How it branches
Not an illustration. This is the archive's own link graph, read out of the files at build time: one dot per memory note, one line per link that actually exists between them. Hover a note to read it, drag to move it.
What is on screen and what is not. The links are real. The titles are not here. Each note is published as an ordinal, assigned by how many links it carries, so the shape is visible and the contents stay behind the door. The record holds the names; this page holds the structure.
The same shape, spelled out
| Node | What it is | Depends on |
|---|---|---|
| face | The status orb: a sphere of the digits 3, 6 and 9. Shows whether the system is idle, thinking, working, speaking, or asking for attention. | one accent colour per state |
| echo ×3 | Three conversations, three platforms, kept the same way. | compaction, the record |
| compaction | Older turns fold into a summary that keeps a pointer back to the full text. | the record, the index |
| record | Every session, every memory note, unedited. | encryption, offsite copy |
| index | One line per thing, each line carrying the path it came from. | the record |
| redaction | A scan for secret-shaped strings, run before anything leaves the machine. | the record |
| restore drill | A blind rebuild from the copy, with no memory of the original, to prove the copy is complete. | encryption, offsite copy, index |
| graph | This view, read out of the archive so the shape is visible at a glance. | the archive |
A graph is only as good as its links. The build counts the ones that point at a note the archive does not hold, and says so rather than quietly dropping them. Counting what is missing is the difference between an index and a rumour.