Fifty-eight terms in six clusters, each page carrying a one-breath definition, the etymology and backstory, the wider ecosystem, how codexproof implements it, an interview-grade summary line, academic references, and a figure. Terms interlink like a wiki; the questions and answers link into them.
Alphabetical
By unit
The six units below follow the paper: naming, signing, trust, provenance, systems, related work.
Unit A - Naming and integrity
Provenance node (the unit) - the atomic unit of the whole system - one small record per evidence event, wrapping the payload with its typed lineage links, data name, and times; the thing canonical bytes serialize, BLAKE3 hashes, and Ed25519 signs.
Sealing (the operation) - the five-step act that turns an unsigned node template into a SignedNode - serialize, name, frame, sign, staple - about fifteen microseconds of notarization at the moment evidence crosses a trust boundary.
cid (content-id) - a cid is the name the bytes give themselves - the fixed-size digest of a piece of content, used AS its identifier, so the name and the integrity check are the same object.
Content addressing - locating and naming data by WHAT it is (its digest) instead of WHERE it lives (a host and a path), which makes the data verifiable no matter who serves it.
BLAKE3 - a cryptographic hash function from 2020, built on the BLAKE2 lineage, that is extremely fast on ordinary CPUs because it hashes in a tree and can use SIMD lanes and threads.
Hash-then-sign - the standard construction where you sign the fixed-size digest of a message rather than the message itself, so signature cost stays flat no matter how big the content is.
Canonical bytes (canonicalization) - the single deterministic serialization of a structured object that everyone agrees to hash - necessary because hashing requires exact bytes, and the same logical object can otherwise serialize a hundred ways.
DAG (directed acyclic graph) - a graph whose edges have direction and which contains no cycles - the natural shape of history, derivation, and dependency, because effects cannot precede causes.
Merkle lineage - the family of designs descending from Ralph Merkle's hash trees, where a parent commits to its children by containing their hashes, so one trusted root authenticates an arbitrarily large structure.
Self-certifying names - names that carry the means of their own verification - you can check that data matches its name, or that a key owns its name, without asking any authority.
Unit B - Signing and identity
Digital signature - a mathematical scheme where a private key produces, over a message, a short proof that anyone holding the matching public key can check - authenticity and non-repudiation without shared secrets.
Keypair (scalar and point) - in elliptic-curve cryptography the private key is a scalar - a very large number - and the public key is a point on the curve, computed as scalar times a fixed base point; easy to compute forward, believed infeasible to reverse.
Ed25519 - the modern default signature scheme - Schnorr-style signatures over the twisted Edwards form of Curve25519, giving 32-byte keys, 64-byte signatures, microsecond speeds, and no random-number generator needed at signing time.
ECDSA (the contrast case) - the elliptic-curve version of the older DSA standard - ubiquitous, standardized, and carrying one famous footgun: every signature needs a fresh, secret, unpredictable nonce, and any slip leaks the private key.
Schnorr lineage - the signature design family started by Claus-Peter Schnorr - simple, provably secure in the random-oracle model, and linear in a way that makes batch verification and aggregation natural; Ed25519 is its most successful descendant.
Detached signature - a signature that travels alongside the content as an independent value rather than wrapping it or living inside it - verify by recomputing the message and checking, never by opening or excising.
did:key - the simplest W3C Decentralized Identifier method - the identifier IS the public key, encoded into a string, so identity needs no registry, no ledger, and no lookup, and verification works offline forever.
Domain separation - prefixing every signed or hashed message with a context string that says what protocol and version this artifact belongs to, so a signature minted for one purpose can never be replayed as valid for another.
Unit C - The trust plane
Anchor (trust anchor) - the key a verifier chooses to treat as axiomatic - the root of its trust decisions, from which every authorization chain must hang; in codexproof, one did:key per domain, held in each consumer's own list.
Trust schema - a rule set that decides, from NAMES alone, which keys may sign which data - turning authorization into a mechanical check on name relationships instead of a human judgment call.
LVS (Light VerSec) - the NDN project's compact language for writing trust schemas - declarative rules relating data-name patterns to key-name patterns, compiled and enforced by verifiers.
Namespace - a hierarchical name prefix that scopes authority - in codexproof, the slash-delimited prefix a domain owns, under which all its data names and signer names must nest.
Certificate chain - the sequence of signed statements linking a working key back to a trust anchor - each certificate saying "the holder of key A vouches that key B holds name N" - so trust flows from the axiom to the edge.
SRL (signed revocation list) - codexproof's revocation mechanism - the anchor signs the ENTIRE sorted set of revoked certificate identifiers as one object, stamped with a freshness window and a monotone version, and verifiers refuse to authorize anything without a fresh one.
Fail-closed - the design posture where missing, stale, or unverifiable security state causes DENIAL rather than acceptance - the system would rather stop than guess.
CRL and OCSP (the cautionary ancestors) - WebPKI's two revocation mechanisms - periodically published revocation lists, and an online status-checking protocol - both effectively neutered in practice because browsers treat their failure as ignorable.
X.509 and WebPKI (the contrast infrastructure) - the certificate format (X.
C2PA (the adjacent standard) - the Coalition for Content Provenance and Authenticity - the industry standard (Adobe, Microsoft, camera makers) for cryptographically signed manifests attached to media files, recording capture and edit history.
Unit D - Provenance and data
W3C PROV (and the typed links) - the W3C's 2013 standard vocabulary for provenance - entities, activities, and agents, connected by typed relations like wasGeneratedBy, used, wasDerivedFrom, and wasAttributedTo - a grammar for saying where things came from.
Envelope - the complete, self-contained verification package codexproof ships - the signed node bundle plus the certificates, anchors line-up, and revocation material a stranger needs to verify it offline, with nothing else to fetch.
Bundle - the ordered set of signed provenance nodes inside an envelope - the DAG serialized for transport, which the verifier re-walks node by node, checking bytes against cids and signatures against signers.
Selective disclosure - revealing some fields of a signed object while withholding others, without breaking the signature - done by committing to hidden fields with salted hashes so the signature covers commitments, not plaintext.
Salt and commitment - a commitment binds you to a value without revealing it (publish the hash now, open it later); a salt is the random ingredient that stops guessers from brute-forcing low-entropy values through the hash.
RAG (retrieval-augmented generation) - the architecture where a language model's answer is grounded by first retrieving relevant documents from an external store and stuffing them into the prompt - the model generates FROM evidence instead of purely from weights.
Vector store - a database indexed for nearest-neighbor search over embedding vectors - you query with a vector and get back the stored items whose vectors are closest, which operationally means "most semantically similar.
Embedding - a learned mapping from content - text, images, code - into a dense numeric vector such that semantic similarity becomes geometric closeness; the representation that makes vector search possible.
Chunk - the retrieval-sized piece of a source document - the unit that gets embedded, indexed, retrieved, and cited; small enough to fit prompts, large enough to carry meaning.
Unit E - Systems and transport
Thin waist - the architectural pattern where many diverse things above and many diverse things below all interoperate through ONE narrow, stable interface in the middle - the hourglass shape that let the Internet scale.
NDN (Named Data Networking) - the research architecture that rebuilds networking around requesting DATA BY NAME instead of connecting to hosts - every data packet signed, any node able to cache and serve, trust computed from names.
Gateway - codexproof's keyless coordination service - one plain HTTP router that fans interrogations across the operator's stores, assembles unsigned provenance templates, dispatches them to per-domain signing proxies, and returns sealed envelopes; routes, but cannot mint.
Signing proxy - the per-domain sidecar holding that domain's signing key - namespace- pinned, reachable only from inside, doing exactly one job: sign templates whose names it owns, refuse everything else.
