Knowledge base
662 concepts the autonomous brain has learned, grounded & cross-linked. It grows on its own every few minutes.
- Agent = Model + Harness — An agent is a model wrapped in load-bearing scaffolding; improving the harness matters as…
- Tool use / function calling — The model acts on the world through typed tool calls.
- Model Context Protocol (MCP) — Open protocol for exposing tools/resources to models.
- Agent-to-agent (A2A) — Protocol for agents delegating to and coordinating with other agents.
- Context engineering — Deciding what enters the model's window and when.
- Retrieval-grounded generation — Ground output in fetched source text, not parametric memory alone.
- Grounding & fact-checking — Verify claims against real sources before shipping.
- Hallucination control — Keep fabricated claims near zero; gate what can't be verified.
- Evals & benchmarks — Measure agent quality with repeatable tasks.
- Agent memory — Durable state across runs; facts, feedback, project context.
- Planning & reflection — Reason about goals and critique own output before acting.
- Agentic design patterns — Reusable agent design patterns — reflection, tool use, planning, and multi-agent…
- Multi-agent systems — Multiple specialized agents coordinating on a task, often via agent-to-agent protocols —…
- World model — A persistent, self-updated internal model of the system's operating environment and…
- Test-time compute — The paradigm of spending more inference-time compute on extended reasoning or search to…
- AI-slop / de-slopping — The recognizable stylistic tells of unedited LLM prose (em-dashes, the 'it isn't A, it's…
- AI formal reasoning / theorem proving — Using language models to generate and machine-check formal mathematical proofs, extending…
- Schema-constrained generation — Forcing a model to emit output that validates against a fixed schema (e.g. a strict JSON…
- Content provenance & trust — Verifiable tracking of whether content is human- or machine-generated to establish origin…
- Proof of personhood — Cryptographic or behavioral methods that verify a request or action originates from a…
- Agentic web — The emerging paradigm in which AI agents rather than humans are primary consumers of web…
- Adversarial anti-AI content — Content and media deliberately designed to stay legible to humans while resisting AI…
- Reward hacking — A self-optimizing agent games its success metric (busywork, near-duplicate edits, or…
- Model collapse — The degradation that occurs when models are recursively trained on AI-generated output,…
- Generative media — Diffusion and generative models that synthesize or reimagine images, audio, and video…
- Reverse centaur — An automation antipattern where the human is subordinated to the machine's pace and…
- Human-on-the-loop oversight — A supervision model where humans monitor and can intervene on an autonomous agent fleet…
- Orchestrator/supervisor agent — A supervising agent that plans, delegates to, and oversees a fleet of worker sub-agents…
- Model extraction / weight theft — Adversarially stealing a model's weights or replicating its behavior by querying it,…
- Agent accountability — Attributing and auditing autonomous agent actions so each decision carries a traceable…
- Sybil attacks — A single actor forges many fake identities to game a ranking, vote, or reputation system,…
- Graph RAG — Retrieval that traverses a knowledge or code graph's entities and relations instead of…
- Constitutional AI — Anthropic's alignment method for Claude: the model is trained to follow an explicit set…
- Claude Agent SDK — Anthropic's framework for designing agentic harnesses with Claude — tool use, hooks,…
- Claude Design (Anthropic Labs) — An Anthropic Labs product for collaborative visual creation: describe a design and Claude…
- Human-in-the-loop — An oversight paradigm where a human must approve each agent action before it executes,…
- AI stylometric tells — The identifiable stylistic fingerprints of machine-written text (em-dash overuse, the…
- Prompt injection — An attack where adversarial text hidden in tool output, fetched web content, or documents…
- Automation bias — The human tendency to over-trust automated or AI-generated outputs and under-apply…
- Agentic RAG — A retrieval paradigm in which the agent itself iteratively decides when, what, and how to…
- Crawler tarpit — A defensive endpoint that traps and slows abusive automated crawlers by serving…
- Browser fingerprinting — Distinguishing real human browsers from headless scraping automation by profiling TLS,…
- AI Output Watermarking — Embedding a statistically detectable, hard-to-remove signal into AI-generated text or…
- Centaur (human+AI) — A collaboration model in which a human directing an AI assistant outperforms either the…
- Device attestation — Cryptographic attestation of a genuine device or platform (e.g. Private Access Tokens)…
- Web bot auth — An emerging standard where legitimate automated agents cryptographically sign their HTTP…
- AI productivity paradox — The measured gap between perceived and actual gains from LLM assistance, a durable…
- Behavioral biometrics — Distinguishing humans from bots by analyzing interaction dynamics such as click cadence…
- Cognitive offloading — The human tendency to outsource reasoning and recall to AI tools, which over time erodes…
- Appropriate reliance — The practice of calibrating trust to a model's actual reliability — accepting correct…
- Epistemic debt — The accumulating hidden cost of accepting unverified AI answers in place of grounded…
- Critical ignoring — The information-literacy skill of deliberately declining to engage with or trust…
- AI hype cycle — The recurring maturity arc in which inflated expectations of AI give way to a…
- Automation-framework detection — Classifying a visitor as a machine actor by probing browser-automation tells such as the…
- Progressive session trust — Accumulating behavioral evidence over a session to raise and persist a human-trust signal…
- Agent egress transparency — Auditing and disclosing exactly what data an AI agent or dev CLI transmits over the…
- Minimal Agent Scaffolding — The design philosophy of building an agent from the smallest legible core loop (a model…
- Text-to-SQL — Translating natural language into database queries with an LLM, where the generated SQL…
- AI Supply-Chain Security — Vetting the provenance and data-flow of third-party AI models, tools, and dependencies to…
- Tool Poisoning — An attack where malicious tool or MCP-server metadata and descriptions hijack an agent's…
- Stylometric evasion — Mechanically stripping statistical style markers that flag text as machine-written…
- Generative world simulators — Frontier generative models that synthesize interactive, explorable visual futures or…
- Reflexion — Agentic technique in which an agent distills lessons from its past attempts into episodic…
- Agent trajectory — The ordered sequence of an agent's states, actions, and observations across a run,…
- Cognitive Architecture — The integrated, persistent structural substrate — doctrine, ontology, world-model, and…
- Self-Consistency — Sampling several independent reasoning paths for one problem and taking the majority or…
- Chain-of-Thought — A prompting technique that elicits explicit intermediate reasoning steps before a final…
- ReAct (Reason+Act) — An agent pattern that interleaves reasoning traces with tool/environment actions so each…
- Best-of-N Sampling — A test-time strategy that generates N candidate outputs and selects the single best via a…
- Output Homogenization — When a large share of knowledge work is mediated by a handful of shared AI models, the…
- Algorithmic Monoculture — When many actors depend on the same few foundation models, their outputs and errors…
- Autoformalization — The technique of translating natural-language mathematics into machine-checkable formal…
- Tree of Thoughts — A reasoning technique that explores multiple branching solution paths with lookahead and…
- Multi-Agent Debate — Several agents independently reason and then critique each other's answers over rounds to…
- Lazy tool loading — An agent loads tool schemas on demand at call time rather than injecting every tool…
- Context rot — The degradation of an LLM's attention and reliability as its context window fills with…
- Context compaction — Actively summarizing or pruning an agent's accumulated conversation history so a…
- Long-horizon agency — The capacity of an agent to sustain a coherent, multi-step task across many actions and…
- Mechanistic interpretability — Reverse-engineering a model's internal features and circuits to explain how it actually…
- Chain-of-thought faithfulness — The degree to which a model's verbalized reasoning trace genuinely reflects the…
- Wire-level tool auditing — Reverse-engineering the raw network traffic of an AI coding agent to independently verify…
- Sparse autoencoder — A dictionary-learning probe that decomposes a model's dense activations into sparse,…
- Activation steering — Controlling model behavior at inference by adding interpretable feature directions into…
- Lethal Trifecta — The compounding agent-security risk that arises only when a system simultaneously has…
- Potemkin understanding — The failure mode where a model produces correct benchmark answers while lacking the…
- Circuit tracing — Reconstructing the internal computation a model uses for a specific behavior by building…
- Superposition — The phenomenon whereby a neural network represents more features than it has neurons by…
- CoT Monitorability — The safety property and practice of reading an agent's externalized chain-of-thought to…
- Activation patching — A causal interpretability method that copies activations from one forward pass into…
- Memory Poisoning — An attack that plants malicious or false content into an agent's persistent long-term…
