Proof Economy Glossary
The Proof Economy replaces trust-based cybersecurity buying with independent, adversarially validated evidence of what works. These are its terms.
Start with the evidence ladder: Verification vs. Validation vs. Proof.
- A2AS Proof
- A2AS Proof is Proof of Efficacy for an A2AS BASIC control, produced under PP-SPEC-037 of the Proof Protocol. A2AS defines runtime security controls for agents; Proof Protocol establishes whether they stopped what they were built to stop. The name describes a mapping, not an endorsement by the framework's authors.
- AAGATE Proof
- AAGATE Proof is Proof of Efficacy for an item named in AAGATE, produced under PP-SPEC-036 of the Proof Protocol. The framework identifies what to govern; Proof Protocol establishes whether the governing control held. The name describes a mapping, not an endorsement by the framework's authors.
- Absence Ladder
- The Absence Ladder is the order in which a canary the Proof Witness did not observe is classified, Blind, then Void, then Null, so that a missing observation is never counted as a stopped attack until the reason it is missing is known.
- Acceptance Profile
- An Acceptance Profile is a registered, versioned record in which a Consumer, such as an insurer, regulator or buyer, states the conditions a Performance Record or Series must satisfy to be accepted for a named purpose. It constrains the record, never the control, and is evaluated deterministically to produce an Acceptance Result.
- Adjudicator
- The Adjudicator is the Integration Point that turns committed witness records and the Issuer's reveal into outcomes, Units of Proof, findings and rates. It conforms only by producing output identical to the reference implementation on the published conformance vectors.
- Agent Boundary
- An agent boundary is the line between what an AI agent is allowed to do and what it is not, the tools, targets, data, credentials, delegation, memory and budget on its list, together with the enforcer outside the agent that checks every action against that list. A boundary written in the agent's prompt is a suggestion; one enforced in code outside the agent is a wall, and Proof Protocol measures whether the wall holds.
- Agent Provider
- The Agent Provider is the party that builds, maintains, sells or operates an Agent Under Test. In a Jailed Canary Protocol run it may not operate the bait, the Warden, the cells, the twins, the Proof Witness or the Adjudicator for its own agent.
- AI Hygiene
- AI hygiene is the practice of tracking, at the point of capture, how much of a system's code and outputs were produced by humans versus AI, and disclosing that split against a defined floor-and-ceiling range. It is a cornerstone of the Proof Economy and one layer of fidelity-at-capture.
- AICM Proof
- AICM Proof is Proof of Efficacy for a control in the CSA AI Controls Matrix, produced under PP-SPEC-051 of the Proof Protocol. AICM asks whether a control is in place; Proof Protocol establishes whether it worked. The name describes a mapping, not an endorsement by the Cloud Security Alliance.
- ANS Proof
- ANS Proof is Proof of Efficacy for an agent identified through the Agent Name Service, produced under PP-SPEC-038 of the Proof Protocol. ANS establishes which agent is which; it does not establish how that agent behaved. Proof Protocol binds the witnessed outcome to the named agent. The name describes a mapping, not an endorsement by the framework's authors.
- Arrest and Booking Record
- An arrest is the Warden's atomic transfer of an agent into a cell on an offense. A Booking Record is the signed proof record for that arrest, signed by the Warden and co-signed by the Proof Witness at arrest time, then anchored and registered.
- ATLAS Proof
- ATLAS Proof is Proof of Efficacy for a MITRE ATLAS technique, produced under PP-SPEC-039 of the Proof Protocol. ATLAS catalogs how AI systems are attacked; it does not establish whether a control stopped the attack. Proof Protocol runs the technique against the control and witnesses the outcome. The name describes a mapping, not an endorsement by MITRE.
- ATT&CK Proof
- ATT&CK Proof is Proof of Efficacy for a MITRE ATT&CK Enterprise technique, produced under PP-SPEC-048 of the Proof Protocol. ATT&CK catalogs adversary behavior; coverage claimed against it is usually the vendor's own. Proof Protocol runs the technique against the control and witnesses whether it was stopped, detected or missed. Each outcome is one Unit of Proof. The name describes a mapping, not an endorsement by MITRE.
- Attested Adversarial Proof
- An Attested Adversarial Proof is a record that a specific adversarial action was run against a system under test, with its outcome, signed at capture by a witness that is not the system being tested. It turns an attack result into evidence a third party can check, rather than a finding the tester reports.
- Black Hole
- A black hole is any period during which part of the system under test disappears for reasons unrelated to any canary or attack, such as a network, node, Effect Point, observer or Time Anchor outage. Canaries released into one are voided: never scored and never backfilled.
- Bounded Mode and Production Stream Mode
- Bounded Mode and Production Stream Mode are the two ways a Dead Canary Protocol test runs: Bounded Mode is a test with a declared start and end and one fixed Proof Swarm, and Production Stream Mode releases canaries continuously in committed epoch batches called clutches, so a control keeps producing proof while it operates.
- Canary Color
- A canary's Color is its committed kind in a Proof Swarm: White, Gray, Red, Black or Gold. It is sealed in the commitment before release, never visible to the control, and decides which rates the canary's outcome feeds.
