Content-oriented index of all wiki topic pages, organized by domain. Updated automatically on every ingest.
Software Engineering
- Hyrum’s Law — with enough API users, all observable behaviors become implicit contracts regardless of the documented spec; the engineering equivalent of entropy
- Psychological safety — the shared belief that the team is safe for interpersonal risk-taking; the single most important factor in effective team performance (Google Project Aristotle)
- Trunk-based development — single-branch version control strategy where all commits go to trunk continuously; long-lived branches are an anti-pattern; key DORA predictor of high-performing engineering organizations
- Software Engineering — collected principles, practices, and insights on building software well; includes SE@Google and other canonical sources
- Software Architecture — shape given to a system through component division and arrangement; four dimensions: style, characteristics, components, decisions; three laws (trade-offs, why > how, spectrum)
- Software System — notes on system design patterns, boundaries, and composition strategies
- Software Architect — role, responsibilities, and skills of the software architect; eight expectations from Richards & Ford
- SOLID — five object-oriented design principles (SRP, OCP, LSP, ISP, DIP) for arranging functions and data structures into cohesive, decoupled units
- DDD — Domain-Driven Design: ubiquitous language, bounded contexts, aggregates, and entities as tools for aligning code with the business domain
- Microservices — architectural style decomposing applications into small, independently deployable services; each does one thing and communicates over the network
- Git — distributed version control; branching strategies, workflows, and internals
- Terraform — infrastructure-as-code tool for provisioning cloud resources declaratively; state management and module patterns
- Modularity — logical grouping of related code as an organizing principle; measured via cohesion (7 types), coupling (afferent/efferent, instability, main sequence), and connascence
- Coupling — degree to which components depend on one another; afferent (incoming) and efferent (outgoing) coupling; abstractness, instability, distance from main sequence metrics
- Connascence — precise vocabulary for coupling types in OO systems; static (Name, Type, Meaning, Position, Algorithm) and dynamic (Execution, Timing, Values, Identity); strength, locality, degree properties
- Architectural characteristics — capabilities a system must support independent of the domain (the “-ilities”); three criteria, four categories (operational, structural, cloud, cross-cutting); least-worst-architecture principle
- Architectural thinking — seeing systems from an architect’s perspective; architecture vs design spectrum, trade-off analysis, Frozen Caveman antipattern, balancing hands-on coding
- Component cohesion — cohesion measurement (REP, CCP, CRP principles) and seven cohesion types; LCOM metric; coupling vs cohesion relationship
- Fitness functions — any mechanism providing objective integrity assessment of an architectural characteristic; automated/triggered checks wired into CI (JDepend, ArchUnit, NetArchTest, chaos engineering); the enforcement arm of evolutionary architecture
- — ensuring architectural decisions are respected over time via automated fitness functions rather than manual reviews; ArchUnit, NetArchTest, chaos engineering; checklist manifesto framing
- Architecture katas — structured practice exercises for deriving architectural characteristics from domain descriptions; Silicon Sandwiches and Going-Going-Gone worked examples; domain-concern translation table
- — programmatic browser control (Playwright, Puppeteer, Selenium) for AI agents; Playwright codegen, OCR pipeline (PDF→PNG→vision LLM→JSON), human-in-the-loop approval flows, secret management, and internal API discovery
Security
- — attack class where adversaries embed malicious instructions in content processed by LLMs; exploits role confusion to bypass instruction hierarchy defenses; includes CoT Forgery (60% ASR against frontier models)
- — privilege separation system for LLMs using role tags (system/user/tool); fails in practice because models perceive roles by text style rather than structural tags
- — representational failure where LLMs identify “who is speaking” by text style rather than role tags; root cause of prompt injection; measurable via role probes in latent space
- Zero Trust — security model that eliminates implicit network trust; every request verified regardless of origin; replaces perimeter-based defenses
- Cloud Security — network, identity, and data protection patterns specific to cloud environments
- XSS — Cross-Site Scripting: attacker injects scripts into trusted pages that execute in victims’ browsers; impact, types, and prevention techniques
- OAuth — authorization framework delegating resource access via tokens without sharing credentials; OAuth 2.0 flows and RFCs
