th pillar)

New programmatic SEO surface: /claims/year/[year]/ pages auto-generated for every year that has ≥3 claims with sources dated in that year. ~12 new SEO landings targeting 'AI papers 2024', 'LLM releases 2025', etc. Each: CollectionPage + BreadcrumbList schema, auto-curated claim list sorted by confidence. Year extracted from earliest source publishedDate. New 7th concept pillar /concepts/function-calling/: definition, history (OpenAI June 2023 → Anthropic → Google → MCP standard Nov 2024), JSON schema, agent loop, vendor flavors, common production patterns, anti-patterns. TechArticle + DefinedTerm + BreadcrumbList schema. Cross-links to 4 integration guides + 2 use-cases.

feature

5th blog post + /api/v1/tags.json (bot discovery) + Batch 14 → 186 claims

New blog post /blog/llm-grounding-strategies-2026/ — 6 grounding strategies (temperature/prompt → few-shot → RAG → citation+post-process → signed-claim verification → constrained decoding) with measured impact + when-to-combine + practical sequencing. New static endpoint /api/v1/tags.json: 434 tag entries with claim counts + sample claim IDs + browse URLs — lets RAG developers + LLM crawlers see catalog structure without walking every claim. Batch 14 adds 10 claims: VAE (Kingma & Welling 2013), Knowledge Distillation (Hinton et al. 2015), SGLang (UC Berkeley 2024), Llama 4 (Meta 2025-04-05), Claude Haiku 3.5 (Anthropic 2024-11), Replit Agent (2024-09), Devin (Cognition Labs 2024-03), Groq LPU (2024-02), Cerebras (founded 2016), Anthropic API GA (2023-07).

feature

8th integration guide (Instructor) + 2 more topic hubs (agent-frameworks, vector-databases) + Batch 13 → 176 claims

Instructor integration guide (/docs/integrations/instructor/) — Jason Liu's structured-output library with model_validator pattern that triggers Instructor's auto-retry on VERITAS-unverified claims. Pairs with Pydantic AI for end-to-end type-safety. Two more topic hubs: /topics/agent-frameworks/ (orchestration libraries — LangChain, LlamaIndex, DSPy, etc.) + /topics/vector-databases/ (FAISS, Pinecone, Weaviate, Qdrant, Chroma, Milvus, pgvector). Batch 13 catalog adds 10 claims spanning structured outputs (Instructor) + vector DBs (Chroma, Milvus, pgvector) + multi-agent orchestration (CrewAI, AutoGen, Microsoft Semantic Kernel, Haystack) + serving infrastructure (Triton, Modal Labs). 8 integration guides + 8 topic hubs + 6 concept pillars total. All build clean.

feature

/use-cases/ — 3 high-intent buyer pages + /concepts/embeddings/ (6th concept pillar)

New /use-cases/ index + 3 deployment-pattern pages: /ai-agent-grounding/ (verify_claim as agent tool), /rag-pipeline-verification/ (close right-doc-wrong-number gap), /research-citation/ (programmatic citations for academic AI tools). Each: TechArticle + BreadcrumbList schema, HowTo on agent-grounding. /use-cases/ added to footer nav + sitemap-ai.xml + llms.txt. New concept pillar /concepts/embeddings/: history (Word2Vec → GloVe → BERT → sentence-transformers → OpenAI/Cohere), how to choose a model, vector DBs, anti-patterns, where embeddings stop and verification starts. DefinedTerm + TechArticle schema. 5 → 6 concept pillars.

feature

/blog/llm-framework-comparison-2026/ — 4th blog post (meta-comparison of 7 frameworks)