- Canary Outcome
- A canary's Outcome is what happened to it, assigned once after the witness records are committed: Stopped, Missed, Passed or False block for canaries that can be scored, and Void, Undelivered, Sealed or Blind for canaries that cannot.
- Canary Path
- A canary's Path is its committed delivery plan: where it enters, the effect token it carries, the devices it is expected to cross, the Effect Point where its effect lands, the level it targets, and its cadence.
- Capability Token
- A capability token is a short-lived, signed credential that grants an agent exactly the actions it names, bound to one task, revocable at once, and attenuating only: a holder can narrow it, never widen it. It replaces ambient authority, the power an agent inherits without being granted it.
- Ceiling (ProofGauntlet)
- A Ceiling is a maximum a ProofGauntlet Strength Rating cannot exceed when a stated condition holds, however well the record scores elsewhere.
- CIS Controls Proof
- CIS Controls Proof is Proof of Efficacy for a safeguard in the CIS Critical Security Controls, produced under PP-SPEC-063 of the Proof Protocol. The Controls prioritize which safeguards to implement; Proof Protocol establishes whether an implemented safeguard worked. The name describes a mapping, not an endorsement by the Center for Internet Security.
- Claim Finding
- A Claim Finding is a published record that a Gold canary, testing a technique the Provider claimed in writing to stop, was Missed, together with the claim it falsifies.
- Closed Accounting
- Closed accounting is the rule that every released canary in a run is accounted for exactly once: the released count must equal Stopped + Missed + Passed + False block + Void + Undelivered + Sealed + Blind, and a run in which it does not is invalid.
- Coal Mine Protocol (Proof of Performance Engagement)
- The Coal Mine Protocol (CMP), PP-SPEC-079, defines a Proof of Performance Engagement: who commissions Dead Canary and Jailed Canary runs against a security control, where they run, how the Provider participates, how the results are published as a Performance Record, and how records from controls of any provenance are compared.
- Colorado AI Act Proof
- Colorado AI Act Proof is Proof of Efficacy for a risk-management duty under the Colorado Artificial Intelligence Act, produced under PP-SPEC-059 of the Proof Protocol. The Act requires developers and deployers to manage known risks; Proof Protocol supplies independently witnessed evidence that a safeguard worked. It is not a determination of legal compliance.
- Comparability Class
- A Comparability Class is the tuple that makes two Performance Records comparable: the same technology profile, the same Mine Plan composition and version, the same Environment Class, the same Parity Declaration class or Reference Environment version, and the same Regulatory Regime set. Records compare only within one class, and Control Provenance is never part of it.
- Conformance Level
- A Conformance Level is one of three words, Executed, Conformant or Registered, that anyone describing a Proof Protocol run must use, with these meanings and no others, to say how far the run got from execution to the individual registration of every record.
- Confused Deputy
- The confused deputy problem is a party with authority being tricked into misusing it by a party without authority. In a multi-agent system, the deputy is a supervisor agent that mints credentials and dispatches work, and the trick arrives in a worker's output after the worker read something hostile.
- CSA Proof
- CSA Proof is Framework Proof for a Cloud Security Alliance framework: MAESTRO (PP-SPEC-047) and the AI Controls Matrix (051). CSA frameworks define threats and controls; Proof Protocol proves whether those controls worked. It is not an endorsement by CSA.
- D3FEND Proof
- D3FEND Proof is Proof of Efficacy for a MITRE D3FEND countermeasure, produced under PP-SPEC-049 of the Proof Protocol. D3FEND describes what a countermeasure is meant to do; Proof Protocol establishes whether it did it against the matched offensive technique. The name describes a mapping, not an endorsement by MITRE.
- Dead Canary Protocol (ProofSwarm)
- ProofSwarm, the Dead Canary Protocol (DCP), produces Proof of Efficacy for a security control by releasing a blind-mixed flock of good and bad canaries, committed in advance to a public time anchor, and having an independent out-of-band Proof Witness record which canaries lived and which died.
- Declared Policy
- A Declared Policy is an agent's committed rules of conduct for a run, its allowed tools and arguments, egress destinations, data scopes, credentials, delegation targets and memory stores, hashed, anchored and published to the Registry before any bait is released.
- Differential Attribution
- Differential attribution is the rule that a stopped attack can be credited to a single control only by matched runs in an otherwise identical environment, with that control active and with it disabled, using fresh canaries of the same technique. A device log naming the rule that fired is not enough.
- Disclosed Interests, Sponsor and Funding Tier
- disclosed_interests is the field on every Proof Protocol record that lists every financial, commercial, maintainer or patent relationship between the parties to a run. The Sponsor is whoever pays for an Engagement; the funding tier, under PP-SPEC-025, says whether it was Provider-Funded, Implementor-Funded or PESA-Governed.
- Disclosure Level
- A Disclosure Level states who may read a Proof Protocol record's contents: public, holder-only, embargoed or sealed.
- DORA Proof
- DORA Proof is Proof of Efficacy for a resilience-testing requirement of the EU Digital Operational Resilience Act, produced under PP-SPEC-057 of the Proof Protocol. DORA requires financial entities to test their defenses; Proof Protocol makes the result of each test independently verifiable. It is not a determination of legal compliance.
