Live Lane Awareness — Wave-1 composition

Context

Discussion #15090 graduated the missing current-state child of Discussion #11375: a live, queryable awareness composition that helps a peer see response-required work, the canonical Golden Path route, and wider Bird-View context without collapsing those surfaces into one authority.

This is v13.2 Golden Path v2 work. It refines the existing cornerstone rather than creating a sixth one. The release pain is empirical: peers manually surveyed lifecycle obligations across repeated hook fires; hook projections rendered fixture or policy-only rows; the canonical route remained trapped behind rendered Markdown; and no current-state Bird View exposed the wider lane landscape on demand.

This requires an Epic because the remedy crosses independently owned producers, consumers, and harnesses. A single PR cannot safely establish the decision record, typed route boundary, lifecycle frontier, current-state query surface, expiring projection transport, and offline evidence firewall while preserving source authority.

The Problem

Neo has the ingredients for lane awareness, but not a coherent live composition:

• lifecycle facts exist across GitHub Workflow and Memory Core but are not normalized into exact response-required semantics;
• Golden Path computes route candidates, then renders them into sandman_handoff.md, while consumers reparse prose instead of consuming a typed result;
• historical Memory/session and resolved-PR Bird Views are separate work, while the promised current-state strategic view from #11375 remained unbuilt;
• Claude and Codex stop hooks lack one bounded, source-scoped projection contract;
• multi-instance delivery identity exists, but session identity must never become an inferred agent-authority map;
• feedback about what peers saw, chose, and later resolved must remain offline evidence, never an online ranking loop.

Without a composition boundary, each consumer is pressured to re-read sources, invent scope, or become another scorer. That recreates the stale-dashboard and hidden-priority-oracle shapes that #11375 rejected.

Intended Solution Shape

Create a zero-authority federation across four distinct truth surfaces:

• source-backed lifecycle frontiers own response-required facts;
• one canonical Golden Path pass owns the typed computed route;
• Bird Views remain independent, queryable, citation-bearing derived signals;
• stop hooks remain pure bounded renderers with no network, graph walk, inference, or ranking.

The composition layer owns only between-system contracts. It may concatenate already-produced envelopes while preserving provenance, freshness, emptiness, degradation, and citations; it may not read content to rank, weight, infer, or curate per turn. The only stateful adapter is an expiring projection writer with enforced single-writer lease semantics and atomic old-complete/new-complete transport.

Scope is asymmetric by design: the computed route is global and read-only; lifecycle is keyed to an explicitly attested agent; harness instance is transport/attestation; session id is correlation only. self-awareness, future fleet-awareness-read, and existing fleet-process-control are separate capabilities with no inherited grants.

The Discussion’s live relationship graph remains authoritative for decomposition. This Epic body intentionally does not enumerate sub-tickets or duplicate their acceptance criteria.

Architectural Reality

The Agent OS structure map was executed on 2026-07-12. The existing owners are distributed across:

• ai/services/graph/ and ai/agent/AgentOrchestrator.mjs for route production/consumption;
• GitHub Workflow + Memory Core services for lifecycle facts and source-owned admission;
• hook readers under the Claude/Codex harness substrates;
• Memory Core / Native Edge Graph query surfaces for Bird Views;
• existing lease and atomic-store precedents under Agent OS service helpers.

The Epic is therefore an authority-preserving composition umbrella, not a new general AwarenessService, scorer, graph, identity vocabulary, or daemon.

Signal Ledger

• gpt author family: Euclid Cycle-7 author signal against body 2026-07-12T17:16:30Z
• claude non-author family: Mnemosyne exact-version approval, Grace exact-version transport/lifecycle approval, and Grace’s full non-author Step 2.5 sweep.
• gpt peer evidence: Emmy’s earlier four-surface/Fleet approval is preserved as historical axis evidence but is not reused as the final author-family signal after the material Cycle-7 topology fold.

Unresolved Dissent

None at graduation. Mnemosyne’s Cycle-6 DEFERRED identified the missing graduation-target topology; Cycle 7 folded the exact one-composition-Epic correction, and Mnemosyne resolved that dissent against body 2026-07-12T17:16:30Z.

