When someone claims their AI platform is reliable, you should ask one question: How do they know?
Most AI platforms answer vaguely. They point to uptime percentages, response time dashboards, or qualitative assurances. Few can show you a complete, auditable record of every behavioral verification that has been run against every component of their system.
RCT Labs can: 4,849 automated tests, 0 failures, 0 errors — verified on v5.4.5 (March 21, 2026).
This article explains what those 4,849 tests cover, how the 8-level test pyramid is structured, and why this testing discipline is not just a quality metric — it is the mechanism that makes the platform's performance claims verifiable.
The 8-Level Test Pyramid
The RCT Ecosystem uses an 8-level pyramid that progresses from unit isolation to mathematical property verification:
Level 1: Unit Tests — 1,343 tests
Unit tests verify individual functions in complete isolation. Each function is tested with:
- Happy path inputs — expected inputs that produce expected outputs
- Edge cases — boundary conditions (empty arrays, zero values, max values)
- Adversarial inputs — malformed data, injection attempts, type mismatches
- Performance bounds — each unit test has a maximum execution time
Every algorithm in the 41-algorithm library has its own unit test suite. Combined with FDIA scoring units, Delta Engine compression units, and JITNA packet handling units, Level 1 alone covers 1,343 cases.
Level 2: Integration Tests — 34 tests
Integration tests verify that components work correctly when combined. Key integration suites:
- FDIA + JITNA: Does a JITNA packet correctly trigger FDIA validation?
- Delta Engine + RCTDB: Does a delta write correctly reconstruct to full state on read?
- HexaCore + SignedAI: Does the consensus system correctly aggregate 7-model outputs?
- Intent Loop + Memory: Does warm recall correctly bypass LLM computation?
Level 3: Service Tests — 1,889 tests (62 services × ~30 tests each)
Each of the 62 microservices has its own service test suite. Service tests verify:
- Input validation and rejection
- Output format compliance
- Error handling and circuit breaker activation
- Timeout behavior
- Health check endpoint correctness
Level 4: Contract Tests
Contract tests verify that the API contract between any two services matches both the producer's implementation and the consumer's expectation. If Service A changes its response format, contract tests catch the breaking change before it reaches integration testing.
Level 5: Performance Tests
Performance tests measure latency and throughput against defined SLAs:
- Cold start: must complete in <5 seconds
- Warm recall: must complete in <50ms
- Throughput: minimum 18 req/s aggregate across HexaCore models
- Memory: Delta Engine must not grow unboundedly under sustained load
Level 6: Security Tests (OWASP A01–A10)
Security tests cover the OWASP Top 10 AI security risks:
- A01 Prompt Injection: JITNA validates all input through the Normalizer before LLM contact
- A02 Insecure Output Handling: All outputs are SignedAI-verified before return
- A03 Training Data Poisoning: Not applicable (no fine-tuning in production)
- A04 Model Denial of Service: Circuit breakers protect all model endpoints
- A05–A10: Access control, supply chain, data disclosure, and integrity tests
Level 7: Chaos Tests
Chaos tests deliberately introduce failures to verify recovery:
- Circuit breaker tests: Does the system correctly route away from a failed model?
- Network partition tests: Does the JITNA protocol handle dropped packets correctly?
- Memory pressure tests: Does the Delta Engine correctly evict hot-zone entries under pressure?
- Cascade failure tests: Does one failing microservice correctly isolate without affecting others?
Level 8: Property Tests (Mathematical)
Property tests verify mathematical invariants using Hypothesis-style automatic input generation:
- FDIA Constitutional: For all inputs where A=0, F must equal 0. Tested over 10,000 random A/D/I combinations.
- Delta Losslessness: For all delta chains, full state reconstruction must exactly equal original full state. Tested over 10,000 random delta sequences.
- Determinism: SHA-256 output check — 10 identical runs of the same query must produce identical results.
Why This Matters for Enterprise Buyers
Verifiable Claims
When RCT Labs claims 0.3% hallucination rate or 99.98% uptime, these claims are backed by a complete, auditable test suite. An enterprise buyer can ask "show me the test" — and the answer exists.
Most AI vendors cannot do this. They offer benchmarks run on cherry-picked datasets, not continuous verification of their production systems.
Regression Prevention
4,849 tests running on every code change means that a change to the FDIA scoring algorithm cannot accidentally break JITNA packet validation without immediate detection. This is how a solo developer can maintain a 62-microservice platform with zero production failures.
Compliance Evidence
For regulated industries (healthcare, finance, legal), a complete test suite with documented results is often required for vendor evaluation. The 4,849-test suite, structured across 8 levels with OWASP security coverage, is compliance evidence that arrives ready-made.
The Test Coverage Breakdown
| Level | Count | Coverage | |---|---|---| | Unit | 1,343 | 41 algorithms + core components | | Integration | 34 | 17 component pairs | | Service | 1,889 | 62 microservices × ~30 each | | Contract | ~200 | Service API contracts | | Performance | ~500 | Latency and throughput SLAs | | Security | OWASP A01-A10 | All 10 categories | | Chaos | ~300 | Circuit breaker, partition, cascade | | Property | 10,000+ | Mathematical invariants (random input) | | Total | 4,849 | 0 failures, 0 errors |
Summary
4,849 tests with 0 failures is not a vanity metric. It is the mechanism that makes every performance claim verifiable:
- 0.3% hallucination: Property tests verify FDIA gating prevents hallucination-prone outputs
- <50ms warm recall: Performance tests verify Delta Engine latency
- 99.98% uptime: Chaos tests verify circuit breaker recovery
- Constitutional AI guarantees: Property tests verify
A=0 → F=0over 10,000 random inputs
If you cannot show the test, you cannot make the claim. RCT Labs can show the test.
This article was written by Ittirit Saengow, founder and sole developer of RCT Labs.
What enterprise teams should retain from this briefing
The RCT Ecosystem runs 4,849 automated tests — and passes all of them, with 0 failures and 0 errors. This article explains the 8-level test pyramid, the 62-microservice verification strategy, and why this testing discipline is a direct SEO and enterprise trust signal.
Move from knowledge into platform evaluation
Each research article should connect to a solution page, an authority page, and a conversion path so discovery turns into real evaluation.
Previous Post
PDPA and AI Compliance in Thailand: A 2026 Enterprise Guide
Thailand's PDPA (Personal Data Protection Act) imposes strict requirements on AI systems that process personal data. This guide explains the key obligations, common compliance gaps, and how a Constitutional AI framework like RCT Labs addresses PDPA requirements architecturally.
Next Post
RCTDB v2.0: The 8-Dimensional Universal Memory Schema for AI Systems
RCTDB is the universal memory architecture of the RCT Ecosystem — an 8-dimensional schema designed for structured AI memory, full provenance tracking, and PDPA-compliant right-to-erasure. This article explains the schema, three storage zones, and why traditional vector databases fall short for enterprise AI.
Ittirit Saengow
Primary authorIttirit Saengow (อิทธิฤทธิ์ แซ่โง้ว) is the founder, sole developer, and primary author of RCT Labs — a constitutional AI operating system platform built independently from architecture through publication. He conceived and developed the FDIA equation (F = (D^I) × A), the JITNA protocol specification (RFC-001), the 10-layer architecture, the 7-Genome system, and the RCT-7 process framework. The full platform — including bilingual infrastructure, enterprise SEO systems, 62 microservices, 41 production algorithms, and all published research — was built as a solo project in Bangkok, Thailand.