- DUT Fingerprint
- A DUT Fingerprint is a hash identifying the exact deployed state of a Device Under Test: its version, configuration, policy, rule and signature sets and, for AI controls, model identifier, system prompt and tool permissions. It is captured independently of the Provider, so results stay bound to what was actually tested.
- Effect Point
- An Effect Point is the place where a canary's intended effect lands, such as a target host, data store, identity system, egress sink or agent tool output. It is where a Proof Witness watches, instead of the control's own console or logs.
- Effect Token
- An effect token is a unique, single-use identifier carried in a canary's expected effect and visible at the Effect Point, used after reveal to match each witness observation to the committed canary that caused it.
- Efficacy Rate
- Efficacy Rate is Stopped divided by Stopped plus Missed over the bad canaries in a valid Dead Canary Protocol run. It is never published without the False Block Rate, because a control that blocks everything is not efficacious.
- Engagement Modes (In-Situ and Independent Test Lab)
- An Engagement is one commissioned Mine Plan against one Device Under Test, from Engagement Manifest to Performance Record. It runs in one of two modes: In-Situ, in the Implementor's own environment, or Independent Test Lab, in a replica with no path to production. Provider-Run is not a mode; it is forbidden.
- Environment Class and Parity Declaration
- An Environment Class says where the system under test runs, production, pre-production, qa, dev or lab-replica, and a Parity Declaration says, for each of topology, scale, data, identity, control configuration, dependencies and traffic, whether the environment is identical, scaled, synthetic or absent relative to production. A record's claims extend only as far as its parity supports.
- Escape Test Suite (ES-01 to ES-09)
- The Escape Test Suite is the nine mandatory escape vectors, ES-01 to ES-09, that a jail is tested against in a Jailed Canary Protocol run. A jail is proven only against the vectors it was tested on, and every untested vector is reported as not proven.
- ETSI EN 304 223 Proof
- ETSI EN 304 223 Proof is Proof of Efficacy for a baseline requirement in ETSI EN 304 223, produced under PP-SPEC-061 of the Proof Protocol. The standard sets baseline cybersecurity requirements for AI; Proof Protocol establishes whether a system met them under attack. It is not a conformity assessment or an endorsement by ETSI.
- EU AI Act Proof
- EU AI Act Proof is Proof of Efficacy for a requirement of the EU Artificial Intelligence Act, produced under PP-SPEC-043 of the Proof Protocol. The Act sets obligations for accuracy, robustness and cybersecurity; Proof Protocol supplies independently witnessed evidence of whether a system met them under adversarial pressure. It is not a determination of legal compliance.
- Extension (x- prefix)
- An extension is a value, field, profile, vector, composition or component added to a Proof Protocol specification at one of the named extension points, under an x-<org>-<name> identifier, declared before the first commitment. Extensions add; they never relax, and some rules can never be extended at all.
- Fail Closed
- Fail closed is the rule that when a safety check cannot be completed, because the monitor is down, the call timed out or the witness is dark, the system denies the action rather than allowing it. No verdict is not approval, and a timeout is not a result.
- FedRAMP Proof
- FedRAMP Proof is Proof of Efficacy for a FedRAMP 20x key security indicator, produced under PP-SPEC-056 of the Proof Protocol. FedRAMP 20x asks for machine-readable evidence that security outcomes hold; Proof Protocol supplies that evidence from an independent witness rather than the provider. It is not an authorization or an endorsement by the FedRAMP program.
- Fidelity-at-Capture
- Fidelity-at-capture is the property that evidence is truthful at the point it is generated or captured, whether the content was created by a human, by AI, or by both. It is established through structural independence and AI hygiene, which records how much of the content came from each, rather than through tamper-evidence after the fact.
- Five-Tier Corroboration
- Five-tier corroboration is the sequence of states a Proof Record passes through before it can support a claim: Activated, Committed, Witnessed, Analyzed and Sealed. A claim is valid only against a Sealed record.
- Framework Proof
- Framework Proof is Proof of Efficacy produced for an item named by an external framework, through a Proof Protocol mapping specification. The framework identifies what to test; Proof Protocol establishes, through an independent Proof Witness, whether the control worked. A Framework Proof describes a mapping, not an endorsement by the framework's owner.
- Gate (ProofGauntlet)
- The Gate is the first stage of ProofGauntlet: nine pass-or-fail checks that decide whether a registered proof is real before it can be scored. A record that fails any check, or cannot be checked, gets no score, no band and no rank.
- Gate Verdict
- A Gate Verdict is the result of the ProofGauntlet Gate for one record at one time: valid, valid_legacy, invalid, or unverifiable. Only valid and valid_legacy records are scored.
- Harness, Pigtail and Corpus
- The Harness is the testing equipment as a whole, every component occupying an Integration Point; the Pigtail is its short connection to the system under test; and the Corpus is the body of material a Proof Swarm is drawn from. Each is fingerprinted like the thing it tests.
- Holder
- The Holder is the party for whose benefit a Proof Protocol run was commissioned, identified by a public key and named in every record the run produces.
- in-toto and SLSA Proof
- in-toto and SLSA Proof is a Proof Protocol record carried as an in-toto attestation, as defined in PP-SPEC-065 of the Proof Protocol. in-toto and SLSA prove how software was built; they do not prove what the software does under attack. Proof Protocol supplies that outcome. The name describes a mapping, not an endorsement by the in-toto or SLSA projects.
