Run's manifest: its loopback origin, health and ready paths, and its real lifecycle (Network's status showed Run unknown) (0.214.0) (#216)

ddiff --git a/CHANGELOG.md b/CHANGELOG.md
iindex df575583..238082de 100644
---- a/CHANGELOG.md
++++ b/CHANGELOG.md
@@@ -6,6 +6,10 @@ the notes were in the tag and commit messages (`git tag -n1`).
  
  ## 0.214.0 — 2026-10-26
  
++- **Run's manifest names its loopback origin** (`internalOrigin` http://127.0.0.1:4920, `health` /api/health, `ready`
++  /api/ready): Network's status page showed Run as unknown ("no internal address known to Network") although it runs.
++  Its `lifecycle` said "not built yet" throughout; it now says what Run does (shutdown drains, what it resumes at boot,
++  rollback, no leases).
  - **Vibe by voice on a person's own machine** (the owner, 2026-10-24: "a call word on Live and on Node"; consumer plan
    §8). `bot.device-message@1` gains two frames and two config fields, all additive:
    - `voice_ask` (device to server): `id` and `text`, what the person asked after the machine's call word, transcribed
ddiff --git a/manifests/services/run.json b/manifests/services/run.json
iindex dd123968..38c1f4f3 100644
---- a/manifests/services/run.json
++++ b/manifests/services/run.json
@@@ -11,6 +11,9 @@
      "openvibe.run"
    ],
    "publicOrigin": "https://openvibe.run",
++  "internalOrigin": "http://127.0.0.1:4920",
++  "health": "/api/health",
++  "ready": "/api/ready",
    "exposure": {
      "state": "live",
      "publicSite": "service",
@@@ -36,22 +39,50 @@
    "namespacesOwned": [],
    "lifecycle": {
      "liveness": {
--      "none": "not built yet; nothing listens on a loopback address"
++      "endpoint": "/api/health",
++      "means": "the process is up and answers HTTP; the database, Events and Bot are what ready checks"
      },
      "shutdown": {
--      "none": "not built yet; nothing runs or drains"
++      "signal": "SIGTERM",
++      "deadlineSeconds": 5,
++      "drains": [
++        "Network key refresher stopped",
++        "stream tickets stopped",
++        "dispatcher poller stopped (no new tick; the one in flight finishes)",
++        "job event streams closed (a client reconnects)",
++        "events outbox relay stopped",
++        "HTTP server drained",
++        "telemetry flushed",
++        "store closed"
++      ]
      },
      "startupRecovery": {
--      "none": "not built yet; nothing starts"
++      "resumes": [
++        {
++          "kind": "state",
++          "what": "every job and its frames are committed rows: the first poller tick after boot places queued jobs, mirrors placed and running ones from Bot, and expires those past their ttl"
++        },
++        {
++          "kind": "outbox",
++          "what": "run.job.* events and usage readings written before the stop wait in their outboxes and are relayed at boot"
++        }
++      ]
      },
      "rollback": {
--      "none": "not built yet; nothing is released or rolled back"
++      "conditions": [
++        "automatic: ovhost deploy puts the previous sha back and restarts again when the ready URL does not answer 2xx within ready.timeoutSeconds after the restart (exit 3)",
++        "manual: ovhost rollback <service> --to <sha>"
++      ],
++      "window": "automatic only while ovhost waits for readiness after the restart; manual afterwards, as long as no blocker applies",
++      "blockers": {
++        "none": "migrations 0001-0004 are expand-only (Run's own job, usage outbox and telemetry tables): an older release runs on the newer schema"
++      }
      },
      "contracts": {
        "range": ">=0.90.0 <1.0.0"
      },
      "leases": {
--      "none": "not built yet; nothing runs"
++      "none": "no claims: one process places jobs, one poller tick at a time; a job moves on by conditional row updates in Run's own database"
      }
    },
    "notes": "The job service of the network (plan T14, ADR-034): accepts jobs (run.job-create-request@1), places each on a node that advertises worker:<class> (platform.resource-offer@1, platform.runtime-offer@1, platform.workload-requirements@1), sends it as platform.job@1 over the node link and follows its platform.job-frame@1 frames. Consumes billing.usage.record to write one usage reading per wall-clock second (run:<job id>:<n>, service run, operation function.invoke, unit s) and media.object.read to resolve digest-pinned inputs. Serves its job API on loopback :4920 (Run #1 and #2, 2026-10-07).",