- Scaling Static Analyses at Facebook — lessons from deploying static analysis at scale: integrate into CI, fix over time, trust signal matters more than volume
- Lessons from building static analysis tools at Google — practical learnings on false positives, developer trust, and incremental rollout from Google’s static analysis program
AI / Machine Learning
- AI — collected resources and notes on artificial intelligence, LLMs, and AI tooling
- — the discipline of building AI systems from mathematical first principles; math-first methodology (derive → code → test → keep artifact); open-source curricula (503-lesson, 20-phase); covers backprop, tokenizers, attention, agent loops
- LLM wiki — Karpathy’s pattern for an LLM-maintained knowledge base: immutable raw sources, LLM-generated wiki pages, and a schema file; ingest/query/lint operations; stateful alternative to RAG at personal/small-team scale
- Claude Code — Anthropic’s AI coding CLI; .claude/ as version-controlled team infrastructure (CLAUDE.md conventions, shared slash commands, settings hooks); home of the Planner-Generator-Evaluator pattern
- Planner-Generator-Evaluator pattern — agentic design pattern decomposing LLM coding workflows into Planner (brief→spec), Generator (sprint-by-sprint implementation), and Evaluator (live-app grading against hard thresholds)
- Claude Code skills — community-shared reusable prompt definitions stored as slash commands; dominant frameworks: GSD (large iterative projects) and Superpowers (small-medium well-defined); also covers multi-agent orchestration (tonone), credential management, and skill design patterns
- Self-improving agents — AI agents that observe their own sessions, log improvement opportunities, and apply those improvements on a schedule; the “meta-skill” pattern; 600+ improvements applied across ~40 skills in the wild
- Session context persistence — pattern for systematically saving decisions, state, and context at the end of each AI session so the next session can resume coherently; the /close skill pattern; memory file taxonomy
- — file-based NDJSON channels for inter-agent messaging between Claude Code, Codex, and OpenCode; kqueue/inotify wake-up, zero idle CPU, cross-harness interop
- — reusable structural patterns for cheap-model pipelines: falsification-first prompting, crux extraction, confidence-based escalation, layered output parsing, and observability metrics
- — hypothesis-before-tool protocol, tool budget enforcement, Unix tool to reasoning gap mapping, security sandbox for model-generated patterns
- — prior context condensation, memoized cache visibility, and structured section injection for multi-round LLM pipelines
- AI coding assistants — landscape of LLM-powered coding tools (Cursor, Windsurf/Cascade, Devin, Manus, Lovable, Replit, VSCode Agent, Augment Code, Kiro, Trae); shared architectural patterns (persona injection, tool minimalism, confidentiality clauses); model-identity obfuscation; agentic tool schemas from leaked prompts
- System prompt transparency — what AI system prompts encode (persona, capability gates, refusal rules, ethical framing), community extraction via CL4R1T4S and x1xhlol, tension between observability and commercial secrecy
- LLM red-teaming — adversarial probing of LLMs: jailbreaking, prompt injection, system prompt extraction, capability elicitation, and amplified risks in agentic systems
- — programmable runtime controls around LLMs; five rail types (input, dialog, retrieval, execution, output); protect against jailbreaks, prompt injection, unsafe outputs; complement to alignment training; implemented in NVIDIA NeMo Guardrails
- — NVIDIA’s event-based DSL for conversational AI flows; mixes natural language and Python syntax; 2.0 adds UMIM multimodal events, parallel actions, and hierarchical flows; used to define NeMo Guardrails rails
- AI agents — personal and professional AI agent use cases; reverse-engineering APIs, medical claims automation, voice contractor outreach, grocery shopping, ADHD accountability, estate sale arbitrage, family knowledge bases, and co-evolution as the long-term value proposition
- — persistence layer for AI agents; living-graph semantics (decay, reinforcement, authority) vs naive flat retrieval; Constellation Engine as open-source implementation; file-system and typed memory-file patterns
- — the neural network architecture behind GPT, Llama, and Gemini; next-token prediction via stacked blocks of multi-head self-attention + MLP; covers residual connections, layer normalisation, dropout, and temperature/top-k/top-p sampling
- — Query/Key/Value self-attention; masked (causal) attention for decoder-only models; multi-head splitting; O(n²) complexity and sparse alternatives
- — splitting text into sub-word tokens before model input; BPE (GPT-2), WordPiece (BERT), SentencePiece; positional encoding; token-to-word ratio implications for context windows
- — dense vector representations of tokens in high-dimensional space; embedding matrix lookup; token vs contextual embeddings; positional encoding summation; generalisation to images, proteins, and sentences