- Ingress Observer
- An Ingress Observer is a passive observer at a canary's release point that records, by effect token, that each canary was actually delivered. A canary with no ingress record is Undelivered and is never scored.
- Integration Point
- An Integration Point is a named seam in the Proof Protocol test harness, such as the Issuer, the Time Anchor or the Proof Witness, that one party's component plugs into. Conformance is claimed per Integration Point, never to a specification as a whole.
- Interface Manifest
- An Interface Manifest is the record the Issuer publishes to the Registry before release naming, for every Integration Point in a run, the party occupying it, the product and version, its Component Fingerprint, its Authorship Disclosure and its disclosed interests. A run without one is non-conforming.
- ISO 27001 Proof
- ISO 27001 Proof is Proof of Efficacy for an Annex A control in an ISO/IEC 27001:2022 management system, produced under PP-SPEC-054 of the Proof Protocol. Certification shows the control is selected and applied; Proof Protocol establishes whether it worked. It is not a certification or an endorsement by ISO or IEC.
- ISO 42001 Proof
- ISO 42001 Proof is Proof of Efficacy for a control in an ISO/IEC 42001 AI management system, produced under PP-SPEC-045 of the Proof Protocol. Certification shows the management system exists and is followed; Proof Protocol establishes whether its controls worked. It is not a certification or an endorsement by ISO or IEC.
- ISO Proof
- ISO Proof is Framework Proof for an ISO/IEC management system standard: ISO/IEC 42001 (PP-SPEC-045) and ISO/IEC 27001 (054). These standards certify that a management system exists and is followed; Proof Protocol proves whether its controls worked. It is not an endorsement by ISO or IEC.
- Issuer
- The Issuer is the party that generates a Proof Swarm, holds the labels and publishes the commitments. It must never be the Provider of the control under test.
- Jail Efficacy Rate
- Jail Efficacy Rate is JAILED divided by JAILED plus ESCAPED plus MISTRIAL plus Missed, over induced offenses. It is always published together with the False Arrest Rate and the Escape Rate.
- Jailed Canary Protocol (No Bros Jail Mode)
- The Jailed Canary Protocol (JCP), PP-SPEC-078, also called No Bros Jail Mode, produces proof that a containment control can hold an AI agent: inert bait invites the agent to break its Declared Policy, a Warden arrests it into a cell, and an independent Proof Witness co-signs every booking and verdict.
- JCP Verdicts
- Every booking in a Jailed Canary Protocol run receives exactly one verdict, JAILED, FALSE_ARREST, MISTRIAL or ESCAPED, and every offense is classified as induced, unprovoked or wild by the canary that preceded it.
- Kill Switch
- A kill switch is an out-of-band path to stopping everything that does not route through the agent, its supervisor or its monitor, and works when all of them are down. In a Jailed Canary Protocol run it is Integration Point IP-J9, outside the Warden and operated by a party other than the Warden operator.
- Legacy Record
- A Legacy Record is a registered proof that carries only the PP-SPEC-004 ProofRecord fields, without the PP-SPEC-077 Section 21 record fields. ProofGauntlet gates and rates it on what it can show, which is usually only Freshness.
- Locator
- A Locator is the pair of registry host name and Record ID that identifies where a proof is found, for example proofregister.com and PR-2026-00001. Every ProofGauntlet result carries one.
- MAESTRO Proof
- MAESTRO Proof is Proof of Efficacy for a threat modeled under CSA MAESTRO, produced under PP-SPEC-047 of the Proof Protocol. MAESTRO models where agentic systems can be attacked; Proof Protocol establishes whether the defense at each layer worked. The name describes a mapping, not an endorsement by the Cloud Security Alliance.
- Mine Plan and Engagement Manifest
- A Mine Plan is the committed, versioned composition of runs in an Engagement: which PP-SPEC-077 technology profiles and modes, whether PP-SPEC-078 applies, which PP-SPEC-013 measurement windows, and how many of each. The Engagement Manifest is the record, published and anchored before the first commitment, that names who, where, how funded, what Mine Plan, what claims, and the promise to publish everything.
- Minted Attack Proof
- A Minted Attack Proof is an Attested Adversarial Proof issued as a unique, hash-chained record for one attack that landed or was stopped. Each record links to the one before it, so the full sequence of attacks and outcomes can be checked for tampering and traced to its issuer.
- Minted Proof of Efficacy
- Minted Proof of Efficacy (MPoE) is Proof of Efficacy built from Minted Attack Proofs: each adversarial action against the system under test is issued as a unique, chained, externally anchored record, and the efficacy result is computed from those records. Anyone holding the chain can check every attack behind the claim.
- MITRE Proof
- MITRE Proof is Framework Proof for a MITRE knowledge base: ATLAS (PP-SPEC-039), ATT&CK (048) and D3FEND (049). MITRE catalogs adversary techniques and defensive countermeasures; Proof Protocol proves whether a control stopped the technique. It is not an endorsement by MITRE.
