The site's own provenance
This site is the provenance layer of a network — it exists to say where a body of work's thinking came from. A site that asks that question of somebody else owes you the same answer about itself. Where the register came from, how an entry is written, what the release pipeline refuses to publish, and the four things the whole apparatus cannot do.
None of this is on the front page, on purpose. The front page is for the ideas; this section is for how they got there. Both matter and they are not the same reader.
Where the register came from
The site was commissioned by a nine-document pack — the brief, the format, the influence map with its evidence, the architecture, the boundaries, the gaps, a source manifest, the licence, and a summary of the working document the whole format was canonised from. It is published in full, unredacted, because a site that argues from a document should let you read the document.
| Step | What happened |
|---|---|
| 1 · the list | Dinis Cruz named the people, works, topics and things he believed had shaped his thinking — individuals, topics, eras, things, including books and including musicians. |
| 2 · the format, already invented | He had already written an immediate-connection register for Bret Victor's Inventing on Principle, as a working document about the estate's viewer. The format existed before the site did, which is why it is proven in use rather than designed on spec. The seven blocks → |
| 3 · the mining | The corpus — the estate, the docs site, the issue stores — was searched for every name on the list and every external name the search itself surfaced. That second half is where the discovered entries came from. |
| 4 · the tiers | Entries were sorted by what the evidence supports, not by importance. TRACED, STATED, DISCOVERED → |
| 5 · the register | Everything became one JSON file, and the site became a projection of it. data/influences.json |
The nine documents, in full How an entry is written The three tiers Tier movement — the changelog
The register is the site
Everything under /register/, all of the map, and every count on every page comes from one file. Nothing about this site's own shape is typed by a person.
# one file, four projections data/influences.json ├─▶ register/<slug>/index.html the entry, seven blocks ├─▶ register/<slug>/index.md its markdown twin ├─▶ register/<slug>/trace/ the table as page, markdown and JSON └─▶ map/index.html + graph.json the graph, computed not drawn # and the counts, everywhere they appear chrome.py writes every tier tally into every page validate.js recomputes them from the same file and compares
The prose is authored once. Every sentence of resonance and every trace row is written in the JSON, in a deliberately tiny markdown — links, bold, italic, code — and rendered twice: as the HTML page and as its markdown twin. Two renderings of one string cannot drift. Writing the page by hand and the twin by hand is the arrangement that always does.
Two rules, enforced rather than stated
The release gate runs six checks. Four are the ones every site in this network runs — version agreement, internal links and fragments, canonical host, and a credential-shape tripwire. Two exist only here, and both turn an editorial commitment into something that can fail a build.
The no-verbatim gate
This site explains why a work resonated and traces where it was applied. It does not reproduce the work. Every talk, book, essay, keynote and record named here belongs to its author and stays where they put it.
That was Dinis Cruz's own rule, and it happens to be exactly what copyright law rewards: the analysis, the resonance and the trace tables are original work and carry CC BY 4.0 cleanly, while a talk transcript or a verse of lyrics inside a CC BY page would be relicensing what the site does not own.
# every quotation declares whose words it carries <blockquote data-quote="founder">…</blockquote> # unrestricted: his own writing <blockquote data-quote="Steve Jobs">…</blockquote> # a third party: 40 words <blockquote>…</blockquote> # fails the release ✗ 62-word quotation attributed to "X" exceeds the 40-word cap the fix is to cut the quotation, never to raise the cap
What it cannot do, stated plainly: it cannot see an unmarked quotation woven into a paragraph, and it cannot tell you whether a 39-word quotation was fair. What it can do is make the marked case cheap and the unmarked case a review finding, and stop the failure mode that actually ships — a long passage pasted in whole because it was useful at the time.
Two corollaries the site lives by. Lyrics are the sharpest edge: publishers enforce aggressively even over a single verse, so the Rush entry describes and links and quotes at most a fragment under analysis. And likenesses are original artwork only — never scraped photographs, never AI renderings presented as portraits. Several of these people are alive and known to this project's community, which makes it a matter of respect before it is one of personality rights.
The register is the data
Twenty-five entries in three tiers, and the counts appear on the front page, the register index, the tiers page and llms.txt. None of those numbers is typed. A page writes an empty slot; chrome.py fills it from the register; validate.js recomputes the same counts from the same file and fails the release on disagreement. Both halves have to be wrong in the same direction to ship a wrong number.
The reason this earns a gate at all: tier movement is this site's changelog. An entry going STATED → TRACED because a briefing document landed is the most interesting event the site records, and a hand-typed seven stated on three pages is exactly what makes that event invisible. The same check catches both orphan cases — an influence in the data with no page, and a page with no influence behind it.
The pipeline, end to end Release history Comms: the briefing queue
What this site cannot do
Stated as a section rather than a footnote, because a register that only ever confirms itself would be an autobiography with citations.
A trace table cannot show causation
It shows that a pattern from an anchor work appears at a named place in the estate. It cannot show that the anchor work put it there. The Victor register's own finding is the honest framing and the weaker one: half the estate's strongest viewer features turned out to be unknowing implementations of the talk's demos — evidence that the instinct and the principle agree, rather than evidence of transmission.
The corpus is cited, not resolved
This repository holds the website, not the estate, so every evidence path is quoted from a dated mining run rather than checked at build time. Nothing on the site can disagree with the pack; the site can and will drift from the estate. Seven entries have no trace table for exactly that reason, and the entries say so instead of filling the gap with plausible rows. R1 →
The tiers partly measure what has recently been written down
Two decades of earlier material — an older blog, conference talks, decks, the earlier books — were unreachable when the mining ran, and that is precisely where the stated influences' evidence will be. An entry sitting at STATED may be the deepest influence on the list. R3 →
Nothing here has ever been retracted
Publishing your gaps is the cheap half of falsifiability; withdrawing a claim in public, with a date, is the expensive half. There is no DISCOVERED → removed row, no retraction, and no counter-influence entry. Until one of those exists, this section is evidence of ambition rather than of practice.
And the conflict of interest is the sharpest in the network. A project publishing the list of thinkers who shaped its founder is doing something where association is flattery, and it costs nothing to claim. Saying that Victor, Popper and Alexander are in your intellectual lineage makes your work sound better without making it better. Every incentive on this site points at overstating.
The defences are the ones available: every entry lists the files it stands on, the tier system separates evidence from assertion visibly, the discovered entries are claims about the founder that he is asked in public to strike, and the numbers are computed rather than asserted. None of that removes the incentive. It makes the site checkable by somebody who does not share it. The full disclosure, including the five places this approach is weakest →
For agents
An agent reading this site inherits the same split the licence makes. llms-full.txt is safe to ingest wholesale: it carries the site's own analysis, every entry's markdown twin and every source document, and nothing that belongs to a third party. That holds by construction rather than by promise — every part of it is assembled from pages the no-verbatim gate has already passed.
llms.txt llms-full.txt the register, as data the graph, as data