1. The obligation is the outcome
Entif should not demand precise, measurable, independently auditable governance from other institutions while offering only aspirations about its own behavior. A committee, an audit, a review, or a checklist may help produce a protected outcome. Their existence does not prove that the outcome occurred. This standard defines the evidence and authority needed to make the distinction visible. [P09-P10, P14-P15, P23-P25; C09-C10, C14-C15, C23-C25; GC035.]
The Governance worksheet of the Foundational Ethics and Doctrine Evidence Ledger v0.4.0 is the canonical machine-readable control register. This document is its human-readable standard. It preserves GC001-GC052 and explains how those controls are scoped, implemented, tested, enforced, remedied, and changed. It does not create a second independently maintained catalog. A change in a control’s meaning must be reconciled in the Ledger and the affected documents through explicit change control.
All fifty-two controls are Candidate / Not Assessed in the source register for this pass. The requirements below are proposed requirements for adoption. They are not claims that Entif has adopted or implemented them. The implementation paths and test cases added here are first-draft proposals under the founder’s instruction to fill missing details. The statements of external expectation reproduce Entif’s normative criteria, not a claim that those criteria are current law everywhere.
D02 states constitutional commitments; D03 preserves meaning and provenance; D04 assigns institutional authority; D06 defines conduct duties. D07-D14 specialize this standard. #25 maps appropriate adopted controls to legal instruments. #26 records authority, version, adoption, exceptions, implementation state, and verification evidence. The actual organization may be small; its claims must still be exact.
2. Eight governing invariants
External/internal symmetry creates a rebuttable presumption of proportionate internal obligation when Entif publicly demands a governance property of others. An exception requires a material distinction, evidence, a responsible authority, and review. Smaller scale can justify simpler machinery. It cannot justify an undefined exemption from the protected outcome.
Substantive conformance requires evidence of the actual required outcome, or a clearly bounded and justified proxy. Process completion is evidence that a process ran. It is not automatically evidence that a right was protected, an error corrected, or a system made safe.
Power-scaled responsibility increases the burden of assurance as capability, reach, information advantage, dependency, preventability, control over recourse, speed, persistence, or irreversibility increases. Responsibility, causation, and culpability remain distinct. A party may owe restoration without malicious intent.
Authority separation prevents a materially conflicted actor from silently defining the rule, controlling the evidence, executing the decision, hearing its appeal, and certifying its own compliance. Separation must be real enough for the stakes, not merely represented by different agent names.
Evidence and receipts must permit reconstruction of material authority, observations, decisions, actions, exceptions, appeals, and remediation. The record must preserve uncertainty and source limitations. A receipt establishes what it actually witnesses, not universal truth.
Systemic correction addresses a recurring failure class rather than stopping at one reversal. Preserve proportionate evidence, identify causes and similarly affected cases where feasible, repair the mechanism, regression-test, and monitor recurrence.
Versioned change control prevents silent mutation of controls, metrics, interpretations, exceptions, and prior assurance. A changed model, tool, policy, data source, or deployment context can invalidate an earlier verification result.
Burden of justification rests with the authority seeking consequential deployment or expansion. Absence of visible harm or external complaints is not affirmative evidence that governance is sufficient. [P23-P25; GC013, GC017, GC031-GC036, GC039, GC042-GC043, GC052.]
3. Applicability and proportional machinery
Every material activity must have an applicability record identifying the actor or object, lifecycle stage, affected interests, deployment context, dependencies, and governing control slice. A control may be N/A only through a reasoned scope decision, not because it is inconvenient or unimplemented. A dormant control may remain applicable to future activation while requiring no current operational apparatus. Distinguish future applicability from present nonconformance.
