Anchor generated Sandman handoff in v13 Dream beat

Context

The v13 release note already has a dedicated Dream Pipeline / Golden Path chapter via #12716 / PR #12717. The operator then pointed at a sharper evidence angle for the remaining release-note weighting pass: the Native Edge Graph is not just a planned architecture. It exists as repo-local substrate, has active Hybrid GraphRAG / mutate_frontier control surfaces, and produces a concrete generated handoff at resources/content/sandman_handoff.md.

This ticket is a narrow follow-up, not a rewrite of #12716: anchor the generated Sandman handoff artifact and its current graph-audit caveats in the v13 Dream beat so the release note shows live substrate, not only architecture prose.

V-B-A evidence before filing:

• Live open PR check on 2026-06-08: GPT PRs #12778 and #12781 remain open, approved, clean, and green; Claude draft #12784 remains open. This ticket is created as a future leaf, not an immediate competing PR.
• Live latest-open sweep: checked 60 open issues with number, title, assignees, labels, and URL. Found #12696, #12782, #12740, #12450, and older Dream/REM parents; no equivalent generated-handoff release-note follow-up was present.
• Live exact searches for Native Edge Graph Dream Pipeline release notes and Golden Path Gemma4 31B release notes found only broad/older items: #12696, #12065, and #10120, not this narrow evidence-anchor leaf.
• Semantic fallback sweep with query_documents(type='ticket') for Native Edge Graph Dream Pipeline Gemma4 31B release notes v13 beat surfaced older graph/Dream work, not a duplicate release-note follow-up. ask_knowledge_base is intentionally not used as authority here because the current MCP runtime is stale and still attempts invalid Gemini synthesis despite merged source fixes in #12738 / #12759.
• Source inspection confirmed the current release note already has a Dream chapter and digest section, but its representative anchors cite learn/agentos/DreamPipeline.md, GoldenPathSynthesizer.mjs, and sandman-handoff-format.md rather than the generated resources/content/sandman_handoff.md artifact.
• resources/content/sandman_handoff.md currently shows a generated Native Edge Graph audit captured on 2026-06-08, including guide disconnects for golden-path and native-edge-graph plus an orphan-concept warning for golden-path. That caveat is valuable: it proves the graph can audit itself while preventing release-note overclaim.

The Problem

The existing Dream chapter is directionally right but still slightly abstract. It explains the pipeline and cites the format surface, yet the operator's stronger point is that the graph database and Dream output are already producing live, inspectable artifacts.

The generated handoff proves two things the release note should preserve:

1. The Native Edge Graph / Dream Pipeline is active enough to audit the repository and emit a priority handoff.
2. The graph is honest enough to report its own remaining guide / implementation-edge gaps, so public prose must not claim the ontology is complete.

Without this follow-up, the Dream beat risks sounding like architectural intent rather than shipped substrate with live output and visible residuals.

The Architectural Reality

• README.md frames Neo's Brain as Memory Core, Knowledge Base, Native Edge Graph, DreamService, and Active Hybrid GraphRAG.
• learn/agentos/DreamPipeline.md documents the six-phase REM pipeline, Native Edge Graph ingestion, Golden Path math, and advisory handoff boundary.
• ai/services/graph/GoldenPathSynthesizer.mjs implements Hybrid GraphRAG traversal over semantic embeddings plus structural graph weight and writes the handoff.
• resources/content/sandman_handoff.md is the generated handoff artifact from the local issue sync + Native Edge Graph run.
• resources/content/release-notes/v13.0.0.md already has a Dream chapter and a curated digest section, so this work should be a targeted evidence anchor, not a broad rewrite.

The Fix

Update the v13 release note with a small, source-backed refinement:

• Add resources/content/sandman_handoff.md as the generated handoff evidence anchor, either in the Dream chapter representative anchors or the Dream digest bullet set.
• Add one concise caveat that the generated handoff currently reports native-edge-graph / golden-path guide and implementation-edge gaps; that is an honest graph-audit signal, not a release blocker by itself.
• Keep the existing advisory-forecast framing: Golden Path informs peer maintainers, it does not assign workers.
• Keep Gemma/local-model wording precise. If the prose names the local max model, use the valid configured model form (for example gemma4:31b / 31B local model as appropriate to the source) rather than an invalid bare gemma4 runtime name.

Contract Ledger Matrix

Decision Record impact

none. This is release-note evidence tightening over already-merged substrate. It does not change Dream, graph, provider, or AiConfig behavior.

Discussion Criteria Mapping

Acceptance Criteria

• resources/content/release-notes/v13.0.0.md explicitly references the generated resources/content/sandman_handoff.md artifact or describes its current generated-output role with a stable source anchor.
• The Dream beat includes the honest residual: the generated handoff currently reports native-edge-graph / golden-path guide or implementation-edge gaps, so v13 ships the audit substrate without claiming ontology completion.
• The advisory forecast boundary remains intact: Golden Path informs peer maintainer self-selection and does not become hierarchical auto-assignment.
• Any 31B local-model wording uses a valid configured model name or a source-backed generic phrase; no invalid bare gemma4 fallback is introduced.
• The diff is limited to release-note evidence tightening unless a reviewer explicitly asks for a related digest-line adjustment.

Out of Scope

• Rewriting the whole Dream chapter delivered by #12716 / PR #12717.
• Final narrative-arc polish for the v13 release note.
• Runtime changes to DreamService, GoldenPathSynthesizer, Memory Core, or MCP tools.
• Closing #12065 or claiming the REM/Sandman epic is fully resolved.
• Updating the future release-notes skill #10321.

Avoided Traps

• Treating the existing #12716 work as missing. It is not missing; this is a narrower evidence-anchor follow-up.
• Overclaiming the generated handoff as proof that all graph ontology gaps are solved. The current output says the opposite for native-edge-graph and golden-path guide/implementation coverage.
• Turning this into a broad final-polish PR while two GPT PRs are still at the human merge gate.
• Reintroducing invalid provider fallback names while trying to highlight the 31B local-model path.

Related

Parent: #12696 Related: #12716, PR #12717, #12065, #12782, #10321

Origin Session ID: e8f07ef9-ef7e-4815-8ff4-7abe13720621 Retrieval Hint: v13 release notes generated sandman_handoff Native Edge Graph Golden Path guide disconnect ontology caveat Dream Pipeline 31B local model