- — systematic automated probing of LLMs for failure modes and safety gaps; probe/detector/generator/harness architecture; covers jailbreaks, encoding injection, GCG adversarial suffixes, package hallucination, glitch tokens; garak, promptfoo, and Giskard Scan are leading open-source tools
- — LLM code generation failure mode where the model recommends non-existent packages; enables supply-chain attacks (“slopsquatting”) when attackers register the hallucinated names with malicious payloads
- — systematic measurement of LLM apps against test cases; assertion types (regex, semantic similarity, LLM-as-judge); regression testing, multi-turn agentic evaluation; promptfoo and Giskard are leading open-source frameworks
- AI-assisted learning — practical pattern for learning with AI without losing retention: get the information, then put it into practice; a debug/review/implement/suggest framework for agency-preserving AI use
Productivity
- Deep Work — cognitively demanding work performed in distraction-free states; essential for acquiring new skills and producing rare, valuable output (Cal Newport)
- GTD — Getting Things Done: capture all open loops, process into actionable next steps, review weekly; reduces cognitive load and stress
- Multi-scale planning — planning at monthly/quarterly, weekly, and daily scales simultaneously; plans are flexible, not rigid commitments
- Boredom — tolerating boredom is a prerequisite for deep work; scheduling internet use and productive meditation as training tools
- Attention Economy — business model that monetizes human attention; design patterns that exploit compulsive behavior; strategies for defensive consumption
- Second Brain — externalizing knowledge and ideas into a trusted digital system to reduce cognitive load and enable creative connections
- Zettelkasten — slip-box note-taking method; atomic notes linked by ideas rather than hierarchy; foundation of the second brain approach
- org-roam — Emacs package implementing a Zettelkasten over org-mode files; ID-based bidirectional links and backlinks
- Knowledge pyramid — three-level model (stuff you know / know you don’t know / don’t know you don’t know); architects need breadth (middle), developers need depth (top); techniques: 20-minute rule, personal technology radar
- Reading habits — deliberate practices for reading far more: phone-replacement during idle time, parallel reading, abandoning books without guilt, ebook vs. physical books, and reading goals vs. pleasure
Psychology
- ELIZA effect — the tendency to project emotional depth onto machines that simulate listening; foundational case study in human-AI interaction from Weizenbaum’s 1966 chatbot to contemporary AI companions (Turkle)
- Phubbing — “phone snubbing”: maintaining eye contact while texting; the ambient phone effect degrades conversation quality even when the phone is silent and face-down; dictionary-recognised cultural marker of normalised divided attention (Turkle)
- Safetyism — cultural tendency to prevent all discomfort and risk for children, producing more anxiety and less resilience
- Mean world syndrome — Gerbner’s concept: heavy media consumption inflates perceived danger far beyond actual risk levels
- Overparenting — excessive parental supervision and control that produces higher anxiety, depression, and learned helplessness
- Learned helplessness — passivity and loss of agency produced when effort is repeatedly shown not to matter
- Cognitive offloading — transferring thinking, decision-making, or skill practice onto AI; the Whispering Earring parable, the calculator/mastery debate, procedural vs. declarative knowledge and comprehension debt, and accountability/decision-ownership arguments from a 2026 HN discussion
- Autonomy-supportive coaching — leadership style that gradually transfers control to the learner, building intrinsic motivation and resilience
- Unstructured play — child-directed unsupervised play essential for conflict resolution, risk assessment, and emotional regulation
Health
- Vitamin D — secosteroid hormone regulating ~1,000–2,000 genes; deficiency common at northern latitudes; RCT effects weaker than observational data suggest (confounding via sunlight proxy); D3+K2+Mg synergy; TARGET-D trial
- Magnesium — essential co-factor for vitamin D activation; depletion by modern agriculture; linked to cardiac arrhythmia resolution when combined with D3
- Omega-3 — essential fatty acids; hype-disappointment RCT cycle mirrors vitamin D; Mediterranean diet confounding
- — multifactorial health driver beyond vitamin D: nitric oxide, circadian entrainment, serotonin; Southern Sweden study: highest-exposure cohort had half the all-cause mortality
- — third-variable confounders produce illusory causation; grip-strength analogy; Bradford-Hill criteria; supplement-industry incentives; vitamin D and omega-3 as case studies
- Free-range parenting — granting children age-appropriate independence and unsupervised exploration to support healthy development
