Documentation · Architecture Deep Dive Series

[contexa73]

Documentation · Architecture Deep Dive Series

Labels: documentation
Assignee: @contexa73
Type: Documentation expansion

Background

The current architecture documentation at https://docs.ctxa.ai/docs/reference/architecture/overview.html provides a high-level orientation. To help integrators, security reviewers, and academic readers understand the engine more deeply, we want a structured Architecture Deep Dive series.

Planned articles

#	Title	Estimated length	Priority
1	The 9-dimension behavioral signal model — weights, gates, and baseline learning (HCAD)	3000 words	High
2	The Canonical Security Context — 22 fields, 6 profiles, 17 prompt sections	4000 words	High
3	Tiered LLM analysis — Layer 1 vs Layer 2, escalation criteria, model selection	2500 words	High
4	The 5 dynamic actions — TTL, HTTP status, authority swap mechanics	2000 words	Medium
5	Sealed evidence packages — SHA-256 integrity, 8 JSON sections, replay path	3000 words	High
6	Prompt Quality Assurance — 8-stage workflow, 5-category 15-metric catalog	4000 words	High
7	Closed-loop remediation — finding, fix, reverification comparison	2500 words	Medium
8	MCP delegation — 31-claim envelope, 8-guard chain, action approval plane	3000 words	Medium

Style and audience

• Primary audience: senior security engineers and security architects evaluating Contexa for adoption
• Secondary audience: academic readers studying AI-driven decision governance
• Tone: declarative, no narrative storytelling (per documentation policy)
• Diagrams: Mermaid preferred, hand-drawn architecture acceptable for complex flows

Acceptance for each article

• At least one Mermaid diagram per article
• Cross-links to source files (contexa-core/..., contexa-iam-enterprise/...) where applicable
• Concrete code excerpts (≤30 lines each) at least 3 times per article
• Reviewed by @contexa73 before publication

How to contribute

If you want to co-author an article, comment with the article number you want to take and your background. Authors retain attribution in the published article.

Reference

• Existing overview: https://docs.ctxa.ai/docs/reference/architecture/overview.html
• HCAD: https://docs.ctxa.ai/docs/reference/architecture/hcad-early-detection.html
• Zero Trust flow: https://docs.ctxa.ai/docs/reference/architecture/zero-trust-flow.html
• Prompt engineering: https://docs.ctxa.ai/docs/reference/architecture/prompt-engineering.html