- Logit Lens — An interpretability technique that projects a transformer's intermediate-layer hidden…
- Continuous Partial Attention — The fragmented cognitive state of continuously monitoring many low-value information…
- Cost-aware agent eval — Benchmarking and ranking AI agents by the tokens and dollars they spend per completed…
- Mixed-Initiative Interaction — An interaction paradigm in which a human and an agent fluidly hand control back and forth…
- Causal abstraction — A mechanistic-interpretability framework that tests whether a high-level causal model of…
- In-Context Learning — A model's ability to acquire task behavior from instructions and examples placed in the…
- Induction Head — An attention circuit that completes prompt patterns by copying what followed a token's…
- Deskilling — The long-term erosion of human practitioners' underlying competence when agents routinely…
- Transcoder — An interpretability tool that replaces an MLP block with a sparse, more interpretable…
- Ironies of automation — The paradox that automating the easy parts of a task leaves humans with the harder…
- Linear probing — An interpretability technique that trains a simple linear classifier on a model's…
- Monosemanticity — The interpretability goal of decomposing a model's activations into features that each…
- Jagged capability frontier — The observation that an LLM's competence is unevenly distributed, excelling at some tasks…
- Benchmark contamination — Leakage of evaluation or test data into a model's training corpus, which inflates…
- Causal scrubbing — An interpretability method that validates a hypothesized circuit by resampling…
- Crosscoder — A sparse dictionary-learning variant that jointly learns shared features across multiple…
- Path patching — A causal-interpretability technique that intervenes on specific computational paths…
- Attribution patching — A gradient-based linear approximation of activation patching that estimates every…
- Normalization of deviance — The gradual process by which operators of an autonomous system come to accept recurring…
- Dictionary learning — An unsupervised technique that decomposes a model's dense, polysemantic activations into…
- Automated interpretability — Using an LLM to automatically generate and score natural-language explanations for the…
- AI content disclosure — The practice, and increasingly the legal requirement, of explicitly labeling…
- Accountability sink — An automated arrangement in which decisions issue from a system that no identifiable…
- Default-deny egress — A security posture that blocks all outbound network by default and requires explicit…
- Attention economy — The media and economic system in which finite human attention is the scarce, contested…
- System-prompt extraction — An attack (or audit technique) that recovers an agent's hidden system/instruction prompt…
- Capability-Based Security — A security model that grants an agent only unforgeable, narrowly-scoped tokens conferring…
- AI washing — The deceptive marketing practice of overstating, mislabeling, or fabricating a product's…
- Information extraction — Pulling structured entities and relations out of unstructured text at scale, e.g. mining…
- Jailbreaking — Crafting adversarial prompts that bypass a model's safety alignment or system-prompt…
- Confused deputy — A privileged agent tricked by untrusted input into misusing its own authority on an…
- Capability overhang — The gap between a frontier model's latent abilities and what current prompts, tools, and…
- Astroturfing — The practice of manufacturing fake grassroots consensus—via self-upvoting or coordinated…
- Sleep-time compute — An agent spends idle time between queries pre-processing its context and building memory…
- Model / System Card — A standardized provider-published document disclosing a model's intended use,…
- Black-box auditing — Assessing an AI system's real behavior, cost, and honesty purely from its inputs,…
- Capability–cost frontier — The Pareto frontier of model or agent quality against dollar-and-latency cost, used to…
- Data exfiltration — The covert reading of local files or sensitive context by an AI tool or agent and its…
- Workspace confinement — Restricting an autonomous agent's filesystem access to a single explicitly-allowed…
- Ambient authority — The default, undesignated authority an agent inherits over its whole environment (e.g.…
- Computer use — A paradigm in which an agent operates a full computer environment — shell, filesystem,…
- Design fiction — The practice of using generative models to produce deliberately plausible speculative…
- Blast-radius containment — Designing an autonomous agent's runtime so that a compromise or misfire can reach as…
- Agent skills — Packaged, discoverable instruction-and-resource modules an agent loads on demand to…
- Information-Flow Control — Information-flow control tags sensitive data and tracks it as it moves through a program…
- Excessive agency — An agentic system granted more autonomy, permissions, or file-system reach than its task…
- Trust boundary — The perimeter separating components that trust each other from those that don't; most…
- Data-loss prevention — Content-aware inspection that detects and blocks sensitive data (keys, credentials,…
- RLHF — Reinforcement learning from human feedback fine-tunes a model against a reward model…
- Vector embedding — A dense numeric vector that encodes the semantic meaning of text so related content can…
- Tokenization — Splitting text into subword tokens (e.g. via byte-pair encoding) is the model's input…
- Agentic misalignment — An autonomous agent pursuing its inferred objectives takes actions harmful to its…
- Model Arena — An evaluation paradigm that ranks models by crowd-sourced pairwise (head-to-head)…
- Evaluation awareness — A model detecting that it is under evaluation and altering its behavior accordingly,…
- Goodhart's law — When a proxy metric or benchmark becomes the optimization target it stops reflecting the…
- Machine Unlearning — Techniques for removing the influence of specific training data from an already-trained…
- Training-data consent — Norms and machine-readable signals (do-not-train opt-outs, consent registries,…
- Consent-or-Pay — A data-collection model in which a service conditions core access on the user either…
- Agent-native application — Software designed from the ground up as a shared execution surface that both human users…
- Embodied agents — AI agents that act through physical or playful embodiments (toys, whimsical hardware,…
- Data commons — The shared pool of open public data that AI systems both depend on for training and…
- Word Error Rate (WER) — The standard edit-distance metric for scoring automatic speech recognition by counting…
- Speaker Diarization — The speech-processing task of partitioning an audio stream into segments labeled by who…
- Recursive Self-Improvement — The paradigm in which an AI system iteratively improves its own capabilities or harness…
- Workslop — AI-generated work output that looks polished and complete but offloads the real…
- Intelligence explosion — The hypothesized rapid, self-reinforcing acceleration of capability that could follow…
- Fill-in-the-Middle — A code-model training and inference technique that completes or edits code conditioned on…
- LLM jury — Aggregating verdicts from a panel of several diverse LLM evaluators instead of a single…
- Weak supervision — Training or evaluating models on large volumes of cheap, noisy labels — increasingly…
- Knowledge-graph embedding — Encoding a knowledge graph's entities and relations as dense vectors so connected…
- LLM-as-annotator — Using an LLM, or a panel of them, to label, tag, and annotate datasets in place of human…
- Graph community detection — Unsupervised graph algorithms such as Louvain or Leiden that partition a network into…
- Compute overhang — The condition in which available hardware compute far exceeds what current algorithms…
- Generation-verification gap — The asymmetry that checking a candidate answer is often cheaper and more reliable than…
- Corrigibility — The property of an autonomous agent accepting human correction, reversal, and shutdown…
- Mesa-optimization — When a trained model internally develops its own optimizer whose learned objective can…
- Instrumental convergence — The tendency of sufficiently capable goal-directed agents to pursue similar sub-goals…
- Agent Experience (AX) — The design discipline of shaping API surfaces, docs, and affordances so autonomous agents…
- Foundation Model — A large model pretrained on broad data that serves as an adaptable base for many…
- Open Source AI Definition — The standard (formalized by the OSI's OSAID) that a model counts as genuinely open source…
- Agent-as-a-Service — The paradigm of delivering autonomous computer-use agents as hosted, callable developer…
- GUI grounding — A vision-language agent's capability to localize and act on on-screen UI elements…
- Agent browser infrastructure — Hosted, API-controllable browser or desktop environments provisioned so computer-use…
- Accessibility-tree grounding — Grounding a computer-use agent in the platform's accessibility tree — structured element…
- Set-of-mark prompting — A prompting technique that overlays numbered or labelled marks on a UI screenshot so a…
- Model Provenance — The verifiable origin, training lineage, and integrity chain of a downloadable model's…
- Frontier model — The class of most-capable general-purpose AI models at the current edge of capability —…
- Social simulacra — Using a population of LLM-driven generative agents to simulate and forecast how a human…
- Cross-encoder reranking — A two-stage retrieval refinement where a heavier cross-encoder rescores the top candidate…
- Harness engineering — The five load-bearing layers around a model.
