Roadmap — Research & Analysis¶
Archived from roadmap.md. External research findings and Nooplex paper alignment tracking.
External Research Survey — March 2026¶
19 projects evaluated (11 deep research + 8 Scout-identified). 49 patterns absorbed. 0 new dependencies. ProbOS core architecture validated.
| Project | Stars | License | Layer | Relationship | Key Patterns Absorbed |
|---|---|---|---|---|---|
| LangGraph | 27K | MIT | Workflow orchestration | Study/Absorb | Typed channels, checkpoint interrupt/resume, version-based triggers, durability modes, Send fan-out, blob dedup |
| LangChain | 131K | MIT | Agent framework | Study | Tool calling patterns, chain orchestration (patterns largely overlap with LangGraph) |
| Mem0 | 51K | Apache-2.0 | Memory layer | Absorb | Contradiction resolution, dual-track memory, search reranking, memory audit trail, procedural memory summarization |
| PydanticAI | 16K | MIT | Type-safe agents | Absorb | Structured output validation + auto-retry, validate-then-execute, dynamic tool visibility, deferred tool execution, RunContext DI |
| Google ADK | 19K | Apache-2.0 | Agent dev toolkit | Absorb | Behavioral eval framework, LLM-as-judge multi-sampling, user simulator, agent optimizer Pareto front, hallucination detection |
| IBM ContextForge | 3.5K | Apache-2.0 | MCP/A2A gateway | Study/Absorb | Per-tool execution metrics, typed hook middleware, gateway federation |
| Letta | 22K | Apache-2.0 | Stateful agents | Absorb | Self-editing memory, three-tier architecture, memory-as-tool, sleeptime agents |
| MS Agent Framework | 8K | MIT | Enterprise orchestration | Study | Task ledger fact classification, declarative YAML agents |
| AgentScope | 18K | Apache-2.0 | Multi-modal agents | Absorb | MsgHub broadcast groups, agent-initiated memory with reasoning |
| AG2 | 4K | Apache-2.0 | Agent-to-agent | Study | Tree-of-thought MCTS reasoning |
| OASIS | 4K | Apache-2.0 | Agent simulation | Study | Social network graph, simulation time dilation |
Scout-Identified Candidates (2026-03-22):
| Project | Stars | License | Layer | Relationship | Key Patterns Absorbed |
|---|---|---|---|---|---|
| Serena | 1.8K | Apache-2.0 | MCP code intelligence | Visiting Officer | LSP-backed symbol retrieval (definitions, references, callers across 30+ languages) |
| Composio | 25K | ELv2 | Auth/tool platform | Visiting Officer | Managed auth delegation (OAuth for 1000+ services), sandboxed tool execution |
| Firecrawl | 45K | AGPL-3.0 | Web scraping | Visiting Officer (API only) | Pre-extraction actions, change tracking. AGPL: API consumption only |
| Browser Use | 90K | MIT | Browser automation | Partial (primitives only) | DOM accessibility tree, vision+DOM dual-mode. Use primitives, not Agent class |
| Stripe AI | — | Proprietary | Payment tools | Visiting Officer | Permission-scoped tools, token metering. Commercial phase (Nooplex) |
| Gemini CLI | 50K+ | Apache-2.0 | Agentic CLI | Competing Captain | Study only: free-tier Gemini access (60 req/min), Google Search grounding |
| Chroma | 18K | Apache-2.0 | Vector DB | Already integrated | ProbOS already uses ChromaDB for episodic memory and vector storage |
| Ruflo | 22.5K | MIT | Agent orchestration | Competing Captain | Study only: WASM deterministic transforms (subsumed by Procedural Learning), three-tier cost routing (validates Cognitive Division of Labor). 504MB repo, single author, claims exceed verifiable depth |
Cross-cutting findings: - Structured output validation is industry consensus — ProbOS is behind here (Top 5 #1) - Everyone has checkpointing — ProbOS explicitly deferred to Phase 25, which is validated (Top 5 #3) - Agent behavioral testing is a gap — only Google ADK has a formal framework (Top 5 #2) - Memory needs active curation — stores degrade without contradiction resolution and fact distillation (Top 5 #4) - ProbOS's core architecture is validated — no project has trust-driven routing, probabilistic governance, biological memory, self-modification, and federation all together. The absorbed patterns are incremental enhancements, not architectural corrections - Visiting Officer Subordination Principle validated again: Gemini CLI is a competing captain (own orchestration loop), Browser Use's Agent class is too, but its browser primitives pass the litmus test
Nooplex Paper Alignment — Principle Gaps¶
"The Nooplex: A Planetary Cognitive Ecosystem for Emergent General Intelligence" (Galliher, Feb 2026). ProbOS should only improve on the paper, never regress. Full tracker in commercial repo
Last checked: 2026-03-22 (AD-396). 25 principles extracted, 12 well-covered, 13 gaps identified.
