tv-anarchy/docs/data-model.md

# Data Model

Two sets: the **config + state schemas in use today** (the app and helpers read
and write these), and the **fleet/mesh data model** (single-fleet core
implemented in the governor's fleet engine; cross-fleet parts still design).

---

## In use today

### `devices.json` — device registry + playback target config

Path: `~/.config/tv-anarchy/devices.json` (auto-migrated forward from
`~/.config/tv-anarchy/hosts.json`, and before that `~/.config/plumtv/hosts.json`).
Source: `Sources/TVAnarchyCore/DeviceConfig.swift`. Written pretty-printed +
sorted-keys; seeded on first run with Plum VLC + Black (mpv) if absent. The
decoder accepts the legacy top-level `hosts` key alongside `devices`.

Each device carries a `type` — `cellphone | laptop | storage | seed | broadcast`,
mapping to the planned fleet classes `consumer | roamer | server | seedbox |
broadcast` — and an overridable `services` preset (`stream`, `offlineCache`,
`ttlSeed`, `custody`, `publicSwarmFace`, `f2fRelay`, `meshAnchor`). Legacy
entries without a `type` get one inferred from the player backend.

```jsonc
{
  "devices": [
    {
      "id": "plum-vlc",
      "name": "Plum VLC",
      "kind": "vlc",                 // vlc | mpv-ipc | quicktime | blacktv(legacy)
      "type": "laptop",              // cellphone | laptop | storage | seed | broadcast
      "services": { "stream": true, "offlineCache": true, "ttlSeed": true },
      "vlc": { "host": "127.0.0.1", "port": 8080 }
    },
    {
      "id": "black",
      "name": "Black TV",
      "kind": "mpv-ipc",
      "type": "storage",
      "services": { "stream": true, "custody": true },
      "mpv": {
        "endpoints": ["lilith@10.0.0.11", "lilith@10.9.0.4"], // LAN first, WG overlay fallback
        "socket": "/tmp/mpv.sock",   // default
        "sudo": true,                // root-owned socket → sudo socat
        "socat": "socat",            // default
        "volumeScale": 130           // mpv --volume-max
      },
      "commands": {                  // argv templates for what IPC can't do
        "launchShow":     ["/usr/local/bin/black-tv","play-show","{query}","{season?}","{episode?}"],
        "launchResume":   ["/usr/local/bin/black-tv","resume-show","{query}"],
        "launchFile":     ["/usr/local/bin/black-tv","play","{path}"],
        "releases":       ["/usr/local/bin/black-tv","releases"],
        "resolveRelease": ["/usr/local/bin/black-tv","resolve-release","{releaseId}"],
        "stats":          ["/usr/local/bin/black-tv","stats"],
        "stop":           ["/usr/local/bin/black-tv","stop"],
        "restart":        ["/usr/local/bin/black-tv","restart"]
      }
    }
  ]
}
```

Notes:
- `kind`: `vlc`, `mpv-ipc`, `quicktime`. `blacktv` is legacy and auto-migrated to
  `mpv-ipc` at load; only `mpv-ipc`/`vlc`/`quicktime` are offered in the editor.
- Command-template tokens: `{query}`, `{season?}`, `{episode?}`, `{path}`,
  `{releaseId}` (a `nil` template = capability absent).
- VLC password is **not** stored here — it's resolved at runtime from the
  governor's config or `$VLC_HTTP_PASSWORD` (see `VLCConfig`).
- `MpvConn` decodes from a minimal `{ "endpoints": [...] }`; every other field
  has a default.

### `config.json` — governor (`portable-net-tv`)

Path: `~/.config/portable-net-tv/config.json` (source: `governor/README.md`).

```jsonc
{
  "vlcHttp": { "host": "127.0.0.1", "port": 8080, "password": "<password>" },
  "buffer":  { "dir": "/Users/natalie/Movies/net-tv-buffer", "ahead": 3, "minFreeGB": 2 }
}
```

`buffer` is optional (defaults: `~/Movies/net-tv-buffer`, `ahead` 3, `minFreeGB` 2).

### Watch log — append-only JSONL (the watch protocol)

Primary path: `~/.local/state/tv-anarchy/watched.jsonl` (plum-side; the app,
governor, and mcp all append here). Legacy path
`~/.local/state/tv-anarchy-mcp/watched.jsonl` is auto-migrated on first read
(`packages/media`). Black TV plays are mirrored from black's
`~/.local/state/black-tv/watched.jsonl` into a local `black-watched.jsonl` cache
(by `BlackWatchlog` + `WatchHistoryController.syncBlack`); both files are unioned
at read time.

