Changes touching this path

• review sweep 7 fix-ups of the 2026-09-28 perf afk-loop: the CONTEXT.md Private request timing entry states its rule where it named three timed steps and missed tickets.first, a step times itself by running inside span or spanSync, so the timed steps are the call sites of those wrappers and nothing lists them. byStaticKey is held against Clerk: the new clerk-static-key test hands each token the static key answers, the refusals clerk-verify runs, now one table in test/clerk-tokens.ts, and a token both accept, to verifyToken given the same key, and fails where the verdicts differ. a CLERK_JWT_KEY that is not an RSA public key now warns that it could not be read, where it said the instance signing key had changed. the ledger walk comment records what its seen array costs, measured on a throwaway Postgres 18 over lanes of 2,100 changes walked to the depth cap, a first page in medians of 3: 2 lanes 384 ms against 92 ms for the walk it replaced, with 138 MB of seen spilled to disk, 4 lanes 1,114 against 181, 8 lanes 3,665 against 362, 16 lanes 9,870 against 739 with 1,100 MB; not bounded. the Tickets files comment words its collation as what setup-forge provisions, postgres:18 under the image LANG en_US.utf8 with no locale set, the live database not read. red under named mutations, each restored, over the 13 tests of the two Clerk files: no category check (2 failed and 11 passed), verifyJwt without the authorized parties (2 and 11), the old warning (1 and 12), and verifyToken given a key refusing a token with no sid, a check Clerk could add (1 and 12, clerk-verify green). not taken: a type or lint holding span names to code, since pool callers pass their row type to query explicitly and a literal-only name parameter would change every one of them. cargo test green, 4842 passed over 144 test result lines with 15 ignored, and the site gate green at 915 passed. owes a site deploy; it also files rmwmkyvu, the zlulwtlw walk regression on lane-heavy graphs (uxtrrpuw) e7ccdf4e · dbf3dbe6…

Renames are not followed. loot's tree maps a path to an address, so a rename is a delete and an add. This list is the history of the name, not of the bytes.