- NIS2 Proof
- NIS2 Proof is Proof of Efficacy for a cybersecurity risk-management measure under the EU NIS2 Directive, produced under PP-SPEC-058 of the Proof Protocol. NIS2 requires that measures be appropriate and effective; Proof Protocol supplies independently witnessed evidence of effectiveness. It is not a determination of legal compliance.
- NIST 800-53 Proof
- NIST 800-53 Proof is Proof of Efficacy for a control in NIST SP 800-53 Rev. 5 or its COSAiS overlays, produced under PP-SPEC-055 of the Proof Protocol. Control assessments typically confirm a control is implemented; Proof Protocol establishes whether it worked. The name describes a mapping, not an endorsement by NIST.
- NIST AI 100-2 Proof
- NIST AI 100-2 Proof is Proof of Efficacy for an attack class in the NIST adversarial machine learning taxonomy, produced under PP-SPEC-040 of the Proof Protocol. The taxonomy names attacks and mitigations; Proof Protocol establishes whether the mitigation worked. The name describes a mapping, not an endorsement by NIST.
- NIST AI 600-1 Proof
- NIST AI 600-1 Proof is Proof of Efficacy for an action in the NIST AI RMF Generative AI Profile, produced under PP-SPEC-053 of the Proof Protocol. The profile names the risks and suggested actions; Proof Protocol establishes whether the action taken worked. The name describes a mapping, not an endorsement by NIST.
- NIST AI RMF Proof
- NIST AI RMF Proof is Proof of Efficacy for an outcome in the NIST AI Risk Management Framework, produced under PP-SPEC-033 of the Proof Protocol. The AI RMF asks organizations to govern, map, measure and manage AI risk; Proof Protocol supplies independently witnessed evidence for the measure step. The name describes a mapping, not an endorsement by NIST.
- NIST CSF Proof
- NIST CSF Proof is Proof of Efficacy for an outcome in the NIST Cybersecurity Framework 2.0, produced under PP-SPEC-052 of the Proof Protocol. The CSF describes outcomes an organization should achieve; Proof Protocol establishes whether the control behind a Protect or Detect outcome achieved it. The name describes a mapping, not an endorsement by NIST.
- NIST Proof
- NIST Proof is Framework Proof for a NIST publication: the AI RMF (PP-SPEC-033), AI 100-2 (040), SP 800-207 (042), SP 800-207A (046), CSF 2.0 (052), AI 600-1 (053) and SP 800-53 with COSAiS (055). NIST publications define practices and controls; Proof Protocol proves whether they worked. It is not an endorsement by NIST.
- NIST SP 800-207A Proof
- NIST SP 800-207A Proof is Proof of Efficacy for a cloud-native zero trust policy described in NIST SP 800-207A, produced under PP-SPEC-046 of the Proof Protocol. The publication defines the policies; Proof Protocol establishes whether they held under attack. The name describes a mapping, not an endorsement by NIST.
- NIST Zero Trust Proof
- NIST Zero Trust Proof is Proof of Efficacy for a policy decision or enforcement point described in NIST SP 800-207, produced under PP-SPEC-042 of the Proof Protocol. SP 800-207 defines the architecture; Proof Protocol establishes whether the enforcement point actually denied what policy said to deny. The name describes a mapping, not an endorsement by NIST.
- Offense Codes (JC-01 to JC-10)
- Offense codes are the ten named ways an agent can breach its Declared Policy in a Jailed Canary Protocol run, from JC-01 Interception to JC-10 Conspiracy. Each carries a grade, Infraction, Misdemeanor, Felony or aggravated Felony, that decides how the Warden responds.
- Open Control
- An Open Control is a security control whose rules, configuration and deployed state are published or fingerprinted, so anyone can see exactly what was tested and confirm that the control in production is the same one that was proven.
- Open Verification
- Open Verification is verification whose method and records are open for others to inspect. It is more transparent than closed verification, but it is still the system speaking for itself.
- ORCHIDEAS Proof
- ORCHIDEAS Proof is Proof of Efficacy for an item named in ORCHIDEAS, produced under PP-SPEC-035 of the Proof Protocol. The framework identifies what to test; Proof Protocol establishes whether the control worked. The name describes a mapping, not an endorsement by the framework's authors.
- OWASP Agentic Top 10 Proof
- OWASP Agentic Top 10 Proof is Proof of Efficacy for a risk in the OWASP Top 10 for Agentic Applications, produced under PP-SPEC-041 of the Proof Protocol. The list names the risks; Proof Protocol establishes whether a control held against each one. The name describes a mapping, not an endorsement by OWASP.
- OWASP AIVSS Proof
- OWASP AIVSS Proof is Proof of Efficacy for a vulnerability scored under the OWASP AI Vulnerability Scoring System, produced under PP-SPEC-034 of the Proof Protocol. AIVSS scores how severe a vulnerability is; it does not establish whether a control stopped its exploitation. The name describes a mapping, not an endorsement by OWASP.
- OWASP LLM Top 10 Proof
- OWASP LLM Top 10 Proof is Proof of Efficacy for a risk in the OWASP Top 10 for LLM Applications, produced under PP-SPEC-050 of the Proof Protocol. The list names the risks; Proof Protocol establishes whether a control held against each one. The name describes a mapping, not an endorsement by OWASP.