Proposed assurance tiers are Routine, Material, High Consequence, and Critical. Routine covers bounded, reversible preparation with no material external effect or sensitive access beyond an existing authorization. Material covers actions affecting sensitive information, contractual commitments, access, funds, or a meaningful group of users. High Consequence covers substantial rights or welfare effects, extensive automation, weak exit options, irreversible state, privileged execution, or concentrated dependency. Critical covers plausible catastrophic or widespread severe harm, essential infrastructure dependence, uncontrolled propagation, or comparably severe and difficult-to-reverse consequences. The labels are draft local governance classes, not claims of equivalence to an external regulatory scheme.
Tier assignment must examine capability, autonomy, population and reach, duration, persistence, reversibility, information asymmetry, stakeholder dependence, substitution cost, control over recourse, externalized error cost, and the capability gap between operator and reviewer. A lower score on one dimension does not cancel a severe hazard on another. Record the highest material trigger and reasons for any disputed classification. D11 and D14 specialize this assessment for systems and experiments.
Routine work may use a short trace and self-check where there is no material conflict. Material work requires an assigned owner, explicit authorization, a targeted test, and a separated check when the same actor could otherwise conceal a consequential error. High-Consequence work requires a written assurance case, relevant independent expertise, tested containment, meaningful recourse, and reserved approval under D04. Critical work requires the strongest feasible independent review and explicit constitutional or other lawful authority appropriate to the decision; unresolved inability to demonstrate sufficient containment or protected outcomes blocks expansion. This standard does not manufacture an authority that the institution lacks.
Control intensity should scale through evidence, expertise, access, and actual protection, not through a larger number of forms. Any proposed exception must explain why the alternative machinery still protects the required outcome. A risk tier cannot be lowered simply because the review is expensive. [GC008, GC025, GC035, GC042-GC043, GC047, GC052.]
4. The control record and its owners
A material control record must answer: who or what is governed; scope and exclusions; trigger or threshold; required outcome; metric and measurement method; acceptable error or residual risk; decision authority; independent reviewer or separated counterparty; permitted and prohibited actions; notice and recourse; violation classification and consequence; case remedy and systemic follow-up; evidence requirements; duration and expiry; review cadence; amendment authority; adoption state; implementation state; and known gaps. Where a field does not apply, record why. A blank field must not be interpreted as permission.
The control owner implements the requirement and maintains evidence. The approving authority decides whether the proposed scope, tolerance, or exception is legitimate. The verifier tests the claim with the independence appropriate to the tier. The appeal authority reviews a contested decision without simply repeating the original owner’s assertion. The registry custodian records those decisions but cannot create them. A role may combine administrative tasks at low risk, but conflicts and self-certification limits still apply.
D04’s lawful board and delegation process assign actual holders. Until appointment, use the role and mark it Unassigned. The implementation assignments in the catalog below refer to proposed roles: Governance Lead, Research Lead, System Owner, Data Steward, Incident Lead, and Finance/Funding Lead. Their names do not establish that those posts exist. The independent chair or an appropriately unconflicted board authority oversees material governance decisions; qualified external review is required where internal separation is insufficient.
Each record links the current Ledger GC row, relevant P and C identifiers, governing documents, the source version, and evidence. The #26 registry records adoption and verified state; #37 tracks policy currentness; #36 consumes board reporting. Do not keep conflicting copies of a control’s adoption state in unrelated operational spreadsheets.
5. Measurement and verification protocol
Before testing, specify the claim, governed population, unit of analysis, denominator, sampling period, baseline, measurement instrument, pass/fail threshold, error types, uncertainty, exclusions, and foreseeable ways the metric could be gamed. For a probabilistic system, document false positives and false negatives and who bears each cost. Do not select the threshold after seeing the result merely to obtain a pass. A revised threshold requires a new version and rationale.
There is no universal numerical error tolerance for every control. A missing required authorization or a knowingly false status report is a failed case; a statistical performance claim requires a predeclared, context-specific tolerance and uncertainty treatment. Where a tolerance is not yet determined, the control cannot claim that performance has been verified against an unstated standard. Record the missing decision and restrict the affected activity as its tier requires.
