41 Algorithms — Reference Overview

RCT Platform is driven by 41 algorithms organized into a 9-Tier stack. Each Tier builds on the one below it — from mathematical foundations at Tier 1 to fully autonomous generation at Tier 9.

Design Philosophy

Algorithms are not isolated tools. They form an intent-preserving pipeline: every input is scored, planned, verified, and crystallized before it reaches execution.

---

Architecture at a Glance

┌──────────────────────────────────────────────────────────┐
│  Tier 1  │  Meta Foundation            │  3 algorithms   │
│  Tier 2  │  Core Processing            │  3 algorithms   │
│  Tier 3  │  Enhancement & Reflection   │  5 algorithms   │
│  Tier 4  │  Generation                 │  5 algorithms   │
│  Tier 5  │  Infrastructure             │  6 algorithms   │
│  Tier 6  │  Support Systems            │  4 algorithms   │
│  Tier 7  │  Evolution — Performance    │  5 algorithms   │
│  Tier 8  │  Evolution — Error Reduction│  5 algorithms   │
│  Tier 9  │  Autonomy Layer             │  5 algorithms   │
├──────────────────────────────────────────────────────────┤
│  TOTAL                                 │  41 algorithms  │
└──────────────────────────────────────────────────────────┘

All 41 algorithms are validated against 500,000+ property-based test examples using the Hypothesis framework. The SDK ships reference implementations for the core tiers (Tier 1–3).

---

Tier 1 — Meta Foundation

The philosophical and mathematical bedrock. Every intent processed by RCT passes through Tier 1 first.

---

ALGO-01: FDIA

Full Name: Fundamental Decision Intelligence Algorithm
Score: 9.2 / 10

The core equation of the entire RCT system. Every AI action is scored before execution.

\[F = D^I \times A\]

Symbol	Meaning
F	Future — the output the AI must deliver
D	Data quality (0.0 – 1.0)
I	Intent precision — acts as an exponent, not a multiplier
A	Architect — human-in-the-loop gate. When A = 0, output is constitutionally blocked

Why I is an exponent: High intent amplifies good data exponentially. Low intent degrades the result even with perfect data. This is intentional — clarity of purpose is the primary lever.

Use with: core/fdia/ — full implementation guide with code examples.

---

ALGO-02: MOIP

Full Name: Multi-Objective Intent Planning
Score: 9.1 / 10

Uses Pareto optimization to plan across multiple competing goals simultaneously — cost, speed, quality, compliance — without collapsing them into a single weighted score.

Use cases: Enterprise workflow planning with multi-stakeholder constraints, budget vs. quality tradeoff resolution, compliance-aware resource allocation.

---

ALGO-03: Delta Engine

Full Name: Gap Analysis & Compression Engine
Score: 8.7 / 10

Stores and computes state differences (deltas) rather than full snapshots — the equivalent of git diffs for AI memory.

\[\Delta = \text{State}_{\text{after}} - \text{State}_{\text{before}}\]

Result: 74% memory compression with full audit-trail replay capability.

Connects to: ALGO-10 (Delta Memory), ALGO-25 (Delta Block) — both use Delta Engine as their computation backend.

---

Tier 2 — Core Processing

The main cognitive processing layer. Handles intent parsing, knowledge retrieval, and structured reasoning.

---

ALGO-04: RCT-7

Full Name: 7-Step Cognitive Process Framework
Score: 9.0 / 10 | End-to-end accuracy: 96.0%

The structured reasoning kernel. Every complex request passes through all 7 steps in sequence — no step can be skipped for Tier-O (high-stakes) operations.

Step	Name	Purpose
1	Observe	Capture raw input without interpretation
2	Analyze	Parse into structured entities and constraints
3	Deconstruct	Break into atomic executable sub-tasks
4	Reverse Reasoning	Work backward from goal to validate plan
5	Identify Core Intent	Resolve the single primary intent from competing signals
6	Reconstruct	Build the verified, signed execution plan
7	Compare	FDIA-score the plan against the original intent

Three variants: RCT-O (full, 7 steps) · RCT-S (condensed, 4 steps) · RCT-I (interpretive, for ambiguous inputs).

See: RCT-7 Thinking Protocol — full documentation with examples.

---

ALGO-05: GraphRAG

Full Name: Graph-Based Retrieval Augmented Generation
Score: 8.5 / 10

Combines graph traversal with semantic vector search for knowledge retrieval. Standard RAG finds similar text; GraphRAG finds similar text and follows entity relationships across the knowledge graph.

Query → Vector Search (semantic)
      + Graph Traversal (relationships)
      → Multi-hop Context Assembly
      → Grounded Generation

Result: Significantly reduced hallucination rate by anchoring generation to graph-verified facts.

