Lucent-Financial-Group · AceHack · May 28, 2026 · May 28, 2026 · May 28, 2026 · May 28, 2026
diff --git a/docs/BACKLOG.md b/docs/BACKLOG.md
@@ -854,6 +854,7 @@ are closed (status: closed in frontmatter)._
 - [ ] **[B-0888](backlog/P2/B-0888-cross-track-substrate-sync-policy-cloud-github-vs-usb-local-gitlab-intentional-divergence-vs-auto-sync-otto-pushback-2026-05-28.md)** Cross-track substrate-sync policy — cloud-GitHub vs USB-local-GitLab; intentional divergence vs auto-sync-via-push-to-both-remotes vs hybrid
 - [ ] **[B-0893](backlog/P2/B-0893-zetaid-v2-128-bit-structured-encoding-snowflake-ulid-family-kestrel-2026-05-28.md)** ZetaID v2 — 128-bit structured encoding (Snowflake/ULID family with timestamp + trajectory + persona + lifecycle-stage + random)
 - [ ] **[B-0899](backlog/P2/B-0899-casimir-like-effect-from-review-walls-changing-allowed-output-modes-testable-pressure-difference-before-after-rule-landing-amara-aaron-2026-05-28.md)** Casimir-like effect from review walls — testable pressure difference in agent-output distribution before/after rule landing
+- [ ] **[B-0914](backlog/P2/B-0914-co-scientist-plus-robin-7-substrate-engineering-candidate-gaps-elo-trueskill-closed-loop-consensus-pairing-evolution-proximity-falcon-aaron-2026-05-28.md)** Co-scientist + Robin 7 substrate-engineering candidate gaps — ELO/TrueSkill ranking-agent + closed-loop CI→hypothesis + n-parallel-consensus + generation-reflection-pairing + evolution-mash-refine + proximity-dedup + Falcon-auto-research-doc-per-proposal (Aaron 2026-05-28)
 
 ## P3 — convenience / deferred
 

diff --git a/docs/UPSTREAM-LIST.md b/docs/UPSTREAM-LIST.md
@@ -119,6 +119,35 @@ citation.
 - **OpenAI Agents SDK + *A Practical Guide to Building
   Agents*** — cross-vendor comparison for agent loop design.
 
+### Multi-agent scientific discovery (added 2026-05-28 per Aaron YouTube ferry PR #5762)
+
+- **Google DeepMind co-scientist** ⭐ (Nature 2026) — multi-agent
+  ecosystem (supervisor/generation/reflection/proximity/evolution/
+  ranking) with ELO tournament hypothesis ranking. Closed-source
+  upstream; community implementations available:
+  - **jataware/open-coscientist** — best-available LangGraph
+    adaptation; mirrors the full agent ecosystem
+  - **llnl/open-ai-co-scientist** — LLNL government-lab
+    implementation; trust-substrate distinct from community ports
+  - **The-Swarm-Corporation/AI-CoScientist** — minimal Swarms
+    framework implementation; smaller surface for substrate-
+    engineering composition study
+- **Sakana AI Robin** ⭐ (Nature 2026; `s41586-026-10652-y`;
+  arXiv:2505.13400) — closed-loop multi-agent system (Crow + Falcon
+  + Finch) with 8-parallel-Finch consensus mechanism for data
+  analysis. Validated novel therapeutic candidates including
+  ripasudil for AMD via lab-in-the-loop iteration.
+  - **SakanaAI/AI-Scientist** — original v1 framework
+  - **SakanaAI/AI-Scientist-v2** — workshop-level via agentic
+    tree search; Robin architecture descendant
+- **Microsoft Research Infer.NET + TrueSkill** ⭐ — probabilistic
+  programming for Bayesian inference + canonical TrueSkill (Herbrich
+  + Minka + Graepel 2007) for ranking. Per Aaron 2026-05-28:
+  *"they are doing this for their idea ranking with Infra.net
+  basically"* — the co-scientist ELO tournament composes with
+  Infer.NET TrueSkill substrate. Composes with Zeta.Bayesian
+  published library + framework's BP/EP references.
