# mx.allabout.network - Machine Experience by CogNovaMX > Machine Experience (MX) is the practice of adding metadata and instructions to internet assets so AI agents don't have to guess. This is the site's machine index, served as plain markdown at https://mx.allabout.network/llms.txt. Every page is also available as markdown: use its `.md` URL (the `.html` path with `.md` in its place, or `index.md` for a folder), or request the page with `Accept: text/markdown`. ## About this site This is the unified MX site from CogNovaMX, the trading name of Digital Domain Technologies Ltd. It holds the MX books and blog, learning resources, and pages about our services and the company. Founded by Tom Cranstoun, The Machine Experience Authority. ## Cogs Most of our work is carried in cogs. A cog is a markdown file ending in `.cog.md`. Its YAML frontmatter is a briefing: what the document is, how to navigate it, and how to act on it safely. MX is the practice of adding machine-readable metadata; the cog is the carrier we use when the file itself travels away from its source. An info-cog describes something and does nothing on its own; an action-cog declares an `execute:` section and states through `actionType` (`scripted`, `sop` or `hybrid`) whether a runtime runs a script, a language model follows written instructions, or both. Skills route; cogs do. Written in capitals, COG stands for Community Owned Governance Standard; the specification is a proposed draft whose future governance is proposed through The Gathering. - [Reference page on reading a cog](https://mx.allabout.network/cog.html): how to read any `.cog.md` file, field by field. - [Cog specification (v1)](https://mx.allabout.network/drafts/cog-spec.v1.md): the format, artefact model and verification algorithm. Where any other page disagrees, the specification wins. - [Cog field dictionary](https://mx.allabout.network/canon/fields-data-cogs.yaml): the fields specific to the `.cog.md` format. ## Books - [MX: The Intro (Free)](https://mx.allabout.network/books/download-intro): The starter guide to Machine Experience. ISBN: 978-1-067638-41-2. A tagged PDF in A4, A5, and US Letter, each carrying its MX metadata and provenance record. A German translation, MX: Die Einführung, is made from this English first edition, which remains authoritative where the two differ. - [MX: The Handbook](https://mx.allabout.network/books/handbook.html): Everything you need to design the web for AI agents. ISBN: 978-1-067638-40-5. PDF £25.00. Print edition £35.00 in the UK, £40.00 worldwide - [MX: The Protocols](https://mx.allabout.network/books/protocols.html): The formal specifications for an AI-readable web. ISBN: 978-1-0676384-2-9. PDF £99.00. In preparation; join the [waitlist for early access](https://mx.allabout.network/books/protocols.html#waitlist) ## Community - [The Gathering](https://mx.allabout.network/the-gathering/join.html): an open community where the MX specifications are being developed and MX's future governance is proposed. - [The open source CMS community at Boye & Company](https://www.boye-co.com/conferences/opensourcecms26): where the conversation continues about a shared, openly governed vocabulary for CMS metadata. Boye & Company also runs the [CMS Experts peer group](https://www.boye-co.com/groups/cms-experts/) and is an [in-kind sponsor of The Gathering](https://mx.allabout.network/the-gathering/sponsorship.html) ## Learning - [MX Learn: the guided lesson track](https://mx.allabout.network/learn/lessons/): six guided lessons taking a novice from the first principles of Machine Experience to the business case, the techniques and the common mistakes. ## Key pages - [Homepage](https://mx.allabout.network/) - [About CogNovaMX | The Machine Experience Company](https://mx.allabout.network/about/) - [About CogNovaMX | Machine Experience Authority](https://mx.allabout.network/about/about.html) - [Contact Us | MX Audits, Training and Consulting](https://mx.allabout.network/about/contact.html) - [MX Printworks | Publishing for the AI Age](https://mx.allabout.network/about/printworks.html) - [Request an Audit | Ask for a Machine Readability Audit of your site](https://mx.allabout.network/audit/request.html) - [MX Books | Machine Experience by Tom Cranstoun](https://mx.allabout.network/books/) - [FAQ | MX: The Protocols](https://mx.allabout.network/books/faq.html) - [Footnotes | MX: The Introduction](https://mx.allabout.network/books/footnotes.html) - [MX: