A selvedge is the self-finished edge of woven cloth: the threads that turn back on themselves instead of fraying out. These forty-two glyphs did the same thing. Every one of them was ordinary — on shop ledgers, prescriptions, road signs, almanacs, account books, amulets, the margins of borrowed books — and every one of them now lives in a single specialist register: pharmacy, mathematics, astronomy, astrology, liturgy, palaeography, the occult. The dashboard's argument is that most of them did not die. They were captured: penetration collapsed by three to five orders of magnitude while persistence and retention stayed near the top of the scale. That combination has no status in the Loom's crosswalk, so this corpus proposes one.
Solid bar: the glyph in ordinary, non-specialist use. Hairline: its residual life in a contained register, dashed where retention has effectively stopped. The tick marks the exit event where one is dated. The year axis is deliberately non-linear — five segments, compression factors printed beneath — because a linear axis from 3000 BCE would render the entire print era as a smear.
The argument, plotted. Horizontal: orders of magnitude lost in penetration. Vertical: persistence inside the residual register. Death sits bottom-right. Nearly everything is top-right instead.
What actually removed the glyph from ordinary use. Countable within the corpus; not generalisable beyond it, per FD-GL01.
Ordered by source script, red to violet. Status is encoded as form rather than colour — hue belongs to the seven script dyes and nothing else.
This corpus is scored against the term_record contract in claude/Scraper-Metrics-Build-1.md, with one addition. Staying power is the conjunctive weighted geometric mean of amplitude, persistence and retention — any axis at zero zeroes the score — at weights 0.25 / 0.40 / 0.35, persistence carrying the most because persistence is the register-capture signal.
The corpus is curated backwards from glyphs that have a residual habitat. There is no row for the roughly four thousand Tironian notes, the thirteen thousand of the Carolingian expansion, the bulk of the alchemical Decknamen cipher, or the lunette alphabet of Shams chapter 19 — glyphs that left ordinary use and landed nowhere at all. Any claim of the form “glyphs tend to survive in specialist registers” is unavailable from this corpus in either direction. Identical mechanism to FD-SC03 in the metrics build. Blocks any survival-rate chart; the uneven band sizes on this page are a curation artifact, not a finding.
Every exit in this corpus predates 2006 and all but two predate 1950. The metrics build establishes that penetration, retention and community spread are undefined before roughly 2006 because there is no account-level record. Every score on this page is therefore editorial and permanently so. The lifespan bars are the only layer with a documentary basis; the scatter is an argument rendered as a chart, and is labelled as one.
Visibility here is partly a function of Unicode, which encoded most of these glyphs because a specialist community asked. Two entries in the corpus have no code point at all — the Shams seals and the Budūḥ square as a unit — and Siddhaṃ will not reliably render in a browser. A glyph with no code point produces no corpus and is structurally invisible to any scraper, which means measured amplitude has a floor set by the Unicode Consortium rather than by usage. Consequence for the Loom: the orthographic-carrier rule in standing law 3 needs an unencoded flag, and an unencoded thread can only ever enter by hand.
Standing law 4: no myth etymologies. These are the attractive stories this dashboard declines to carry, kept visible so the refusal is auditable rather than silent.