Not a criticism of the work: the work is the whole reason this page exists. This is the full version of the audit summarized on the review, plus a concept of what the site could look like once every publication, patent, and talk is a machine-queryable record instead of a paragraph of prose in a nested HTML table. Every number below was freshly measured; every fix is labeled concept, not a claim of anything already built for him.
evidence.json. This page is staged on our own turf, unlisted and noindex: never deployed to his infrastructure, never sent without the operator's say-so.
No jargon: what the small fixes below are worth in the only currency that matters for a corpus this good: getting found, read, and shared.
Most first-time visitors arrive from a podcast, a tweet, or an email on their phone, and today your whole corpus opens shrunk to about 40% scale, unreadable until they pinch-zoom, so many simply leave; the fix keeps the reader who was already interested.
Right now pasting your homepage into Slack, X, or a message renders a bare blue link with no title or image, so a mention that could pull people in reads as an afterthought: a small change makes each share carry your title, a line, and a picture.
Twenty-four papers, twenty-six patents, and 284 press mentions live only as prose an assistant can't query, so a journalist's or a researcher's AI asked “what has he published on causal entropy?” can't answer from your site: making it machine-readable puts your own work in those answers.
Five things your site already gets right, measured the same way the gaps below were.
| Finding | |
|---|---|
| Credit | The bio paragraph alone passes a real 5-second test: who you are, what you lead, who you back, what you publish, all in the first screen, with zero marketing filler. Most personal sites fail this; yours doesn't. |
| Credit | Every agent-discovery path we checked (robots.txt, llms.txt, sitemap.xml, five .well-known/* paths) returns a true, byte-identical 404 matching a nonsense-path control — no soft-404 deception. Honest absence beats a fake 200. |
| Credit | The corpus is real and dense: 24 publications, 26 patents, and, by our own count, 284 press mentions across 216 outlets (BBC News and Boston Globe among them), every one a working, direct link. Spot-checked, nothing broken. |
| Credit | The Innermost Loop's RSS feed (theinnermostloop.substack.com/feed) is real and live: the one piece of this whole surface that's already machine-readable today, and the newsletter with 100,000+ daily readers is exactly the content an agent would most want to subscribe to. |
| Credit | No cookie interstitial, no popup, one lightweight analytics tag. Nothing fights the visitor. That restraint is worth keeping in any redesign, not "fixing." |
Five things we measured that are costing you, quietly, on the two audiences that matter most now: phones, and agents.
| Finding | |
|---|---|
| Gap | No <meta name="viewport"> tag, on any page. Measured with a real 390px-wide mobile viewport: the browser falls back to a 980px desktop-width layout and auto-shrinks it to fit: window.innerWidth reports 980, visualViewport.scale reports 0.398. On the channel most first-time visitors arrive from, a link shared from a podcast, a tweet, an email, your entire corpus opens pre-shrunk to about 40% scale, unreadable until pinch-zoomed. |
| Gap | Zero machine-readable markup, on any of the four pages checked. No JSON-LD or schema.org markup (the invisible labels search engines and AIs read), no Open Graph tags: paste your homepage URL into Slack, X, LinkedIn, or iMessage and no title, description, or image card renders; it shows a bare blue link. (Your own company site, psi.inc, ships proper og:title/og:description/og:image tags today: the practice is already native to your orbit, just not on your personal domain yet.) |
| Gap | The headshot is 1.66MB, displayed at 185px tall. No responsive srcset, no modern format (WebP/AVIF), no width attribute (only height, inviting layout shift as it loads): roughly 9,000% more bytes than the display size needs, on a page with no other art direction to justify it. |
| Gap | No HTTP compression, no cache-control, on the HTML documents. The server sends identity encoding even when a client explicitly offers gzip, br, and ships no cache-control header on the page itself: a repeat visitor re-downloads the same bytes every time. |
| Gap | The corpus exists only as prose in nested HTML tables. Zero <div> elements, no <!DOCTYPE html> declared, three of the four corpus tabs (/publications, /patents, /press) 301-redirect to a trailing-slash URL before they render. An agent, or a journalist, or a due-diligence LLM, can't ask "what has he published on causal entropy" or "list patents about environmental measurement" without parsing 24+26+284 free-text entries by hand. |
The everclear version of what a first-time visitor, human or agent, actually needs in the first five seconds. Counts are your own site's stated numbers, cited inline.
Intelligent-seeming behavior emerges from a system maximizing causal path entropy: the diversity of futures reachable within a horizon T. The theoretical anchor of your later work.
The public explainer of the paper above: 2M+ views, 27 languages, per your own site. Intelligence as the drive to keep future options open.
