# Permissions, the console, audit and stats

How people manage what they control on the hub: products, catalogs, importers, hosted rooms, and the hub itself.
The products contract is `products.md`, imports are `imports.md`; this is who may do what, and what they see.

## Permissions (**Decided**, 2026-11-04)

- **Code checks permissions, never roles.** Every check is `can(person, permission, target)`. Permissions are
  `resource:action` strings (list below).
- **A role is a named bundle of permissions**, given to a person at a **scope**: the hub (everything), a catalog, a
  product or a hosted room.
- **Broader scopes cover narrower ones.** A permission held hub-wide covers every catalog, product and room. One held
  on a catalog covers that catalog's members, importer, grants and hosting (not the products themselves: delisting
  a product from the catalog is `hosting:delist` on the catalog; removing it everywhere is `product:remove`, which
  only hub-wide roles have).
- **Owners are derived**, not stored as assignments: a product's creator (`product.creator_id`), a catalog's owner
  (`catalog.owner_id`), a hosted room's owner. They can't fall out of step with those columns; they change only by
  transfer.
- **Giving roles**: you may give (or take away) a role at a scope only if you hold every permission in it at that
  scope, so nobody hands out more than they have. Built-in roles can't be deleted or changed (except by admins,
  hub-wide). Every change is audited.
- **Custom roles** (later): the model allows them (a catalog owner making "Member approver" with only
  `member:approve`); built-in roles come first.

```
role(id, name, scope_type, permissions[], builtin)
role_assignment(person, role, scope_type, scope_id)   -- scope_id null for the hub
```

### Built-in roles

| Role | Scope | Permissions |
| --- | --- | --- |
| Admin | hub | all |
| Moderator | hub | `product:*` (not `transfer`), `revision:retract`, `catalog:suspend`, `catalog:delete-review`, `catalog:approve`, `hosting:delist`, `import:grant-on-behalf`, `import:withdraw`, `import:pause`, `case:*`, `person:approve`, `person:ban`, `person:link-cap`, `person:notes`, `audit:read`, `stats:read`, `room:*` |
| Catalog owner (derived) | catalog | `catalog:edit`, `catalog:rotate-key`, `catalog:delete-request`, `catalog:transfer`, `team:manage`, `member:*`, `hosting:delist`, `import:*` (not `grant-on-behalf`), `case:read`, `case:resolve`, `audit:read`, `stats:read` |
| Catalog operator | catalog | as the owner, without `catalog:delete-request`, `catalog:transfer`, `team:manage` |
| Creator (derived) | product | `product:edit`, `product:publish`, `product:delete`, `product:restore`, `product:transfer`, `revision:upload`, `revision:retract`, `hosting:choose`, `case:read`, `audit:read`, `stats:read` |
| Room owner (derived) | hosted room | `room:*` |
| Room moderator | hosted room | `room:kick`, `room:ban`, `room:mute` |

Hub-wide `admin` and `moderator` keep living in Better Auth's `user.role` (its admin plugin needs it); the permission
check maps them to the bundles above. A creator with `product:edit` can't change fields a moderator locked
(`product:edit-locked`), and lowering a rating is `product:rate-lower` (moderators only; creators ask for a review).

### Permissions

| Resource | Actions |
| --- | --- |
| `product` | `edit`, `edit-locked`, `lock`, `rate-lower`, `publish`, `delete`, `restore`, `remove`, `unremove`, `transfer` |
| `revision` | `upload`, `retract` |
| `catalog` | `approve`, `edit`, `rotate-key`, `suspend`, `delete-request`, `delete-review`, `transfer` |
| `team` | `manage` |
| `member` | `approve`, `invite`, `remove`, `ban` |
| `hosting` | `choose`, `delist` |
| `import` | `grant`, `grant-on-behalf`, `accept`, `run`, `pause`, `withdraw` |
| `case` | `read`, `resolve`, `assign` |
| `audit` | `read`, `read-sensitive` |
| `stats` | `read` |
| `person` | `approve`, `ban`, `link-cap`, `notes`, `role`, `view-as` |
| `room` | `edit`, `close`, `kick`, `ban`, `mute` |
| `settings` | `edit` |

`audit:read-sensitive` (IP addresses, sessions), `person:role`, `person:view-as` and `settings:edit` are admins'.