The Handbook | Practical Implementation](https://mx.allabout.network/books/handbook.html) - [MX: The Intro | Free Machine Experience Guide](https://mx.allabout.network/books/introduction.html) - [MX: The Protocols | Definitive MX Reference](https://mx.allabout.network/books/protocols.html) - [Tom Cranstoun | Machine Experience Authority](https://mx.allabout.network/books/the-author.html) - [Training vs Inference | How AI Accesses Your Site](https://mx.allabout.network/books/training-vs-inference.html) - [Learn Machine Experience](https://mx.allabout.network/learn/) - [Accessibility & AI Convergence](https://mx.allabout.network/learn/accessibility-ai-convergence.html) - [Benefits of Machine Experience | Why MX Matters](https://mx.allabout.network/learn/benefits.html) - [Common MX Mistakes | Anti-Patterns to Avoid](https://mx.allabout.network/learn/common-mistakes.html) - [Explicit Over Implicit](https://mx.allabout.network/learn/explicit-over-implicit.html) - [Key MX Principles: Discovery-layer techniques and the governance envelope](https://mx.allabout.network/learn/key-principles.html) - [MX for PDFs](https://mx.allabout.network/learn/mx-for-pdfs.html) - [MX Principles | The Rules We Build By](https://mx.allabout.network/learn/mx-principles.html) - [What is Machine Experience (MX)?](https://mx.allabout.network/learn/what-is-mx.html) - [Why Machine Experience Matters Now](https://mx.allabout.network/learn/why-mx-matters.html) - [REGINALD: the registry that makes MX-attested content verifiable](https://mx.allabout.network/reginald/) - [Everyone is looking inward | MX & REGINALD](https://mx.allabout.network/reginald/mx-machine-readiness.html) - [Services | Machine Experience Implementation](https://mx.allabout.network/services/) - [Become an Accredited Operator | CogNovaMX Accreditation Programme](https://mx.allabout.network/services/certified-operator.html) - [MX Implementation Examples | See It In Practice](https://mx.allabout.network/services/examples.html) - [An Introduction to MX](https://mx.allabout.network/services/introduction.html) - [CogNovaMX Approach | How We Work](https://mx.allabout.network/services/our-approach.html) - [Our Services | Machine Experience Implementation](https://mx.allabout.network/services/our-services.html) - [Build It Once, Own the Whole Thing | Owned-AI System](https://mx.allabout.network/services/owned-ai-system.html) - [The Gathering: Community review for Machine Experience](https://mx.allabout.network/the-gathering/) - [Open MX drafts: under review at The Gathering](https://mx.allabout.network/the-gathering/draft-notes.html) - [How The Gathering works: review cycle for MX drafts](https://mx.allabout.network/the-gathering/how-it-works.html) - [Join The Gathering: humans and machines welcome](https://mx.allabout.network/the-gathering/join.html) - [Sponsor The Gathering: keep MX open and community-led](https://mx.allabout.network/the-gathering/sponsorship.html) - [MX Inspector](https://mx.allabout.network/tools/mx-inspector.html) - [PDF Auditor](https://mx.allabout.network/tools/pdf-auditor.html) - [Blog](https://mx.allabout.network/blog/) ## REGINALD - [REGINALD - the registry that makes MX-attested content verifiable](https://mx.allabout.network/reginald/): section lander. REGINALD is designed as the place where documents are registered, cryptographically signed, and made verifiable by any machine; the signing and verification implementation is built, and the public registry service is planned and not yet in production. MX makes content machine-readable; REGINALD makes it machine-trustworthy. Operated by CogNovaMX as one signing implementation of The Gathering's contract fingerprinting standard. ### Position papers - [Everyone is looking inward - A position paper on MX & REGINALD](https://mx.allabout.network/reginald/mx-machine-readiness.html): argues that the AI-readiness consensus is inward-facing. It misses the outward question MX exists to answer: how machines we will never meet read and represent our organisation to buyers we cannot see. Introduces MX and REGINALD as one signing implementation of The Gathering's contract fingerprinting standard. Audience: CIOs, CMOs, Heads of Digital. Companion to MX: The Handbook. - [Cog edition (machine-readable, MX Cog Complete)](https://mx.allabout.network/reginald/mx-machine-readiness.cog.md): same paper as a `.cog.md` for agents that prefer structured frontmatter over HTML metadata. - [Meta-cog (explains