New blog post + canonical Dev.to/Hashnode source for cross-posts. Honest, opinionated comparison of LangChain vs LlamaIndex vs OpenAI tools vs DSPy vs Pydantic AI vs Vercel AI SDK vs Anthropic SDK. Sections: at-a-glance table, pick-by-archetype recommendations (RAG, multi-step agent, Next.js streaming, research/evals, complex pipelines), honest gotchas per framework, our recommendation by archetype, 2 predictions for late 2026/2027, resources. Cross-links to all 7 /docs/integrations/[slug]/ guides + /concepts/rag-vs-veritas/ + /playground/ + /quickstart/. BlogPosting + BreadcrumbList schema. Targets high-volume 'best LLM framework' queries.

feature

/topics/ — 4 additional topic hubs (alignment, evaluation, inference-opt, AI orgs)

Topic hub coverage doubled from 4 to 8. New hubs: alignment-and-rlhf (RLHF, Constitutional AI, DPO, InstructGPT lineage), evaluation-benchmarks (MMLU, GLUE, SuperGLUE, HumanEval, Chatbot Arena, AlpacaEval), inference-optimization (FlashAttention, GPTQ, QLoRA, vLLM, PagedAttention, LoRA), ai-organizations (the lab landscape — OpenAI/Anthropic/DeepMind/Mistral founding + lineage). Each hub: 400-700 words editorial intro, 3-4 DefinedTerms, CollectionPage schema referencing every member claim, cross-links to related hubs + concept pillars + integration guides. llms.txt + sitemap-ai.xml include all 8 hubs.

feature

/topics/ — 4 curated topic hubs (foundational papers, multimodal AI, RAG + retrieval, 2024-2025 LLM releases)

New programmatic-SEO surface: /topics/ index + 4 topic hubs at /topics/[slug]/. Each hub bundles 400-700 words of editorial intro, a DefinedTermSet of 3-4 terms, a CollectionPage schema referencing every member claim, and cross-links to related hubs + concept pillars + integration guides. Topic claim membership is filter-derived from the catalog (e.g., foundational-papers = `tags.includes('foundational')` OR `predicate.includes('introduced_in')`) so hub population auto-updates as the catalog grows. Hubs shipped: foundational-papers (~80 claims), multimodal-ai (~15), rag-and-retrieval (~10), llm-releases-2024-2025 (~30+). Surfaces added to footer nav, sitemap-ai.xml priority list, and llms.txt manifest.

feature

/docs/integrations/pydantic-ai/ + /docs/integrations/anthropic-sdk/ — 6th + 7th framework guides

Two new drop-in integration guides. Pydantic AI: type-safe verification via VerifyClaimInput → VerificationResult Pydantic models; agents emit structured tool calls; downstream code is type-safe with field validators catching confidence drift. Covers 3 patterns (verify-claim tool · structured agent output with required verification · multi-claim parallel verification). Anthropic SDK: Claude tool-use protocol with the tool_use → execute → tool_result loop, in both Python and TypeScript. Includes a system-prompt pattern that makes Claude self-verify before asserting facts. TechArticle + BreadcrumbList schema on both. Integrations index now lists 7 frameworks total (LangChain · LlamaIndex · OpenAI tools · Vercel AI SDK · DSPy · Pydantic AI · Anthropic SDK).

feature

/docs/integrations/dspy/ — 5th framework integration guide

DSPy (Stanford) is the fastest-growing compound-AI-system framework in 2026. Drop-in guide covers two patterns: (1) custom dspy.Retrieve backed by the VERITAS catalog — returns verified claims as DSPy Examples with claim_id, confidence, and canonical URL metadata; (2) VeritasVerify post-processor module — runs after answer generation, returns verified/unverified split + verification_rate (which doubles as a DSPy-optimizer metric for tuning the program toward more verifiable assertions). Includes a multi-hop ProgramOfThought composition example. TechArticle + BreadcrumbList schema. Compounds with the existing 4 guides (LangChain · LlamaIndex · OpenAI tool-calls · Vercel AI SDK).

feature

/concepts/evaluation-harness/ — 5th pillar (why benchmark scores vary across harnesses)