### "Why can't I?"

A refused action answers 403 `forbidden` with `missing: {permission, scope}` (e.g. `{permission: "member:approve",
scope: {catalog: "c1"}}`), and the website says which permission it needed. **View as** (admins,
`person:view-as`): the console as another person sees it, read-only, audited, with a banner; nothing can be changed
while viewing.

### Two-factor sign-in

Anyone holding a sensitive permission must have two-factor sign-in on (Better Auth's `twoFactor` plugin, TOTP with
backup codes): `product:remove`, `person:ban`, `person:role`, `audit:read-sensitive`, `catalog:transfer`,
`catalog:suspend`, `import:grant-on-behalf`, `settings:edit`. Without it, those actions answer 403
`two_factor_required` and the console asks them to set it up. Giving someone such a role reminds them.

## The console

One console at `/console` for everyone who controls something: a creator with a product, a catalog's team, a
hosted room's owner or moderators, moderators and admins. Others don't see the link. Every page and action asks
for a permission, and the API checks it again (the website only hides what can't be used).

Sections (each shown only to those with something in it):

- **Overview**: charts and to-dos (open cases, member requests, failed imports, drafts) for what the person
  controls.
- **Products**: their products (everyone's for moderators), drafts, info history, locks, stats.
- **Catalogs**: the catalogs they're on the team of (all for admins and moderators): settings, key, team, members
  and invites, hosted products, cases, the importer (grants, queue, requests) when it has one, health and storage.
- **Rooms**: their hosted rooms.
- **Cases**: reports, appeals and reviews about what they control.
- **People**: moderators and admins.
- **Audit**: scoped (below).
- **Settings**: admins (registration, announcements, roles).

No catalog is special: importer pages are part of any catalog with imports on; what only admins do (granting on a
creator's behalf) is a hub-wide permission usable on every such catalog.

## Audit log

- Every change is an entry: `{at, actor, action, tags, detail}`. `tags` are the resources it touches, e.g.
  `{catalog: "c1", product: "p9"}` for "operator delisted p9 from c1", `{person: "x"}` for a ban.
- **Who sees an entry**: anyone with `audit:read` at a scope covering any of its tags. Admins and moderators see
  everything.
- **Fields by permission**: IP addresses and sessions only with `audit:read-sensitive`. People appear by hub
  username; linked IMVU accounts appear only to moderators (alts stay private). Catalog owners see moderators'
  actions on their catalog and their products, with the moderator's name and reason.
- **Filters**: resource, actor, action, date range, text; admins also get "everything a person did" and
  "everything that happened to X". Every detail page links to its entries.
- Retention (later): entries kept indefinitely for now; a rule per action type later.

```
audit(id, at, actor, action, tags jsonb, detail jsonb, ip)   -- GIN index on tags
```

The existing `target` column becomes tags (`target` kept for old entries).

## Cases: reports, appeals and reviews

One queue for everything someone has to decide. A case has a kind, the resources it's about (tags, as the audit
log), who opened it, a status (`open`, `resolved`, `declined`), an assignee, a resolution and a thread of notes.

| Kind | Opened by | Decided by |
| --- | --- | --- |
| `report` | anyone signed in, about a product, catalog, room or person | the catalog's team (for products it hosts) and moderators |
| `rating-review` | a creator lowering adult to general | moderators (`product:rate-lower`) |
| `removal-request` | a catalog's team asking to remove a product everywhere | moderators |
| `catalog-deletion` | a catalog's owner (deleting goes to review for now) | moderators (`catalog:delete-review`) |
| `appeal` | the person affected by a removal, ban, delisting or decline (one per action) | a different moderator than the one who acted |

- People see cases they opened and cases about what they control (`case:read` at a covering scope).
- Reports are limited per person (10 an hour), one open report per person and resource.
- **Canned reasons** (admins edit the list) for removals, bans, declines and edits; free text can be added.
- **Moderator notes**: private notes on people and products (`person:notes`), shown in the console only.
- **Second approval** for the most destructive actions: banning a person, removing a catalog and transferring a
  catalog need a second moderator's approval (a case the first opens); admins can act alone in an emergency, which
  is flagged in the audit log.

