Corpus Overview
This one-shot Safari Reading List build preserves 112 sources: 102 complete extraction records and 10 explicit insufficient_source records. The navigable layer now has 34 synthesis pages. 99 complete records contribute to a substantive page; 3 remain intentionally source-only, with rationales in _meta/source-routing.json. Counts, exact source keys, raw-body hashes, extraction provenance, and quality flags live in _meta/; this layer does not invent claims for unavailable captures.
Reading map
Agent systems, software, and AI safety
- Agent Harness Design — the control system around a model
- Context And Tool Orchestration — context as a bounded operational resource
- Long Running Agent Systems — handoffs, evaluation, and coordination at duration
- Software Quality With Agents — deterministic checks, judgment, and recovery
- Agent Security And Human Oversight — authority boundaries and review
- Ai Safety And Governance — normative and organizational safety concerns
- Ai Enabled Product Prototyping — turning ambiguity into testable interactions
- Spec Driven Development — executable documentation as a shared interface
- Version Control And Agent Workflows — workspaces, reversibility, and handoffs
- Large Codebase Reasoning — accountable progress under incomplete models
- Deep Learning Foundations — code-first reconstruction of model mechanisms
- Agent Control Patterns — a side-by-side comparison of harness choices
Systems, infrastructure, privacy, and media
- Distributed Systems Correctness
- Incident Debugging And Observability
- Database Performance And Scaling
- Web And Cloud Infrastructure
- Privacy And Security Boundaries
- Media Technology And Open Standards
Photography
- Photographic Exposure And Sharpness
- Photographic Composition And Practice
- Photographic Lighting And Studio Practice
- Camera Systems And Gear Choice
Markets, organizations, and professional practice
Recurring entities
Provenance boundary
Every factual synthesis claim points to immutable raw article records; source-specific caveats remain attached to extraction records and are reflected in page confidence and contested status. The corpus intentionally mixes source types—practitioner accounts, technical analyses, vendor material, journalism, and personal essays—without treating them as equivalent evidence. [src] [src] [src] [src]
For article-level provenance and the deliberate route for every complete record, see Source Coverage.