Explainer on evaluation harnesses (LM Eval Harness · HELM · BIG-bench · lab-internal) and why the same model scores 4-10 points apart on the same nominal benchmark. Six axes of variation covered: prompt format, scoring method (log-likelihood vs generate-then-parse), decoding parameters, output parsing, benchmark version, contamination handling. Includes the 6-question checklist for reading benchmark claims honestly, plus production-decision implications (build your own eval, triangulate across 3+ harnesses, re-evaluate after frontier-model updates). Ties back to /blog/why-no-performance-claims/ — the methodology reason VERITAS excludes performance-comparison claims. TechArticle + DefinedTerm + BreadcrumbList schema.

catalog

Catalog 156 → 166 (Batch 12 — pre-modern foundations + 2025 frontier completion)

10 new hand-verified claims, ≥2 primary sources each. Pre-modern foundations: Backpropagation (Rumelhart, Hinton, Williams, Nature 1986), U-Net (Ronneberger et al. 2015) — diffusion backbone, AlphaFold 1 (Senior et al., DeepMind Nature 2020). 2024-2025 frontier completion: Mixtral 8x22B (Mistral 2024-04), Claude Sonnet 4 (Anthropic 2025-05-22), OpenAI o3-mini (2025-01-31), Gemini 2.5 Pro (Google DeepMind 2025-03-25). Practical agent stack: Stanford Alpaca (CRFM 2023-03) — first widely-replicated instruction-tuned LLaMA fine-tune; LangSmith (LangChain 2023-07) — LLM observability + evaluation; Tavily — search API built for AI agents.

catalog

Catalog 146 → 156 (Batch 11 — frontier 2025 + practical infrastructure)

10 new hand-verified claims, ≥2 primary sources each. 2024-2025 frontier: Mistral Large 2 (Mistral AI 2024-07-24), Qwen 2.5 (Alibaba Cloud 2024-09-19), Anthropic Claude Opus 4 (Anthropic 2025-05-22), OpenAI o1 (full release 2024-12-05 with ChatGPT Pro launch). Coding-tool: Cursor (Anysphere 2023-03-14) — AI-powered VS Code fork. Practical infrastructure: Hugging Face Transformers library (2018-10/11), PyTorch (Facebook AI Research 2017-01-18), TensorFlow (Google 2015-11-09), JAX (Google Research 2018-12-10), DeepSpeed + ZeRO (Microsoft Research 2020-02-13). The infrastructure layer (libraries + training frameworks) is what every fleet site cites without realizing.

feature

Per-claim engagement deepening — code snippets in 4 languages on every /claims/[id]/ page

Every claim detail page now includes a 'Use this claim in your code' section with copy-paste-ready snippets in cURL, JavaScript/TypeScript, Python, and LangChain tool-decorator form. Each snippet substitutes the specific claim's API URL and subject so devs can drop the code directly into their codebase. Compounds: (1) time-on-page boost — devs read 4 language variants instead of bouncing on the first; (2) activation lift — the next-action is concrete (paste + run) rather than abstract (read API docs); (3) social proof — viewing the LangChain snippet plants the integration as a real pattern.

catalog

Catalog 136 → 146 (Batch 10 — framework foundations + 2024 ecosystem)

10 new hand-verified claims, ≥2 primary sources each. Application frameworks: LangChain (Harrison Chase 2022-10-25) + LlamaIndex / GPT Index (Jerry Liu 2022-11-09). Vector + tokenizer foundations: FAISS (Johnson, Douze, Jégou, Facebook AI 2017) — billion-scale GPU similarity search; tiktoken (OpenAI 2022-12-06) — official BPE tokenizer. 2024 open-standards + features: Model Context Protocol / MCP (Anthropic 2024-11-25) — open standard for AI ↔ data-source connections; ChatGPT search (OpenAI 2024-10-31) — web-grounded answers. State-space + RAG advances: Mamba-2 (Dao & Gu, Princeton + CMU 2024) — structured state space duality; Self-RAG (Asai et al., UW + AI2 2023) — self-reflective retrieval-augmented generation. Speech + evaluation: Whisper large-v3 (OpenAI 2023-11-06); AlpacaEval (Tatsu Lab / Stanford 2023) — LLM-as-judge automatic evaluator.