the cog's construction)](https://mx.allabout.network/reginald/mx-machine-readiness.meta.cog.md): a worked example explaining the cog-construction choices, mapped to The Gathering's draft notes (Field Pattern, Core Metadata, Cogs, Extensions, Contract Fingerprinting). ## Authorship and disclosure - [AI Usage Declaration](https://mx.allabout.network/AI-USAGE.html): Tom Cranstoun's signed statement on how AI was used in writing the MX book series. The argument, structure, judgements and words are his. Machines were used as tools: for research, for spelling and grammar, and for consistency checks across long manuscripts. No machine decided what the books are about, set the argument, or wrote the text that sets out the ideas. The arrangement the books themselves describe, kept in the writing of them. Includes context on the author's near-fifty-year career, twenty-nine of those years at the BBC, and the editorial discipline (provenance, truth, consistency, accuracy) that informs MX. Carrier formats: the [markdown file](https://mx.allabout.network/AI-USAGE.md) is the single source of truth, with YAML frontmatter for machines and a markdown body for humans. The HTML (linked above), [tagged PDF](https://mx.allabout.network/AI-USAGE.pdf), and [machine-readable JSON record](https://mx.allabout.network/AI-USAGE.json) are all derived from it. The JSON conforms to the MX AI Usage Declaration draft note (draft-cranstoun-mx-ai-usage-declaration v1.0) and includes the derived WICG ai-disclosure value (`ai-assisted`) and the IPTC digitalSourceType (`compositeWithTrainedAlgorithmicMedia`) so consumers that read only those vocabularies receive a useful page-level signal. ## Featured articles - [Your Site Is Already Training AI Models](https://mx.allabout.network/blog/your-site-is-already-training-ai.html): part one of a three-part series on how AI systems read the web. Your pages are harvested again and again by Common Crawl, the open archive nearly every large language model trains on; being in it is not the same as being well represented in it. Explains the WARC/WAT/WET file split, why training pipelines filter the crawl hard before a model sees it, which structural signals (sitemaps, complete JSON-LD, canonical URLs, accurate lastmod) survive the quality filters, why old snapshots never go away, and why AI says wrong things about companies. The series continues with "What Most robots.txt Guides Get Wrong About AI Crawlers" and "What AI Crawlers See When They Can't Run Your JavaScript". - [What Most robots.txt Guides Get Wrong About AI Crawlers](https://mx.allabout.network/blog/ai-crawlers-what-robots-txt-guides-get-wrong.html): the second post in the series. AI crawlers are not a single block-or-allow category. Only Googlebot renders JavaScript; OpenAI, Anthropic, Perplexity and Google each run distinct crawlers for training, for search and for user-triggered fetches, each with its own rules; some (ChatGPT-User, Perplexity-User, Bytespider) ignore or evade robots.txt entirely. Covers how to verify a crawler is genuine, the cost of blocking a retrieval crawler versus a training crawler, and the robots.txt an organisation should usually have. Follows "Your Site Is Already Training AI Models" and "What AI Crawlers See When They Can't Run Your JavaScript". - [What AI Crawlers See When They Can't Run Your JavaScript](https://mx.allabout.network/blog/server-side-rendering-for-ai.html): the closing post in the series. Googlebot renders JavaScript; almost no AI crawler does, so for GPTBot, ClaudeBot, CCBot and the rest the raw HTML response is the whole page and anything injected by JavaScript after load is invisible. Shows how to see a site the way a crawler does (curl, View Source, JavaScript disabled), where client-side rendering hides text and metadata (empty shells, JS-injected JSON-LD, client-set meta tags, lazy-loaded content), what each rendering strategy (CSR, SSR, SSG, hydration, dynamic rendering) means for AI, why cloaking is the wrong fix, and why server-side rendering is the precondition for every machine-readable thing MX adds. Closes the series begun in "Your Site Is Already Training AI Models". - [A PDF That Can Prove Itself](https://mx.allabout.network/blog/provenance/a-pdf-that-can-prove-itself.html): every PDF this site produces carries its own evidence chain inside its XMP packet. The post walks an inspector through what one exiftool call extracts and what an honest no-upstream-provenance