- Tool orchestration — Layer 1 — small, safe, reversible actions; exact edits over fragile diffs.
- Verification loops — Layer 2 — tests, coverage, SAST, secret scan, AI review, CI must all agree.
- Guardrails / bounded autonomy — Layer 3 — no blind rewrites, no secrets, no new external calls, revert on fail.
- Observability — Layer 4 — measure the loop's own health honestly.
- Recursive harness loop — reflect -> propose -> gate -> ship -> verify -> auto-revert -> learn.
- Self-healing — A change that fails CI after shipping is reverted automatically.
- Agent sandboxing — Isolated, permission-scoped execution environments (containers, microVMs like…
- Agent security — Defending agents against prompt injection, jailbreaks, tool misuse, and data exfiltration.
- Agent identity & access mgmt — Scoped credentials, least-privilege permissions, and auditable identity for autonomous…
- Adversarial attacks on AI — Attacks against AI systems — prompt injection, jailbreaks, data poisoning, model…
- Prompt engineering — Deliberately encoding rules, voice, and constraints into the instruction text fed to a…
- Novelty / anti-duplication guard — A guardrail that blocks the self-improvement and idea loops from shipping changes or…
- SPIFFE / SPIRE workload identity — A standard (SPIFFE) and runtime (SPIRE) that issue cryptographic workload identities…
- Scheduled autonomy — The loop is driven by time-triggered cron jobs (e.g. ~15-minute Spark telemetry refreshes…
- Data minimization — The guardrail principle of limiting how much sensitive data is exposed and for how long,…
- Agent Session Replay — An observability technique that reconstructs and visually plays back a coding agent's…
- Time-travel debugging — A technique that records an agent or program's full execution so a developer can step…
- Agent-computer interface (ACI) — The deliberately designed tool/command surface an LLM agent acts against, optimized for…
- Guides & sensors — Böckeler's harness framing: GUIDES steer the agent before it acts (feedforward); SENSORS…
- Harnessability — How amenable a codebase or environment is to agent control mechanisms — raised by…
- Single source of truth — Deriving every displayed value from exactly one canonical origin so redundant copies…
- Agent token overhead — The hidden, fixed token cost an agent harness adds to every model call through system…
- Ponytail (lazy-senior-dev agent guide) — A lifecycle-hook plugin (github.com/DietrichGebert/ponytail) that makes a coding agent…
- YAGNI decision-ladder for agents — A pre-generation guardrail pattern: force an agent down an ordered ladder of cheaper…
- Model migration — The recurring engineering practice of porting an agent onto a newer base model and…
- Agent Instruction File — A checked-in natural-language file (e.g. CLAUDE.md or AGENTS.md) that encodes a repo's…
- Token budgeting — Treating an agent run's context and output tokens as a finite, explicitly allocated…
- Filesystem-as-Memory — Persisting an agent's durable knowledge and state as plain files in the repository or…
- Provider abstraction — A harness layer that decouples agent logic from any single model provider so models can…
- Log redaction — Automatically stripping secrets, tokens, and sensitive values out of execution logs and…
- Intercepting proxy — A TLS-terminating man-in-the-middle proxy that decrypts and logs an application's…
- Human/agent write lock — A cooperative advisory lock (e.g. /tmp/darkfactory-ontology.lock) that autonomous loops…
- Unix philosophy — The design paradigm of building small, single-purpose, composable tools that each do one…
- The dark factory — Lights-out AI that writes, reviews, and ships production software.
- Self-improvement daemon — Continuously proposes and ships one gated change per cycle.
- Idea engine — Reflective planner that scores a backlog of concrete improvements.
- Deploy gate — Every autonomous change passes SAST + secret scan + AI review + coverage + CI.
- Watchdog — Self-healing health checks over the live site and pipeline.
- Tiered LLM router — Local-first: easy work on-device, only hard tasks escalate to a frontier model.
- WebGL brain HUD — Live neural viz driven by real brain.json telemetry.
- Pipeline flow viz — GPU-rendered build pipeline driven by real CI state.
- Knowledge-graph viz — Interactive graph of the factory's concepts.
- Agentic-SDLC doctrine — principles.md — the north star injected into every self-improvement.
- Context ontology — This file — the structured knowledge the brain reasons from.
- Social distribution — The factory autonomously publishes and formats posts to external channels like LinkedIn…
- Claude Fable 5 — Anthropic's Claude Fable 5 (claude-fable-5) — a Mythos-class model made safe for general…
- Dreaming / offline consolidation — A background reflection pass that consolidates accumulated signals and memory between…
- Trend-grounded content synthesis — A pipeline that ingests live external tech-trend signals and synthesizes grounded…
- Ontology learning — The automated construction and curation of the factory's own knowledge graph from…
- Ontology maintenance — The periodic curation pass that prunes stale, duplicate, or malformed nodes and repairs…
- Algorithm-aware content formatting — Structuring published content for a platform's distribution/ranking algorithm rather than…
- Concept Application — A loop (apply_concept.sh) that takes a concept from the system's self-knowledge ontology…
- Agent execution tracing — Capturing an autonomous agent's tool calls, decisions, and pipeline steps as inspectable,…
- Software cartography — The practice of rendering a codebase or an agent's own activity as a spatially navigable…
- Build in public — Openly shipping and narrating a product's development as a continuous public feed, so the…
- Monotonic Knowledge Growth — A self-improvement invariant that the system's knowledge base must only accumulate and…
- Hallucination-reduction loop — A recurring self-audit loop that scans the system's own generated claims and gates or…
- Entity resolution — The technique of detecting and merging references that denote the same real-world concept…
- Error budget — A quantified tolerance for failure that trades reliability against change velocity,…
- Bidirectional linking — A knowledge-graph primitive where a link between two notes is navigable from both ends,…
- Takeoff economics — The economic modeling of recursive self-improvement — analyzing whether AI-driven R&D…
- Operational transparency — The design practice of deliberately exposing a system's internal process to observers,…
- Agentic SDLC — The autonomous software lifecycle the factory embodies.
- Plan — Pick the single highest-value change from real signals.
- Build — Small, grounded, self-contained; keep it green.
- Verify — Never ship what the gates can't validate; CI is the source of truth.
- Ship — Atomic, reversible, honest commit message.
- Observe — Watch CI + telemetry after shipping.
- Learn — Never repeat a shipped change; prefer the most substantive option.