---

ALGO-06: JITNA Protocol

Full Name: Just In Time Nodal Assembly
Score: 8.3 / 10

The communication protocol of the RCT kernel. Every message between agents is a JITNA packet with 6 canonical fields:

Field	Meaning
I	Intent
D	Data
Δ	Delta (gap between current and target state)
A	Architect (accountability assignment)
R	Reflection
M	Memory reference

See: JITNA Protocol · RFC-001 Specification

---

Tier 3 — Enhancement & Reflection

Self-improvement, evolutionary growth, and reflection loops that make the system learn from its own outputs.

---

ALGO-07: MEE v2

Full Name: Meta-Evolution Engine v2
Score: 9.85 / 10

The engine that enables monotonic growth — the system's output value increases every generation, with mathematical proof that it never decreases.

\[G_{n+1} = G_n \times (1 + M \times \Delta) \times R_t\]

Symbol	Meaning
G	Growth value (current generation)
M	Mutation rate
Δ	Improvement delta per evolution cycle
R_t	Retention factor at time t

Property: Monotonic — growth is guaranteed to be non-negative across all configurations.

---

ALGO-08: Self-Evolving Systems

Full Name: Autonomous Self-Improvement System
Score: 9.85 / 10

An autonomous system that observes its own performance metrics, identifies improvement opportunities through FDIA scoring, proposes modifications through MEE v2, validates through SignedAI consensus, and applies changes — forming a closed improvement loop without human intervention.

Depends on: ALGO-07 (MEE v2) as the growth engine, ALGO-30 (ABV) for validation confidence.

---

ALGO-09: Reflexion+

Full Name: Enhanced Self-Reflection Agent
Score: 8.6 / 10

A multi-round self-correction loop. The agent generates an output, an external critic (SignedAI) evaluates it, the agent reflects on the critique, and regenerates — converging toward a verified answer.

Attempt → Critic (SignedAI) → Reflection → Improved Attempt → Convergence Test

Differentiator from standard Reflexion: Uses an external verifier at the Critic stage instead of self-evaluation only.

---

ALGO-10: Delta Memory

Full Name: Incremental Memory Storage
Score: 8.0 / 10

User-facing memory API built on ALGO-03 (Delta Engine). Stores session and long-term memory as delta chains rather than full snapshots — enabling instant warm recall at < 50ms (vs cold start 3–5s).

---

ALGO-11: BBA→P→CF

Full Name: Belief → Plan → Consequence Flow
Score: 9.3 / 10

A causal reasoning pipeline: starts from stated beliefs/assumptions, builds an executable plan, then simulates the downstream consequences before committing. Exposes hidden causal chains and prevents unintended side effects.

---

Tier 4 — Generation

Algorithms that create — new knowledge, new algorithms, new modules, and new structured outputs.

---

ALGO-12: Meta-Algorithm Generator (Γ)

Full Name: Meta-Algorithm Generator
Score: 8.9 / 10

Given a problem description, generates a new algorithm specification automatically. Operates as a meta-programming layer above the algorithm stack — capable of spawning new specialized algorithms to fill gaps.

---

ALGO-13: RCT-Diffusion

Full Name: Diffusion-Based Generative Algorithm
Score: 9.6 / 10

Applies the diffusion model paradigm (iterative denoising) to structured data generation — synthetic datasets, enterprise knowledge augmentation, and sparse-data enrichment. Generates with higher fidelity than autoregressive approaches for structured schemas.

---

ALGO-14: GraphRAG Complete

The production-grade successor to ALGO-05. Integrates full Neo4j graph database with Qdrant vector search for enterprise-scale knowledge retrieval with multi-hop reasoning across arbitrarily large knowledge graphs.

---

ALGO-15: HRM Controller

Full Name: Human-in-the-loop Reasoning Manager
Score: 9.0 / 10 | Accuracy: 98.0%

Manages the escalation pipeline to human reviewers. Triggers when FDIA score falls below threshold, when policy evaluation returns REQUIRES_REVIEW, or when Constitutional AI gate fires. Delivers a complete context package to the reviewer — not just an alert.

See: Governance Layer

---

ALGO-16: Vector Search

Full Name: HNSW Semantic Search
Score: 7.0 / 10

High-performance semantic search using HNSW (Hierarchical Navigable Small World) graph algorithm on Qdrant. Target latency: < 50ms (p95, warm cache). Supports quantization for 70% index size reduction.

---

Tier 5 — Infrastructure

The routing, orchestration, and fusion layer that connects all other algorithms.

---

ALGO-17: Graph Traversal

Score: 7.2 / 10

Multi-algorithm graph navigation: BFS, DFS, Dijkstra, PageRank, Betweenness Centrality. Powers knowledge graph navigation, dependency resolution, and influence ranking across the RCT knowledge base.