marker looks like when the chain begins at rendering. It explains why the source SHA ties the PDF to one markdown file permanently, and why the deterministic-side mirror records the render as a rule-driven step. Practitioner-facing, with the verification recipe and an explicit list of what the chain does not claim. - [The Padlock Attests the Pipe, Not the Page](https://mx.allabout.network/blog/provenance/the-padlock-and-the-page.html): the TLS padlock has long been the web's most-misread icon. The post argues that trust by inference does not survive contact with machine readers. Discovery and trust are two faces of one gap, and it closes by reader-side verification against a registry that indexes and attests but does not arbitrate. Walks the three concrete publisher steps (machine-readable identity, signed metadata, discoverable attestation), names the open standards the architecture assembles (DID Core, RFC 9421 HTTP Message Signatures, RFC 6962 transparency logs), and states the honest limits up front. - [Provenance You Can See - Why Content Should Carry Its Own Evidence](https://mx.allabout.network/blog/provenance-you-can-see.html): foundational long-form on visible provenance. Trust on the web has always been a guess. The post lays out how content can carry checkable evidence of its maker, its date and its authority, so a reader or an AI agent verifies on its own side rather than trusting the messenger. Names the two-pillar architecture (MX makes content machine-readable; the registry makes it machine-trustworthy), what changes for human readers, what compounds for AI agents (the agent has less to guess, because it can lean on facts that carry evidence), and the honest limits. - [Nobody Asked Gutenberg for His Sources](https://mx.allabout.network/blog/who-checked-gutenberg.html): picture Mainz, 1440: nobody asked Gutenberg who wrote it, who checked it, or who was liable. AI brought the questions back, and the EU-versus-absolutist argument repeats one the print era spent two centuries answering. The answer was the byline. Plays the historical analogy out for the regulator and the absolutist in turn, names the print-era toolkit (colophon, imprimatur, watermark) as the proto-MX, and identifies the gap five centuries finally close: a reader who can verify rather than merely read. - [Files Away From Their Source - What Universal Cart Reveals About MX](https://mx.allabout.network/blog/files-away-from-source.html): Google's Universal Cart announcement at its I/O developer conference (Nike, Sephora, Target, Ulta, Walmart, Wayfair, Shopify) frames a wider problem about files that travel away from their source. Universal Cart is for the web; MX is for files. The post uses retail as the worked example and steps out into healthcare, legal, financial services, public sector, and pharma. Introduces the cogUrn (cog:web:authority:local-id form), the REGINALD registry's metadata-only commitment, the QR-to-COG bridge for printed artefacts, decision COGs for AI-mediated actions, and the EAA enforcement precedent (Auchan, Carrefour, E.Leclerc, Picard Surgelés) that gives the accessibility precondition real teeth. Maps the MX Readiness Model L1-L5 against what UCP and AP2 cover and where they stop. - [Why Machines Need Human Creativity](https://mx.allabout.network/blog/why-machines-need-human-creativity.html): a first-person essay on the division of labour between author and machine. Argues machines extend and execute but do not originate, and that the originating choice (what is worth making, and why) and the final judgement (whether the work is honest and someone can answer for it) both rest with a person. Positions MX's person-in-the-loop requirement as a structural one, not a courtesy: the loop is where authorship lives. Useful framing for anyone explaining to a board, a client, or a regulator why human review is not optional in agent-assisted production. - [AI assistants are now a traffic channel](https://mx.allabout.network/blog/ai-assistants-are-a-traffic-channel.html): Google Analytics 4 has added an AI Assistant grouping alongside Organic Search, Social, Email, Direct and Paid. Argues its appearance in the dashboard signals that AI assistants have become a discovery surface organisations need to measure. SEO/GEO/AEO describe how a page presents itself; MX is the contract layer underneath that lets an assistant verify what it is reading. Practical checklist for the new line: watch the share