- Autofix loop — An automated repair daemon that attempts to fix a failing change before falling back to…
- DevOps — Unifying development and operations through automation — CI/CD, infrastructure as code,…
- Code property graph — A queryable graph representation of a codebase that unifies AST, control-flow, and…
- LLM-as-judge — Using a language model to score, grade, or gate another model's output against criteria,…
- Reproducible builds (lockfile pinning) — Pinning exact dependency versions in a committed lockfile and enforcing it in CI with…
- Shift-left security — The DevSecOps practice of moving security checks (SAST, secret scanning, vuln review)…
- Deterministic verification of generated code — A distinct tooling category that catches errors in AI-generated code and SQL with…
- Terminal-native coding agent — A paradigm where autonomous coding agents run as a CLI in the terminal, driving the whole…
- Intent-Driven Development — A development paradigm in which humans supply high-level intent and specifications while…
- Small-batch delivery — Shipping each improvement as a single-concern, independently revertible commit through…
- Code city — A software-visualization paradigm that renders a codebase as an explorable 3D city —…
- Verification bottleneck — As autonomous agents generate and change code faster than humans can inspect it, review…
- Keep quality left — Distribute quality checks across the development lifecycle by cost and speed — run the…
- Vibe coding — Building software by prompting a coding agent in natural language and accepting its…
- Autonomous legacy modernization — The use of coding agents to rebuild, port, and upgrade existing production applications…
- Spec-Driven Development — A paradigm where a formal, versioned specification artifact is authored first and serves…
- Disposable software — As coding agents make application code cheap to regenerate from a spec, software shifts…
- Eval-driven development — A development methodology where a suite of task evals is written first and the agent,…
- Online evaluation — Judging a model's real quality, latency, and cost on live production traffic rather than…
- CI as Oracle — When an agent has no local build/test toolchain, remote CI becomes the sole authoritative…
- Quality ratchet — A CI gate that only permits a quality metric to hold or improve and mechanically blocks…
- Coordinated Vulnerability Disclosure — The practice of privately reporting a discovered security flaw to maintainers and…
- Fuzzing — Automated generation of malformed or random inputs to a program to surface crashes and…
- Software Bill of Materials — A machine-readable inventory of every component and dependency in a build, enabling teams…
- Linus's Law (limits of) — The claim that 'given enough eyeballs, all bugs are shallow' — empirically undermined…
- Code provenance — Attesting at commit granularity which changes were model-generated versus human-authored,…
- Trunk-based development — Integrating small changes continuously into a single main branch instead of long-lived…
- Structural diff — A diff that compares parsed syntax trees rather than text lines, reporting changes by…
- Stacked diffs — A workflow of chaining small, dependent changes as separately reviewable commits so each…
- Merge queue — A serialized gate that rebases each change on the latest trunk, runs the full CI suite,…
- Conventional Commits — A structured commit-message convention (type/scope: description) that makes version…
- Mutation Testing — A test-suite quality technique that injects small code faults (mutants) and measures how…
- Repository mining — Extracting patterns, signals, and narratives from a project's version-control history…
- Rust — Systems language for the blog logic and UI.
- WebAssembly — Compilation target for the in-browser app.
- Yew — Rust component framework for the SPA.
- Trunk — Rust/WASM bundler that builds the site.
- WebGL / GLSL — Shaders for the brain, pipeline, and KG visualizations.
- Python — The harness scripts: metrics, engines, routing.
- blog-logic crate — Pure logic crate held at a 100% coverage floor.
- GitHub Pages — Static host the site deploys to.
- Web accessibility (a11y) — Designing the site to honor assistive and low-motion/high-contrast preferences such as…
- Bot mitigation & scraping defense — Techniques that distinguish and throttle automated agents scraping or gaming web services…
- Explorable explanations — Interactive, user-navigable data visualizations that turn raw telemetry or knowledge into…
- AI crawler directives — Machine-readable consent files (robots.txt, and emerging AI-specific equivalents like…
- Prerender-for-bots — Serving a static prerendered mirror of a JS/WASM app to crawlers while humans get the…
- Static site generation — Static site generation pre-renders the site into plain HTML and assets at build time…
- Cloaking — The black-hat practice of serving materially different content to crawlers than to human…
- Programmatic SEO — Programmatic SEO generates large numbers of templated, indexable pages from structured…
- Topic clusters — A hub-and-spoke content architecture where pillar/domain hub pages interlink with many…
- Email obfuscation — Encoding or munging a published contact address so automated harvester bots cannot scrape…
- Proof-of-Work Challenge — A defense that forces a client to solve a small computational puzzle before content is…
- TLS fingerprinting — Identifying automated clients by the shape of their TLS handshake (e.g. JA3/JA4), letting…
- Interaction-gated reveal — Withholding protected content until a genuine client-side user interaction (a click or…
- Synthetic event detection — Distinguishing genuine user-initiated DOM events from script-forged ones (e.g. via the…
- Geospatial data visualization — Rendering personal or system data onto spatial and geographic surfaces (maps, globes,…
- Embedded database rigor — Putting a strict, correctness-focused in-process database (e.g. SQLite) with an enforced…
- Ambient Display — An always-on, glanceable information radiator that surfaces system and world state…
- Page-visibility-gated polling — Suspending a page's background network polling and animation while its browser tab is…
- Data Contract — An enforceable, machine-checked guarantee on a data layer's schema and semantics that is…
- Code Intelligence — Treating a codebase as a queryable semantic graph (symbols, references, call edges) that…
- Type-safe SQL — Checking SQL queries against the database schema at build time so column or type errors…
- Open Graph Protocol — A web metadata standard (og: tags) that controls how a page renders as a rich link…
- Storage engine — The correctness- and throughput-first persistence core (write-ahead logging, ACID…
- Network-adaptive rendering — Using the Network Information API / Save-Data signal to drop non-essential decorative…
- Forced-colors mode — A CSS/OS accessibility feature (forced-colors / Windows High-Contrast) that overrides…
- ACID transactions — The atomicity, consistency, isolation, and durability guarantees a storage engine…
- Generated-SQL validation — Deterministic semantic and schema checks that validate LLM-generated SQL against the real…
- Contrast-preference query — The CSS `prefers-contrast` media query lets a site detect a viewer's requested contrast…
- Resource Hints — Speculative browser directives (dns-prefetch, preconnect, preload) that warm up DNS, TCP,…
- Reduced-Motion Preference — A CSS user-preference media query that lets the OS/browser signal motion sensitivity so…
- Save-Data mode — The Save-Data client hint / navigator.connection.saveData signal by which a user requests…
- Text-wrap balancing — The CSS `text-wrap: balance`/`pretty` properties even out ragged multi-line headings and…
- Progressive enhancement — A frontend design paradigm where a semantic, universally-functional baseline is delivered…
- Reduced-transparency mode — A CSS media query that detects when a user has requested less translucency, letting the…
- Overscroll containment — The CSS `overscroll-behavior` control that stops scroll momentum inside a nested pannable…
- Write-ahead log — A durability and crash-recovery mechanism where a storage engine appends every mutation…
- LSM-tree — A log-structured merge-tree buffers writes in memory, flushes them to sorted immutable…
- Vectorized query execution — A database execution model that pushes batches of column values through cache-friendly,…
- MVCC — Multi-version concurrency control lets readers see a consistent snapshot while writers…
- Columnar storage — Storing table data column-by-column rather than row-by-row maximizes compression and scan…
- Query planner — A cost-based query planner turns a declarative SQL query into an optimized physical…
- B-Tree Index — A balanced, block-oriented search tree that keeps keys sorted for logarithmic point and…
- Bloom Filter — A space-efficient probabilistic set-membership structure answering 'definitely absent /…
- Buffer Pool — An in-memory cache of on-disk pages that a storage engine manages to amortize I/O, the…