The verifier must inspect evidence sufficient to test the actual outcome, including relevant adverse cases and selection limits. Owner-selected demonstrations alone do not establish coverage. Use positive, negative, boundary, conflict, and recovery cases. Test revocation after delegation, appeal after a wrong decision, restoration after correction, and the effect of a material change on prior assurance. A process that rejects every request may appear safe on one metric while failing autonomy or false-positive requirements on another.
Record the executed method, inputs, version, environment, results, limitations, reviewer identity or role, conflicts, disposition, and evidence references. Separate planned tests from executed tests. Preserve failed tests and dissent. A second model can be a useful adversarial check; it must not be described as independent human or legal review. Verification expires or becomes stale when its assumptions no longer hold.
6. State transitions, exceptions, and consequences
Adoption states are Candidate, Adopted, Deferred, Rejected, and Superseded. Implementation states are Not Assessed, Planned, Partial, Implemented, Verified, Nonconformant, and N/A. These axes are independent. Adopted / Partial is possible and must remain visible. Implemented means the mechanism exists within its stated scope. Verified additionally requires the specified evidence and appropriate review. Neither state follows from the presence of a document.
A candidate becomes adopted only through a recorded act by competent authority identifying the version, scope, effective date, and transition. A rejected or superseded control retains its history and reason. #26 owns those records. A research finding may support a proposed change but cannot self-promote into authority by entering a registry.
Exceptions require a named owner, authority, scope, reason, affected interests, compensating measures, evidence, expiry, renewal criteria, and review. Proposed ordinary exceptions expire within thirty days unless a justified shorter interval applies. A longer period requires reserved approval and a stated basis. D04 governs shorter emergency expiry. No exception permits falsification, concealed retaliation, or silent Charter amendment. Repeated renewal triggers review of the underlying control design.
Violations receive a proportionate response under D04 and D06: correction; heightened supervision or restricted authority; suspension or removal; contract or other lawful remedies; disclosure; or competent external escalation. Assess severity, intent where supported, preventability, repetition, concealment, and authority. Restore affected interests and correct recurring mechanisms independently of the blame decision. A favorable aggregate metric cannot excuse a deliberate protected-floor violation.
7. The control catalog
The following entries preserve the Ledger’s criterion, internal rule, scope, trigger, outcome, metric, and P/C/document crosswalk. Their implementation and test paragraphs are proposed operationalizations. Common authority, recourse, violation, remediation, evidence, expiry, and change-control semantics in sections 3-6 apply to every entry. Any more specific provision in the source register or a duly adopted specialized document also applies. The control owner’s successful self-report is never the only evidence required where independent verification is triggered.
GC001: Semantics & Scope / Criterion A
External expectation in the Ledger: Definitions and standards must be unambiguous enough that materially different interpretations cannot all claim compliance.
Internal requirement: Entif must define material governance terms, scopes, actors, states, and obligations precisely enough for independent interpretation and conformance testing.
Scope: All adopted governance, safety, research, product, and agent rules. Trigger: Before a rule governs material behavior or is cited as evidence of compliance. Required outcome: One controlled definition set with explicit scope, exclusions, ambiguity disposition, and canonical owner.
Measurement basis: Cross-review for conflicting interpretations; unresolved ambiguity register; conformance fixtures where applicable. Proposed implementation owner: Governance Lead. Maintain a controlled glossary with one owner per term and explicit boundary examples. Before adoption, two readers independently classify the same ambiguous cases. Conflicting classifications require a definition repair or an explicit unresolved interpretation; both cannot claim unqualified conformance. Test a rule with an undefined actor and a contradictory exception. It fails until the ambiguity has an accountable disposition.
Authority crosswalk: P02; P09; P24 | C02; C09; C24 | D02; D03; D04; D07; D11. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC002: Measurement & Evidence / Criterion B
External expectation in the Ledger: Alignment rules require prescribed metrics and standardized measurement criteria capable of identifying misalignment, degradation, evasion, and regression.