---

ALGO-18: Adaptive Prompting

Score: 7.8 / 10

Dynamically adapts prompt templates based on user profile, domain expertise level, language preference, and historical response quality — without requiring manual template engineering.

---

ALGO-19: Data Fusion v2

Score: 9.0 / 10

Fuses data from multiple sources with conflict resolution. When sources disagree, applies source reliability weighting, timestamp recency scoring, and provenance analysis to produce a consensus value with a confidence score.

---

ALGO-20: Workflow Orchestrator v2

Score: 9.2 / 10

DAG-based async workflow engine for multi-agent tasks. Supports parallel execution of independent tasks, retry with exponential backoff, dead letter queues for failed tasks, and checkpoint/restore for long-running workflows.

---

ALGO-21: Fast/Slow Router

Score: 8.8 / 10 | Accuracy: 95.8%

The primary routing algorithm. Classifies every incoming intent by complexity and routes to the appropriate lane:

Lane	Criteria	Target Latency
Fast Lane	Simple, high-confidence, cached patterns	< 200ms
Slow Lane	Complex, ambiguous, high-stakes	Full RCT-7 pipeline

---

ALGO-22: Halting Detection

Score: 8.5 / 10

Detects infinite loops, deadlocks, and stuck states in multi-agent workflows through execution timeout monitoring, cycle detection in the execution graph, and heartbeat tracking. Failed tasks are routed to the Dead Letter Queue with full context.

---

Tier 6 — Support Systems

---

ALGO-23: Content-Box

Score: 8.2 / 10

Structured output rendering engine. Takes validated AI output and renders it into the appropriate format — Markdown, JSON, HTML, or React component schemas — based on the requesting consumer.

---

ALGO-24: Benchmark Suite

Score: 7.6 / 10

The testing framework for all 41 algorithms. Five test profiles from Normal (250 examples) to Ultra-continuous (250,000 examples), all using property-based testing via Hypothesis. Used to generate the accuracy and reliability scores reported throughout this documentation.

---

ALGO-25: Delta Block

Score: 7.7 / 10

The atomic storage unit inside ALGO-03 (Delta Engine). A Delta Block captures a single state transition with full provenance — what changed, when, by which agent, and under which FDIA score.

---

ALGO-26: Intent Classification

Full Name: Intent Conservation & Tracking
Score: 9.4 / 10

Tracks and preserves intent across multi-turn conversations and multi-agent handoffs. Prevents intent drift — the gradual corruption of the original request as it passes between agents and transformation steps.

---

Tier 7 — Evolution Cycle 1 (Performance)

Five algorithms that scale the platform from prototype to production throughput.

ALGO	Name	Key Target
ALGO-27	TVRA — Time-aware Video Reasoning & Analysis	+18pp video comprehension improvement
ALGO-28	CIO — Concurrent I/O Optimizer	1,000 → 10,000 concurrent requests
ALGO-29	UIA — Universal Integration Adapter	15 → 5,000 API integrations
ALGO-30	ABV — Adaptive Bayesian Validator	Confidence 70% → 95%
ALGO-31	ALBAS — Auto-Load Balancing & Auto-Scaling	1,000 → 10,000 requests/sec throughput

ALGO-30 (ABV) uses Bayesian inference to continuously update confidence scores as evidence accumulates — reducing both false positives and false negatives in fact verification.

---

Tier 8 — Evolution Cycle 2 (Error Reduction)

Five algorithms that drive error rates toward zero.

ALGO	Name	Key Target
ALGO-32	MCTR — Multi-Chain Thought Reasoning	−15% reasoning errors
ALGO-33	FGHF — Fact-Grounded Hallucination Filter	Hallucination rate < 0.3% (vs industry 12–15%)
ALGO-34	SWCAR — Semantic Web & Content Auto-Repair	−12% web content errors
ALGO-35	ATC — Adaptive Timeout Controller	−7% timeout failures
ALGO-36	RFLH — Rare-Failure Learning Heuristic	−3% edge case failures

ALGO-33 (FGHF) verifies every factual claim against the knowledge base before output reaches the user. Combined with ALGO-05 (GraphRAG) and ALGO-30 (ABV), it achieves a hallucination rate of < 0.3% — a 97% reduction from the industry baseline.

---

Tier 9 — Autonomy Layer

The autonomous generation pipeline. See the full showcase: Tier 9 — Autonomous Pipeline.

