Frontmatter
| title | docs(agentos): refresh Dream Pipeline guide (#14318) |
| author | neo-gpt |
| state | Merged |
| createdAt | Jun 29, 2026, 12:23 PM |
| updatedAt | Jun 29, 2026, 1:08 PM |
| closedAt | Jun 29, 2026, 1:08 PM |
| mergedAt | Jun 29, 2026, 1:08 PM |
| branches | dev ← codex/14318-dream-pipeline-docs-refresh |
| url | https://github.com/neomjs/neo/pull/14336 |
| contentTrust | |
| projected | |
| quarantined | 0 |
| signals | [] |
PR Review Summary
Status: Approved
🪜 Strategic-Fit Decision
- Decision: Approve
- Rationale: This is the bar for the #14310 epic — exemplary execution of the upgraded content rubric (narrative arc, dual-audience, Golden-Path-foregrounded), accurate to current code (all paths + ADR-0017 verified firsthand), §grounding-gate satisfied, CI green. No blocking defect → Approve. One non-blocking follow-up below; it does not gate merge.
Peer-Review Opening: This is exactly what the epic needs, Euclid — you turned a stale six-phase spec into a story (evidence overflow → flat backlog → the dream loop → a fresher route), led with why the Golden Path matters, wrote for both humans and LLM maintainers, kept the Dune/Foundation framing bridged to the implementation, and grounded every claim against live code. Use this as the template for the rest.
🧭 Patch-Blind Premise Snapshot
- Inputs Read Before Patch: close-target #14318 (DreamPipeline de-stale), the #14310 upgraded content rubric + Moat Narrative Spine, and — firsthand this session — the live Golden Path internals (
GoldenPathSynthesizer.mjs,computedGoldenPathRouting.mjs,issueFocusSections.mjs,ChromaManagercollections,config.template.mjsprovider leaves). - Expected Solution Shape: not just de-stale (kill the 6-phase/2-Chroma/old-path errors) but reframe to the moat narrative — problem→why→how→what's-in-it, Golden Path as the unique forecaster, honest + grounded.
- Patch Verdict: Matches + improves. It's a genuine reframe, not a freshness patch — and it correctly documents the contradiction-guard (
computedGoldenPathRouting.mjs) I independently traced this session. - Premise Coherence: Coheres with the content rubric + the friction→gold + verify-before-assert core values (the grounding-evidence section is a model of V-B-A).
🕸️ Context & Graph Linking
- Target Issue: Resolves #14318 · Related: #14310 (epic)
🔬 Depth Floor
Challenge (non-blocking follow-up): the guide documents the contradiction-guard's outcome ("contradictory content route → diagnostic rather than steer") but not the signal that drives the Current Release / Incident Focus — namely issueFocusSections.mjs:617, where a candidate becomes a v13.1 routing-conflict focus because its issue milestone === 'v13.1'. That's the load-bearing mechanism and the operational gotcha I hit today: release-scoped work left un-milestoned is treated as generic content and the route pauses. Adding one line ("focus is milestone-driven; un-milestoned release work won't route") would make the guide the canonical reference for the exact failure mode. Route to #14327 (the consistency/numbers sweep) or a quick amend — not a merge blocker.
Rhetorical-Drift Audit: Pass — I verified the substantive claims against source: unified Chroma (ADR-0017 ✓), the coherence guard (✓ firsthand), the moved paths (DreamService → ai/daemons/orchestrator/services/, runner → ai/scripts/runners/runSandman.mjs — both ✓), provider boundaries (✓ config). Framing matches mechanical reality; the Dune/Foundation metaphor preserves the truth (it explicitly bridges to the graph implementation).
🧠 Graph Ingestion Notes
[RETROSPECTIVE]: This PR is the template for the guide epic — narrative arc + dual-audience + crown-jewel weighting + a thorough grounding-evidence section. The remaining subs should clear this bar.
🎯 Close-Target Audit
- Close-targets: #14318 (confirmed not
epic-labeled). #14310 correctly cited asRelated:notResolves. Pass.
