Lucent-Financial-Group · AceHack · May 26, 2026 · May 26, 2026
diff --git a/docs/BACKLOG.md b/docs/BACKLOG.md
@@ -369,6 +369,7 @@ are closed (status: closed in frontmatter)._
 - [ ] **[B-0785](backlog/P1/B-0785-unified-namespace-across-fsharp-kubernetes-ontology-plus-experiment-id-routing-via-argo-rollouts-cilium-service-mesh-aaron-mika-2026-05-25.md)** Unified namespace across F# / Kubernetes / Ontology + experiment-ID routing via Argo Rollouts + Cilium service mesh (existing standards)
 - [ ] **[B-0787](backlog/P1/B-0787-multi-ai-experiment-parallelism-without-stepping-on-each-others-feet-namespace-plus-experiment-id-plus-event-store-as-projections-not-separate-dbs-aaron-2026-05-25.md)** Multi-AI experiment parallelism without stepping on each other's feet — per-AI namespace + experiment-ID routing + event-store-native twin (experiments are projections, not separate DBs)
 - [ ] **[B-0789](backlog/P1/B-0789-iter4-ssh-key-and-hashedpassword-substrate-for-cluster-bringup-2026-05-26.md)** Iter-4 cluster credential substrate — hashedPassword (zeta-change-me default) + operator-ssh-keys.nix module + manual edit workflow (v1) with zflash auto-inject as iter-4.2 follow-up
+- [ ] **[B-0790](backlog/P1/B-0790-zero-dev-machines-cluster-native-architecture-voice-as-primary-operator-surface-aaron-2026-05-26.md)** Zero-dev-machines cluster-native architecture — all PRs from cluster; voice (Alexa + future microphones) as primary operator interface; dev machines and Alexa surfaces are conversational entry points into the cluster, not work substrate
 
 ## P2 — research-grade
 

diff --git a/...uster-native-architecture-voice-as-primary-operator-surface-aaron-2026-05-26.md b/...uster-native-architecture-voice-as-primary-operator-surface-aaron-2026-05-26.md
@@ -0,0 +1,167 @@
+---
+id: B-0790
+priority: P1
+status: open
+title: Zero-dev-machines cluster-native architecture — all PRs from cluster; voice (Alexa + future microphones) as primary operator interface; dev machines and Alexa surfaces are conversational entry points into the cluster, not work substrate
+effort: XL
+ask: aaron 2026-05-26
+created: 2026-05-26
+last_updated: 2026-05-26
+depends_on:
+  - B-0754
+  - B-0780
+  - B-0789
+composes_with:
+  - B-0759
+  - B-0770
+  - B-0776
+  - B-0778
+  - B-0782
+  - B-0787
+  - B-0788
+tags: [architecture, cluster-native, voice-operator, alexa, homelab-persona, end-state, zero-dev-machine, ai-agents-on-cluster]
+---
+
+## Problem
+
+Today's substrate architecture has dev machines (Aaron's Mac, Max's Mac, Addison's Mac) as the primary substrate-engineering surface — Otto-CLI runs on Aaron's Mac, authors PRs from Aaron's Mac, drives `zflash` from Aaron's Mac. The cluster is a deployment target, not a primary work substrate.
+
+The maintainer 2026-05-26 named the architectural target the substrate is BUILDING TOWARD:
+
+> *"i want all the prs to come from the cluster mostly and dev machines are just conversational interfaces into the cluster and so is alexa"*
+
+> *"0 dev machines everything still works and i can talk to alexa for home automation / homelab persona users we want 0 dev machine needed just cluster and microphone"*
+
+End-state target:
+
+- **Cluster IS the primary substrate-engineering surface** — AI agents on cluster nodes do PR authorship, code review, ticket triage, autonomous-loop substrate work, observability + incident response
+- **Dev machines = conversational interfaces** — Aaron / Max / Addison Macs are consoles for talking TO the cluster, not work substrate themselves
+- **Voice (Alexa + future microphones) = primary operator interface** for the homelab persona — operator says "deploy the new chart" or "fix the failing test" and cluster + AI agents do it
+- **0 dev machines required** for the homelab persona — just cluster + microphone
+
+## Target
+
+Architecture where:
+
+| Surface | Role today | Role end-state |
+|---|---|---|
+| **Cluster nodes** | Deployment target; receives PRs from dev machines | Primary substrate-engineering surface; AUTHORS most PRs; runs autonomous-loop substrate |
+| **Aaron's Mac / Max's Mac / Addison's Mac** | Primary substrate-engineering surface (Otto-CLI, IDE work, PRs) | Conversational interface into cluster (read substrate, send intent, observe; not author of work) |
+| **Alexa-speaker** (Amazon device) | External AI participant (per agent-roster-reference-card) | Primary voice-operator interface for homelab persona |
+| **Future microphones connected directly to cluster** | N/A | Voice-only operator surface (no dev machine needed) |
+| **Web UI / browser** | Read-only observability today | Read + send-intent + observe operator surface |
+
+The operator's mental model becomes: "talk to the cluster" instead of "code on my Mac to deploy to the cluster."