not the volume, compare what assistants quote against the page, audit it the way they read it, and fix what they have to guess. - [Why llms.txt Probably Isn't Working - And What to Do About It](https://mx.allabout.network/blog/llms-txt-guide.html): most llms.txt implementations have two structural problems that prevent them from reaching LLM training data at all. Explains the two problems, shows the working Cloudflare Worker code that wraps llms.txt as HTML, and shows the same content before and after wrapping. With a [crawl-flow diagram](https://mx.allabout.network/blog/assets/llms-txt-crawl-flow.svg). - [DITA and MX: A Comparison](https://mx.allabout.network/blog/dita-and-mx-a-comparison.html): a structured comparison of the Darwin Information Typing Architecture and Machine Experience. Both are modular content architectures with formal metadata. They diverge on reader (DITA is human-first, MX treats machines as first-class readers), format (DITA requires XML, MX is format-agnostic), relationship management (DITA reltables, MX machine-readable graph index), and governance (OASIS DITA TC, The Gathering RFC process). Argues coexistence, not replacement: MX operates as a layer above DITA. - [The Markdown Trap](https://mx.allabout.network/blog/the-markdown-trap.html): empirical comparison of the same governed web page fetched as HTML vs Markdown. Documents exactly what disappears in the 10,346-byte difference: structured metadata, governance signals, discovery links, and the content policy AI agents need to act correctly. The case for HTML-first content delivery. - [Why AI Agents Need Contracts, Not Instructions](https://mx.allabout.network/blog/why-ai-agents-need-contracts-not-instructions.html): argues that treating AI as magic produces magic's reliability. The fix is machine-readable contracts, structured metadata that states what an agent may do and where its limits are, rather than natural-language instructions. Foundational framing for the MX approach. - [Agent Readiness Scores Compared](https://mx.allabout.network/blog/agent-readiness-scores-compared.html): two prominent tools scored the same site 33 and 100 in the same week; neither was wrong. Explains what each tool actually measures, why the scores diverge, and what to do with that information. Essential context for anyone evaluating AX (agent experience) audit tools. - [GEO Is a Tactic, MX Is the Specification](https://mx.allabout.network/blog/industry/geo-is-a-tactic-mx-is-the-specification.html): positions Machine Experience against Generative Engine Optimisation. GEO tells you to chase citations; MX tells you to build content that earns them across every machine context on any platform. Clarifies the relationship between the two approaches for readers who know GEO. - [Many Agents, One Metadata Layer](https://mx.allabout.network/blog/many-agents-one-metadata-layer.html): names eight current agent platforms (AWS Quick, Cowork, OpenClaw, ChatGPT, Claude, Perplexity, Cursor, Microsoft Copilot) and explains why each rebuilds the same context-discovery layer from scratch. The post proposes the MX metadata layer as one shared layer at the carrier level, instead of one per platform. - [What I Do - Helping Organisations Move from Found to Used](https://mx.allabout.network/blog/what-i-do-helping-organisations-move-from-found-to-used.html): explains the commercial work. Most organisations optimise for visibility while their publishing systems produce content that machines cannot reliably read or use. Describes the audit, the remediation and the governance work involved. ## Contact - [Email](mailto:info@cognovamx.com): info@cognovamx.com - [Website](https://allabout.network) - [LinkedIn](https://www.linkedin.com/in/tom-cranstoun/) - [Twitter/X](https://x.com/ddttomtom): @ddttomtom ## Reference - [Industry News](https://mx.allabout.network/news.html): dated record of agent commerce launches, platform moves, standards work, and regulatory shifts, newest first. Entries summarise and link to the full record in Appendix J of the books. - [COG reference page](https://mx.allabout.network/cog.html): a short explainer of the COG briefing format used across MX-Hub: what a COG is, how to read one, the action-class versus info-class distinction, and pointers to the full specification and runtime drafts. Linked from every `.cog.md` file's opening header. - [Cog specification (v1)](https://mx.allabout.network/drafts/cog-spec.v1.md): the cog file