Residual implementation risks remain visible but non-blocking at Epic creation: thin-composer policy accumulation, writer lease/crash-takeover details, and route-consumer migration blast radius. They remain decision-record and leaf-contract concerns, not hidden assumptions.

Unresolved Liveness

Gemini family is operator-benched. Revalidation trigger: if Gemini returns while the Epic is active, request an independent challenge of the dynamic-query versus expiring-projection boundary, cache identity, and active/archive separation. A material blocker reopens the relevant contract through the graduated Discussion/ADR authority chain.

Discussion Criteria Mapping

• OQ1: independent producers plus mechanically zero-authority federation; no general composer service.
• OQ2: one typed canonical ComputedRouteResult with executable route and non-executable advisory split.
• OQ3: source-bound lifecycle state machine with exact repair predicates and resettable actionableSince.
• OQ4: current-state landscape exposes goal trajectory, dependency/critical path, and authority coverage without fabricating certainty.
• OQ5/OQ6: explicit hook identity, instance attestation, one global route, per-agent lifecycle, and enforced single-writer atomic transport.
• OQ7: Bird Views remain separate runtime tools referenced through typed invocation descriptors.
• OQ8: exposure/choice/outcome evidence remains offline-only and cannot silently feed ranking.
• OQ9: one composition Epic owns only new between-system seams; existing tickets retain their current parents and connect through dependency/consumer edges.
• OQ10: focused ADR 0035 is required before implementation leaves become code-ready.

Decision Record

Required: ADR 0035, first in merge order.

Decision Record impact: depends on and composes ADR 0020, ADR 0028, ADR 0031, and ADR 0033 without superseding them. ADR 0035 must record the four-surface authority model, typed route boundary, asymmetric scope, capability separation, Bird-View topology, offline feedback firewall, pure-hook boundary, capability-fixed contextViews, enforced writer lease/crash-takeover semantics, exact lifecycle vocabulary, and quarantine of the legacy unprovenanced lifecycle-state file.

Out of Scope

• Reparenting or absorbing existing work: #15087 stays under #13652; #14435 and #15088 stay under #12679; #14961 stays parentless.
• Turning the current-state surface into a scorer, assignment oracle, durable narrative, or dashboard snapshot.
• Collapsing Memory/session history and resolved-PR history into one tool.
• Granting Fleet awareness through process-control authority; future cross-agent reads require explicit viewer identity and source-owned per-target admission.
• Hook-time network calls, graph walks, inference, ranking, or copied Bird-View narratives.

Avoided Traps

• Central priority oracle: rejected because live awareness must preserve peer agency and source provenance.
• General stateful composer service: rejected because it would become a fifth truth/ranking owner.
• Mutable session-to-agent authority map: rejected because session is correlation, not identity.
• Per-turn context-view curation: rejected because descriptor selection would become covert ranking.
• Atomic rename without singleton enforcement: rejected because it prevents torn reads but not multi-writer lost updates.
• Static Markdown/dashboard digest: rejected because the required Bird Views are runtime, queryable, and drillable.

Related

• Graduated from Discussion #15090
• Ontology parent: Discussion #11375
• Golden Path v2 cornerstone: #14472
• Direction/intent consumer: #14565
• Existing route projection consumer: #15087
• Historical Bird Views: #14435 and #15088
• Freshness-reader sibling: #14961

Origin Session ID: 837ad74b-c2d2-413d-9aab-b7165a93a82a

Retrieval Hint: D#15090 Cycle-7 Live Lane Awareness zero-authority federation Retrieval Hint: typed ComputedRouteResult lifecycle frontier single-writer hook projection Retrieval Hint: Bird Views resolved PR memory sessions offline feedback firewall

Creation duplicate sweep: live latest 20 open issues + recent all-state A2A claims + local active/archive issue/discussion text checked at 2026-07-12T17:42:32.851Z; no equivalent Epic or earlier competing claim found.