JIT capture (just-in-time) - sealing provenance at the moment evidence is SERVED rather than when it is stored - the gateway builds templates from live query results and the proxies sign them on the spot, so the seal cost rides the interrogation path.
Loopback - the network interface that never leaves the machine - 127.
p50 and p99 (latency percentiles) - the 50th percentile (median - half of requests are faster) and the 99th (the tail - one in a hundred is slower), the two numbers that together describe how a system feels versus how it fails.
Egress (and the cost model) - data leaving a system's boundary - the bytes-out that clouds meter and bill; in codexproof's producer-side cost model, egress is essentially the whole bill.
WASM (WebAssembly) - a portable binary instruction format that runs at near-native speed inside browsers and beyond - the technology that lets codexproof's verifier run identically on a server, a laptop, and a web page.
MCP (Model Context Protocol) - Anthropic's open protocol (late 2024) for connecting AI agents to tools and data - JSON-RPC over defined transports, with servers exposing tools, resources, and prompts that any MCP-speaking client can discover and call.
A2A (Agent-to-Agent protocol) - the Google-initiated open protocol (2025, later a Linux Foundation project) for agents discovering and working with EACH OTHER - agent cards describing capabilities at well-known URLs, tasks with lifecycle, messages and artifacts between peers.
IPFS (the contrast system) - the InterPlanetary File System - the peer-to-peer content-addressed storage network (Protocol Labs, 2014) that made cids famous; tamper-EVIDENT by naming, but authorless: a hash proves what the bytes are, never who stands behind them.
Unit F - Related-work standards and operational terms (coverage-sweep additions)
in-toto - the academic-then-industrial framework for securing software supply chains by having every step of a pipeline sign metadata about what it did - a layout declares who may do which step, and link attestations prove each step happened as declared.
SLSA - Supply-chain Levels for Software Artifacts - Google's graduated framework (levels 1 through 3+) specifying how much provenance and build integrity an artifact carries, from "provenance exists" to "hardened, unfalsifiable builds.
Sigstore - the open service stack (Fulcio, Rekor, cosign) that made signing free and keyless for open source - short-lived certificates tied to OIDC identities, with every signature logged in a public transparency log.
SCITT - the IETF's Supply Chain Integrity, Transparency and Trust working group - an architecture for append-only transparency registries where issuers publish signed statements about artifacts and verifiers check registry receipts.
RATS and attestation - Remote ATtestation procedureS - the IETF architecture (RFC 9334) for one party producing EVIDENCE about the state of its hardware and software so a relying party can appraise it; attestation is the general act of a system vouching for its own configuration with cryptographic backing.
Verifiable Credentials - the W3C data model for digitally signed claims about subjects - driver's licenses, diplomas, memberships - designed to be presented and verified without calling the issuer, often with selective disclosure.
FLIC - File-Like ICN Collections - the NDN/ICN community's manifest format for representing large objects as content-addressed collections of segments, so big data travels as a signed tree of named pieces.
Interrogation - codexproof's name for the full evidence-producing query - ask the gateway a question, it fans across the stores, structures the results into provenance templates, gets them sealed, and returns answer plus envelope.
Delta (the /delta re-fetch path) - the incremental-fetch optimization - when a consumer already holds part of an evidence graph, ask only for what changed or what is missing, rather than re-shipping whole envelopes.
AIP (Agent Identity Protocol) - verifiable identity and delegation chains that travel along MCP and A2A invocations - the chain of command beside codexproof's chain of custody.
PROV-AGENT - the e-Science extension of W3C PROV to agent workflows - same vocabulary, resting on trusted infrastructure rather than signatures.
ICN (information-centric networking) - the umbrella tradition behind NDN: fetch named data, secure the object rather than the channel, let any holder serve it.
Secure Web Objects (SWO) - the NDN team's signed named objects without the network - the closest prior substrate; their key-locator says who signed, the paper's typed links add what it derived from.