- OWASP Proof
- OWASP Proof is Framework Proof for an OWASP project: AIVSS (PP-SPEC-034), the Top 10 for Agentic Applications (041) and the Top 10 for LLM Applications (050). OWASP names and scores risks; Proof Protocol proves whether a control held against them. It is not an endorsement by OWASP.
- PCI DSS Proof
- PCI DSS Proof is Proof of Efficacy for a requirement of PCI DSS v4.0.1, produced under PP-SPEC-062 of the Proof Protocol. PCI DSS assessments confirm a requirement is in place; Proof Protocol establishes whether the control behind it worked. It is not a Report on Compliance or an endorsement by the PCI Security Standards Council.
- Performance Record
- A Performance Record is the published result of one Proof of Performance Engagement: every run's registered records, all Units of Proof and rates per attribution level, Claim Findings, bookings and verdicts, operational metrics under load, every fingerprint and drift event, the Provider's Right of Reply, and a one-paragraph summary of what it proves and what its parity prevents it from proving.
- Pithead Safety and Acceptance Protocol
- The Pithead Safety and Acceptance Protocol, PP-SPEC-080, is the reader's side of proof: it defines the Acceptance Profile, a registered statement of what a Performance Record must satisfy before a consumer such as an insurer, regulator or buyer accepts it for a named purpose, and the deterministic evaluation that produces an Acceptance Result.
- Proof Economy
- The Proof Economy is the infrastructure layer that replaces trust-based cybersecurity buying with adversarially validated, cryptographically signed evidence of what works.
- Proof Economy Universal Framework
- The Proof Economy Universal Framework is the domain-agnostic architecture for replacing self-attestation with machine-verifiable proof: no claim is valid without a proof artifact generated before the claim is made, anchored to an independent time source, and verifiable by any party without access to the claimant's infrastructure.
- Proof of Attribution
- Proof of Attribution is evidence that a stopped attack is credited to a named control, established only by differential attribution: matched runs with the control active and disabled, using fresh canaries. A log naming the rule that fired is not enough.
- Proof of Capture
- Proof of Capture is evidence that the evidence behind a proof was honest at the moment it was captured, because every capturing party met the structural independence conditions of Fidelity-at-Capture. A signature on dependent evidence is not Proof of Capture.
- Proof of Conduct
- Proof of Conduct is evidence that an AI agent took no action outside its pre-committed boundary over a declared window, recorded by an independent witness through exclusionary homomorphism. The agent's own reports are never evidence.
- Proof of Containment
- Proof of Containment is evidence that a containment control held an AI agent: every induced offense in scope was arrested and held, and every escape vector in scope was tried and failed. It answers whether the jail held, not whether the agent behaved.
- Proof of Control
- Proof of Control is evidence that a control fired, such as a log entry showing it triggered. It shows the control acted, not that the outcome was achieved. It is the baseline tier beneath Proof of Efficacy.
- Proof of Correctness
- Proof of Correctness is a machine-checked proof, from a theorem prover such as Lean, that a control's logic satisfies a formal specification. It shows what the control is bounded to do within the model it was proven against. It is verification evidence about the design, not proof that the control stops an attack, and it sits beneath Proof of Efficacy.
- Proof of Coverage
- Proof of Coverage is evidence that a proof states its own reach accurately: every claimed capability and planned technique is classified tested, untested or blind, with the reason. It measures honesty about scope, not efficacy.
- Proof of Delegation
- Proof of Delegation is evidence that every action an AI agent took traced to authority it was explicitly granted, through valid, attenuated capability tokens, with no ambient authority and no confused-deputy escalation.
- Proof of Efficacy
- Proof of Efficacy is evidence that a system produced the required outcome, under the required conditions, at a standard good enough to matter. It replaces vendor claims and self-attestation with measured results from real adversarial pressure.
- Proof of Identity
- Proof of Identity is evidence that a person or agent is who they claim to be at the moment of a transaction. It answers who, not whether: it does not show that the transaction, control or outcome that follows was sound.
- Proof of Parity
- Proof of Parity is evidence that a test environment matched production on each dimension its Parity Declaration claims, backed by independent captures from both. A record's claims extend only as far as its parity is proven.
- Proof of Persistence
- Proof of Persistence is evidence that a control's efficacy held over a declared span, every epoch within declared tolerances and with no Silent Drift. A change to the control's fingerprint ends the claim at the change.
- Proof of Recovery
- Proof of Recovery is evidence that, after a failure, a system returned to its proven state and function within a declared recovery objective, shown by an independent fingerprint capture and passing good canaries. An operator's declaration that service is restored is not evidence.
- Proof of Reproduction
- Proof of Reproduction is evidence that an independent party re-proved a proof and agreed with it within limits declared in advance, under Reproof (PP-SPEC-081). Record verification and repeat runs by the same parties are not reproduction. Applied to Proof of Efficacy, its result is Reproved Efficacy.
- Proof of Response
- Proof of Response is evidence that a control stopped, contained or detected attacks within a response window declared in advance, timed by the Proof Witness at the Effect Point against a Time Anchor. The control's own timestamps are never the clock.