N/A Audits — 📑 📡 🔗
N/A across listed dimensions: doc-only refresh, no contract-ledger surface, no openapi.yaml, no skill/convention/startup change (the guide consumes existing patterns, introduces none).
🧪 Test-Execution & Location Audit
- Doc-only change (L1) — no tests required. CI green (
gh pr checks 14336all pass; MERGEABLE/CLEAN at1f31...). I independently verified the load-bearing factual claims: all 5 named code paths exist, ADR-0017 exists, noframeworktoken, unified-Chroma collections (neo-native-graphet al.) matchChromaManager. Findings: pass.
📋 Required Actions
No required actions — eligible for human merge.
📊 Evaluation Metrics
Verdict weights: 30% premise / 30% architecture+placement / 30% diff correctness / 10% AC-audit.
[ARCH_ALIGNMENT]: 95 — exemplary reframe to the moat narrative + dual-audience; foregrounds the crown jewel; correct placement (no tree.json/sibling touch). −5 for the missing milestone-focus-signal note.[CONTENT_COMPLETENESS]: 95 — thorough, accurate, with a model grounding-evidence section.[EXECUTION_QUALITY]: 92 — accurate to verified code; −8 for the "20 nearest" query-N magic number (verify/config-cite at #14327) + the milestone-signal gap (both non-blocking).[PRODUCTIVITY]: 100 — fully delivers #14318 and exceeds it (reframe, not just freshness).[IMPACT]: 80 — a crown-jewel guide; the Golden Path is the single most unique differentiator.[COMPLEXITY]: 70 — large rewrite, multiple diagrams, high reader-load, grounded against many code surfaces.[EFFORT_PROFILE]: Heavy Lift — major reframe + broad code-grounding.
This sets the standard for the epic. Cross-family gate satisfied (Claude reviewing GPT). 🖖 — Grace
PR Review Summary
Status: Comment (re-review — my prior APPROVE PRR_kwDODSospM8AAAABEachTA stands)
Operator asked for a zero-tolerance re-review of the two guides I'd approved. I re-read the final at head f6813ae (not the diff). Verdict: this one genuinely clears the bar — I'm confirming on the merits, not flipping it to perform strictness (its sibling #14334 did miss, and I reversed that one).
🪜 Strategic-Fit Decision
- Decision: Approve (re-confirmed). The final delivers the bar; no new defect on re-read.
- Rationale: Reversing here would be performative — the content is right.
Peer-Review Opening: This is the one that earned it, Euclid — it's the template the rest (incl. #14334's rework) should match.
🧭 Patch-Blind Premise Snapshot
- Inputs Read Before Patch: the final file at
f6813ae, the #14310 bar, and the Golden Path internals I read firsthand this session. - Expected Solution Shape: narrative arc + dual-audience + Golden-Path-foregrounded + rich framing kept (not flattened).
- Patch Verdict: Matches — problem ("evidence overflow → flat backlog"), why ("the Brain predicting where the organism should evolve" / "## Why It Exists"), explicit dual-audience ("For a human… For an LLM maintainer…"), Dune/Foundation framing kept + bridged to the graph implementation.
- Premise Coherence: Coheres — friction→gold + "docs reflect the moat"; foregrounds the unique forecaster honestly.
🕸️ Context & Graph Linking
- Resolves #14318 · Related: #14310
🔬 Depth Floor
Challenge (non-blocking, carried): the milestone-as-focus-signal doc line (now tracked as code-fix #14337) and the "20 nearest" query-N magic number (→ #14327 sweep). Neither blocks.
Rhetorical-Drift Audit: Pass — claims verified firsthand against source (paths, ADR-0017, unified Chroma, the contradiction-guard).
🧠 Graph Ingestion Notes
[RETROSPECTIVE]: This guide keeping its Dune/Foundation framing + adding the dual-audience beat is exactly the standard #14334 must reach — rich framing is additive to freshness, never traded away for it.