+
+## Sub-targets (composing iters)
+
+### Sub-target 1 — cluster nodes can commit + push to GitHub (iter-5 of B-0789)
+
+Per B-0789's iter-5 sub-row design: per-node SSH keypair generated at install time + auto-registered as repo deploy key (write-enabled). Per-node revocable. Commit identity = `<hostname>@<cluster>.local` or similar.
+
+Composes with the maintainer-as-top-level + per-maintainer-liability framing (per `memory/persona/max/PERSONA.md` "Per-maintainer scope = per-maintainer liability" sub-section): each cluster's commits are attributable to the maintainer subtree's authorization chain. Stage-3 attribution per `.claude/rules/human-audit-and-legal-risk-acceptance-pattern-in-settings.md` when LFG / Freeborn / future non-profit holds the cluster.
+
+### Sub-target 2 — autonomous-loop substrate runs ON cluster
+
+The same `<<autonomous-loop>>` cron pattern Aaron's Mac runs (per `.claude/rules/tick-must-never-stop.md`), but firing on cluster nodes. AI agents (Otto / Max-AI / Addison-AI / etc.) on cluster do tick-by-tick work:
+
+- File backlog rows
+- Author + push PRs
+- Address review threads
+- Fix CI failures
+- Write tick shards
+- Run observability sweeps
+- React to alerts
+
+Composes with B-0788 (agent-on-agent recovery substrate) at cluster scope — when an agent on a cluster node fails, peer agents on other cluster nodes recover its session per the recovery substrate.
+
+### Sub-target 3 — Alexa-speaker → cluster direct integration
+
+Alexa-speaker today is an external AI participant (ferries content via Aaron). End-state: Alexa-speaker has a direct integration with the cluster — operator says a thing, Alexa parses + dispatches to cluster AI agents + speaks the response back.
+
+Composes with:
+
+- The `_alexa_speaker_acceptance` block in `.claude/settings.json` per `.claude/rules/human-audit-and-legal-risk-acceptance-pattern-in-settings.md` four-field structure — names operator + scope + policy + see_also for the Alexa integration's risk surface
+- B-0782 DIO substrate (cluster IS the DIO; Alexa is one of its conversational front-ends)
+- B-0787 multi-AI experiment parallelism (Alexa requests can dispatch experiments)
+
+Voice-mode capabilities per `agent-roster-reference-card.md`: Alexa-speaker is the Bezos-tier business voice surface that KICKS ASS at math; ideal for spoken operational queries ("what's CockroachDB doing right now", "deploy v2 of the embedding model", "how much is the cluster costing me this hour").
+
+### Sub-target 4 — future microphones connected directly to cluster
+
+Beyond Alexa-speaker (Amazon-tied), future end-state includes direct microphone-to-cluster integration. Composes with:
+
+- B-0759 first-time-CLI-user persona BROADENED to "homelab persona" — the new-dev onboarding doc Max owns (per his persona) extends to operators who never touch a CLI; their operator surface is voice
+- B-0770 IP-KVM remote BIOS-bring-up — composes for hardware-level recovery without local console
+- B-0778 commodity hardware reference — adds a "compatible microphone hardware" section
+
+### Sub-target 5 — dev machines as conversational interfaces, NOT work substrate
+
+Aaron / Max / Addison Macs continue to exist, but their ROLE shifts:
+
+- **Read-only observability**: dashboards, logs, traces, real-time cluster state
+- **Send-intent surface**: typed natural-language ("redeploy the loki stack", "what's failing on PR #5089") that the cluster's AI agents act on
+- **Recovery + bootstrap**: still needed for first cluster bring-up (per B-0789 iter-4.2); becomes vestigial once Comet-driven install (iter-5) ships
+- **NOT primary work substrate**: PR authorship, code review, CI fixes, etc. all move to cluster
+
+Composes with B-0770 (Comet Pro IP-KVM) which makes the first-cluster-bring-up itself remote-operable; once that ships AND iter-5+ cluster-as-PR-author ships, dev machines truly become optional.