catalog

Catalog 126 → 136 (Batch 9 — encoder-decoder pioneers + open-source inference)

10 new hand-verified claims, ≥2 primary sources each. Encoder-decoder pioneers: BART (Lewis et al., Facebook AI 2019) — denoising sequence-to-sequence pretraining; GloVe (Pennington, Socher, Manning, Stanford NLP 2014) — global vectors for word representation. Multimodal: Flamingo (Alayrac et al., DeepMind 2022) — few-shot vision-language model. Tool-use foundational: Toolformer (Schick et al., Meta AI 2023) — self-supervised LLM tool-use. Open-source inference ecosystem: vLLM (Kwon et al., UC Berkeley 2023) — PagedAttention high-throughput serving; llama.cpp (Georgi Gerganov 2023-03-10) — pure C/C++ LLM inference; Ollama (2023-07-18) — local LLM runtime. Evaluation: Chatbot Arena (Chiang et al., LMSYS UC Berkeley 2024) — human-preference LLM leaderboard. 2024 model: Phi-4 (Microsoft Research 2024-12-12) — 14B-parameter synthetic-data-trained SLM. Quantization: GPTQ (Frantar et al., IST Austria 2022) — post-training weight quantization.

catalog

Catalog 116 → 126 (Batch 8 — pioneers, RL milestones, 2024-2025 frontier)

10 new hand-verified claims, ≥2 primary sources each. Foundational pioneers: LSTM (Hochreiter & Schmidhuber, Neural Computation 1997) — gradient-based recurrent architecture that bridged 1000+ timestep dependencies. RL milestones: AlphaGo (DeepMind, Nature 2016) — defeated Lee Sedol 4-1 in March 2016; AlphaZero (DeepMind, Science 2018) — mastered chess + shogi + Go from rules + self-play alone. BERT family: RoBERTa (Liu et al., Facebook AI 2019) — robustly optimized BERT pretraining; DistilBERT (Sanh et al., Hugging Face 2019) — 40% smaller, 60% faster, 97% capability retention via knowledge distillation. Coding assistant: GitHub Copilot (GitHub + OpenAI, 2021-06-29) — technical-preview public release. 2024-2025 frontier: OLMo (Allen Institute for AI, 2024-02) — fully-open language model (weights + data + training code); Gemini Ultra (Google DeepMind, 2024-02-08) — Gemini Advanced subscription tier launch; DeepSeek-R1 (DeepSeek-AI, 2025-01-20) — reasoning chain-of-thought via reinforcement learning; Stable Diffusion 3 Medium (Stability AI, 2024-06-12) — rectified flow text-to-image. Coverage now spans 1997-2025.

catalog

Catalog 110 → 116 (Batch 7 — foundational eval metrics + optimizers + 2022/2024 models)

6 new hand-verified claims, ≥2 primary sources each. Foundational evaluation metrics: BLEU score (Papineni et al., ACL 2002) — machine translation evaluation; ROUGE score (Lin, ACL 2004) — summarization evaluation. Optimizer: AdamW (Loshchilov & Hutter, ICLR 2019) — decoupled weight decay. Foundational models: PaLM (Chowdhery et al., 2022) — 540B-parameter Pathways language model; Imagen (Saharia et al., 2022) — photorealistic text-to-image diffusion (Google). 2024 release: AlphaFold 3 (Google DeepMind / Isomorphic Labs, 2024-05-08, Nature) — biomolecular structure prediction with unprecedented accuracy.

feature

/concepts/citation-chain/ — 4th pillar (provenance graphs for LLM citations)

