docs/the-demo.md

The demo

tests/render-city.ts — point it at a repository the system has never seen, and it becomes a city that states its own reasoning.

─── Enterprise XI
  115 files · 662 facts → 43 districts, 113 buildings
  refused: 38× import dependency · 2 actors unlocated · 2 unplaced
  pipeline: 6 stages executed, 0 declared-only
  probes: symmetrise=confirmed embed=confirmed distance=survived weight=survived

What a viewer can tell apart

The question this artifact exists to answer is not "does it look good". It is whether a human can separate fact, interpretation, and refusal.

Click any building. It answers all five questions, per mark:

height — 21.92 world units
  from repo/bytes = 50000 · divided by the largest value currently observed,
    square-root transformed, scaled by 26, offset by 2, floored at 0.4
  scope world · warrant evidence + convention — floored at 0.4
  actually claims: tall relative to the current population of the world
  discarded (hidden-basis): the measured magnitude itself …

footprint — 3 · decorative
  from nothing · a constant per subtype, carries no semantic claim

Claim-changing marks sort first; decorative ones last. A reader who stops after two entries has seen the marks most likely to mislead.

What it refuses, in the picture

38 import dependencies — the fact vocabulary has no binary state family, so there are no facts to draw. Named, counted, and absent.
2 actors, 0 located — a commit licenses author → file and nothing about place. No position was invented.
2 files unplaced — never co-changed, so no metric licenses a position. Omitted rather than parked at the origin, which would read as central.

What it admits about its own arrangement

co-change      compression   removed distinctions irrelevant to the claim
weight         compression
distance       lossless
shortest-path  convention    selects a reference its input does not contain
symmetrise     convention    removes distinctions relevant to its own claim
embed          convention

Three of six stages are choices, not measurements — and they are the last three, closest to the pixel. Position is evidential right up until it becomes a picture.

And the losses are witnessed, not asserted:

symmetrise  confirmed   both produce d(a,b) = 1. Two opposite relationships are one number.
embed       confirmed   the tetrahedron's four equal distances embed as 0.745, 0.972, 0.901

survived appears twice and the page says what it means: one probe passed. Absence of contradiction is not proof of correctness.

Evidence versus interpretation

460 evidence · 202 convention

File sizes, containment, commit and author counts are measured. "A commit is a transfer of work from an author to a file" is a reading. The 90-day staleness threshold is a stipulation. Both are declared rather than absorbed.

Standing


Runs on an unprepared repository?	yes
Every mark traceable to a fact?	yes — asserted by the manifest tests
Refusals visible?	yes — three classes, counted
Ontology expanded?	no

What is still not established

This is not the product surface. It is a standalone artifact. The 3-D app still renders deriveStructure's fixed estate, and the inspector panel wired into WorldView remains visually unverified — the app is passkey-gated and signing in is the user's action.
Position carries no information about coupling — 54.5% of proximity comparisons reversed, measured earlier. The page renders it and says so; it does not fix it. A viewer reading nearness as relatedness is wrong more than half the time, and the page's own stage table is what tells them.
Comprehension is untested. Whether a viewer reads the caveats and understands them is condition 3, and no test here can establish it.
One repository, TypeScript only, 115 files.

301 tests, 26 DOM, 40 mutations, tsc clean.