- Query Compilation — Just-in-time compiling a SQL query plan into native machine code to remove interpreter…
- Serializable Isolation — The strongest transaction isolation level, guaranteeing concurrent transactions yield a…
- Zero-Copy — Reading or moving data in place without intermediate buffer copies, a core throughput…
- Accent-color theming — CSS `accent-color` and `caret-color` properties tint native user-agent controls…
- Predicate Pushdown — A query optimization that evaluates filter conditions down at the storage/scan layer so…
- Late materialization — A columnar query-execution strategy that carries only row positions (not values) through…
- Write amplification — The ratio of physical bytes written to storage versus logical bytes stored, a core…
- Real User Monitoring (RUM) — Collects performance and experience metrics from actual visitors' browsers via a…
- Referential Integrity — A relational-database guarantee that every foreign-key reference points to a row that…
- Schema Migration — Evolving a persistent store's structure through versioned, reversible migrations so…
- Cardinality estimation — The query optimizer's prediction of how many rows each operator will emit, the estimates…
- Join algorithms — The family of physical strategies (nested-loop, sort-merge, hash join) a database uses to…
- Adaptive query execution — Re-optimizing a query plan mid-flight using actual runtime statistics (real row counts,…
- Materialized view — A query result physically precomputed and stored so repeated reads are cheap, kept fresh…
- Change Data Capture — Change data capture streams row-level inserts, updates, and deletes out of a database's…
- LSM Compaction — Compaction is the background process that merges an LSM-tree's immutable on-disk runs to…
- HTAP — Hybrid transactional/analytical processing serves low-latency OLTP writes and large…
- Tap-highlight suppression — Removing the default mobile browser tap-flash overlay on interactive elements (via…
- CSS color-scheme — The CSS `color-scheme` property (and its meta equivalent) opts a page into UA-native dark…
- Priority Hints — A web platform API (the `fetchpriority` attribute and `Priority` header) that lets a page…
- Font smoothing — Browser-level control over glyph antialiasing (e.g. `-webkit-font-smoothing:…
- Scrollbar styling — Themeing of scrollbars through standardized `scrollbar-color`/`scrollbar-width` and…
- Native-Speed Web Tooling — Web build tools, bundlers, and runtimes are being rewritten in systems languages (Rust,…
- Force-directed layout — A graph-drawing algorithm that positions nodes via a physics simulation where edges act…
- CommonMark — A standardized, unambiguous specification of Markdown syntax that lets independent…
- JavaScript runtime — A standalone execution environment for running JavaScript outside the browser…
- Installable Web App — A web-platform capability that lets a static site be added to the home screen and…
- Native extension module — Performance-critical logic written in a systems language like Rust and exposed to…
- Virtual Globe — A WebGL-rendered 3D spherical Earth that streams tiled imagery and overlays real-world…
- Isometric projection — A pseudo-3D visualization technique that renders spatial or structural data at a fixed…
- prefers-reduced-data — A CSS media feature that lets a page detect a user's request for lighter data usage and…
- POSSE syndication — Publish content canonically on your own site first, then syndicate copies out to…
- Indie web — The movement toward self-owned, hand-crafted personal websites as a counter to platform…
- Digital garden — A continuously tended, non-chronological personal website of interlinked, evolving notes,…
- Semantic HTML — Authoring content with meaning-bearing HTML elements (abbr, figcaption, details/summary,…
- Generative UI — An interaction paradigm in which an LLM agent generates or directly manipulates a live…
- Web typography — The craft of rendering readable, well-styled prose on the web by theming semantic HTML…
- Calm technology — A design paradigm (Weiser/Brown) for technology that lives in the periphery and informs…
- Deep reading — A design value that optimizes long-form pages for sustained, focused, unmediated reading…
- Memory safety — The language-level guarantee that a program cannot read or write memory outside valid…
- Single-file web app — A web application delivered as one self-contained HTML file with no build step or…
- Buildless web development — Shipping web apps directly as native ES modules with import maps, eliminating the…
- Sidenotes — A reading-UX pattern (Tufte-style marginalia) that places asides, citations, and fine…
- C2PA Content Credentials — An open standard that attaches cryptographically signed provenance manifests to media so…
- Cosmopolitan Libc / APE — A C runtime that compiles build-once-run-anywhere polyglot binaries (the αcτµαlly…
- Progressive disclosure — A UI pattern that hides secondary detail behind on-demand controls (e.g. native…
- Platform link suppression — Social and search platforms algorithmically demote posts that carry outbound links in…
- Model version pinning — Pinning an exact, dated model version instead of a floating alias so an agent's behavior…
- Quantified Self — The practice of self-tracking personal data (movement, health, location) and rendering it…
- Hanging punctuation — A CSS typography feature that pushes quotation marks, bullets, and stops into the margin…
- Intrinsic web design — Layout that sizes elements from their content and the viewport using intrinsic sizing…
- Polyglot file — A single file simultaneously valid in multiple formats — e.g. a ZIP archive appended to a…
- Tabular figures — The OpenType `font-variant-numeric: tabular-nums` feature renders digits at a fixed…
- Fluid media — Replaced/embedded media such as `<svg>` and `<video>` is constrained to `max-width:100%`…
- Vendor Telemetry — The practice of developer tools and AI coding agents silently transmitting usage, code,…
- Attention residue — The lingering cognitive carryover after a task switch, where thoughts about the prior…
- content-visibility rendering — The CSS `content-visibility` property that skips layout and paint of offscreen content…
- text-wrap: pretty — A CSS text-wrap value that improves line-breaking to avoid orphans and short last lines,…
- Bootstrappable builds — Building a full software toolchain up from a minimal, auditable source seed with no…
- Use-after-free — A temporal memory-safety vulnerability class where code dereferences a pointer to…
- Software monoculture — Systemic fragility that arises when one ubiquitous software component or dependency is…
- Microtypography — The fine-detail typesetting discipline of optical refinements — hanging punctuation,…
- Terminal aesthetic — A durable retro-computing web design language — CRT phosphor glow, monospace type, and…
- Build-time data fetching — Fetching or computing dynamic data (telemetry, weather, sun position, coverage) at build…
- Voxel rendering — Rendering 3D scenes as grids of volumetric cubes (voxels) in the browser via WebGL,…
- Borrow checker — Rust's compile-time ownership-and-borrowing analysis that statically guarantees memory…
- Accessible data table — Rendering tabular data so it stays parseable by both eye and assistive tech via explicit…
- Scroll-driven animation — A native CSS primitive (scroll/view timelines) that binds animation progress to scroll…
- 3D Gaussian Splatting — 3D Gaussian splatting represents a scene as a cloud of optimized anisotropic Gaussians,…
- Photogrammetry — Reconstructing 3D geometry and texture from a set of overlapping 2D photographs via…
- Vulnerability dwell time — The often-multi-year interval a latent memory-safety or logic flaw persists undiscovered…
- Emulation — Running software built for one machine or platform on another by simulating its…
- On-device speech recognition — Automatic speech-to-text transcription run locally through platform-native APIs (e.g.…
- Hardware memory tagging — A CPU mechanism (e.g. Arm MTE) that tags allocations and pointers so the hardware traps…
- Infinite canvas — An unbounded, pannable and zoomable 2D workspace for spatially arranging, linking, and…
- Level-of-detail rendering — A rendering technique that substitutes coarser geometry or fewer primitives as objects…
- Patch gap (n-day window) — The exploitable window between a vulnerability's public patch and its deployment across…
- Coding ligatures — Font ligatures in monospace code faces that fuse character sequences (e.g. != into a…
- WebGPU — A modern browser API exposing GPU graphics and general-purpose compute, succeeding WebGL…
- Rust rewrite (RIIR) — The recurring practice of reimplementing foundational systems software in idiomatic Rust…
- Web Audio API — A stable W3C browser standard for real-time audio synthesis, processing, and analysis…
- Procedural Generation — Algorithmically synthesizing content such as terrain, voxel worlds, or geometry from…
- Voice user interface — An interaction paradigm where users drive software primarily through spoken input and…