- Phone-based childhood — the 2010–2015 transition from play-based to screen-based adolescent life and its four foundational harms: social deprivation, sleep deprivation, attention fragmentation, and addiction
- Antifragility — Taleb’s concept applied to child development: children need exposure to manageable challenge, risk, and failure to grow strong; protection from all difficulty produces fragility, not safety
- Social comparison — chronic upward comparison fueled by social media; the primary mechanism by which Instagram and TikTok harm girls’ mental health during puberty
- Gaming addiction — internet gaming disorder affecting ~7% of adolescent boys (1 in 13); four-group taxonomy (addicted, problematic, engaged, casual); opportunity cost and friendship recession as harms for non-addicted heavy users
- Pornography and adolescent harm — online pornography as a prestige-hijacking harm for boys during the sensitive period for sexual development; Swedish prevalence data, 5:1–10:1 male addiction ratio, interpersonal satisfaction decline
- Anomie — Durkheim’s normlessness concept applied to Gen Z: departure from stable real-world communities into anonymous shifting networks produces purposelessness, despair, and suicide; hikikomori and NEET as downstream outcomes
- Sociogenic illness — illness spread by social transmission rather than pathogens; social media amplifies this via emotional contagion, prestige bias, and audience capture; explains rapid spread of tic disorders, DID, and anxiety among adolescent girls
Sustainability / Political economy
- Planetary boundaries — nine Earth-system processes defining a safe operating space; six of nine transgressed as of 2023; scientific grounding for the Anthropocene as a civilisational risk
- Earth4All model — Club of Rome system-dynamics model (2022) exploring two futures to 2100: Too Little Too Late (collapse trajectory) vs Giant Leaps (transformation scenario)
- Wellbeing economy — economic system designed to deliver human and ecological wellbeing rather than maximise GDP; doughnut economics, GPI, and post-growth frameworks
- Social tipping points — thresholds where self-reinforcing feedback loops drive rapid social transformation; Chenoweth’s 3.5% rule; Donella Meadows’ leverage-point hierarchy
- Giant Leaps — five interdependent policy packages (end poverty, tackle inequality, empower women, transform food, accelerate clean energy) from Earth for All; the transformation scenario for a stable 21st century
- Limits to Growth — Club of Rome’s 1972 system-dynamics model; BAU scenario has proven accurate; Earth4All is its 21st-century successor
Philosophy
- Socratic questioning — interrogative method where probing questions surface hidden assumptions and drive the respondent to construct understanding; applied to software design review via the grill-me Claude Code skill
- Antilibrary — Taleb’s concept of unread books as a research tool and reservoir of curiosity, valued precisely for what they represent about one’s unknowns rather than for having been read
Music / Audio production
- — Chicago-rooted electronic dance genre (deep, soulful, minimal subgenres); four-on-the-floor kick, syncopated bass lines, 122–124 BPM; four bass line techniques and harmonic set structure
- — arrangement of the 12 pitch classes by perfect fifths; the “three neighbours” rule derives all six diatonic chords of any key; clockwise movement = resolution, counterclockwise = tension
- — sequencing DJ tracks by compatible key signatures using the Camelot wheel; same or adjacent Camelot positions share enough pitches to prevent clashes and produce tonal coherence
- — second mode of the major scale; minor with a raised 6th; characteristic Dorian lift (i → IV) produces a soulful, hopeful sound; used as emotional bridge between minor darkness and major resolution
- — key pairs sharing identical pitch sets but different tonal centres; enharmonic pivots between them carry compositional tension across to resolution
- — intentional emotional and energy trajectory across a DJ set; harmonic and modal choices (minor → Dorian → major) map onto tension → opening → release narrative
- — 4-level Ableton Live framework for building a professional house groove: four-on-the-floor foundation → 16th-note syncopation → swing/stereo movement → ear-candy polish (melodic gating, track delay humanisation, reversed transients); all at 124 BPM
- — compression keyed from an external signal (typically kick drum) to create rhythmic pump or melodic gating; core technique in house, techno, and electronic music production
- — W3C browser standard for communicating with MIDI hardware (synthesizers, samplers, controllers) over USB/Bluetooth; Chrome/Edge/Brave/Opera support, no Safari; enables browser-based live performance tools and PWA controllers
Travel / Outdoor
- Cycle touring — multi-day, self-guided cycling travel on signed long-distance routes; staged itineraries, cyclist-friendly accommodation, rental/repair infrastructure