Four Core Principles

Implemented Technical Practices

• ✓Methodology page (10 languages) — complete documentation of knowledge sources, chart engine, and quality workflow
• ✓System Card (Model Card) — the applicable/non-applicable scope and known limits of Starnum Logic Engine v5.0
• ✓Quality benchmark — evaluation results on a fixed 37-chart test set, re-run periodically for public comparison
• ✓CHANGELOG — detailed record of each system update, following the Keep a Changelog format
• ✓Explicit school-of-thought declaration — the entire site uses the Lu Binzhao (Zhongzhou) school; no mixing, clearly marked on the model card
• ✓Explicit limitation statement — forbidden domains (death prediction, medical diagnosis, political figures) are published on the acceptable-use page
• ✓Claim-Evidence system — every metaphysics claim in every article is tagged to a knowledge-base source; no assertions out of thin air
• ✓Security disclosure policy — RFC 9116 standard, with public 30/60/90-day tiered remediation timelines

Implemented Technical Practices

• ✓SHA256 content-integrity hash — each article embeds a SHA256 value injected into the JSON-LD hasHash field, so any content change is detectable
• ✓iztro open-source chart engine — the chart calculation logic is fully open source; any researcher can independently verify chart results
• ✓Machine-readable JSON-LD Schema — every article's Schema is validated by automation and conforms to schema.org
• ✓Public benchmark set (eval-set) — reference analyses for 37 fixed test charts, reproducible and verifiable
• ✓Git version history — Git version control maintains complete change history, fully traceable
• ✓Automated validation of 117 metaphysics hard rules — enforced before publication; any violation blocks the release
• ✓Multi-model cross-validation — metaphysics logic reviewed in parallel by 4 AI tools; any anomaly flags the article for re-review
• ✓Dependency vulnerability scanning — local tools periodically scan npm dependencies to ensure third-party package security

Implemented Technical Practices

• ✓Lu Binzhao knowledge base — 2,758 files / 1,537,711+ lines of structured metaphysics text, cross-referenced across 8 sources to build a reliable literature basis
• ✓School consistency statement — the entire site uses the Zhongzhou (Lu Binzhao) Four-Transformation table; no mixing schools for layout convenience
• ✓Classical source citations — major claims are explicitly attributed to named works such as the Zi Wei Dou Shu Quan Shu series
• ✓multi-provider governance audit — Anthropic / OpenAI / Google Gemini governance benchmarks review metaphysics logic in parallel; fixes are decided by vote
• ✓117 metaphysics hard rules — core rules such as impossible Four-Transformation events and pattern-judgment conditions are machine-enforced
• ✓GraphRAG knowledge graph — 343 entity nodes / 679 relationships, anchoring each claim in a consistent structural position
• ✓Author credentials — metaphysics research background and rationale for the chosen knowledge system are published on the about page
• ✓22 standard analysis blocks — each chart reading follows a fixed structure, ensuring completeness and consistency

Implemented Technical Practices

• ✓Automated quality workflows — quality checks, sitemap updates, credential-expiry monitoring, and more, all fully automated
• ✓Rule lifecycle management — rules untouched for 90 days are auto-proposed for retirement, preventing unbounded rule-base growth
• ✓Failure-driven evolution — when a class of quality issue occurs 3 times, a new rule is auto-proposed; mistakes never repeat
• ✓Six-layer quality closed loop (L1–L6) — automated defenses from article output up to strategy, each layer reducing manual intervention
• ✓Content freshness tracking — stale articles are auto-detected and queued for refresh, maintaining long-term validity
• ✓Resource-cap protection — GPU lockfile prevents out-of-memory failures, ensuring stable long-running operation
• ✓Credential-expiry tracking — API keys and service credentials trigger alerts before they expire, avoiding outages
• ✓Low-compute design — Cloudflare Pages static hosting plus Workers edge compute, operating costs stay controlled

How We Resolve Principle Conflicts

When the four principles conflict, we prioritize in this order:

Resolution Examples

• P4 vs P3: An automation feature boosts efficiency but might introduce metaphysics-logic errors → P3 wins, feature does not ship until logic is verified
• P1 vs P4: Publishing certain internal data improves transparency but adds maintenance cost → P1 wins, we absorb the cost
• P2 vs P4: Adding a validation mechanism slows the system down → P2 wins, validation is non-skippable

Feature × Principle Matrix

Every core feature must map to at least 2 principles; otherwise its necessity is re-evaluated.

Related Transparency Resources

These resources show how the four principles are implemented at every layer: