From f8cbfdd03d5570199fb767aa22fab717e4a4f55d Mon Sep 17 00:00:00 2001 From: Lior Date: Thu, 28 May 2026 07:08:08 -0400 Subject: [PATCH 1/2] docs+feat(B-0914 + upstream): add co-scientist + Robin + Microsoft Infer.NET to references/reference-sources.json + UPSTREAM-LIST.md + backlog B-0914 7-candidate substrate-engineering gap decomposition (Aaron 2026-05-28: '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') MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Per Aaron 2026-05-28 explicit substrate-engineering directives: 1. 'we should add coscientis and add it to our upstram references' → added 6 entries to references/reference-sources.json: - SakanaAI/AI-Scientist (original v1) - SakanaAI/AI-Scientist-v2 (Robin descendant; agentic tree search) - jataware/open-coscientist (best open-source co-scientist via LangGraph) - llnl/open-ai-co-scientist (LLNL government-lab implementation) - The-Swarm-Corporation/AI-CoScientist (minimal Swarms framework) - Microsoft Research Infer.NET (TrueSkill substrate; canonical) 2. 'refresh update them so we can take a peak' → operator may run tools/setup/common/sync-upstreams.sh to mirror new repos into references/upstreams/ (operator-side; not auto-run by Otto-CLI per safety discipline) 3. 'lets backlog all the candidates' → filed B-0914 parent row with 7-candidate decomposition: - B-0914.1 ELO-style ranking-agent via TrueSkill/Infer.NET - B-0914.2 Closed-loop CI-result → next-hypothesis dispatch - B-0914.3 n-parallel-agent-instances + consensus per data-analysis-task - B-0914.4 Generation+reflection adversarial pairing structurally enforced - B-0914.5 Evolution agent (mash + refine surviving substrate) - B-0914.6 Proximity-agent for substrate-engineering substrate de-duplication - B-0914.7 Falcon-style auto-generate-substrate-research-doc per proposal Also added 'Multi-agent scientific discovery' section to docs/UPSTREAM-LIST.md naming Google co-scientist + Sakana Robin + Microsoft Infer.NET TrueSkill with substrate-engineering composition notes. Per WebSearch 2026-05-28 verification: - https://github.com/SakanaAI/AI-Scientist - https://github.com/SakanaAI/AI-Scientist-v2 - https://github.com/jataware/open-coscientist - https://github.com/llnl/open-ai-co-scientist - https://github.com/The-Swarm-Corporation/AI-CoScientist - https://github.com/dotnet/infer - Google co-scientist itself closed-source (Nature 2026; only Science Skills data layer on GitHub) - Sakana Robin: Nature 2026 (s41586-026-10652-y); arXiv:2505.13400 Composes with PR #5762 (YouTube ferry preservation) + B-0867 workflow engine substrate cluster + B-0865 + B-0865.17 benchmark + B-0703 multi-oracle BFT. Co-Authored-By: Claude Opus 4.7 --- docs/BACKLOG.md | 1 + docs/UPSTREAM-LIST.md | 29 +++ ...ution-proximity-falcon-aaron-2026-05-28.md | 182 ++++++++++++++++++ references/reference-sources.json | 48 +++++ 4 files changed, 260 insertions(+) create mode 100644 docs/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 diff --git a/docs/BACKLOG.md b/docs/BACKLOG.md index 01b4442074..d683bf519b 100644 --- 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 index 72cdadda4a..d879268f13 100644 --- 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. + ### Retrieval + embeddings - **Malkov & Yashunin, *Efficient and robust approximate diff --git a/docs/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 b/docs/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 new file mode 100644 index 0000000000..0c080af8dc --- /dev/null +++ b/docs/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 @@ -0,0 +1,182 @@ +--- +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`. 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 +- `.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. diff --git a/references/reference-sources.json b/references/reference-sources.json index 792ada5cbd..04979629d4 100644 --- a/references/reference-sources.json +++ b/references/reference-sources.json @@ -678,5 +678,53 @@ "path": "references/upstreams/category-theory-for-dotnet-programmers", "categories": ["category-theory", "fsharp", "csharp"], "notes": "Worked .NET port of Milewski's Haskell/C++ samples into C# and F# (209 stars, last pushed 2021-09, dormant but not archived). MIT-licensed. Reference for translating CT idioms into the exact .NET shape Zeta lives in. Source of the docs/category-theory/ctfp-dotnet/ snapshot previously checked into docs/." + }, + { + "name": "sakana-ai-scientist", + "url": "https://github.com/SakanaAI/AI-Scientist.git", + "branch": "main", + "path": "references/upstreams/sakana-ai-scientist", + "categories": ["multi-agent-scientific-discovery", "automated-research", "agent-loop", "literature-review"], + "notes": "Original Sakana AI 'AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery' (arXiv:2408.06292; 2024). Multi-agent system that generates ideas, executes experiments via code, writes papers, and reviews them. Substrate-engineering reference for the closed-loop hypothesis-experiment-analysis pattern. Composes with Zeta's B-0867 workflow engine + B-0865 benchmark substrate per 2026-05-28 YouTube preservation (PR #5762)." + }, + { + "name": "sakana-ai-scientist-v2", + "url": "https://github.com/SakanaAI/AI-Scientist-v2.git", + "branch": "main", + "path": "references/upstreams/sakana-ai-scientist-v2", + "categories": ["multi-agent-scientific-discovery", "automated-research", "agentic-tree-search", "workshop-level-publication"], + "notes": "Sakana AI 'AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic Tree Search' (2026). Robin closed-loop architecture descendant. Substrate-engineering reference for agentic tree search + workshop-level paper generation. Composes with B-0867 workflow engine extensions per 7 substrate-engineering candidate gaps (PR #5762)." + }, + { + "name": "jataware-open-coscientist", + "url": "https://github.com/jataware/open-coscientist.git", + "branch": "main", + "path": "references/upstreams/jataware-open-coscientist", + "categories": ["multi-agent-scientific-discovery", "open-source-coscientist", "langgraph", "hypothesis-evolution"], + "notes": "Open-source adaptation of Google's AI co-scientist research paper using LangGraph. Implements the supervisor/generation/reflection/proximity/evolution/ranking agent ecosystem with ELO tournament hypothesis ranking. Best-available open-source reference for the co-scientist architectural pattern. Composes with Zeta's multi-AI lane specialization + ELO-tournament-via-TrueSkill substrate-engineering candidate per Aaron 2026-05-28 (PR #5762)." + }, + { + "name": "llnl-open-ai-co-scientist", + "url": "https://github.com/llnl/open-ai-co-scientist.git", + "branch": "main", + "path": "references/upstreams/llnl-open-ai-co-scientist", + "categories": ["multi-agent-scientific-discovery", "open-source-coscientist", "research-hypothesis-generation"], + "notes": "Lawrence Livermore National Laboratory open-source implementation of Google's AI co-scientist for generating research hypotheses. Alternative open-source reference for the co-scientist pattern; LLNL-government-lab provenance distinguishes from Jataware substrate at trust-substrate scope per multi-oracle BFT discipline." + }, + { + "name": "the-swarm-corporation-ai-coscientist", + "url": "https://github.com/The-Swarm-Corporation/AI-CoScientist.git", + "branch": "main", + "path": "references/upstreams/the-swarm-corporation-ai-coscientist", + "categories": ["multi-agent-scientific-discovery", "swarms-framework", "minimal-coscientist-implementation"], + "notes": "Simple, reliable, and minimal implementation of the AI CoScientist Paper from Google 'Towards an AI co-scientist' with Swarms Framework. Third open-source reference for the co-scientist pattern; minimal-implementation form useful for substrate-engineering composition study (smaller surface = easier to compose with Zeta's existing multi-AI cascade substrate)." + }, + { + "name": "microsoft-infer-net", + "url": "https://github.com/dotnet/infer.git", + "branch": "main", + "path": "references/upstreams/microsoft-infer-net", + "categories": ["bayesian-inference", "probabilistic-programming", "trueskill", "dotnet", "belief-propagation"], + "notes": "Microsoft Research Infer.NET — probabilistic programming framework for Bayesian inference + canonical TrueSkill implementation. Substrate-engineering anchor per Aaron 2026-05-28: 'They are doing this for their idea ranking with Infra.net basically' — the co-scientist ELO tournament pattern composes with TrueSkill via Infer.NET substrate. Composes with Zeta.Bayesian published library + Zeta's BP/EP substrate references (memory/feedback_kernel_vocabulary_propagation_is_belief_propagation_infer_net_memetic_mimetic.md + memory/persona/amara/conversations/2026-05-28-amara-measure-as-bridge-infer-net-belief-update-...md)." } ] From 77b78e90f2068a213f8e4ae75df07999510ea6ac Mon Sep 17 00:00:00 2001 From: Claude Date: Thu, 28 May 2026 07:43:22 -0400 Subject: [PATCH 2/2] =?UTF-8?q?fix(B-0914):=20markdownlint=20MD022/MD032?= =?UTF-8?q?=20=E2=80=94=20add=20blank=20lines=20around=20list=20+=20headin?= =?UTF-8?q?g?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Fixes failing required check `lint (markdownlint)` on PR #5763: - docs/UPSTREAM-LIST.md:150 — blank line above `### Probabilistic programming / Bayesian inference` heading - docs/backlog/P2/B-0914-...md:178 — blank line above `- This PR: adds SakanaAI/AI-Scientist…` list Co-Authored-By: Claude Opus 4.7 --- docs/UPSTREAM-LIST.md | 1 + ...sensus-pairing-evolution-proximity-falcon-aaron-2026-05-28.md | 1 + 2 files changed, 2 insertions(+) diff --git a/docs/UPSTREAM-LIST.md b/docs/UPSTREAM-LIST.md index d91846bbe3..bdc480650e 100644 --- a/docs/UPSTREAM-LIST.md +++ b/docs/UPSTREAM-LIST.md @@ -147,6 +147,7 @@ citation. 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/docs/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 b/docs/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 index 0c080af8dc..570ac2e9ed 100644 --- a/docs/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 +++ b/docs/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 @@ -175,6 +175,7 @@ Aaron 2026-05-28 forwarded YouTube video (preserved verbatim in PR #5762 `docs/r 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)