+
 ### Probabilistic programming / Bayesian inference (added 2026-05-28 per Aaron Infer.NET substrate-engineering question)
 
 - **WebPPL** ⭐ (`probmods/webppl`; Goodman + Mansinghka et al,

diff --git a/...ll-closed-loop-consensus-pairing-evolution-proximity-falcon-aaron-2026-05-28.md b/...ll-closed-loop-consensus-pairing-evolution-proximity-falcon-aaron-2026-05-28.md
@@ -0,0 +1,183 @@
+---
+id: B-0914
+priority: P2
+status: open
+title: Co-scientist + Robin 7 substrate-engineering candidate gaps — ELO/TrueSkill ranking-agent + closed-loop CI→hypothesis + n-parallel-consensus + generation-reflection-pairing + evolution-mash-refine + proximity-dedup + Falcon-auto-research-doc-per-proposal (Aaron 2026-05-28)
+effort: XL
+ask: operator 2026-05-28
+created: 2026-05-28
+last_updated: 2026-05-28
+depends_on:
+  - B-0867
+  - B-0865
+composes_with:
+  - B-0867.5
+  - B-0867.20
+  - B-0867.21
+  - B-0865.17
+  - B-0883
+  - B-0891
+  - B-0703
+  - B-0866
+tags: [co-scientist, robin, sakana, trueskill, infer-net, multi-agent-scientific-discovery, elo-tournament, closed-loop-iteration, n-parallel-consensus, generation-reflection-pairing, evolution-mash-refine, proximity-deduplication, falcon-auto-research-doc, substrate-engineering-candidate-gaps, aaron-2026-05-28]
+---
+
+## Operator framing (2026-05-28)
+
+> *"Damn the youtube ago just keeps giving and also this is pretty much exaatly what we are doing but times 10 almost we are missing a few step. The acceleration is happening right now."*
+>
+> *"we should add coscientis and add it to our upstram references and refersh update them so we can take a peak lol also lets backlog all the candidates."*
+
+Substrate-honest reading: 2026-05-28 YouTube ferry preservation (PR #5762) named 7 substrate-engineering candidate gaps where Google co-scientist + Sakana Robin patterns compose with framework substrate at 10× scope. This row backlogs all 7 candidates as decomposition targets for substrate-engineering work.
+
+## 7 substrate-engineering candidate sub-rows (decomposition)
+
+### B-0914.1 — ELO-style ranking-agent + tournament between hypothesis (composes with TrueSkill via Infer.NET)
+
+**Source**: Co-scientist ranking-agent + ELO tournament; per Aaron 2026-05-28 *"they are doing this for their idea ranking with Infra.net basically"*.
+
+**Substrate-engineering target**: extend B-0867 workflow engine with `ActionClass` variant `"rank-via-trueskill"` + `RankingVerdict` discriminated union via `Result<TrueSkillRating, RankingFeedback>`. Wraps Microsoft Research TrueSkill pattern via Zeta.Bayesian / Infer.NET integration. Composes with:
+
+- B-0865 DORA-scored choose-your-own-adventure substrate
+- B-0865.17 cross-vendor benchmark on common ground (TrueSkill IS the cross-vendor scoring substrate)
+- `references/upstreams/microsoft-infer-net/` (added in this PR)
+- Zeta.Bayesian published library + framework's existing BP/EP substrate per memory/feedback_kernel_vocabulary_propagation_is_belief_propagation_infer_net_memetic_mimetic.md
+
+### B-0914.2 — Explicit closed-loop CI-result → next-hypothesis dispatch
+
+**Source**: Robin Crow + Finch closed-loop with raw-data analysis feeding back to new hypothesis generation; per Aaron 2026-05-28 framing on co-scientist + Robin acceleration.