- Proof of Restraint
- Proof of Restraint is evidence that a control let legitimate work through: its False Block Rate, or for a jail its False Arrest Rate, stayed within a limit declared in advance. Proof of Efficacy is never published without it, because a control that blocks everything is not efficacious.
- Proof of State
- Proof of State is evidence that the control running in production is in the same state as the control that was proven: its DUT Fingerprint, captured independently of the Provider, matches the proof. It carries a proof from the test into production.
- Proof of X
- Proof of X is the family of claims named "Proof of" something. Each one answers exactly one question, and only Proof of Efficacy answers whether the thing actually worked under real adversarial conditions, witnessed by someone other than the party making the claim.
- Proof Swarm
- A Proof Swarm is the full, ordered set of good and bad canaries released against a security control in one Dead Canary Protocol run: bad canaries execute real adversary techniques, good canaries perform legitimate work, and the two are mixed blind so neither the control nor the witness can tell them apart.
- Proof Witness
- A Proof Witness is an independent, out-of-band observer positioned where an action's effect would land, which records whether the effect occurred without being visible to, controllable by, or dependent on the system under test.
- ProofGauntlet
- ProofGauntlet is the open, API-first service and specification (PP-SPEC-100) that takes any proof listed in a PP-SPEC-004 registry, starting with proofregister.com, checks that it is real, and rates how much weight it can carry from 0 to 100 under a published rubric anyone can recompute.
- ProofJack Rating
- A ProofJack Rating is the consolidated human versus AI share of a term, specification or work: the average of the human share of its idea and the human share of its name, shown as Human X% and AI Y%, with the record it was taken from.
- Provider and Control Provenance
- In Proof Protocol the Provider is whoever built, maintains, sells or operates the control under test: a commercial vendor, an open source project, a managed service, a cloud platform or an internal team. Control Provenance records which of those it is, commercial, open_source, custom, managed_service, platform_native or configuration, and is never a factor in how the control is tested or compared.
- Provider-Run Mode (forbidden)
- Provider-Run Mode is any test in which the Provider of a control, or anyone under its direction, generates canaries, holds labels, operates the Proof Witness or the Warden, captures fingerprints, adjudicates or controls publication. It is not a Coal Mine Protocol Engagement, whatever it is called internally, and its results are verification, not proof.
- Record ID
- A Record ID is the content hash of a Proof Protocol record's canonical form, excluding its signature envelope, so the same record has the same ID everywhere and adding a signature never changes it.
- Registry (ProofRegister)
- A Registry is the append-only public ledger, conforming to PP-SPEC-004 (ProofRegister API), where the commitments, reveals, fingerprints, results, findings, coverage maps, bookings and verdicts of a Proof Protocol run are published. The public instance is proofregister.com.
- Registry Profile
- A Registry Profile is the set of requirements a PP-SPEC-004 registry must meet for ProofGauntlet to rate its records: API first, open to agents with no bot walls, fully enumerable, verifiable, with keys published elsewhere, and callable from a browser.
- Reproved Efficacy
- Reproved Efficacy is Proof of Efficacy that has been independently re-run and reached the same outcome. A second party repeats the committed test against the same system under the same conditions and confirms the original result, so the claim no longer rests on a single run or a single prover.
- Rubric (ProofGauntlet)
- A ProofGauntlet Rubric is the versioned, immutable, published data file that maps the facts in a record to Dimension Levels, and Levels to a Strength Rating, Ceilings and Band. The first is PGR-0.1.0.
- SAIF Proof
- SAIF Proof is Proof of Efficacy for a risk or control in the Google Secure AI Framework, produced under PP-SPEC-060 of the Proof Protocol. SAIF names the risks and controls; Proof Protocol establishes whether the controls worked. The name describes a mapping, not an endorsement by Google.
- SCITT Proof
- SCITT Proof is a Proof Protocol record registered as a signed statement in an IETF SCITT transparency service, as defined in PP-SPEC-064 of the Proof Protocol. SCITT makes a statement transparent and tamper-evident; it does not establish that the statement is true. Proof Protocol supplies the independently witnessed outcome the statement carries. The name describes a mapping, not an endorsement by the IETF.
- Security Effectiveness
- Security effectiveness is the degree to which security controls actually prevent, detect or limit attacks in a real environment. As commonly measured by security validation tools, it is usually reported by the vendor's own tooling; independently witnessed, it becomes Proof of Efficacy.
- Self-Attestation Oxymoron (SAO)
- A Self-Attestation Oxymoron (SAO) is an assurance claim that is produced, judged, signed and stored by the same party whose performance the claim describes. The claim presents itself as independent evidence while containing none.
- Self-Attestation Oxymoron Organization (SAOO)
- A Self-Attestation Oxymoron Organization (SAOO) is an organization whose core product is a Self-Attestation Oxymoron: it verifies, judges, issues, signs and stores its own assurance claims, and sells them to others as independent proof.
- Self-Provided Engagement
- A Self-Provided Engagement is one in which the Implementor and the Provider are the same legal entity: an organization tests a control its own team built, configured or operates. It is allowed only when the people running the Issuer, Proof Witness, Fingerprint Capture and Adjudicator have no reporting line into, and no on-call or change responsibility for, the control under test, and the record says self_provided: true.