The engineering counterpart to the PRL result: the theory made into a claimed mechanism, one year before the paper published.
A system that calculates a website's environmental footprint from server, network, and user-terminal emissions data: a literal environmental-measurement instrument, years ahead of most of the industry's own ESG tooling.
Proves optimal physical locations exist for coordinating trades between light-speed-separated exchanges: computation has a geography, and you calculated it.
Discovering and commercializing physics breakthroughs with AI "safely, verifiably, and for broad public benefit": team roots at Google, OpenAI, Harvard, Meta, NVIDIA, MIT, Stanford, Oxford, and more.
One JSON-LD block per publication, embedded on its own page, using schema.org's ScholarlyArticle type: the same markup Nature, arXiv, and Google Scholar already read. Worked example, using only facts already public on your own /publications page:
<script type="application/ld+json"> { "@context": "https://schema.org", "@type": "ScholarlyArticle", "name": "Causal entropic forces", "author": [ { "@type": "Person", "name": "Alexander D. Wissner-Gross", "url": "https://www.alexwg.org/" }, { "@type": "Person", "name": "C. E. Freer" } ], "isPartOf": { "@type": "Periodical", "name": "Physical Review Letters" }, "volumeNumber": "110", "pageStart": "168702", "datePublished": "2013", "url": "https://www.alexwg.org/publications/PhysRevLett_110-168702.pdf" } </script> // schema.org has no stable core type for a granted patent yet: // pragmatic fallback used by most patent-office sites: CreativeWork + additionalType <script type="application/ld+json"> { "@context": "https://schema.org", "@type": "CreativeWork", "additionalType": "https://en.wikipedia.org/wiki/Patent", "name": "Environmental footprint monitor for computer networks", "identifier": "US 8,862,721 B2", "author": ["Alexander D. Wissner-Gross", "Thomas M. Sullivan"], "datePublished": "2014" } </script>
A proposed /llms.txt (the file AI assistants read first): the emerging convention (used by Anthropic, Stripe, Vercel, and Carbonmark/Klima, among others) for pointing an LLM straight at the content worth reading, in the order it matters, in plain markdown a model can consume in one shot instead of crawling 300+ HTML pages.
# Dr. Alexander D. Wissner-Gross > Physicist, computer scientist, entrepreneur, and investor working at the > intersection of AI and the physical world. Founder & Chief Scientist, > Physical Superintelligence (PSI). 24 publications, 26 patents, 128 distinctions. ## Publications - [Causal entropic forces](/publications/PhysRevLett_110-168702.pdf): Phys. Rev. Lett. 110, 168702 (2013). Intelligence as maximizing causal path entropy. - [Relativistic statistical arbitrage](/publications/PhysRevE_82-056104.pdf): Phys. Rev. E 82, 056104 (2010). Optimal physical siting for light-speed-separated trading. - Full index: /publications ## Patents - Causal entropy engine (US 61/738,573, 2012) - Environmental footprint monitor for computer networks (US 8,862,721 B2, 2014) - Full index: /patents ## Talks - A new equation for intelligence: TED / TEDxBeaconStreet (2013-11-17) - Full index: /talks ## Organizations - [Physical Superintelligence](https://www.psi.inc): Founder & Chief Scientist - [021T](https://021t.vc) · [Reified](https://www.reifiedllc.com): investment vehicles - [The Innermost Loop](https://theinnermostloop.substack.com): daily newsletter, 100k+ readers ## Optional - [Full CV](/AWG-CV.pdf) - [Press](/press): 284 mentions, 216 outlets
| Metric | Today (measured) | Concept |
|---|---|---|
| Mobile layout scale | 0.398× forced zoom, no viewport meta | 1.0×, responsive by design |
| Structured data | 0 JSON-LD, 0 Open Graph tags, 4 pages checked | JSON-LD per work + full Open Graph |
| Agent discovery | true 404 on every path checked | llms.txt indexing every work |
| Homepage weight | ~1.78MB (1.66MB in one uncompressed PNG) | <200KB first paint, target |
| Corpus queryability | linear HTML prose; no feed but the newsletter | one JSON index, 24+26+284 records |
None of this is a request. It's the same standard applied to our own surfaces, cite the source, measure the claim, never invent a number, turned, for one page, onto the site holding the work this whole package is actually about.
This is the actual engagement: a real, measured review, because his corpus deserved one. The gentle pitch further into this package is the appendix, not the point. If nothing else lands from any of it, a ten-minute fix to the viewport tag and a real Open Graph block would probably do more for how this corpus is discovered and shared, today, than anything else in this note.