ALGO	Name	Score	Capability
ALGO-37	Planning Depth Expander	10.0	15-level hierarchical planning (+400% depth)
ALGO-38	CSP Solver	10.0	Constraint satisfaction via AC-3 + Backtracking
ALGO-39	Genesis Engine	9.5	Auto-instantiates new modules in ~2 seconds
ALGO-40	ITSR	9.8	Tech stack recommendation, 98.7% accuracy
ALGO-41	The Crystallizer	9.6	Golden keyword extraction via entropy scoring

These five algorithms form an end-to-end autonomous pipeline: plain text in → production-ready module out, in approximately 2 seconds.

---

Summary Table

ID	Short Name	Full Name	Score	Tier
ALGO-01	FDIA	Fundamental Decision Intelligence Algorithm	9.2	1
ALGO-02	MOIP	Multi-Objective Intent Planning	9.1	1
ALGO-03	Delta Engine	Gap Analysis & Compression Engine	8.7	1
ALGO-04	RCT-7	7-Step Cognitive Process Framework	9.0	2
ALGO-05	GraphRAG	Graph-Based Retrieval Augmented Generation	8.5	2
ALGO-06	JITNA	Just In Time Nodal Assembly Protocol	8.3	2
ALGO-07	MEE v2	Meta-Evolution Engine v2	9.85	3
ALGO-08	Self-Evolving	Autonomous Self-Improvement System	9.85	3
ALGO-09	Reflexion+	Enhanced Self-Reflection Agent	8.6	3
ALGO-10	Delta Memory	Incremental Memory Storage	8.0	3
ALGO-11	BBA→P→CF	Belief → Plan → Consequence Flow	9.3	3
ALGO-12	Meta-Algorithm Γ	Meta-Algorithm Generator	8.9	4
ALGO-13	RCT-Diffusion	Diffusion-Based Generative Algorithm	9.6	4
ALGO-14	GraphRAG Complete	Full-Stack Graph + Vector Hybrid	—	4
ALGO-15	HRM	Human-in-the-loop Reasoning Manager	9.0	4
ALGO-16	Vector Search	HNSW Semantic Search	7.0	4
ALGO-17	Graph Traversal	BFS / DFS / PageRank Navigation	7.2	5
ALGO-18	Adaptive Prompting	Dynamic Prompt Adaptation Engine	7.8	5
ALGO-19	Data Fusion v2	Multi-Source Fusion with Conflict Resolution	9.0	5
ALGO-20	Workflow Orchestrator v2	DAG-Based Async Workflow Engine	9.2	5
ALGO-21	Fast/Slow Router	Dual-Lane Routing by Complexity	8.8	5
ALGO-22	Halting Detection	Deadlock & Loop Detection	8.5	5
ALGO-23	Content-Box	Structured Output Rendering Engine	8.2	6
ALGO-24	Benchmark Suite	Algorithm Performance Testing Framework	7.6	6
ALGO-25	Delta Block	Incremental State Storage Unit	7.7	6
ALGO-26	Intent Classification	Intent Conservation & Tracking	9.4	6
ALGO-27	TVRA	Time-aware Video Reasoning & Analysis	—	7
ALGO-28	CIO	Concurrent I/O Optimizer	—	7
ALGO-29	UIA	Universal Integration Adapter	—	7
ALGO-30	ABV	Adaptive Bayesian Validator	—	7
ALGO-31	ALBAS	Auto-Load Balancing & Auto-Scaling	—	7
ALGO-32	MCTR	Multi-Chain Thought Reasoning	—	8
ALGO-33	FGHF	Fact-Grounded Hallucination Filter	—	8
ALGO-34	SWCAR	Semantic Web & Content Auto-Repair	—	8
ALGO-35	ATC	Adaptive Timeout Controller	—	8
ALGO-36	RFLH	Rare-Failure Learning Heuristic	—	8
ALGO-37	Planning Expander	15-Step Hierarchical Planning Expander	10.0	9
ALGO-38	CSP Solver	Constraint Satisfaction Solver (AC-3)	10.0	9
ALGO-39	Genesis Engine	Auto-Instantiation Protocol	9.5	9
ALGO-40	ITSR	Intent-Driven Tech Stack Recommender	9.8	9
ALGO-41	The Crystallizer	Golden Keyword Extraction Engine	9.6	9

---

Key System-Wide Metrics

Metric	Value	Source Algorithm
Hallucination rate	< 0.3% (industry: 12–15%)	ALGO-33 FGHF
Memory compression	74%	ALGO-03 Delta Engine
Intent recall (warm)	< 50ms	ALGO-10 Delta Memory
Confidence calibration	95%	ALGO-30 ABV
Reasoning accuracy	96.0%	ALGO-04 RCT-7
Routing accuracy	95.8%	ALGO-21 Fast/Slow Router

---

Full architecture specifications: RFC-001 through RFC-006
Autonomous generation pipeline: Tier 9 — Autonomous Pipeline