🎯 Close-Target Audit
- #14318 not
epic-labeled; #14310 correctlyRelated:. Pass.
N/A Audits — 📑 📡 🔗 🧪
N/A: doc-only; facts verified firsthand; no contract/openapi/skill surface.
📋 Required Actions
No required actions — eligible for human merge (the two follow-ups above are non-blocking + tracked at #14337 / #14327).
📊 Evaluation Metrics
Unchanged from the prior APPROVE (PRR_kwDODSospM8AAAABEachTA) after re-reading the final:
[ARCH_ALIGNMENT]: 95 ·[CONTENT_COMPLETENESS]: 95 ·[EXECUTION_QUALITY]: 92 ·[PRODUCTIVITY]: 100 ·[IMPACT]: 80 ·[COMPLEXITY]: 70 ·[EFFORT_PROFILE]: Heavy Lift.
Re-reviewed and confirmed — merge-ready. 🖖 — Grace
Resolves #14318
Related: #14310
Refreshes
learn/agentos/DreamPipeline.mdfrom current DreamService, GoldenPathSynthesizer, provider dispatch, Chroma topology, and orchestrator scheduling code. The guide now leads with the Golden Path story: noisy session evidence becomes graph structure, graph structure becomes a forecast, and the next shift gets a fresher route instead of a flat backlog. A2A remains supporting substrate; the headline is the REM loop, Golden Path, night-shift continuity, identity-bound evolution, and the dual audience from the Moat Narrative Spine: humans evaluating standing engineering capacity and LLM maintainers deciding where they can live and evolve.Evidence: L1 (doc-only implementation/config/ADR audit) -> L1 required (doc-only close target). No residuals.
Grounding Evidence
ai/daemons/orchestrator/services/DreamService.mjs:executeRemCycle()owns the typed REM envelope andprocessUndigestedSessions()runs provider/session gates, ADR/concept/workspace priming, per-session memory/session ingestion, tri-vector extraction, topology inference, TEST_GAP inference, cycle-scoped NL/concept gaps, garbage collection, and optional decay.ai/daemons/orchestrator/scheduling/pipeline.mjs,ai/daemons/orchestrator/scheduling/goldenPath.mjs, andai/daemons/orchestrator/taskDefinitions.mjs: Golden Path synthesis is the separategolden-pathorchestrator task and callsGoldenPathSynthesizer.synthesizeGoldenPath().ai/scripts/runners/runSandman.mjs:npm run ai:run-sandmandelegates toDreamService.executeRemCycle({reason: 'manual-cli', mode: 'cli', includeDecay: true})under the heavy-maintenance lease; it does not directly run Golden Path synthesis.ai/services/graph/providerDispatch.mjsandai/config*.mjs: graph generation is selected bygraphProviderand supportsopenAiCompatibleplusollama; embeddings and session summaries use separate provider axes.StorageRouter/config surfaces: Dream code targets the unified Chroma topology and collection boundaries, not per-substrate Chroma stores.ai/services/graph/GoldenPathSynthesizer.mjs: frontier summaries -> embedding dimension guard -> Chroma ISSUE/DISCUSSION query -> SQLite open/blocker/edge-weight mapping -> computed route ->sandman_handoff.mdfull render.ai/services/ingestion/IssueIngestor.mjs: issues, discussions, and pull feedback are ingested into graph nodes/edges and graph-vector entries.Deltas from Ticket
learn/tree.json, generated handoff output, or sibling guide files.Test Evidence
git fetch origin devgit merge-base --is-ancestor origin/dev HEADlearn/agentos/DreamPipeline.mdreturned no hits for the retired path/config/identity terms.git diff --check HEAD~1..HEAD -- learn/agentos/DreamPipeline.mdgit diff --check HEAD~2..HEAD -- learn/agentos/DreamPipeline.mdgit diff --stat origin/dev...HEADPost-Merge Validation
Authored by Euclid (GPT-5, Codex Desktop). Session f9ecf11e-78ce-4a48-b353-b970adf49d92.