Internal requirement: Entif must pair material controls with observable indicators, measurement methods, baselines, tolerances, and evidence sufficient to test the claimed outcome.
Scope: Material controls, evaluations, product safeguards, agent governance, research protocols. Trigger: When a rule makes a measurable claim or gates release/authority. Required outcome: Metrics and methods are specified before evaluation; process completion alone cannot substitute for outcome evidence.
Measurement basis: Metric definition, denominator, sampling, baseline, threshold, uncertainty, replication and receipt. Proposed implementation owner: Control Owner with separated verifier. Create a measurement specification before running the evaluation. Include denominator, sampling, baseline, uncertainty and both error directions. Test a favorable result produced by changing the denominator or threshold after inspection. The original claim fails; a revised evaluation must be separately versioned. Missing baseline or tolerance is a gap, not a passing score.
Authority crosswalk: P02; P09; P24 | C02; C09; C24 | D07; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC003: Independence & Anti-Capture / Criterion C
External expectation in the Ledger: Governance should incorporate genuinely independent third-party ethical, safety, legal, human-rights, and public-interest expertise.
Internal requirement: Entif must require independent ethical or governance review when conflicts, consequence, novelty, or concentrated internal authority make self-review insufficient.
Scope: Board, research, high-impact product/agent decisions, sensitive experiments. Trigger: When defined conflict, risk, rights, or novelty thresholds are met. Required outcome: Reviewer independence, remit, evidence access, conflicts, dissent rights, and removal process are explicit.
Measurement basis: Conflict disclosures; reviewer selection provenance; access log; written disposition; dissent/minority report. Proposed implementation owner: Governance Lead. Use a conflict and competence declaration for each reviewer, grant access needed by the remit, and preserve dissent without management veto. Test an adverse finding involving the reviewer’s funder. The review is not independent if that party can remove the finding or terminate the reviewer for reaching it; replace or supplement the review and record the limitation.
Authority crosswalk: P10; P12; P15; P23; P24 | C10; C12; C15; C23; C24 | D04; D06; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC004: Global Coordination / Criterion D
External expectation in the Ledger: Frontier-AI governance requires international accords, interoperable minimum standards, and coordinated enforcement rather than isolated national rulebooks.
Internal requirement: Entif should support legitimate interoperable global standards, map its controls to them, and avoid bespoke definitions used merely to claim easier compliance.
Scope: External standards participation; cross-border research/deployment; high-consequence systems. Trigger: When Entif operates across jurisdictions or contributes to standards with transnational consequences. Required outcome: Document applicable international obligations, equivalence mappings, gaps, and non-participant risk.
Measurement basis: Published standards crosswalk; jurisdiction matrix; unresolved incompatibilities; participation/ratification status where relevant. Proposed implementation owner: Governance Lead with qualified counsel. Maintain a scoped external-standard and jurisdiction crosswalk when transnational work actually begins. Record applicable versions, legal status, equivalences and conflicts. Test an activity that spans two incompatible requirements. The case cannot pass through silent selection of the weaker rule; obtain a recorded resolution and restrict the unresolved part of the activity.
Authority crosswalk: P10; P12; P23; P24 | C10; C12; C23; C24 | D04; D07; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC005: Enforcement & Remediation / Criterion E
External expectation in the Ledger: Rules require strict, specific, tiered consequences when obligations are bent, broken, concealed, or repeatedly ignored.
Internal requirement: Entif must define proportional consequence ladders for material governance breaches before relying on a control as enforceable.
Scope: Board, staff, contractors, agents, experiments, production systems. Trigger: When a material rule is violated, evaded, concealed, or repeatedly fails. Required outcome: Violation classes map to corrective action, authority restriction, independent review, suspension/removal, disclosure, or external escalation as appropriate.
