DOG-042 made [/wiki/Bro] fuse one F, and the token then resolved to NOTHING: door.js read a slash-headed ref as another repo's ROOT and tried the literal path, so it looked for a file NAMED wiki/Bro. No such file exists — the page is wiki/Bro.mkd. This is BEE-008's bug one pocket over: a ticket code is a stem and got its ladder, a wiki page is a stem too (wiki: a page is linked bare) and never got one. The reader sees a painted-but-dead reference, which is the worst of both readings — the lexer says "target", the door says "nothing".
Found 2026-08-16 on bee http /bee/cat/pager.js line 1, straight after DOG-042 landed: ([/wiki/Bro]) paints G:[ F:/wiki/Bro G:] and the F carries no <a href>, while ~/src/quickjab/bee-journal/wiki/Bro.mkd sits in a registered repo.
door.js inMount took tries = [String(partial)] for a slash-headed ref — one literal spelling, where a ticket code takes six.ix stayed null), so it was an exact tree descent from the root and nothing else.pager.js is in ///bee, wiki/Bro.mkd is in ///bee-journal, so the fan-out has to carry it.[/meta/wiki] on this very journal's DOG-042 page is the same miss, as is every [/meta/todo] in the tree.A pocket-page reference opens its page, wherever the pocket sits and whichever spelling the page takes.
/wiki/Bro -> wiki/Bro.mkd, /wiki/Nav -> .md, /wiki/Fat -> Fat/README.mkd, .mkd beating .md, /meta/todo, and a spelled-out /wiki/Bro.mkd.///name/path goes on naming its repo outright and descending that root exactly.public_html — still opens the file.///bee — explicit permission for this ticket only; door.js plus its own test leg; NO VC ops of any kind.ninja && ninja -j20 test in quickjab/build green (the bee suites run there as JSRC*).A slash-headed ref is a POCKET PAGE and takes the SAME six spellings a ticket code takes, applied to the path instead of the code. The slash is DROPPED and the segments are looked up like any other ref's: a leading slash likely means that repo's root, but nothing assumes it (gritzko's ruling 2026-08-16) — we look for such a file.
refSpellings branches to pagePaths on a leading /, so every leg that resolves a partial — the reading dir, the ambient repo, the fan-out, and scanPartial's no-git walk — inherits pocket pages at once, with no leg-by-leg edit.docs/wiki/ answers /wiki/Page as readily as one at the root. The cost is that a slash-headed ref can now reach the chooser instead of missing outright, which is the honest answer: these are the files with those segments.public_html dirs, submodules — so assume none). On a TOTAL miss, one leading segment is dropped at a time and the tail re-resolved, so /quickjab/dog/abc/TCP.c — bogus as a path, since within ///quickjab the file is dog/abc/TCP.c — opens anyway. It runs LAST, so nothing that resolves today moves, and the page-ness rides along so a peeled page keeps its spellings.../nowhere/q.txt found some q.txt and stopped being dead) and turned a unique miss into a chooser full of same-named files — both caught by test/http/repo.js and test/choose/pick.js on the first unbounded cut. A basename alone is too weak to vouch for a head nobody recognised.anchored retired as a synonym for "starts with a slash": it now means only what ///name needs — the exact root descent under a NAMED repo — and resolvePartial carries a page flag instead. The two anchored legs page keeps are the ones that made a slash-headed ref reach the registry with no ambient repo, and skip the project-prefixed re-reading.test/http/repo.js's "a leading / … does NOT match a deeper same-named file" pinned the ruling being overturned and was amended to expect the two candidates — on gritzko's say-so 2026-08-16. It is the only assertion in any suite that moved.pagePaths + the one-place slash read in refSpellingsanchored narrowed to ///name; page in resolvePartialDone in the MAIN TREE ///bee on gritzko's explicit say-so (edited, nothing staged, no VC op run). A wiki link now opens its page.
door.js: pagePaths (the six spellings, slash dropped), refSpellings branching on the slash, resolvePartial's anchored -> page, and leg 5, the head peel. inMount is unchanged but for taking false where it took anchored.test/click/page.js: 16 checks through the same door.seatOf a pager click and an http href both ride, over the BEE-008 fixture repo extended with a wiki/ pocket, a meta/ one, and docs/wiki/Deep.mkd — the case that pins segments over roots. All were RED before the change.test/http/repo.js: ONE assertion amended, with permission (a deeper same-named file is now a candidate, not a miss).ninja && ninja -j20 test in quickjab/build: 146/146. JSRCclick carries the new leg, JSRChttp the amended pin./bee/cat/pager.js line 1 links /wiki/Bro -> /bee-journal/wiki/Bro.mkd (200), and DOG-042's own [/meta/wiki] -> /bee-journal/meta/wiki.mkd; an anchored page [/wiki/Bro:12:Yl] lands on #b1130./bee/index/mount.js, all THREE spellings of the same file now link to /quickjab/dog/abc/TCP.c: abc/TCP.c (the submodule boundary), dog/abc/TCP.c (the FSEG suffix) and quickjab/dog/abc/TCP.c (the peel).Suggested commit: BEE-013: pocket pages, and no root assumed