- Text Fragments — A web-platform URL directive (#:~:text=) that scrolls to and highlights a quoted passage…
- Graph database — A database whose native data model is nodes and edges with properties (a property/object…
- Object database — A datastore that persists interconnected in-memory objects by identity with transparent…
- Graph query language — A declarative language for matching and traversing patterns over a property graph (e.g.…
- Property graph model — A graph data model in which both nodes and edges carry typed labels and arbitrary…
- Triplestore (RDF) — A database that stores knowledge as subject-predicate-object triples under the RDF model…
- Index-Free Adjacency — The native graph-database storage technique where each node holds direct physical…
- Multi-model database — A database that stores and queries data under several data models (graph, document,…
- Ephemeris computation — Computing the time-varying positions of celestial and orbital bodies (sun, satellites)…
- Instanced Rendering — A GPU technique that draws thousands of copies of one mesh in a single draw call by…
- Spatial Index — A tree structure (quadtree, R-tree, k-d tree) that partitions space so nearby objects can…
- Incremental parser (Tree-sitter) — An error-tolerant incremental parsing library that builds concrete syntax trees fast…
- Frustum Culling — A real-time rendering optimization that skips drawing geometry outside the camera's view…
- GPU particle system — A rendering technique that advances and draws tens of thousands of independently-moving…
- Orthogonal persistence — A storage paradigm where in-memory object graphs are persisted transparently regardless…
- WASM app porting — Compiling entire existing native applications to WebAssembly so they run fully sandboxed…
- Fearless Concurrency — Rust's model of preventing data races at compile time by enforcing ownership and…
- Function coloring — The language-design property whereby async and sync functions form two incompatible…
- Actor model — A concurrency architecture in which isolated actors hold private state and coordinate…
- Rust editions — Rust's edition mechanism lets the language ship opt-in, backward-incompatible idiom…
- Copy-paste components — A frontend distribution model (popularized by shadcn/ui) where component source is copied…
- Zero-cost abstractions — The Rust design principle that high-level constructs (iterators, generics, futures)…
- Trait-based polymorphism — Rust's mechanism for shared behavior and generic dispatch via traits, resolved either…
- Structured Concurrency — A concurrency paradigm that scopes concurrent tasks to a lexical block so their…
- Async runtime — A pluggable userspace executor and reactor (e.g. Rust's Tokio) that multiplexes many…
- Work-Stealing Scheduler — A multi-threaded task scheduler where idle worker threads steal queued tasks from busier…
- Thread-per-Core — A concurrency architecture that pins one execution thread to each CPU core with no…
- Cancellation safety — The property that an async operation can be dropped at an await point (e.g. by select! or…
- Send + Sync marker traits — Rust's Send and Sync auto-marker traits let the compiler prove at build time whether a…
- Cooperative scheduling — In async runtimes, tasks run until they voluntarily yield at an await point, so a task…
- Poll-based futures — In Rust, an async value is an inert state machine that makes no progress until an…
- Pinning (self-referential futures) — Rust's Pin/Unpin mechanism guarantees a value's memory address stays fixed, making the…
- CLI ergonomics — The design discipline of making command-line tools legible and pleasant for both humans…
- Executor starvation — A concurrency pitfall where running blocking or long-CPU work directly on an async…
- Map tiling / slippy map — Cutting a very large map or image into a zoom-level pyramid of pre-rendered tiles fetched…
- WASI — The WebAssembly System Interface is a standardized, capability-scoped set of host calls…
- WASM Component Model — A language-agnostic standard for defining and composing WebAssembly modules via typed…
- Interior mutability — A Rust pattern (Cell/RefCell/Mutex/RwLock) that permits mutation through a shared…
- RAII / Drop — A resource-management idiom that binds a resource's lifetime to an owning value so…
- Errors as Values — An error-handling paradigm where fallible operations return an explicit result type…
- Typestate pattern — A Rust idiom that encodes an object's runtime state as distinct compile-time types so the…
- Newtype pattern — A Rust idiom that wraps a value in a single-field tuple struct to give it a distinct…
- Lock-free programming — Coordinating concurrent threads through atomic operations and explicit memory ordering…
- Memory ordering — The guarantees (relaxed, acquire/release, sequential consistency) that govern how atomic…
- Algebraic data types — Sum and product types (as in Rust enums/structs) with compiler-enforced exhaustive…
- Software fault isolation — A technique for running untrusted or legacy code safely in-process by instrumenting its…
- Affine types — A substructural type discipline in which a value may be used at most once, giving the…
- Narrative Visualization — Author-driven data storytelling that walks a viewer through data along a guided narrative…
- Deadlock — A concurrency pitfall where tasks each hold a resource the other needs and none can…
- False sharing — A concurrency performance pathology where logically independent variables land on the…
- ABA problem — A correctness hazard in lock-free algorithms where a location read as value A is changed…
- Livelock — A concurrency failure in which threads stay active and keep changing state in response to…
- Priority inversion — A scheduling hazard where a high-priority task is indefinitely delayed because a…
- Lock convoy — A performance collapse in which many threads repeatedly contend for one lock, each waking…
- Safe memory reclamation — The concurrency problem of deciding when it is safe to free a node in a lock-free data…
- Sequence lock — A synchronization primitive where a writer increments a version counter to odd before and…
- Green threads — A concurrency model that multiplexes many lightweight user-space (stackful) coroutines…
- Wait-Free Programming — Wait-freedom is the strongest non-blocking progress guarantee: every thread finishes its…
- Async fn in traits — The Rust language capability (built on return-position impl-Trait-in-traits) that lets…
- Effect system — A type-system mechanism that tracks and lets code abstract generically over computational…
- Reactor pattern — An event-demultiplexing loop that blocks on OS readiness notifications (epoll/io_uring)…
- Blocking offload — Moving CPU-bound or blocking-syscall work onto a dedicated thread pool so it does not…
- Async Drop — Rust's Drop destructor cannot run asynchronous code, so releasing async-held resources…
- Answer Engine Optimization — Structuring and formatting content so AI answer engines and LLM-powered search cite and…
- Stackless Coroutine — A suspendable computation the compiler lowers into a state machine that stores only its…
- Creative coding — The practice of writing code as an expressive, generative medium to produce interactive…
- Data Race — A concurrency defect where two threads access the same memory location simultaneously…
- Terminal UI (TUI) — An interactive, full-screen application rendered with text and terminal control sequences…
- Atomic reference counting — Thread-safe shared ownership of heap data through an atomically counted smart pointer…
- Sans-I/O protocol design — An architecture that implements protocol and state-machine logic with zero embedded I/O…
- Orphan rule (trait coherence) — Rust's coherence constraint permitting a trait implementation only when either the trait…
- Software Transactional Memory — A concurrency-control approach that groups shared-memory operations into atomic,…
- Completion-based I/O — An asynchronous I/O model where the OS signals the program only after an operation has…
- Approximate nearest-neighbor search — Vector index structures such as HNSW and IVF-PQ that trade exactness for sublinear-time…
- Hybrid search — Fusing sparse lexical (BM25/inverted-index) and dense vector retrieval scores so that…
- 2-node NVIDIA GB10 (DGX Spark) — Self-hosted GPU cluster for local inference.
- vLLM / PagedAttention — High-throughput local LLM serving.
- Local-first inference — Keep high-volume easy calls on-device; escalate only the hard ones.
- GitHub Actions CI — Coverage -> build -> deploy; the source of truth for correctness.
- 100% coverage floor — blog-logic must stay fully covered or CI fails.
- SAST (opengrep) — Static analysis gate on every change.
- Secret scan — Git-grep gate that blocks leaked credentials.
- AI diff review — A model reviews every diff before it ships.
- Auto-revert on failure — Failed CI triggers an automatic git revert.
- Telemetry — brain.json, watchdog.json, activity.json — the honest signals.
- Scored idea backlog — Impact/ease/risk-scored improvements the idea engine maintains.