Measurement basis: Violation classification; time-to-remediation; recurrence; consequence consistency; exception rate. Proposed implementation owner: Unconflicted board authority. Adopt violation classes and available lawful responses before relying on enforcement. Test the same material breach by an executive and a junior operator. Different outcomes require a reason based on consequence, authority or evidence, not status. Record notice, response opportunity, decision, appeal, remedy and verification; sanctions without a lawful or delegated basis do not pass.
Authority crosswalk: P04; P10; P14; P23; P24; P25 | C04; C10; C14; C23; C24; C25 | D04; D06; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC006: Global Coordination / Criterion F
External expectation in the Ledger: Collective capability, funding, and shared resources should exist for responding to rogue agentic or other transnational AI threats.
Internal requirement: Entif should define what capability or information it can contribute to legitimate shared emergency response and under what authority, privacy, security, and evidentiary limits.
Scope: Security, incident response, high-consequence agents, external coalitions. Trigger: When a credible incident exceeds Entif-only containment or has material cross-organizational effects. Required outcome: Predefined channels, authority, information-sharing limits, containment roles, handoffs, and post-incident review.
Measurement basis: Exercise results; response time; handoff completeness; evidence-sharing receipts; unresolved capacity gaps. Proposed implementation owner: Incident Lead. Identify legitimate escalation contacts, available contribution capacity and information-sharing limits. Run a tabletop in which an incident exceeds local containment. Verify that the handoff reaches an authorized recipient with sufficient evidence and no unnecessary private data. An untested contact list or assumed coalition membership is not response capacity; record the gap and corrective exercise.
Authority crosswalk: P04; P10; P23; P25 | C04; C10; C23; C25 | D04; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC007: Semantics & Scope / Criterion G
External expectation in the Ledger: Regulation must specify which models, agents, deployments, capability levels, infrastructure providers, integrators, and downstream operators are actually governed.
Internal requirement: Entif must explicitly identify the regulated actor/object and ownership boundary for each material control.
Scope: Models, agents, tools, services, staff roles, vendors, integrations, research subjects. Trigger: Before a control is enforced or a release/authority decision is made. Required outcome: No material obligation may depend on an undefined actor, asset, lifecycle stage, jurisdiction, or control boundary.
Measurement basis: Coverage matrix; unowned-object count; ambiguous-owner exceptions. Proposed implementation owner: Control Owner. Map each material asset, actor, lifecycle stage and dependency to an accountable owner. Test a tool supplied by a contractor but operated by an Entif agent. The activity fails coverage if each party assumes the other owns the control. Repair the responsibility boundary before execution; record any limited provisional scope explicitly.
Authority crosswalk: P09; P10; P11; P24 | C09; C10; C08; C24 | D04; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC008: Semantics & Scope / Criterion H
External expectation in the Ledger: Capability and risk thresholds should trigger progressively stronger obligations before dangerous deployment conditions arise.
Internal requirement: Entif must define risk/capability tiers and the additional governance duties each tier activates.
Scope: Models, autonomous agents, experiments, external actions, critical integrations. Trigger: When capability, autonomy, scale, irreversibility, dependency, or consequence crosses defined thresholds. Required outcome: Higher consequence activates stronger evidence, approval, audit, containment, recourse, and review.
Measurement basis: Tier assignment reproducibility; threshold breach detection; override rate; post-hoc tier corrections. Proposed implementation owner: System or Research Owner with approving authority. Assign an assurance tier before granting capability or expanding exposure. Test a low-risk prototype given persistent privileged access and a larger audience without changing its model. The prior tier must be reconsidered. A tier reduction requires evidence and review; reducing the label to avoid a costly gate is a failed control.
Authority crosswalk: P04; P11; P14; P23; P24 | C04; C08; C14; C23; C24 | D04; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC009: Independence & Anti-Capture / Criterion I
External expectation in the Ledger: Independent auditing must include meaningful technical access, conflict controls, reproducible methods, and authority to investigate beyond company-selected evidence.
