Per `.claude/rules/output-coverage.md`: every output on every layer, in every state.
## Port 0 (process — Grafana upsert messages)
Emitted by the command handler(s) after a `child.register` or `regenerate-dashboard` message. Shape is the same for both; `meta.trigger` distinguishes them.
| Key | Source method | Type | States tested | Test file |
| `meta.title` | `handlers.emitDashboardsFor` | string (child name or id) | populated | `test/basic/slice43-output-manifest.basic.test.js` |
| `meta.trigger` | `handlers.emitDashboardsFor` | `'child.register'` or `'manual'` | both states | `test/basic/slice41-manual-regen.basic.test.js` |
**Degraded-state convention:** missing keys are **absent**, never set to `null`. The `http request` consumer treats absent headers/payload fields as defaults.
## Port 1 (InfluxDB telemetry)
dashboardAPI emits **nothing** on Port 1 by design — it has no measurements, no tick loop, no telemetry. Verified by absence: no `formatForInflux` import, no Port 1 wires in `examples/`.
## Port 2 (registration / control plumbing)
dashboardAPI is a **sink** for `child.register` messages, not a source — it does not register itself with any parent. Nothing emitted on Port 2.
## Structured log outputs
| Event | Level | Triggered by | Fields | Test |
|---|---|---|---|---|
| `regen-emitted` | info | successful composition (auto or manual) | `event`, `trigger`, `dashboardApiId`, `childId`, `dashboardCount` | `test/basic/slice43-output-manifest.basic.test.js` |
| `buildDashboard(opts)` | `{ dashboard, uid, title, softwareType, nodeId, measurementName, bucket }` or `null`; `measurementName` mirrors `outputUtils.formatMsg` (`general.name` \|\| `<softwareType>_<id>`) so the dashboard `_measurement` var matches the telemetry series; `bucket` is the resolved Influx bucket | success (name set + name empty/fallback) | `null` when no template for softwareType | `test/basic/slice43-output-manifest.basic.test.js`, `test/basic/slice46-measurement-name-parity.basic.test.js` |
| `generateDashboardsForGraph(root)` | flat pre-order array of `buildDashboard` results (root first, then full descendant subtree); per-parent dedup + links applied at every level; machineGroup roots additionally get per-pump fan-out panels injected (see below) | 0..N children, 3-level tree, diamond, cycle | empty array when root config missing | `test/basic/slice35-graph-perf-and-uid-uniqueness.basic.test.js`, `test/basic/slice44-recursive-discovery.basic.test.js` |
| `_injectMachineGroupPumpPanels(parentDash, children)` | mutates an MGC dashboard in place: replaces the static Total Flow/Power panels with 3 timeseries panels (Pump % Control, Pump Predicted Flow vs Demand, Pump Predicted Power) whose queries are generated from the child-pump measurement names. Panels carry `meta.emittedFields: []` so they survive the dedup pass | MGC with ≥1 rotatingMachine child | no-op for non-MGC dashboards or MGC with zero pump children (static totals retained) | `test/basic/slice47-mgc-pump-panels.basic.test.js` |
| `subtreeIdsFor(myId, child)` | Set\<string\> | myId + every id in the child's full subtree (recurses all levels, cycle-safe) | myId only when child has no children | `test/basic/slice36-diff-predicate.basic.test.js`, `test/basic/slice44-recursive-discovery.basic.test.js` |
- ❌ Mixing absent vs null for optional fields — `folderUid` / `folderId` are **absent** when unconfigured, never `null`. Verified.
- ❌ Per-call token stamping — token is set on `headers.Authorization` when configured; absent when not. No empty-string sentinel.
- ❌ Tab id over-triggering in diff predicate — predicate only matches against dashboardAPI's own id + child + grandchildren, never tab ids. Verified.
## Migration plan applied
This manifest is created together with slice #43 — the new outputs added in slices #34–#42 are documented here. Other EVOLV nodes still need their own manifests; tracked in `IMPROVEMENTS_BACKLOG.md`.