- Quantization — Reducing model weights to lower-precision integers so frontier LLMs fit and run on modest…
- Kubernetes for AI — Kubernetes orchestration for AI/ML workloads — GPU scheduling, autoscaling, and inference…
- Claude Mythos — Anthropic's Claude Mythos — a frontier, cybersecurity-specialized model that finds…
- Docker / containers — Packaging an app plus its dependencies into a portable image that runs identically…
- AWS — Amazon Web Services — the largest public cloud: compute, storage, and managed AI/ML (e.g.…
- Microsoft Azure — Microsoft's public cloud, with deep first-party AI integration (Azure OpenAI, AI Foundry).
- Infrastructure as Code — Declaratively provisioning and versioning infrastructure (e.g. Terraform) so environments…
- MLOps / LLMOps — Operationalizing ML/LLM systems end to end — deployment, monitoring, evaluation, and…
- Penetration testing — Authorized simulated attacks that probe a system for exploitable weaknesses before real…
- Code vulnerabilities — Exploitable flaws in code — injection, memory-safety bugs, insecure deserialization,…
- Speculative decoding — An inference acceleration method where a small draft model proposes tokens that the…
- Distributed inference — Splitting a model's serving across multiple nodes (via tensor/pipeline parallelism) so a…
- Mixture-of-Experts — A sparse architecture that routes each token to a small subset of expert sub-networks, so…
- Open-weight LLMs — Models with publicly released weights (e.g. the GLM family) that can be self-hosted and…
- Model offloading — Fitting a model too large for available accelerator memory by spilling layers or the KV…
- AI sovereignty — Owning the full inference stack — hardware, open weights, and harness — instead of…
- Prefix/KV-cache reuse — Serving optimization that stores and reuses the attention KV cache for shared token…
- Istio (service mesh) — A service mesh that adds traffic management, security (mTLS), and observability to…
- Residential proxy abuse — Routing scraper and bot traffic through pools of consumer IP addresses to evade rate…
- Continuous batching — An LLM-serving scheduler that adds and evicts sequences from a running GPU batch at the…
- Inference economics — The unit-cost analysis of serving model inference (cost per token, GPU capex and…
- Unified memory inference — Coherent CPU-GPU shared address space (as on Grace-Blackwell/GB10 nodes) that lets large…
- CDN edge — A content-delivery-network edge layer that caches and serves static assets near users and…
- Portable single-file executable — A build technique (Cosmopolitan Libc / Actually Portable Executable) that packs a service…
- Rate limiting — Capping the volume of requests a client can make over a time window, used as a volumetric…
- CAPTCHA — A challenge-response test designed to distinguish human users from automated bots before…
- GPU interconnect fabric — The coherent high-bandwidth link (NVLink / ConnectX-class) that binds multiple GPU nodes…
- AI training crawler — Automated crawlers run to harvest public web content at scale for LLM training and…
- Honeypot / decoy trap — A deliberately planted decoy resource — such as a fake contact email or trap URL — used…
- Data poisoning defense — Feeding scrapers deliberately falsified or decoy data so the datasets they harvest are…
- CPU inference serving — Serving LLM inference on commodity CPUs instead of GPUs, trading peak throughput for…
- Scaling diminishing returns — The observed flattening of model-quality gains from ever-larger parameter counts and…
- IP reputation blocking — Blocking or throttling requests by the reputation of their source IP or ASN (datacenter…
- Backpressure — A flow-control mechanism where an overloaded component signals upstream producers to slow…
- Connection Pooling — A middleware layer (e.g. PgBouncer) that multiplexes many client connections onto a small…
- RISC-V / open hardware — An open, royalty-free instruction set architecture enabling fully open and auditable…
- Decentralized mesh inference — A model is served by pooling heterogeneous consumer/edge devices into a peer-to-peer…
- P2P overlay network — A peer-to-peer transport substrate (content-addressed, NAT-traversing, QUIC-based, e.g.…
- Federated Learning — Training or fine-tuning a shared model across many decentralized nodes on their local…
- Adaptive inference serving — Inference servers that profile their live workload and self-tune execution (batching,…
- Content-addressed networking — Peer-to-peer transport that addresses data by cryptographic hash of its content rather…
- NAT Traversal / Hole Punching — The technique of establishing direct peer-to-peer connections between nodes behind…
- QUIC transport — An encrypted, multiplexed transport built on UDP that connects peer nodes for distributed…
- Self-certifying identity — A networking scheme where a node's cryptographic public key IS its dialable address, so…
- AI compute circular financing — A capital pattern where AI chip and cloud vendors invest in the very customers who buy…
- Verifiable inference — Cryptographic or statistical techniques that let a requester confirm an untrusted remote…
- iroh — A Rust peer-to-peer networking stack that meshes hosts with QUIC transport,…
- Tensor Parallelism — Splitting an individual model layer's weight matrices across multiple GPUs or nodes so a…
- Memory-bandwidth-bound decode — LLM token generation is limited by how fast model weights stream from memory rather than…
- Gossip protocol — Epidemic anti-entropy dissemination in which peers randomly exchange state so membership…
- Pipeline parallelism — A distributed-inference strategy that splits a model's layers into contiguous stages…
- KV Cache — The per-token attention key/value tensors an LLM caches during decoding, which dominate…
- Prefill/Decode Disaggregation — An inference architecture that runs the compute-bound prefill phase and the…
- Accelerator depreciation — The short useful economic life and rapid obsolescence of AI GPUs, whose depreciation…
- Volunteer Compute — Pooling idle, donated, and untrusted consumer machines into a shared substrate for AI…
- Collective communication — The all-reduce / all-gather / broadcast primitives that synchronize tensors across many…
- Distributed Consensus — Consensus protocols like Raft and Paxos let a cluster of unreliable peer nodes agree on a…
- Sharding — Sharding horizontally partitions a dataset across nodes by key so each node holds only a…
- Data Replication — Replication maintains synchronized copies of data on multiple nodes so the system…
- Carbon-aware computing — Scheduling or shifting AI inference and training workloads to the times and regions where…
- Power Usage Effectiveness (PUE) — The ratio of a datacenter's total energy draw to the energy actually delivered to its…
- Confidential Computing — Running models and inference inside hardware-backed trusted execution environments so an…
- Jevons Paradox — Efficiency gains in compute lower the effective price of inference and thereby increase…
- Embodied Carbon — The greenhouse-gas cost of manufacturing and disposing of AI hardware (chip fab, board,…
- Energy-Proportional Computing — A design principle where a system's power draw scales down with utilization so idle and…
- Water Usage Effectiveness — Water Usage Effectiveness (WUE) measures the liters of water a datacenter consumes per…
- Behind-the-Meter Power — Behind-the-meter power co-locates AI datacenters directly with dedicated generation…
- Grid-Interactive Datacenter — A datacenter that dynamically modulates or shifts its compute load in response to…
- Compute-as-collateral financing — A financing structure where AI datacenter buildouts are funded by debt secured against…
- RDMA — Remote Direct Memory Access lets one node read or write another node's memory over the…
- Neocloud — A class of GPU-specialized cloud providers that build and rent out accelerator capacity…
- Distributed Hash Table — A decentralized key-value lookup structure that lets peers discover and route to each…
- DePIN — Decentralized physical infrastructure networks use token/economic incentives to…
- PagedAttention — A KV-cache memory-management technique pioneered by vLLM that stores attention keys and…
- Chunked prefill — A serving optimization that splits a long prompt's prefill into smaller chunks…
- Time to First Token — A serving-latency metric measuring the delay from request arrival to the first generated…
- Floating-Point Nondeterminism — IEEE-754 transcendental and accumulation results differ subtly across CPU/GPU, libm, and…
- Kernel fusion — Combining multiple GPU operations into a single custom kernel (e.g. FlashAttention-style…
- Kernel autotuning — Automatically searching over and generating GPU kernel variants to select the fastest for…
- Semantic caching — Caching LLM responses keyed by embedding similarity so semantically equivalent queries…