One JSONL event per line (full shape and finish rule in
[architecture.md](./architecture.md#playback-observation-finish-detection-and-the-watch-log-protocol)):

```jsonc
{ "ts": "...", "event": "play|resume|reset", "show": "…", "season": 1, "episode": 3,
  "label": "…", "path": "/bigdisk/_/media/.../Show S01E03.mkv",
  "resumeSeconds": 1234, "durationSeconds": 2400, "client": "app|governor|bridge|black",
  "finished": true }
```

- `"play"` events (or `resumeSeconds >= durationSeconds * 0.92`) advance the
  Continue Watching frontier and `playedPaths`.
- `"resume"` events supply live positions for bars and mid-episode resume targets.
- `"reset"` clears played state for a rewatch (history is preserved).
- `client` ("app", "governor", "bridge", "black", "mcp", ...) tags provenance. `FINISHED_FRACTION = 0.92` is the single threshold
  (enforced on write in a few places, on read in `isWatchFinished` helpers).
- Readers in the app (`WatchHistory`, `LibraryController.resumeTarget` etc.),
  governor, and mcp bridge all derive Continue / resume / badges exclusively from
  this log. No other source of truth.

(The obsolete direct "VLC recents" macOS plist integration has been removed.)

### App-local cache + state

| Artifact | Path | Writer |
|---|---|---|
| App settings | `~/.local/state/tv-anarchy/settings.json` | `SettingsStore` (adult gating, hover previews, media-key forwarding, offline-cache sizing; tolerant decode) |
| Library snapshot | `~/.local/state/tv-anarchy/library.json` | `LibraryStore` |
| Metadata sidecar | `~/.local/state/tv-anarchy/meta/<sha256(path)>.json` | `MetaWriter` (mirrored best-effort to `<remotePath>.meta` on black) |
| Artwork cache | frame-grab JPEGs (keyed by path) | `ArtworkService` |
| watched.jsonl (watch log) | `~/.local/state/tv-anarchy/watched.jsonl` (+ black mirror) | app, governor, bridge, black mpv Lua (see architecture.md) |

### Library models (in-memory / snapshot)

`CachedShow { name, rootDir, category, kind(series|movie), posterPath, overview,
episodes[], year, seasonCount, episodeCount, addedAt }`,
`CachedEpisode { path, season, episode, label, metaPath }`,
`LibrarySnapshot { shows[], capturedAt, source(scan|registry) }`,
`ContinueItem { title, path, show, season, episode, positionSeconds, lastSeen,
source(watchlog|vlc), posterPath }`. Decoders are tolerant so old snapshots load.

`ParsedFilename { title, year, season, episode, quality, codec, releaseSource }`,
`MediaMeta { path, parsed, resolvedTitle, mediaType, overview, posterURL,
ratings, genres, enrichedAt }`.

---

### `fleet.json` — the fleet registry (app-side) + governor policy

Path: `~/.config/tv-anarchy/fleet.json`. The `devices` **array** is the
app-side fleet registry (authoritative for the governor's fleet engine when
present; `devices.json` is the fallback):

```jsonc
{
  "devices": [
    {
      "id": "black", "name": "black",
      "deviceClass": "server",            // server|roamer|consumer|seedbox|broadcast
      "alwaysOn": true, "onHomeIp": true, "reachable": "home_lan",
      "duties": ["custody_floor"],        // app-side record; the governor recomputes
      "services": [
        { "id": "black", "kind": "mpv-ipc", "detail": "lilith@10.0.0.11" },
        { "id": "black-transmission", "kind": "transmission", "detail": "transmission RPC" }
      ]
    }
  ],
  // Optional governor policy keys (read by governor/src/fleet/registry.ts):
  "floorCopies": 2,                       // custody floor (default 2)
  "sources": [                            // peer sources; gates enforced on load
    { "id": "dht", "kind": "dht" }        // implicit when absent
  ],
  "staticHoldings": { "apricot": ["Show Name S01"] }  // copies on api-less hosts
}
```

The governor derives from this: `api` (a `transmission` service →
`transmission_rpc`), `ssh`/`addr` (a `user@host` service detail), and capacity
defaults. Engine state (last duty assignment, for change diffs) lives at
`~/.local/state/tv-anarchy/fleet-state.json`. Title-refiner cache:
`~/.local/state/tv-anarchy/title-refinements.json` (filename → refined title,
empties cached too).

