Demos & Benchmarks

36 demo/benchmark pairs covering AI, agents, distributed systems, security, and more. Each feature has a -demo command (interactive demonstration) and a -bench command (performance benchmark).

Running Demos

tri <name>-demo                         # Run interactive demo
tri <name>-bench                        # Run benchmark

Needle Check (Koschei's Immortality Test)

Every benchmark uses the Needle Check — a quality gate based on the golden ratio:

$\text{threshold} = \phi^{-1} = 0.6180339887...$

The benchmark computes an improvement rate and compares it to the threshold:

Status	Condition	Message
PASS (Immortal)	rate $> 0.618$	KOSCHEI BESSMERTEN! Igla ostra. (Koschei is immortal! Needle is sharp.)
WARN (Mortal)	$0 <$ rate $< 0.618$	Uluchshenie est', no Igla tupitsya. (Improvement exists, but needle dulls.)
FAIL (Regression)	rate $\leq 0$	REGRESSIYA! Igla slomana. (Regression! Needle is broken.)

Improvement Rate Formula

Each benchmark calculates the improvement rate as a weighted composite of scenario metrics:

improvement_rate = (metric_1 + metric_2 + baseline_contribution) / 2.0

Where metrics are scenario-specific (hit rate, similarity score, agent count, throughput, etc.) and normalized to the $[0, 1]$ range.

Example Benchmark Output

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Scenario 1: Basic query
    Hit rate:       0.85
    Similarity:     0.92
    Improvement:    0.887

  Scenario 2: Complex reasoning
    Hit rate:       0.72
    Similarity:     0.88
    Improvement:    0.800

  Needle Check (phi^-1 = 0.618):
    Average rate:   0.843
    Status:         KOSCHEI IS IMMORTAL! Needle is sharp.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Pre-Cycle Demos

Early features implemented before the numbered cycle system.

Command	Aliases	Description
tri tvc-demo	tvc	TVC Distributed Chat — 10,000-entry ternary vector corpus
tri tvc-stats	—	TVC corpus configuration and status
tri agents-demo	agents	Multi-Agent Coordination — Coordinator, Coder, Chat, Reasoner, Researcher
tri agents-bench	—	Multi-Agent System Benchmark — 10 task scenarios
tri context-demo	context	Long Context Engine — sliding window (20 messages) + summarization
tri context-bench	—	Long Context Benchmark — 24-turn conversation simulation
tri rag-demo	rag	RAG — Query → Embed → Retrieve → Augment → Generate
tri rag-bench	—	RAG Retrieval Benchmark — similarity scoring
tri voice-demo	voice, mic	Voice I/O (TTS + STT) integration
tri voice-bench	mic-bench	Voice I/O performance benchmark
tri sandbox-demo	sandbox	Code Execution Sandbox — safe execution with 5s timeout
tri sandbox-bench	—	Sandbox Benchmark — Zig, Python, JS, Shell execution
tri stream-demo	stream, pipeline	Streaming Output — token-by-token with 256-token buffer
tri stream-bench	pipeline-bench	Streaming Benchmark — char/token/chunk/SSE modes
tri finetune-demo	finetune	Fine-Tuning Engine — local model adaptation
tri finetune-bench	—	Fine-Tuning Benchmark — learning rate convergence
tri batched-demo	batched	Batched Work-Stealing — parallel batch scheduler
tri batched-bench	—	Batched Stealing Benchmark — throughput/latency
tri priority-demo	priority	Priority Queue — task scheduling
tri priority-bench	—	Priority Queue Benchmark
tri deadline-demo	deadline	Deadline Scheduling — SLA enforcement
tri deadline-bench	—	Deadline Scheduling Benchmark

Cycle 20: Vision

Command	Aliases	Description
tri vision-demo	vision, eye	Local Vision — image → ternary embedding → scene detection → caption
tri vision-bench	eye-bench	Vision Benchmark — 80 COCO semantic categories

Cycle 26: Multi-Modal

Command	Aliases	Description
tri multimodal-demo	multimodal, mm	Multi-Modal Unified Engine — text + vision + voice + code → VSA
tri multimodal-bench	mm-bench	Multi-Modal Benchmark — cross-modal fusion

Cycle 27: Tool Use

Command	Aliases	Description
tri tooluse-demo	tooluse, tools	Multi-Modal Tool Use — function calling with multi-modal inputs
tri tooluse-bench	tools-bench	Tool Use Benchmark — invocation and result handling

