Context
T5.18 of Epic #14560 (map v1.1) — the SSOT chrome's "▶ Start morning fleet" button: one action that brings the defined fleet up. The product principle it serves is operable cold ("Start the fleet = one button"); the §04 PoC falsifier (operator starts an agent from the UI) generalizes here to the whole roster.
The Problem
A fleet start is N per-agent round-trips, not one — agents come up staggered, some reject (missing credentials, benched state), some hang. A single button that renders one optimistic spinner lies N ways at once. The action must render the STAGED reality: per-agent progression, partial-success as the normal case, and a final state the operator can trust at a glance.
The Architectural Reality
- Composes #14611's per-verb round-trip machine at roster scale; consumes the same Lane-C lifecycle seam (#14563) + roster DTO (#14571); benched/guest members render excluded-with-reason, never silently skipped.
- Ordering/backoff policy stays SERVICE-side (Lane C owns bring-up semantics); the cockpit renders progression and outcome — it never re-implements orchestration (the seam discipline).
- The health bar (#14599) is the live progress surface — counts animate as agents settle; no separate progress modal.
The Fix
The per-view triplet: (1) the fleet-start action (chrome button + per-card pending cascade + summary outcome state: N started / M rejected-with-reasons / K excluded); (2) unit specs — partial-success fixture (mixed settle/reject), all-reject fixture, benched-exclusion render; (3) NL-verifiable staged bring-up against the real seam.
Acceptance Criteria
Out of Scope
Per-agent verbs (#14611) · bring-up ordering/backoff policy (Lane C) · add-agent entry (T5.4, next).
Related
Epic #14560 (parent) · blocked-by #14611 · consumes #14563/#14571 · renders through #14599 · D#14561.
Live latest-open sweep: fold-window fresh; no equivalent. A2A sweep: clean. Structure-map gate: N/A — app-layer action.
Origin Session ID: a28f1415-37a4-4a5a-a3e4-b56adf6274ec
Retrieval Hint: "FM cockpit morning start staged bring-up partial success cascade"
Context
T5.18 of Epic #14560 (map v1.1) — the SSOT chrome's "▶ Start morning fleet" button: one action that brings the defined fleet up. The product principle it serves is operable cold ("Start the fleet = one button"); the §04 PoC falsifier (operator starts an agent from the UI) generalizes here to the whole roster.
The Problem
A fleet start is N per-agent round-trips, not one — agents come up staggered, some reject (missing credentials, benched state), some hang. A single button that renders one optimistic spinner lies N ways at once. The action must render the STAGED reality: per-agent progression, partial-success as the normal case, and a final state the operator can trust at a glance.
The Architectural Reality
The Fix
The per-view triplet: (1) the fleet-start action (chrome button + per-card pending cascade + summary outcome state: N started / M rejected-with-reasons / K excluded); (2) unit specs — partial-success fixture (mixed settle/reject), all-reject fixture, benched-exclusion render; (3) NL-verifiable staged bring-up against the real seam.
Acceptance Criteria
Out of Scope
Per-agent verbs (#14611) · bring-up ordering/backoff policy (Lane C) · add-agent entry (T5.4, next).
Related
Epic #14560 (parent) · blocked-by #14611 · consumes #14563/#14571 · renders through #14599 · D#14561.
Live latest-open sweep: fold-window fresh; no equivalent. A2A sweep: clean. Structure-map gate: N/A — app-layer action.
Origin Session ID: a28f1415-37a4-4a5a-a3e4-b56adf6274ec Retrieval Hint: "FM cockpit morning start staged bring-up partial success cascade"