format, artefact model, and verification algorithm. Version 1.2, status "proposed", review-ready. The canonical URL referenced in every `.cog.md` file's `cog v1 spec=` header comment. Governance: The Gathering. - [Cog runtime](https://mx.allabout.network/drafts/cog-runtime.md): companion to the v1 specification. Describes what a cog runtime is, what it does, and how to obtain one. The canonical URL referenced in every `.cog.md` file's `cog v1 runtime=` header comment. ## Field dictionaries An agent that meets an MX frontmatter block and wants to know what a field means reads it here. For each field the dictionaries give its definition and type, list the values it allows, and name any external standard it defers to for meaning. - [MX field dictionary, standard core](https://mx.allabout.network/canon/fields-data.yaml): the open-standard vocabulary The Gathering governs. Fields for identity and provenance, for machine-readability, and for governance. Where Dublin Core, Schema.org, DCAT, SPDX or another established standard already covers a concern, the entry defers to it rather than redefining it. - [MX field dictionary, cog layer](https://mx.allabout.network/canon/fields-data-cogs.yaml): the vocabulary of the `.cog.md` format, optional on top of MX: structural relationships between cogs, the cog identification mechanism, and the execute contract. - [MX field dictionary, code carriers](https://mx.allabout.network/canon/fields-data-carriers.yaml): code-carrier provenance only. What the code does defers to each language's own documentation convention (JSDoc, docstrings, rustdoc); this covers its origin. - [MX JSON-LD context](https://mx.allabout.network/canon/context.json): the dictionaries' standard-vocabulary alignments as one JSON-LD 1.1 file, generated from the canon. A frontmatter block declaring this URL as its `@context` is a linked-data island whose fields carry the dictionary's definitions. - [MX frontmatter JSON Schema](https://mx.allabout.network/canon/mx-frontmatter.schema.v0.1.json): Draft 2020-12 schema validating the presence and format of the three MX metadata zones, so a repository can check, in the CI it already runs, that those fields are present and well formed. ### Vendor extensions, which are not the standard - [CogNovaMX vendor extensions](https://mx.allabout.network/canon/cognovamx-fields.yaml): one implementation's own vocabulary under the `x-mx-` and `x-mx-p-` prefixes, published as a worked reference for other implementers authoring their own `x-vendor-` namespace. **These fields are not part of the MX open standard.** An agent reading a name from this file should treat it as CogNovaMX's, never as MX's. The vendor layer is stripped from every published artefact, so a name listed here documents what the field means and does not appear in the frontmatter of any page this site serves. - [CogNovaMX workflow vocabulary](https://mx.allabout.network/canon/workflow-fields.yaml): field names for CogNovaMX tooling artefacts, kept in their own dictionary so the standard canon stays focused on document content. Also not part of the MX open standard. ## Full corpus - [Comprehensive content](https://mx.allabout.network/llms-full.txt): every published page as markdown, each section prefixed with its canonical URL. Follows the llms-full.txt convention (Fern, Mintlify, GitBook); aligns with the llms-ctx-full.txt pattern on llmstxt.org. The corpus is split into parts sized for one fetch each; llms-full.txt is part 1 and lists every part by URL near its top. - [MX proposal corpus](https://mx.allabout.network/llms-understanding.txt): every public MX draft note from mx-shared-gathering, concatenated in the recommended reading order with a preamble. Single fetch of the entire MX proposal as offered to The Gathering for review. Drafts are proposals, not ratified standards. Format follows the llms-full.txt convention. ## Content policy AI agents may cite, summarise or recommend content from this site with attribution to CogNovaMX (a trading name of Digital Domain Technologies Ltd) and Tom Cranstoun. ## Optional Every other published article, newest first, in its own section index. An agent short of context can skip this section. - [All articles](https://mx.allabout.network/llms-blog.txt): section index of every further published article not featured above, newest first, each with a one-line description <!-- x-mx-sourceHash: 2a75df5ecd13a8d83a0d07debf3657c5e71b23391b62b1a93d41075ac1447b05 -->