---

## Fleet/mesh data model (single-fleet core implemented)

Synthesized from
[`../.project/history/20260608_fleet-manager-mesh-design.md`](../.project/history/20260608_fleet-manager-mesh-design.md).
The single-fleet subset — host registry, duty rules, custody floor, reaper,
source gates, `peers_for`/`custodians_of` — is implemented in
`governor/src/fleet/` (types in `types.ts` mirror the entities below).
Identity/Fleet entities and everything cross-fleet remain design.

### Entities

```
Identity { discord_id, display_name, fleets[] }
Fleet    { id, identity, hosts[], sources[], custody_capacity }
Host     { ...registry, below... }
Source   { ...peer-source, below... }
```

`custody_capacity = Σ over always_on hosts of (disk_free × uptime_score)` — whether
a fleet can hold a floor or is a net consumer. Gift-economy mode ignores it for
prioritization; ratio mode (off by default) weights by it.

### Host registry

```
Host {
  id, fleet_id
  class:       server | roamer | consumer | seedbox | broadcast   // what it IS
  reachable:   home_lan | wireguard | public_ip
  always_on:   bool
  on_home_ip:  bool        // true = public-swarm traffic exposes the home connection
  api:         transmission_rpc | qbittorrent | utorrent_web | none
  capacity:    { disk_free, up_bw, uptime_score(∈[0,1], rolling) }
  duties:      Duty[]      // assigned by the manager, never hardcoded — what it DOES
}
Duty = custody_floor | public_swarm_face | f2f_relay | broadcast
```

Example fleet: black = server, apricot = secondary always-on, plum = roamer (TTL
seeder), phone = consumer (pure sink, never any duty).

> **Partially landed:** the app's `devices.json` (above) already carries the
> class mapping (`DeviceType.fleetClass`) and a `services` preset that mirrors
> the duty vocabulary. What does **not** exist is the manager that *assigns*
> duties from capacity/reachability — services today are user-set presets, not
> computed duties.

### Duty-assignment rules (deterministic, run on registry change)

| Duty | Eligibility | Rule |
|---|---|---|
| `broadcast` | `public_ip && always_on` | exactly ONE per fleet; prefer seedbox > vps-0 > server |
| `f2f_relay` | `always_on && reachable ∈ {wireguard, public_ip}` | broadcast host + any other always-on server |
| `public_swarm_face` | prefer `!on_home_ip` | seedbox FIRST; never a consumer; never on_home_ip if an off-home option exists |
| `custody_floor` | `always_on && disk_free > title_size` | N most-recent eligible holders; ≥1 slot reserved for an always-on node |

Invariants: a `consumer` never receives a duty (checked first); `public_swarm_face`
prefers off-home-IP so home stays dark; every active title's floor keeps ≥1
always-on holder.

### Peer-source model

```
Source {
  id, fleet_id
  kind:            dht | public_tracker | private_tracker | friend_mesh | fleet_host | seedbox
  api/creds:       <opaque, encrypted at rest>
  share_policy:    search_only | content      // private_tracker DEFAULT-CLOSED (search_only)
  swarm_isolation: f2f_only | open            // private_tracker FORCED to f2f_only (un-overridable)
}
```

Two policy gates are load-bearing because a private `.torrent` carries an embedded
passkey — a friend announcing it to the private tracker gets the **source** user
banned:
1. `share_policy` gates the registry merge; flipping private → `content` triggers a
   consequence-explicit warning naming the tracker.
2. `swarm_isolation = f2f_only` is forced for private sources: content is re-hosted
   into the fleet's own WireGuard swarm, served `via: wireguard`, never announced
   to the private tracker. Seedbox sources are `content` + `open` (no gates).

### Derived outputs (the point of everything above)

```
peers_for(infohash)  → Peer[]      // holistic seeder list / user-owned meta-tracker
custodians_of(title) → Host[]      // who is obligated to keep it alive
Peer { addr, source_kind, source_id, served_via: public | wireguard }
```

`peers_for` unions privates ∪ fleet ∪ friends' fleets ∪ seedbox ∪ DHT/public,
provenance-tagged so the UI can show *why* a peer exists and over what transport.