docs(wiki): close the session's loose threads — bay printer question, vocabulary-in-code decision, training note, guard family, OQ #16 settled

Claude-Session: https://claude.ai/code/session_01FWncR69HgGPuei1dLrW3cU

wiki/entities/local-jwt-auth.md

@@ -29,6 +29,8 @@ Authentication and authorization, kept **fully local** — a direct consequence
   `roles` + `role_permissions` tables, composed by an admin from a **code-defined permission grid**
   (`@parking/shared` `PERMISSIONS` = `resource:action`, e.g. `tariff:update`, `payment:create`,
   `event:void`). A `preHandler` `requirePermission(...)` per route checks a PERMISSION, not a role
+  (all of the listed; `requireAnyPermission(...)`, 2026-09-06, for a read two jobs share — the
+  Car Wash price list is the desk's under `carwash:read` and Setup's under `site:read`)
   name. The JWT carries `roleId` (not the permission list); the guard resolves the role's permission
   set per-request from an **in-memory cache** (`bumpPermsCache()` on any role write), so editing a
   role applies immediately — no re-login, no token bloat. No Casbin/engine needed at this scale.

wiki/entities/opencv-anpr-service.md

@@ -282,3 +282,11 @@ read. Composed `model_version` reads `<plate>+yolox:yolox_s.onnx@640`.
   maps to Vetura and no downgrade fires; vans, trucks, buses and motorcycles do separate. Phase B
   (a body-type classifier on the pilot's own frames — every wash order is a labelled frame) is
   what closes that gap; the detector's box is the crop it will classify.
+- **Training (user asked, 2026-09-06): YOLOX-S needs none.** It ships trained on COCO and is a
+  finished detector; its limit is vocabulary, not quality. Phase B is a *different, smaller*
+  model — a classifier over the detector's crop, not a retrained detector — fine-tuned on a small
+  Apache-licensed backbone. Data: a few hundred crops per category to start, a couple of
+  thousand is comfortable, all from the pilot's own lane and camera. Labels are NOT the
+  operator's picks (untrusted — [[threat-model]]) but a trusted reviewer's, gathered through the
+  [[vision-review-outbox]]. Expect 85–95 % on frontal gate views once tuned — enough to flag,
+  never to bill, which is why the flag records and the site threshold exists.