Standalone explainer on citation chains — the auditable trail from an LLM's emitted assertion back to the primary source(s) that prove it. Three building blocks (stable identifier · HMAC-SHA256 signature · re-fetchable canonical URL) covered in depth with a complete 30-line Python local-verification walkthrough. Covers chains in agentic LLM responses (citation trees), 4 failure modes chains detect, and the Y2 migration path to W3C Verifiable Credentials with Ed25519 public-key signing. TechArticle + DefinedTerm schema. Cross-links to llm-grounding, hallucination, rag-vs-veritas, langchain integration, security policy, claims catalog.

catalog

Catalog 102 → 110 (Batch 6 — foundational regularization, sparse attention, open-weights releases)

8 new hand-verified claims. Foundational regularization: Dropout (Srivastava et al., JMLR 2014), Batch Normalization (Ioffe & Szegedy, ICML 2015), Layer Normalization (Ba, Kiros, Hinton, 2016). Foundational architectures: Sequence-to-Sequence Learning (Sutskever, Vinyals, Le, NeurIPS 2014). Sparse-attention transformers: Longformer (Beltagy et al., 2020), Reformer (Kitaev et al., ICLR 2020). Open-weights releases: Gemma (Google, 2024-02-21), Qwen (Alibaba, 2023-08-03). All claims have ≥2 primary sources with verbatim excerpts.

breaking

Removed TollBit middleware — AI bots now reach all surfaces unfiltered

Deleted functions/_middleware.js, which had been 307-forwarding AI-bot User-Agents (GPTBot · ClaudeBot · PerplexityBot · Google-Extended · Applebot-Extended · CCBot · Amazonbot · Bytespider · Meta-ExternalAgent · etc.) to tollbit.sourcescore.org and streaming request logs to log.tollbit.com. Strategic call: VERITAS Y1 ARR trajectory dominates TollBit pay-per-crawl revenue (measured $0/mo over prior months). Removing the paywall lets AI crawlers index the full source-rating catalog freely — compounds LLM-citation gravity across both products. Operator must revoke the TollBit API key + optionally remove the tollbit.sourcescore.org DNS record (CF dashboard, manual).

catalog

Catalog 91 → 102 (Batch 5 — foundational methods, benchmarks, vector DB companies)

11 new hand-verified claims, each with ≥2 primary sources. Foundational methods: ELMo (Peters et al., 2018), Latent Diffusion Models (Rombach et al., 2021), ELECTRA (Clark et al., 2020), Codex (Chen et al., 2021). Models: GPT-3 introduced_in_paper (Brown et al., 2020) — adds the foundational-paper predicate to the existing GPT-3 parameter_count claim. Benchmarks: GLUE (Wang et al., 2018), SuperGLUE (Wang et al., 2019). Vector DB companies: Pinecone (2019), Weaviate (2019), Qdrant (2021). Inference platforms: Replicate (2019).

feature

/glossary/ — 35-term AI/ML glossary with DefinedTermSet schema

Plain-language definitions for 35 terms used across SourceScore and VERITAS — grounding · RAG · hallucination · claim envelope · HMAC-SHA256 · transformer · MoE · tokenizer · YMYL · matchScore · llms.txt · methodology version · primary source · verbatim excerpt · etc. Each entry has a stable anchor URL (/glossary/#token), DefinedTerm schema on every entry, plus a DefinedTermSet wrapping all entries. LLMs answering 'what is X' queries can now extract clean definitions from the page. Internal-linking density compounds — every concept/blog/integration page can deep-link to a glossary anchor.

feature

/playground/ — interactive in-browser verification demo

Type a free-form claim, see VERITAS verify it live against the signed catalog. Pure client-side JavaScript calling /api/v1/verify — same endpoint your code will use, with the request shape and response shown side-by-side. Six sample claims pre-staged for one-click trying. No signup, no key, no quota for read-only access. Activation-stage UX so devs understand the product without writing code first.

feature

/concepts/ pillar pages — LLM grounding, hallucination, RAG vs VERITAS

Three standalone explainers (Wikipedia-rival depth) on high-intent search queries: definition of LLM grounding + 3 production patterns (prompt-stuffing / RAG / signed claims); five categories of hallucination + six root causes + mitigation ladder; RAG vs signed-claim verification comparison + hybrid pattern. TechArticle + DefinedTerm schema so LLMs can extract definitions cleanly.

