2026-06-06 updated 2026-08-31 ● work in progress 260606-sstcorpus

SSTorytime corpus

Every *.n4l knowledge file from Mark Burgess's SSTorytime, imported into ipm and rendered — a whole foreign notation put through the γ(3,4) model at once, to see what carries over and what breaks.

44files

52,239n4l lines

19,899ipm nodes

2,604edges

38rendered

file	lines	nodes	edges	conn	E / T / C / ?	splits	validate	render	artifacts	no defaults	sst native

Click a column to sort, a row to keep it marked while you read across.

The pipeline

.n4l

SSTorytime's own notebook source, parsed by SSTorytime itself (N4L -export) — we do not re-implement the notation.

import & type

The exported graph becomes an ipm model: every node typed event / thing / concept by the node-kind solver, oversized nodes split.

render & validate

A flat diagram per model, checked against the γ(3,4) type table — validate errors should be 0.

zoom canvas

Each model is also baked into a self-contained offline zoom canvas — the zoom artifact link.

Reading the table

conn — edges / nodes, the graph's connectivity. low coverage is conn < 0.2: custom-arrow or annotation-heavy documents (MobyDick, Darwin, NDA) where most of the text never becomes an edge.
E / T / C / ? — the node-kind mix the solver landed on; ? counts what it could not decide. splits counts poly-kinded nodes split in two.
render — too-large and huge are above the render cap, not failures; cyclic and render-crash are.
artifacts — the default stack: solver defaults filled in. no defaults is the strict, constraint-faithful variant, and sst native takes node kinds from SSTorytime's own experimental ETC export instead of our solver — three readings of the same file, side by side.

This is a sandbox measurement, re-run whenever the layout engine moves; a red cell is a finding, not a bug report.

← back to the lab