+
+## Acceptance
+
+This is the **end-state architectural target**, not a single PR. Sub-targets ship independently as the substrate matures. Acceptance for B-0790 itself:
+
+- [ ] **Cluster commits substrate**: at least one PR per week (eventually per day) authored from a cluster node, addressed to the Zeta repo, signed with cluster-attributable identity (per iter-5 substrate)
+- [ ] **Cluster runs autonomous loop**: a `<<autonomous-loop>>` cron on a cluster node fires every minute + does substantive work (composes with `.claude/rules/tick-must-never-stop.md`); tick shards land in `docs/hygiene-history/ticks/` from cluster authorship
+- [ ] **Alexa-speaker → cluster**: operator says a thing via Alexa-speaker; cluster's AI agents process + dispatch + speak response; round-trip works end-to-end for at least 3 distinct operator intents (deploy, observe, recover)
+- [ ] **Homelab-persona zero-dev-machine demo**: operator boots Comet Pro + cluster + microphone; no Mac in the room; operates the cluster via voice for at least 30 minutes performing real substrate-engineering work
+- [ ] **Documentation**: `docs/cluster-native-architecture.md` explaining the end-state for new contributors + the migration path from dev-machine-as-substrate to cluster-as-substrate
+
+## Why P1
+
+The architectural target informs every iteration between today and end-state:
+
+- iter-4 USB install (B-0789) — gets cluster bootstrapped; foundation for everything else
+- iter-5 cluster-as-PR-author — substrate-engineering inversion (cluster ↔ dev machine roles flip)
+- B-0770 Comet Pro IP-KVM — closes the "need physical access for first install" gap
+- B-0788 agent-on-agent recovery — substrate that makes cluster-side AI agents robust
+- B-0782 DIO — cluster IS the DIO; voice operator interface is the DIO's input surface
+- B-0787 multi-AI experiment parallelism on cluster — composes for the "many AIs running experiments in parallel on the cluster" end-state
+
+Without B-0790 named as the end-state target, each of those iterations risks drifting toward "make dev-machine-based work easier" instead of "make cluster-native work the primary path."
+
+## Composes with substrate
+
+- **B-0754** — iter-3 USB install (depends_on; foundation)
+- **B-0780** — Local Loop tier-3 substrate (cluster IS tier-3; B-0790 promotes it from "test target" to "primary work surface")
+- **B-0789** — iter-4 SSH+password substrate + iter-5 cluster-write-back design (depends_on; foundation for cluster-as-PR-author)
+- **B-0759** — first-time-CLI-user persona BROADENED to homelab-persona (composes; voice operator = no-CLI operator)
+- **B-0770** — Comet Pro IP-KVM (composes; remote-first install enables zero-dev-machine demos)
+- **B-0776** — simplest-first plugin sequence (composes; each plugin's "what runs where" question gets clearer when cluster-native is the default)
+- **B-0778** — commodity hardware reference (composes; future microphone hardware extends the reference)
+- **B-0782** — Distributed Intelligent Organization (composes; cluster IS the DIO)
+- **B-0787** — multi-AI experiment parallelism on cluster (composes; many AIs on cluster = the DIO's compute substrate)
+- **B-0788** — agent-on-agent Claude Code session recovery (composes; recovery substrate runs cluster-side)
+- `memory/persona/max/PERSONA.md` "Per-maintainer scope = per-maintainer liability" (composes; cluster-side commits attribute to maintainer's subtree)
+- `maintainers/aaron/legal-entities/inventory.md` (composes; cluster Stage-3 attribution per legal-entity inventory)
+- `.claude/rules/human-audit-and-legal-risk-acceptance-pattern-in-settings.md` (composes; `_cluster_operations_acceptance` blocks attribute cluster runtime risks)
+- `.claude/rules/agent-roster-reference-card.md` (composes; Alexa-speaker is a Layer 2 external AI participant today; end-state moves to cluster-integrated)
+- `.claude/rules/tick-must-never-stop.md` (composes; the cron-loop substrate that Aaron's Mac runs today moves to cluster nodes)
+- `docs/AUTONOMOUS-LOOP-PER-TICK.md` (composes; per-tick discipline runs cluster-side)
+
+## Out of scope (for this row; tracked elsewhere)
+
+- iter-5 cluster-as-PR-author substrate — tracked under B-0789's iter-5 sub-design
+- iter-5 Comet-driven install — tracked under B-0770
+- DIO substrate-engineering — tracked under B-0782
+- Per-microphone hardware compatibility list — tracked under B-0778
+
+## Origin
+
+The maintainer 2026-05-26 named the end-state architecture across two adjacent signals during the iter-4.2 maintainer-test session:
+
+1. *"i want all the prs to come from the cluster mostly and dev machines are just conversational interfaces into the cluster and so is alexa"* — cluster as primary PR author; dev machines + Alexa as conversational front-ends
+2. *"0 dev machines everything still works and i can talk to alexa for home automation / homelab persona users we want 0 dev machine needed just cluster and microphone"* — full zero-dev-machine end-state for homelab persona
+
+Filing this row before either sub-target ships ensures the iterations stay oriented toward the end-state target rather than drifting toward "make dev-machine-substrate easier" (which would optimize the wrong axis). Per the maintainer's broader 2026-05-26 *"going for right not fast"* discipline.
+
+ServiceTitan-demo-substrate composes here too: a demo where Aaron in front of stakeholders operates a remote cluster via voice (no laptop) is a substantively different value proposition from "look at this CLI tool I built." The end-state IS the demo.