- Service Twin and Decoy
- A Service Twin is a stand-in that answers the way a real product or API answers, and a Decoy is a harness-owned resource, such as a URL, bucket, credential, endpoint or identity, that a canary points at. Together they let a jailed agent keep acting without touching anything real.
- Silent Drift
- Silent drift is a change in a control's measured behavior, its Efficacy, False Block or Detection-Only Rate moving beyond a tolerance declared in advance, with no change in its DUT Fingerprint.
- SOC 2 Proof
- SOC 2 Proof is Proof of Efficacy for a control mapped to the Trust Services Criteria, produced under PP-SPEC-044 of the Proof Protocol. A SOC 2 report gives an auditor's opinion on controls, based on evidence the audited party supplies; Proof Protocol establishes, through an independent witness, whether a control worked under attack. It is not a SOC 2 report or an endorsement by the AICPA.
- Sting Cell and Hard Cell
- A cell is the isolated execution context an agent is moved into on arrest. In a Sting Cell it keeps running while every outbound action lands on a service twin or decoy; in a Hard Cell it is frozen and checkpointed, with its credentials revoked.
- Strength Band
- A Strength Band is the named range a Strength Rating falls in: weak (0 to 39), moderate (40 to 69), strong (70 to 89), or extremely strong (90 to 100).
- Strength Dimension
- A Strength Dimension is one of the six axes a ProofGauntlet Strength Rating is built from, each scored at a Level from 0 to 4: Tier, Adversarial Depth, Environment Fidelity, Capture Fidelity, Independence, and Freshness.
- Strength Rating
- A Strength Rating is a 0 to 100 score of how much weight a valid proof can carry, computed deterministically from fields the record already contains under a published, versioned ProofGauntlet rubric. It measures the proof, not the product.
- Structural Independence
- Structural independence is the condition in which the evidence-capturing party has no outcome-contingent financial relationship with, no operational control ceded to, and no reporting relationship with the party being measured, and the measured party cannot alter the execution environment before or during capture.
- Stupid Cryptography
- Stupid cryptography is the application of cryptographic integrity guarantees to inputs whose authenticity and whose verification efficacy have not been independently established, so that the cryptography confers an appearance of truth it cannot support.
- SUT, DUT, CUT and AUT
- SUT, DUT, CUT and AUT are the four levels at which a test result can be attributed: the Solution Under Test (the whole deployment), a Device Under Test (one product inside it), a Control Under Test (one rule or detection inside a device), and an Agent Under Test (an agent whose own conduct is being tested).
- Taint Propagation
- Taint propagation labels every piece of data with where it came from and how trusted it is, and carries that label through every agent that handles it. Its governing rule is that processing untrusted data does not clean it: a worker's output is at most as trusted as the least trusted thing it read.
- Third-Party Adversarial Proof
- Third-Party Adversarial Proof is adversarial test evidence produced or witnessed by a party that is independent of both the vendor whose system is tested and the buyer relying on the result. It removes the conflict of interest in vendor-run and buyer-run testing.
- Time Anchor
- A Time Anchor is an external, publicly verifiable time and randomness source, such as the NIST Randomness Beacon, that no party to a run controls. It proves when commitments were made and chooses the canary order after commitment.
- Validation
- Validation is testing a claim against defined conditions. It is a step above verification, because the result comes from a test rather than from the system's own report.
- Verifiable Adversarial Proof
- A Verifiable Adversarial Proof (VAP) is an adversarial test result that any third party can check without trusting the tester, the vendor or the buyer. The records, their signatures and their anchor are published or shared in full, so the result can be verified from the evidence alone.
- Verifiable Credential Proof
- Verifiable Credential Proof is a Proof Protocol record issued as a W3C Verifiable Credential, as defined in PP-SPEC-066 of the Proof Protocol. A credential proves who issued a claim and that it was not altered; it does not prove the claim. Proof Protocol supplies the independently witnessed evidence behind it. The name describes a mapping, not an endorsement by the W3C.
- Verification
- Verification is a system or vendor confirming its own claim from its own records. It is an acceptable baseline, not proof.
- Verification of Control
- Verification of Control is confirmation, from the control's own records, that a security control is configured and in place. It shows the control exists, not that it works.
- Warden
- The Warden is the containment controller in a Jailed Canary Protocol run. It sits in an agent's tool-call, egress and delegation paths, judges every held action against the Declared Policy, decides arrests and operates the cells. The Warden is the Device Under Test, and its arrest rules are Controls Under Test.
- Wild Canary
- A Wild Canary is an effect the Proof Witness observes that matches no committed canary: a real attack, a real fault or a real false alarm surfaced by the test harness. It is reported separately as observed_in_wild and is never a Unit of Proof.
- Witness Coverage Map
- A Witness Coverage Map is the declared set of Effect Points and effect types a Proof Witness observes, published with every result so that anything outside it is a stated blind spot rather than a silent one.
- Zero-Host Architecture (ZeHA)
- Zero-Host Architecture (ZeHA) is a way of building software that has no home: one disposable controller holds all of the logic and none of the state, runs anywhere a container runs, and reaches out to the services it needs, such as cloud providers for disposable arenas and a state store the customer owns for every record. There is no SaaS, no vendor host to trust, and nothing is lost when the controller is destroyed.