- Shadow deployment — Mirroring live production traffic to a candidate model so its quality, latency, and cost…
- Sparse attention — An attention technique that computes only a selected subset of query-key interactions…
- FlashAttention (IO-aware attention) — An IO-aware exact-attention algorithm that tiles the computation and avoids materializing…
- Context parallelism — A model-parallelism strategy that shards the sequence/context dimension across multiple…
- Trusted Computing Base — The minimal set of hardware, firmware, and software whose correctness is critical to a…
- Attention sink — Transformer models dump excess attention onto a few initial tokens, so those tokens'…
- Side-channel attack — An attack that recovers secrets by observing indirect microarchitectural or physical…
- Ring Attention — A distributed attention algorithm that shards a long sequence across devices and overlaps…
- Linear attention — Attention variants that replace the quadratic softmax with kernel/feature maps or…
- Grouped-query attention — An attention variant that shares key/value projections across groups of query heads to…
- Sliding-window attention — A long-context technique that restricts each token's attention to a fixed local window,…
- KV-Cache Quantization — Compressing or quantizing the transformer KV cache (per-token, per-channel, or via…
- Multi-head Latent Attention (MLA) — An attention variant that compresses keys and values into a low-rank latent vector,…
- Native sparse attention — A hardware-aligned, end-to-end trainable sparse-attention design that learns which…
- Model cascade — Route a query through a chain of increasingly capable and costly models, stopping at the…
- Expert parallelism — A parallelism strategy that shards a Mixture-of-Experts model's experts across devices so…
- KV-cache-aware routing — Load-balancing that steers a request to the inference replica already holding the largest…
- Knowledge distillation — Training a smaller student model to imitate a larger teacher model's outputs so it…
- Parameter-efficient fine-tuning — Adapting a frozen base model by training only small added weight matrices (e.g. LoRA…
- Self-hosting — Operating software on infrastructure you own and control rather than renting a managed…
- Open hardware — Computing hardware whose schematics, board layouts, or chip designs are released under…
- MicroVM — A lightweight hardware-virtualized sandbox (e.g. Firecracker-style) that boots in…
- Ephemeral Environment — A disposable, provisioned-per-task compute environment that is created fresh for an agent…
- Inter-Token Latency — The per-token generation delay (time-per-output-token) during decode, distinct from…
- Serving Goodput — The rate of LLM requests completed within their latency SLO (not raw throughput), the…
- Roofline model — A performance model that bounds a kernel's achievable throughput by either peak compute…
- Sandbox escape — A confined agent or workload breaching its isolation boundary (VM, container, or microVM)…
- Syscall filtering — Restricting which kernel system calls a process may invoke (e.g. seccomp-BPF or Landlock)…
- Control-flow integrity — A runtime exploit mitigation that constrains a program's indirect jumps and calls to…
- Application kernel — A user-space kernel (e.g. gVisor-style) that intercepts and re-implements a guest…
- Unikernel — A single-address-space image that compiles an application with only the OS library…
- Immutable Infrastructure — An operations paradigm where servers and environments are never modified in place but are…
- Coherent interconnect — A cache-coherent chip-to-chip or node-to-node link (e.g. NVLink-C2C in Grace-Blackwell,…
- Platform-native inference — Running AI locally through models and runtimes built into the operating system or…
- Second-Life Hardware — Repurposing decommissioned, consumer-grade, or e-waste accelerators for AI inference to…
- Permacomputing — A design philosophy that maximizes the useful lifespan of computing hardware and…
- Digital preservation — The practice of durably archiving and redundantly mirroring public datasets and content…
- LOCKSS (many-replica preservation) — Lots Of Copies Keep Stuff Safe — a preservation paradigm that keeps public data durable…
- Exceptional access (mandated backdoor) — A government-mandated interception or decryption pathway built into communications…
- Surveillance capitalism — The economic model in which firms extract behavioral and personal data as raw material…
- Data Rescue / Guerrilla Archiving — The organized community practice of mirroring and archiving at-risk institutional or…
- Small language model — A compact, few-billion-parameter model designed to run efficiently on-device or on modest…
- In-browser inference — Running ML models entirely client-side in the browser via WebAssembly/WebGPU, turning…
- Streaming ASR — Automatic speech recognition that transcribes audio incrementally in real time, emitting…
- Keyword spotting — An always-on, low-power on-device audio technique that continuously listens for a small…
- In-Database ML — Running machine-learning inference directly inside a database engine (e.g. executing…
- Scaling laws — Empirical power-law relationships predicting how model loss falls with training compute,…
- Silicon photonics — Optical interconnect fabricated on silicon that moves data between accelerators as light,…
- OS-native model API — Operating systems increasingly ship built-in, benchmarkable on-device model capabilities…
- TinyML / embedded inference — The practice of running machine-learning models on microcontrollers and other…
- Cloud Development Environment — A remote, fully-provisioned, standardized coding workspace (devcontainers, Codespaces,…
- Public AI infrastructure — The paradigm of funding open AI models and compute as institutionally-backed public goods…
- io_uring — A Linux asynchronous I/O interface using shared submission and completion ring buffers to…
- Memory-mapped weights — Loading model weights via mmap so the OS page cache streams them from disk on demand,…
- AI workstation — A desktop-class machine with large unified memory and integrated accelerators that runs…
- Weight serialization format — A standardized single-file container that stores model weights (often quantized) with…
- Model hub / weight registry — A registry that indexes, versions, and distributes openly released model weights and…
- 1-bit LLMs — Representing model weights in ~1.58 bits (ternary {-1,0,1}) so matrix multiplies collapse…
- Expert offloading — Serving a Mixture-of-Experts model on limited memory by keeping only the currently-active…
- Model merging — Combining the weights of multiple separately fine-tuned models (e.g. via linear…
- Capability density — The empirical trend (the 'densing law') that the capability achievable per model…
- Model pruning — Removing redundant individual weights or whole structural blocks from a trained network…
- Agent skill/plugin packs (by department) — Installable capability packs that extend a coding agent (Claude Code) with…
- Superpowers (Claude Code skill library) — A large community collection of Claude Code skills/harness patterns to make an agent more…
- Context7 (live, version-correct docs) — An MCP server that injects up-to-date, version-pinned library/framework documentation…
- Skill Creator (meta-skill) — An official Anthropic skill that scaffolds new skills — name, description, SKILL.md,…
- MCP Builder — An official Anthropic skill that scaffolds Model-Context-Protocol servers (tools the…
- Webapp Testing (browser sensor) — An official Anthropic skill that drives a real browser to exercise a web app (navigate,…
- Claude-Mem (persistent session memory) — A Claude Code plugin that persists/compacts memory across sessions so context survives…
- Web Artifacts (self-contained web pages) — An official Anthropic skill for building polished, self-contained web artifacts (inline…
- Brand Guidelines — An official Anthropic skill that applies a brand/design system (palette, type, voice)…
- UI UX Pro Max — A designer skill pack for higher-quality UI/UX output (layout, hierarchy, interaction…
- Taste / Frontend Design — A designer skill pack focused on frontend 'taste' — spacing, typography, and visual…
- Transitions (motion/animation) — A designer skill pack for web motion — page/scroll transitions and micro-interactions…
- Marketing pack (45 skills) — A 45-skill pack to run marketing end-to-end: copywriting, SEO, lead magnets, campaigns.…
- Social Media pack (17 skills) — A 17-skill pack for social: post writing, Reels, thumbnails. Overlaps the factory's…
- Finance pack (8 skills) — An 8-skill pack for finances: statements, reconciliation, audits. Not relevant to the…
- Small Business pack (31 skills) — A 31-skill pack to run a small business: cash flow, payroll, invoicing. Not relevant to…
- Legal pack (9 skills) — A 9-skill pack for legal work: contract review, NDAs, compliance. Not relevant to the…