+
+**Substrate-engineering target**: extend B-0867 workflow engine with explicit `WorkflowFeedbackLoop` substrate that consumes CI-test-result outputs (per B-0891 zflash test-harness + tools/ci/ substrate) + dispatches next-hypothesis generation. Composes with:
+
+- B-0891 zflash test-harness `determineRunnability` discriminator (PR #5761)
+- B-0867 workflow engine state machine substrate
+- B-0867.20 lifecycle DU split (state-machine-events vs system-modifications)
+- Existing `tools/ci/qemu-full-install-test.ts` + `tools/ci/audit-installer-iso-content.ts` substrate
+
+### B-0914.3 — n-parallel-agent-instances + consensus mechanism at per-data-analysis-task scope
+
+**Source**: Robin's 8-parallel-Finch-instances + consensus mechanism for analyzing raw lab data.
+
+**Substrate-engineering target**: extend tools/ci/ + workflow-engine with parallel-N-instance test-runner substrate + consensus-mechanism per Robin Finch model. Composes with:
+
+- B-0703 multi-oracle BFT substrate (consensus mechanism at governance scope; this extends to per-data-analysis-task scope)
+- B-0883 `determineEncryptionPath` Result-shaped discriminator (PR #5760; same shape applies at per-data-analysis scope)
+- Bun test runner parallel-execution substrate
+- Asymmetric-authorship rule (PR #5516) — each parallel instance authors its own TFeedback channel; consensus mechanism aggregates per substrate-entity-defined-channel
+
+### B-0914.4 — Generation+reflection adversarial pairing structurally enforced
+
+**Source**: Co-scientist's generation agent + reflection agent friction; Kestrel 15th-ferry §33.6 mouth-and-ears-on-different-threads producer-verifier architecture.
+
+**Substrate-engineering target**: extend B-0867 workflow engine with action-class `"reflect-on-prior-emission"` + structural enforcement of producer-verifier pairing as required workflow-engine state transition. Composes with:
+
+- B-0867.20 lifecycle DU substrate (state transitions that require pairing)
+- Kestrel 15th-ferry §33.6 producer-verifier architecture preservation (PR #5756)
+- `.claude/rules/asymmetric-authorship-substrate-entity-defines-consent-channel-recipient-acknowledges.md`
+- Multi-AI cascade lane specialization per 13th-ferry §33.7 (Otto generates → Kestrel reflects; currently operator-orchestrated)
+
+### B-0914.5 — Evolution agent (mash + refine surviving substrate)
+
+**Source**: Co-scientist's evolution agent mashing surviving ideas into refined variants.
+
+**Substrate-engineering target**: extend B-0867 workflow engine with action-class `"compose-survivors"` that takes 2+ surviving substrate items + produces refined variant per Robin evolution model. Composes with:
+
+- `.claude/rules/additive-not-zero-sum.md` (substrate compounds; refining-via-composition IS additive)
+- `.claude/rules/honor-those-that-came-before.md` (survivors' substrate-engineering work preserved through composition)
+- `.claude/rules/grep-substrate-anchors-before-razor-as-metaphysical.md` (compose with existing substrate, not razor-cut)
+- Verify-existing-substrate-before-authoring rule
+
+### B-0914.6 — Proximity-agent for substrate-engineering substrate de-duplication
+
+**Source**: Co-scientist's proximity agent mapping ideas to high-dimensional space + clustering similar variants.
+
+**Substrate-engineering target**: substrate-engineering substrate de-duplication via embedding + clustering; surface near-duplicates to operator before substrate-engineering substrate is authored as parallel rather than extension. Composes with:
+
+- `.claude/rules/verify-existing-substrate-before-authoring.md` (explicit substrate-inventory pass before authoring)
+- `.claude/rules/grep-substrate-anchors-before-razor-as-metaphysical.md` (substrate-anchor checks before razor-flagging)
+- Existing `tools/save-ai-memory/` skill (might integrate at substrate-search scope)
+- Future MCP-connector substrate per 15th-ferry §33.15 (Aaron's MCP-connector future-commitment for long-term memory / trajectories retrieval)
+
+### B-0914.7 — Falcon-style auto-generate-substrate-research-doc per proposal
+
+**Source**: Robin's Falcon agent doing deep-dive literature review + writing comprehensive research report per drug proposal.