## Notifications

One table; shown on the website (a bell, unread count), on the proxy's realtime connection
(`{"type": "notification"}`, so the client can show them later), and by email for types the person opts into.

Types: a moderator edited your product (with the reason); a case about you or yours was opened or decided; your
membership request was decided; your import finished or failed; your catalog's deletion request was decided; a
product in your library has a new revision; a grant was made or withdrawn; someone requested your products
(daily digest).

```
notification(id, person, type, tags, data, created_at, read_at)
notification_setting(person, type, email boolean)
```

## API tokens

Personal tokens for scripts: a name, an expiry (at most a year), and a subset of the person's own permissions at
chosen scopes (e.g. `revision:upload` on two products, `member:approve` on a catalog). Sent as `Authorization:
Bearer t5de_pat_<random>`; only the hash is stored; shown once. A token never has more than its owner holds now: the
check is the intersection, so losing a role takes it from tokens too. Listed and revocable in the account page;
every use updates `last_used_at`; actions are audited as the person, marked "by token <name>".

```
api_token(id, person, name, hash, grants jsonb, expires_at, last_used_at, created_at, revoked_at)
```

## Stats and charts

Charts in the console (shadcn/ui charts, Recharts), following the website's look: neutral, light and dark, color only
for meaning.

| Who | Charts |
| --- | --- |
| everyone | activity on their resources over time, open cases |
| creators | library adds per product per day, people wearing them this week, import requests |
| catalog teams | members over time, uploads per day, products by rating, storage, cases opened and resolved, importer runs and failures |
| admins | sign-ups and approvals, proxies online, products and revisions per day, library adds, cases, catalogs by status, age checks against the provider's monthly budget |

- A nightly job rolls counts up into `stat_daily(scope_type, scope_id, metric, day, value)`; "now" figures are live
  counters. Charts never scan the raw tables.
- **Only totals.** Counts of people (wearers, adds, members) under 5 show as "fewer than 5", so creators can't pick
  out individuals.
- Catalogs may report daily download totals (`POST /api/v1/catalog/stats`); optional, so other people's catalogs
  don't have to.
- Client usage stays in Aptabase; these are the hub's own data.

## Routes

`/api/v1/console/*` (signed in; each checks its permission). Lists take `limit` and a cursor.

| | |
| --- | --- |
| `GET /api/v1/console` | `{sections, counts, todo}`: what this person may see |
| `GET /api/v1/console/permissions?scope=...` | the person's permissions at a scope (the website uses it to show actions) |
| `GET /api/v1/console/audit?tag=&actor=&action=&from=&to=&q=` | entries the person may see, newest first |
| `GET /api/v1/console/stats?scope=&metric=&from=&to=` | `{series: [{day, value}]}` |
| `GET/POST /api/v1/cases`, `GET/PATCH /api/v1/cases/:id`, `POST /api/v1/cases/:id/notes` | cases |
| `GET /api/v1/notifications`, `POST /api/v1/notifications/read`, `GET/PUT /api/v1/notifications/settings` | notifications |
| `GET/POST /api/v1/account/tokens`, `DELETE /api/v1/account/tokens/:id` | API tokens |
| `GET/PUT/DELETE /api/v1/roles/:scopeType/:scopeId/:person` | role assignments at a scope |
| `POST /api/v1/console/view-as/:person` | (admins) start viewing as someone, read-only |

## Later

- Custom roles and their editor.
- Time-limited role assignments (a temporary moderator for an event).
- Announcement banner and maintenance mode (admins).
- Retention rules for the audit log and soft-deleted records.
- Webhooks (e.g. to Discord) for a catalog's cases and uploads.
- Health alerts for catalog teams when the hub can't reach their catalog; storage quotas.
- CSV export of stats.