ADR map
Every site claim → the architecture decision behind it.
The site contains inline per ADR-XXX chips that point to specific architecture decisions. This page lists each cited ADR with title, status, summary, the source file path, and the pages that reference it. The Research Foundations bibliography then links each ADR's design choices to the underlying research.
How to read an ADR citation
An ADR records one architectural decision: title, status, context, decision, consequences. The status is the truth-state of the decision: Active or Accepted means the decision is in force; Proposed means under review and the design is documented but not yet final; Draft is in progress. A claim on the site that points to an Accepted ADR describes built or near-built behaviour. A claim pointing to a Proposed ADR describes documented architecture not yet fully implemented.
The chain is intentional: site claim → ADR title → ADR file → research bibliography. Anyone can verify any architectural assertion in three clicks.
Cited ADRs (alphabetical by number)
Establishes the local-first architecture with optional cloud services. CODITECT runs on a local AI workstation today; managed cloud workstations and hybrid lanes follow the same primitives. The deployment lane is a configuration choice, not a separate codebase.
Defines the unified query interface across the four-tier database. Supports keyword and semantic search across messages, decisions, patterns, error solutions, and tokens. The basis for CODITECT's per-row embedding and RAG-style recall over the audit trail.
Defines how local SQLite stores synchronise to the multi-tenant cloud Postgres SSOT, with conflict resolution, change deduplication, and tenant isolation. The continuous sync layer that keeps every operator's local state in step with the source-of-truth.
Defines a unified, portable component architecture (agents, skills, commands, hooks, scripts) that runs across Claude, Codex, and Gemini with a single source of truth. Foundation for foundation-model-agnostic execution: write the component once, route per task to the model best suited.
Adopts a SaMD-grade validated testing framework using IEC 62304 software lifecycle, GAMP 5 Category 5 validation, and 21 CFR Part 11 electronic records as the structural foundation. The CODITECT Quality Validation System (QVS) maps to NIST AI RMF, ISO 42001, ISO 27001, NAIC, NYDFS Part 500, HIPAA, ISO 13485, and SOC 2. The control register on /compliance is generated from this ADR.
Establishes a strict type-based data model where all CODITECT components are DB-native with schema-per-type validation, deterministic generation, and immutable audit logging. The schema common to local SQLite stores and the cloud Postgres SSOT.
The platform's core engine. A nine-phase loop (DISCOVER, ANALYZE, PLAN, SPECIFY, RESEARCH, IMPLEMENT, REMEDIATE, DEPLOY, OBSERVE) that runs in self-mode (CODITECT applied to itself) and target-mode (CODITECT applied to any customer codebase). Standards-complete: ITIL, ISO/IEC 25010, 5338, 42001, 27001, NIST AI RMF, IEC 62304, ISO 13485, ISO 20000, SOC 2 CC8.1, CMMI. The full description lives at UQDH.
Establishes incidents.db as the system of record for ITIL incident management with cloud sync to my.coditect.ai for system administrator access. The basis for the ITIL Issue Management surface and for the automated incident creation when CI hooks detect a violation.
From ADR to research
Every cited ADR's design choices are grounded in published research, regulatory frameworks, or established engineering patterns. The link from each architecture decision to its evidence base lives in Research Foundations - a Chicago-format bibliography organised by the same architectural areas as the ADRs.
Adding a new ADR to the cited set is straightforward: tag the relevant claim on the page with per ADR-XXX, link the chip here to a new card, and reference any new sources in the bibliography. The chain stays unbroken.