Cycle 30-33: Unified Agents

Command	Aliases	Cycle	Description
tri unified-demo	unified, agent	30	Unified Multi-Modal Agent — all modalities in single agent
tri unified-bench	agent-bench	30	Unified Agent Benchmark
tri autonomous-demo	auto, autonomous	31	Autonomous Agent — self-directed with goal planning
tri autonomous-bench	autonomous-bench	31	Autonomous Agent Benchmark
tri orchestration-demo	orch, orchestrate	32	Multi-Agent Orchestration — task distribution
tri orchestration-bench	orchestrate-bench	32	Orchestration Benchmark
tri mm-orch-demo	mmo, mm-orch	33	MM Multi-Agent Orchestration — multi-modal coordination
tri mm-orch-bench	mm-orch-bench	33	MM Orchestration Benchmark

Cycle 34-37: Memory & Distribution

Command	Aliases	Cycle	Description
tri memory-demo	memory, mem	34	Agent Memory & Cross-Modal Learning
tri memory-bench	mem-bench	34	Memory Benchmark
tri persist-demo	persist, save	35	Persistent Memory & Disk Serialization
tri persist-bench	save-bench	35	Persistence Benchmark — I/O performance
tri spawn-demo	spawn, pool	36	Dynamic Agent Spawning & Load Balancing
tri spawn-bench	pool-bench	36	Spawn Benchmark — spawn latency
tri cluster-demo	cluster, nodes	37	Distributed Multi-Node Agents
tri cluster-bench	nodes-bench	37	Cluster Benchmark — network latency/throughput

Cycle 39-45: Scheduling & Plugins

Command	Aliases	Cycle	Description
tri worksteal-demo	worksteal, steal	39	Adaptive Work-Stealing Scheduler
tri worksteal-bench	steal-bench	39	Work-Stealing Benchmark
tri plugin-demo	plugin, ext	40	Plugin & Extension System
tri plugin-bench	ext-bench	40	Plugin Benchmark
tri comms-demo	comms, msg	41	Agent Communication Protocol
tri comms-bench	msg-bench	41	Communication Benchmark — message throughput
tri observe-demo	observe, otel	42	Observability & Tracing System
tri observe-bench	otel-bench	42	Observability Benchmark — tracing overhead
tri consensus-demo	consensus, raft	43	Consensus & Coordination Protocol (Byzantine-tolerant)
tri consensus-bench	raft-bench	43	Consensus Benchmark
tri specexec-demo	specexec, spec	44	Speculative Execution Engine
tri specexec-bench	spec-bench	44	Speculative Execution Benchmark
tri governor-demo	governor, gov	45	Adaptive Resource Governor
tri governor-bench	gov-bench	45	Resource Governor Benchmark

Cycle 46-52: Advanced Systems

Command	Aliases	Cycle	Description
tri fedlearn-demo	fedlearn, fl	46	Federated Learning Protocol
tri fedlearn-bench	fl-bench	46	Federated Learning Benchmark — convergence
tri eventsrc-demo	eventsrc, es	47	Event Sourcing & CQRS Engine
tri eventsrc-bench	es-bench	47	Event Sourcing Benchmark — event throughput
tri capsec-demo	capsec, sec	48	Capability-Based Security Model
tri capsec-bench	sec-bench	48	Security Benchmark — authorization overhead
tri dtxn-demo	dtxn, txn	49	Distributed Transaction Coordinator (ACID)
tri dtxn-bench	txn-bench	49	Transaction Benchmark — throughput
tri cache-demo	cache, memo	50	Adaptive Caching & Memoization
tri cache-bench	memo-bench	50	Cache Benchmark — hit rates
tri contract-demo	contract, sla	51	Contract-Based Agent Negotiation
tri contract-bench	sla-bench	51	Contract Benchmark — negotiation speed
tri workflow-demo	workflow, wf	52	Temporal Workflow Engine
tri workflow-bench	wf-bench	52	Workflow Benchmark — latency

Summary

Category	Pairs	Cycles
Pre-cycle (AI, scheduling)	11	—
Vision	1	20
Multi-Modal	1	26
Tool Use	1	27
Unified Agents	4	30-33
Memory & Distribution	4	34-37
Scheduling & Plugins	7	39-45
Advanced Systems	7	46-52
Total	36 pairs	72 commands

See Also

• Pipeline — Golden Chain development cycle (where demos are verified)
• TVC Learning — TVC corpus architecture (used in tvc-demo)
• Sacred Constants — The $\phi^{-1} = 0.618$ threshold explained