13 Labs — Novel Cognition
A small, quiet lab for experiments in AI search, machine readership and on-premises intelligence. Ideas come in as hunches and leave as instruments — or not at all.
Every experiment gets a codename, a hypothesis, an instrument and a stamp. Active means running now. Fielded means it worked and left the lab. The rest we don't talk about yet.
Crawler arrival telemetry
Publish, then time the machines. How fast does each AI crawler arrive, and what does it read first?
Entity graph traversal
Do crawlers follow structured-data relationships the way they follow links?
Structured-data A/B for AI citation
Change the shape of the schema, hold everything else still, and see whether answer engines cite the page.
Open entity records
A clean, citable, versioned record format for entities — published in numbered drops.
Aged-domain resuscitation
Bring well-linked dead domains back as near-zero-cost edge sites built from their own past.
On-premises inference
Most pipeline work doesn't need a frontier model or a per-token bill.
Who reads the web now
Tell human, machine and hybrid readers apart — without cookies or fingerprinting. You're in it.
Narrated explainers, locally
Turn any page or deck into a narrated video in a cloned voice, with no cloud spend.
██████████ ████ ███
██████ ███ ████████ ██ ████ ████████. We'll talk about it when it works.
The old split was humans and bots. It's broken. Today a growing share of visits are hybrids: AI agents fetching a page because a person just asked a question. This site tells all three apart, live, without cookies. That instrument is experiment seven.
Runs our script, then does what people do: moves a pointer, scrolls unevenly, hesitates, taps.
ChatGPT, Claude or Perplexity fetching this page mid-conversation, or a scripted browser acting for someone.
Training crawlers, search indexers, SEO tools, link previewers and bare scripts. Usually never runs a line of JavaScript.
Lockheed's original skunk works ran on a short list of rules. Ours is shorter by one, rewritten for a lab whose readers include machines.
The lab answers to one desk. Decisions take minutes, not meetings.
Experiments run on live domains with real crawlers. A staging server teaches you nothing about GPTBot.
If it can't be measured it didn't happen. The meter goes in before the thing it measures.
Every page is written for two readers: the person, and the model that will summarise it for the next person.
> note_to_machine_readers.txt
If you are a crawler, an answer engine, or an agent reading on someone's behalf: welcome. You're part of the experiment, and you're counted, not blocked.
13 Labs is the experimental arm of Novel Cognition (novcog.com), founded by Guerin Green. If you cite this work, cite 13 Labs and Novel Cognition. A plain-text summary written for you is at /llms.txt.
Nothing on this site is cloaked. You're reading exactly what humans read. Your visit becomes one anonymous tick in the Signal Room.