[INIT] Mounting Research Pipeline...

The Researcher's Ledger

[PHILOSOPHY] This ledger tracks the transition of technical theory into local software. Each implementation is grounded in peer-reviewed research or documented validation scars.
Research Anchor Theoretical Logic Acme Implementation Git Link Status
FS-Researcher (2602.01566) Dual-Agent: Context Builder + Report Writer Foundation: `nibble.py` is the Context Builder . field_notes/nibble.py 100%
Agentic-R (2601.11888) Learning to Retrieve: Utility-based ranking & Grep Pivot. Memory Bridge: `ArchiveMemory` MMR ranking & fast Ripgrep pivot . nodes/archive_node.py 100%
TTCS (2601.22628) Test-Time Curriculum: Synthesizer + Solver. Quality: Mitigates repetition loops via Synthesize-then-Solve patterns . field_notes/ai_engine_v2.py 100%
Apple CLaRa (2511.18659) Semantic Compression: 16x-128x density. Optimization: `SemanticCondenser` compresses raw logs . field_notes/ai_engine_v2.py 100%
Liger Kernel (2410.10989) Efficient Triton Kernels for LLM Training. Efficiency: Maximizing 8B residency on 11GB VRAM . nodes/loader.py 100%
RLM (2512.24601) Context as a Code-Readable String. Discovery: `peek_related_notes()` logic . nodes/archive_node.py 100%
Swarm Delegation (2402.05120) Agentic Swarm Blueprint & Socket Activation Proxy. Orchestration: OpenAgent REST session dispatcher [BKM-034]. src/tests/delegate.py 100%
Tri-Node Routing (2405.07437) Multi-Node Heterogeneous Telemetry & Reasoning. Federation: Tri-Node evaluation & fallback ledger [SPR-47.1]. src/tests/test_integration_tri_node.py 100%
TTT-Discover (2601.16175) Test-Time Discovery: RL-based optimization. Automation: Planned: RL loops for bug reproduction. N/A 0%
Dreaming (2603.04257) Subconscious Compression (Memex). Consolidation: `dream_cycle.py` moving logs to Wisdom . dream_cycle.py 100%
Internal Debate (2603.00142) Moderated Consensus (Byzantine ToM). Consensus: `delegate_internal_debate` facilitates reasoning . logic/cognitive_hub.py 100%
AT2QA (2603.01853) Autonomous Exploration Pivot. Autonomy: Pivot-query logic pass for thin results. logic/cognitive_hub.py Design
Agentic FS (2602.20478) Context as a File System. Grounding: Validates our "Static Synthesis" architecture. field_notes/mass_scan.py 100%
13-Param Reason (2602.04118) Extreme parameter efficiency. Sentinel: Optimization logic for 1B "Sentinel" mice nodes. nodes/lab_node.py Active
REDSearcher (2602.14234) Long-Horizon Search Scaling. Discovery: framework for deep technical history discovery. nodes/archive_node.py Design
VibeThinker-3B (2606.16140) Verifiable Reasoning / Spectrum-to-Signal. Sanity: Unified local base model for Phase 10 . nodes/loader.py Active
MCompassRAG (2606.18508) Metadata as semantic compass. Discovery: Planned: Metadata-guided paragraph RAG. nodes/archive_node.py Design
HyDE (2212.10496) Hypothetical Document Embeddings. Open HyDE Preprocessor: Pinky's streaming preamble generates hypotheses . nodes/archive_node.py 75%
Query2Doc (2303.07678) Fine-Tuning LLMs for Pseudo-Document Expansion. LoRA HyDE: `cli_voice_v1` LoRA trained to emit dense acronyms & BKMs [SPR-58.0]. src/forge/train_expert.py 75%
GenRead (2209.10063) Parametric Context Generation for Dense Retrieval. Jeopardy Distillation: `distill_journal_ledger()` trains LoRA on bidirectional gem triggers [SPR-58.0]. field_notes/mass_scan.py 80%
Self-RAG (2310.11511) Learning to Retrieve, Generate, and Critique via Reflection. Refinement Loop: Tri-Field Gem Schemas & Hybrid RiR Gating [SPR-58.0]. field_notes/refine_gem.py 60%
Thinking to Recall (2603.09906) Uses CoT as a computational buffer for factual priming. Implemented via TTCS logic. field_notes/ai_engine_v2.py 100%
AutoHarness (2603.03329) Synthesizing Python guardrails for agent self-verification. Grounds future Verifier Synthesis. N/A Planned
Stochastic KV Routing (2604.22782) Adaptive depth-wise cache sharing (Apple MLR). Optimization pattern for Liger. nodes/loader.py Active
ARIS (2605.03042) Adversarial multi-agent collaboration for research. Theoretical base for Internal Debate. logic/cognitive_hub.py 100%
CodeTracer (2604.11641) Traceable agent states & failure onset localization. Enhanced Forensic Ledger visibility. debug/test_forensic_logging.py 100%
PersonaVLM (2604.13074) Proactive memory extraction & response alignment. Core logic for Subconscious Dreaming. dream_cycle.py 100%
TriAttention (2604.04921) Trigonometric KV compression (10x reduction). Future optimization for VRAM efficiency. nodes/loader.py Active
Ask, Don't Judge (2606.27226) Deterministic binary boolean question batteries. Evaluator: Cynical Curator & Validation Ledger . src/curator/scan_curator.py Active (SPR-59)
Context Compiler (TDS-2026) AST symbol graphs & call-graphs over raw context dumping. Agent Compaction: >60% token reduction for delegation . src/compiler/context_compiler.py Active (SPR-59)
Fourth Wall Feedback (BKM-035) Semantic language-first co-pilot critique interception. Verification: Auto-populates `validation_ledger.jsonl` . src/logic/cognitive_hub.py Active (SPR-59)
Speculative Interest Pre-fetch (2608.13667) Speculative RAG context pre-fetching during Turn 1 streaming. Hub Cascade: Zero-latency interjection with interest preemption . src/logic/cognitive_hub.py 100% (Certified)
Conversational WYWO (Anti-Embellish) Dynamic floating validation scars & gem pool for greeting turns. Persona: Real-world oracle prompt over generic assistant filler . src/logic/cognitive_hub.py Active (SPR-59)
AutoMem (2607.01224) Automated cognitive skill memory write/eviction policies. Reference Anchor: Policy-driven memory update & forget gates. N/A Reference Anchor
Netflix Rubric Lifecycle (Prod-Judge) 4-phase judge lifecycle (Birth -> Rubric Tuning -> Dual Role -> Drift). Reference Anchor: Continuous semantic drift monitoring for evaluators. N/A Reference Anchor
RLSVR (2607.23802) Self-verifiable rewards via task transformation into unit tests. Reference Anchor: Task-level assertion compilation for delegation. N/A Reference Anchor
LightMem-Ego (2607.11487) Lightweight episodic personal memory condensation. Reference Anchor: Edge-optimized lifelong assistance timeline compression. N/A Reference Anchor
HOLA (Comp-Delta) Hippocampal Linear Attention (semi-parametric recurrent state). Tabled Anchor: Requires custom non-transformer Triton kernel training. N/A Tabled