Backed by: @pithy-sh/media, its GET {basePath}/admin/delivery, answered by your own Worker. Scope: media:delivery:read.
What it shows
Media is the capability it reads, and the control-plane scopes say why what was served is a grant of its own. What is stored is Usage.
What your images and video were served: Cloudflare’s count of image requests and video minutes, per asset, joined to your Worker’s own records and read live from it. It appears in the rail under Operations, beside Usage, when your Worker composes media and declares the delivery read, at whatever path you mounted it.
The index is two rows, Images and Video, each with one chip:
| Chip | When |
|---|---|
| Reported | Your Worker reports this type per asset, and the row counts the assets |
| Not reported | Your Worker does not report this type here, such as one you store in R2 |
The record
A reported type lists each asset by name and what it was served: requests per image, minutes per video, most served first, as your Worker sent them. The record names the span it covers, from and to. A type with nothing stored on Cloudflare Images or Cloudflare Stream says so.
A type your Worker does not report is drawn with your Worker’s sentence, never as a zero. Your Worker decides which metric applies to each type, not the pane.
The window is yours to choose
7, 30 or 90 days, opening on 30, the kit’s own default, and the same chooser the Turnstile pane uses. A window is fixed when you choose it, and refreshing a chosen window chooses it again, ending now. Switching environment opens on 30 days again.
A window Cloudflare does not keep for this account is refused, with your Worker’s sentence, and the chooser stays on screen so you can ask for another.
Your Worker’s sentence, not ours
Every analytics refusal is drawn with the message your Worker sent, verbatim, as text, and the pane writes no remedy of its own. There are three:
media/analytics_window: a window Cloudflare does not keep for this account.media/analytics_unavailable: Cloudflare serves this account no delivery analytics, as for a Customer Metadata Boundary outside the US. Your Worker’s sentence names it, in place of an empty chart.media/analytics_forbidden: your Worker’sruntimetoken lacks Account Analytics Read. Your Worker’s sentence names the permission and how to add it.
For those three codes, and only for a pane that names them, the dashboard reads your Worker’s error body: at most 4 KB, kept only when it parses as the kit’s public error payload with one of the named codes, and never on a 401 or a 403. Every other failure is drawn with our own sentence.
Read once, held nowhere
The pane reads when you open it, on the current connection, and keeps nothing. The answer is parsed with the capability’s own schema, not cached, and held in no Pithy table. A response that does not parse fails on screen, naming the environment.
Your Worker audits every read as media/delivery_read, with counts and never a name, and the pane says so.
A connection without media:delivery:read is refused by name, and nothing is sent. It is a separate grant from usage, because naming every asset an environment serves is a second disclosure. Its route is a read, so it is in the default grant; Scopes has how to add one to a connection made before it existed.
It reads, and it never writes
Nothing here changes an asset. The pane counts what was served; stored minutes are not counted, because the kit records no duration.