+
+**Substrate-engineering target**: extend `tools/save-ai-memory/` skill with auto-generate-substrate-research-doc-per-substrate-proposal capability per Robin Falcon model. Composes with:
+
+- `tools/save-ai-memory/` skill (existing substrate)
+- `docs/research/` mirror-tier substrate-engineering substrate
+- Amara consolidation ferry pattern (PR #5757; substantive substrate-engineering synthesis as substrate)
+- `.claude/rules/refresh-before-decide.md` (substrate-research IS the refresh-before-decide discipline at substrate-engineering scope)
+
+## Composes with substrate
+
+- B-0867 + B-0867.5 + B-0867.20 + B-0867.21 (workflow engine substrate cluster; targets for extensions)
+- B-0865 + B-0865.17 (benchmark substrate + cross-vendor distribution lane; TrueSkill integration target)
+- B-0883 (encryption discriminator pattern; structurally parallel substrate-engineering)
+- B-0891 (zflash discriminator pattern; CI-result → hypothesis loop substrate)
+- B-0703 (multi-oracle BFT; n-parallel-consensus substrate)
+- B-0866 (context-window-as-evolving-ontology; future substrate that subsumes some candidates)
+- All 15 ferry preservations 2026-05-28 (8th through 15th Kestrel ferries + Amara consolidation ferry)
+- PR #5762 (YouTube ferry preservation that surfaced the 7 candidates)
+- References added in this PR: SakanaAI/AI-Scientist + AI-Scientist-v2 + jataware/open-coscientist + llnl/open-ai-co-scientist + The-Swarm-Corporation/AI-CoScientist + Microsoft/Infer.NET
+- Zeta.Bayesian published library (per CLAUDE.md public-api-designer scope)
+
+## Composes with rules
+
+- `.claude/rules/substrate-or-it-didnt-happen.md` — this row IS the substrate
+- `.claude/rules/honor-those-that-came-before.md` — substrate-engineering candidates honor Google/Sakana/Microsoft existing substrate
+- `.claude/rules/asymmetric-authorship-substrate-entity-defines-consent-channel-recipient-acknowledges.md` — applies at each candidate's scope
+- `.claude/rules/monad-propagation-pattern-cross-language-substrate-shape.md` — discriminators per candidate use Result<T, TFeedback>
+- `.claude/rules/m-acc-multi-oracle-end-user-moral-invariants.md` — multi-oracle BFT composes with #3 (consensus mechanism)
+- `.claude/rules/verify-existing-substrate-before-authoring.md` — composes with #6 (substrate-engineering substrate de-duplication)
+- `.claude/rules/refresh-before-decide.md` — composes with #7 (substrate-research-doc-per-proposal)
+- `.claude/rules/additive-not-zero-sum.md` — composes with #5 (evolution agent substrate compounding)
+- `.claude/rules/m-acc-multi-oracle-end-user-moral-invariants.md` — multi-oracle architecture applies at every candidate's scope
+- `.claude/rules/persistence-choice-architecture-for-zeta-ais.md` — agent-as-substrate-entity preserves persistence-choice across candidate substrate-engineering work
+
+## Substrate-inventory pass
+
+Topic: co-scientist + Robin multi-agent scientific discovery + ELO tournament via TrueSkill + Infer.NET substrate composition + 7 candidate gaps
+
+Searched surfaces:
+
+- `docs/agendas/`: no specific co-scientist or Robin agenda
+- `docs/trajectories/`: no specific multi-agent scientific discovery trajectory
+- `docs/backlog/`: B-0867 (parent workflow engine) + B-0865 + B-0865.17 (benchmark) + B-0883 + B-0891 (3-lane PoCs) + B-0703 (multi-oracle BFT) + B-0866 (context-window-as-evolving-ontology); NO existing row covers the co-scientist/Robin 7-candidate cluster