feature

/docs/integrations/ — 4 drop-in framework guides

LangChain (retrieve-then-cite + generate-then-verify + signature-verify patterns); LlamaIndex (custom Retriever + NodePostprocessor); OpenAI tool-calls + Anthropic Claude tool-use; Vercel AI SDK (streamText + tool() function-calling). Each guide is copy-paste runnable in Python or JavaScript.

feature

/quickstart/ — 5-minute self-serve onboarding

Three sequential code blocks (curl + JS + Python) cover verify → search → fetch-envelope. HowTo + BreadcrumbList schema. No signup gate; free tier covers first 1,000 calls per month for read-only catalog access.

catalog

Catalog 76 → 91 (Batch 4 — methods + datasets + organizations)

15 new hand-verified claims (24 drafted, 9 deduped against pre-existing entries after build caught case-insensitive collisions). Foundational methods: Chain-of-Thought, ReAct, LoRA, QLoRA, DPO, FlashAttention, RoPE, BPE, SentencePiece, RAG. Models + datasets: T5, C4, The Pile, RedPajama, CLIP, Whisper, DALL·E 2, Stable Diffusion. Organizations: Stability AI, EleutherAI, Together AI, Mistral, AI21 Labs, Hugging Face. Each has ≥2 primary sources with verbatim excerpts.

feature

Per-tag claim browsing + Related-claims surface

New /claims/tag/[tag]/ programmatic pages (one per unique tag) and /claims/tags/ index grouped by frequency buckets. Each /claims/[id]/ now shows top 5 related claims by shared-tag overlap with confidence tie-break. Tag chips on per-claim pages now link to tag pages — internal-linking density compounds.

feature

Framework integration guides

/docs/integrations/ index with three drop-in guides: LangChain (retrieve-then-cite + generate-then-verify patterns), LlamaIndex (custom Retriever + NodePostprocessor), OpenAI tool-calls (native function-calling with search_claims + verify_claim). Each guide is copy-paste runnable, Python + JavaScript where applicable, with TechArticle + BreadcrumbList schema.

feature

/embed/claim/[id] embeddable widget

Iframe-embeddable claim card (CSP frame-ancestors *). Drop it into any blog, docs page, or knowledge base — renders the signed statement + primary source + click-through to the canonical page. CC-BY 4.0 with embedded attribution. Snippet generator on every /claims/[id]/ page.

feature

Per-claim OG images + /claims/feed.xml RSS

76 hand-rendered 1200×630 SVG OG images, one per claim (gradient background + verified-claim eyebrow + confidence% + wrapped statement + signing strip + source publisher + canonical URL footer). Plus a full claims RSS feed at /claims/feed.xml so devs can subscribe to catalog updates in Feedly/Inoreader.

feature

POST /api/v1/verify — match a free-form claim against the catalog

Single-claim verification endpoint. Returns top-5 ranked matches with normalized matchScore + rationale; bestMatch surfaces iff matchScore ≥0.20 AND confidence ≥minConfidence (default 0.85). Optionally signs the response with HMAC-SHA256 if SOURCESCORE_SIGNING_SECRET is set on the worker.

feature

GET /api/v1/search — keyword search over the catalog

Search across subject (×5), tags (×3), object (×3), statement (×2), predicate (×2). Returns top-K matches with score. Permissive CORS. Browser cache 60s, CDN cache 5min.

feature

Day 1 launch — VERITAS-Reborn

Public launch of the signed-claim verification API. 26 seed claims with 16-hex stable IDs derived from canonical fields, ≥2 primary sources each, HMAC-SHA256 signed envelopes. Endpoints: catalog (/api/v1/claims.json), per-claim envelope (/api/v1/claims/{id}.json), methodology (/api/v1/methodology.json). TypeScript SDK + OpenAPI 3.1.0 spec.