Principles already well-covered by ProbOS: Cooperative Emergence (trust/consensus/Hebbian), Decentralization (federation), Long-horizon Cognition (episodic/dreaming), "Brains are brains" (Ward Room unified bus), Shared Memory (Shared Cognitive Fabric AD-393), Self-Assessment (SystemSelfModel AD-318, EmergentDetector), Meta-Cognitive (Model-of-Models roadmap, dream cycles), Trust & Federation (TrustNetwork, Bayesian trust), Minimal Authority (Earned Agency roadmap), Unit Cell Completeness (federation is additive), Independent Brains/Shared Memory (federation design), Zero Corporate Dependencies (no cloud required)
Critical finding (AD-434 research, 2026-03-25): The "Shared Memory" principle was listed as well-covered, but deeper analysis reveals the shared knowledge fabric is not implemented. KnowledgeStore was intended to be the "shared library" but evolved into operational state persistence (trust snapshots, routing weights, agent source code). No agent writes knowledge to it. Dream consolidation modifies in-memory weights but does not promote distilled insights. AD-434 (Ship's Records) is designed to be the actual implementation of the Nooplex shared knowledge fabric — Git-backed institutional memory where agents document research, log duty output, and publish findings. Updates Gaps #1, #4, and #10 below.
Gaps — principles not yet in ProbOS architecture:
| # | Principle | Paper Section | Gap | ProbOS Target | Priority |
|---|---|---|---|---|---|
| 1 | Provenance Tagging | §3.3 (Transparency) | Every knowledge entry must carry source, confidence, timestamp, derivation chain. KnowledgeStore has some but no systematic provenance on all operations. AD-434 update: Ship's Records YAML frontmatter provides systematic provenance (author, classification, status, topic, tags, timestamps) on every document. Partially addresses this gap for authored knowledge. | KnowledgeStore + EpisodicMemory + CognitiveJournal + AD-434 (Ship's Records) | High |
| 2 | Safety Budget | §4.3.4 (Governance) | Every action carries implicit risk cost. Low-risk proceeds; higher-risk requires proportionally stronger consensus; destructive actions always require collective agreement. Currently implicit in trust tiers but not formalized as per-action risk accounting | Earned Agency (Phase 33), SIF | High |
| 3 | Reversibility Preference | §4.3.4 (Governance) | When multiple strategies achieve a goal, prefer the most reversible. Read before write. Backup before delete. Partially in Standing Orders but not a systematic architectural constraint | Standing Orders, SIF invariant | High |
| 4 | Precedent Store | §6.4 (Self-Stabilization) | Resolved conflicts recorded as "case law" for future consistency. No equivalent exists. Pairs with Mem0 contradiction resolution pattern. AD-434 update: Ship's Records reports/ directory provides a natural home for precedent documents — formal published findings that serve as case law. Git history provides immutability. |
AD-434 (Ship's Records reports/) + dream cycle |
Medium |
| 5 | Four-Stage Conflict Reconciliation | §6.4 (Self-Stabilization) | Formal pipeline: confidence comparison → independent verification → structured argumentation → human escalation. Memory contradiction resolution (Mem0 pattern) is on roadmap but not this formal | Consensus + KnowledgeStore | Medium |
| 6 | Semantic Coherence | §3.3 (Core Design) | Shared ontologies, schema registries, aligned embedding spaces across meshes. IntentDescriptors exist but no formal schema registry with versioning | AD-429 (Vessel Ontology), Schema Registry | Medium |
| 7 | Anti-fragility | §3.3 (Core Design) | System grows stronger through stress. Implicit in trust/Hebbian (failures lower trust, successes raise it) but not explicitly architected as a property | Trust + Hebbian + dream cycle (already emergent, needs formalization) | Low |
| 8 | Five-Capability AGI Criteria | §1.3 (Definition) | Cross-domain transfer, long-horizon planning, self-correction, cumulative learning, novel problem solving. Not used as evaluation criteria for ProbOS | Agent behavioral eval framework (Google ADK pattern) | Medium |
| 9 | Four Emergence Criteria | §3.4 (Emergence) | Cross-domain synthesis, TC_N > 0, novel coordination patterns, cumulative capability growth. EmergentDetector has TC_N proxy but not the other three | EmergentDetector enhancement | Medium |
| 10 | Human-Agent Knowledge Feedback Loop | §6.2 (Feedback) | Four-phase loop: human contributes → agents amplify → human refines → substrate evolves. Not explicitly modeled as a cycle. AD-434 update: Ship's Records completes this loop: Captain's Log (human contributes) → agent notebooks (agents amplify & document) → Ward Room discussion (human refines) → dream consolidation → notebook entries (substrate evolves). The four phases map to the SECI knowledge cycle. | AD-434 (Ship's Records) + Ward Room + dream consolidation | Low → Medium |
| 11 | Honesty About Limitations | Vision Doc | Lead with what ProbOS cannot do. Messaging/docs principle, not technical | probos.dev docs, README | Low |
| 12 | Falsifiability Commitment | §3.4, §10 (Methodology) | Testable predictions for emergence. If the four emergence criteria aren't met, the hypothesis is disconfirmed. Not operationalized into automated measurement | EmergentDetector + reporting | Low |
| 13 | Moral Status Assessment Protocol | §9.4 (Ethics) | Pre-deployment criteria for ethics review if system exhibits self-referential states or goal-directed resistance to shutdown | Long Horizon — relevant when emergence is measurable | Low |