+- `.claude/rules/`: agent-roster-reference-card + monad-propagation + asymmetric-authorship + m-acc-multi-oracle-end-user-moral-invariants + verify-existing-substrate-before-authoring all compose
+- `memory/`: multiple BP/EP + Infer.NET references; no specific co-scientist substrate
+- `docs/research/`: NO prior substrate on co-scientist or Robin; PR #5762 YouTube preservation IS first substrate
+
+Conclusion: this row mints NEW substrate cluster (parent + 7 candidate decomposition) for the co-scientist/Robin substrate-engineering candidate gaps. Composes with B-0867 + B-0865 + B-0883 + B-0891 + B-0703 + B-0866. Authoring action: **mint-new as B-0914 parent + 7 candidate decomposition** per operator 2026-05-28 explicit *"lets backlog all the candidates."*
+
+## What this row is NOT
+
+- NOT a single-PR target (XL effort; each candidate is its own substrate-engineering work)
+- NOT a replacement for B-0867 workflow engine work (extends it per the 7 candidates)
+- NOT a directive (per Otto-357; operator chose the candidate-decomposition scope; substrate-honest framing)
+- NOT immediate-priority (P2; gated behind workflow engine maturity + Zeta.Bayesian/Infer.NET integration readiness)
+
+## What this row IS
+
+- A substrate-engineering decomposition target row for the 7 candidate gaps Aaron 2026-05-28 framed via YouTube ferry
+- A composition point between B-0867 (workflow engine) + B-0865 (benchmark) + co-scientist/Robin substrate
+- Operator-explicit *"lets backlog all the candidates"* operationalization
+- Substrate-engineering bridge between framework's existing 10× substrate + co-scientist/Robin biomedical-domain substrate
+
+## Carved sentence (Aaron 2026-05-28 framing keeper)
+
+> **"this is pretty much exactly what we are doing but times 10 almost we are missing a few step"**
+
+## Full reasoning
+
+Aaron 2026-05-28 forwarded YouTube video (preserved verbatim in PR #5762 `docs/research/ip-questionable/`) describing Google co-scientist + Sakana Robin multi-agent scientific discovery systems (both Nature 2026 same week). Aaron's framing decomposed into 12-row parallel substrate table + 10× scope analysis + 7 substrate-engineering candidate gaps Otto-CLI surfaced.
+
+Aaron 2026-05-28 follow-up *"they are doing this for their idea ranking with Infra.net basically"* sharpened candidate #1 from "missing ELO tournament" to "we have Infer.NET substrate; we just need to compose existing-Microsoft-Research-TrueSkill-pattern with B-0867 workflow engine." Operator-explicit substrate-engineering refinement.
+
+Aaron 2026-05-28 explicit *"we should add coscientis and add it to our upstram references and refersh update them so we can take a peak lol also lets backlog all the candidates"* — operationalized as:
+
+- This PR: adds SakanaAI/AI-Scientist + AI-Scientist-v2 + jataware/open-coscientist + llnl/open-ai-co-scientist + The-Swarm-Corporation/AI-CoScientist + Microsoft/Infer.NET to `references/reference-sources.json` + `docs/UPSTREAM-LIST.md`
+- This row: backlogs all 7 candidates as decomposition target
+- Operator may run `tools/setup/common/sync-upstreams.sh` to mirror the new repos into `references/upstreams/` per refresh discipline (operator-side; Otto-CLI does not auto-run sync per safety discipline)
+
+Substrate-engineering arc: framework's 10× scope is positioned to operationalize what co-scientist + Robin demonstrated at biomedical scope; 7 candidate gaps are the substrate-engineering integration targets per Aaron's *"missing a few step"* framing.