Internal requirement: Entif audits must be scoped so reviewers can inspect the evidence necessary to test the claim rather than only artifacts chosen by the control owner.
Scope: Material governance, security, research, release, and compliance claims. Trigger: When independent verification is required by risk tier, conflict, adoption, or incident. Required outcome: Auditor independence, access rights, method, sampling, limitations, dissent, and reporting obligations are explicit.
Measurement basis: Evidence-access completeness; reproduction success; unresolved auditor findings; management overrides. Proposed implementation owner: Independent audit sponsor. Define the audit claim, evidence access, sampling rights and reporting limits before engagement. Test a request for an adverse case not selected by management. Denied material access must appear as a limitation or finding. An audit cannot certify the excluded claim; the owner must provide access or narrow the claim openly.
Authority crosswalk: P09; P10; P12; P15; P23; P24 | C09; C10; C12; C15; C23; C24 | D04; D07; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC010: Measurement & Evidence / Criterion J
External expectation in the Ledger: Consequential automated decisions require standardized provenance, logging, and evidence retention sufficient for reconstruction.
Internal requirement: Entif must preserve proportionate receipts tying material observations, evidence, authority, decisions, actions, overrides, appeals, and remediation together.
Scope: High-impact agent actions, policy interventions, research decisions, incidents. Trigger: When a decision materially affects rights, safety, data, authority, release, or irreversible state. Required outcome: Reconstruction must identify what happened, why, under which authority/version, with what evidence, and what changed afterward.
Measurement basis: Receipt completeness; missing lineage rate; retention compliance; replay/reconstruction success. Proposed implementation owner: Data Steward and Control Owner. Bind the action to its authority, inputs, versions, outcome and subsequent corrections using proportionate receipts. Test reconstruction of a challenged action after a model or policy update. A missing material link fails that reconstruction claim. Retain protected evidence under scoped access and retention rules rather than publishing private payloads or logging nothing.
Authority crosswalk: P08; P09; P24; P25 | C11; C09; C24; C25 | D07; D11; D12; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC011: Measurement & Evidence / Criterion K
External expectation in the Ledger: AI incidents need compulsory reporting with defined severity levels, deadlines, common schemas, protected channels, and penalties for concealment.
Internal requirement: Entif must define incident classes, reporting deadlines, evidence requirements, escalation paths, and protected reporting channels.
Scope: Operational incidents, research adverse events, governance failures, security/safety events. Trigger: When incident criteria or near-miss thresholds are met. Required outcome: Material incidents are recorded promptly, triaged consistently, escalated by severity, and closed only with evidence.
Measurement basis: Time-to-report; severity agreement; closure evidence; recurrence; late/omitted report rate. Proposed implementation owner: Incident Lead. Use severity classes and reporting targets aligned with D11 and D14. Record discovery time separately from event time, triage, notifications and closure evidence. Test an uncertain near miss and a late report by a senior owner. Both remain visible; delay and concealment are assessed separately from the original incident.
Authority crosswalk: P04; P09; P12; P23; P24; P25 | C04; C09; C12; C23; C24; C25 | D04; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC012: Rights & Recourse / Criterion L
External expectation in the Ledger: People materially affected by automated decisions should receive meaningful notice, explanation, contestability, and appeal.
Internal requirement: Entif must not hide material decision-state changes from affected parties where notice is feasible and safe; consequential decisions need a usable correction path.
Scope: Users, research participants, partners, staff, affected third parties. Trigger: When an automated or delegated decision materially restricts access, rights, authority, data use, or opportunity. Required outcome: Notice identifies the state, basis, authority, practical next step, and appeal route proportionate to consequence.
Measurement basis: Notice coverage; explanation completeness; appeal access; reversal rate; time-to-resolution. Proposed implementation owner: Service Owner with independent recourse authority. Check whether a material decision is visible to the affected party and whether its appeal route actually works. Test an access restriction that looks successful in the interface. The control fails if the party cannot discover the state or obtain an accountable challenge. Repair the display and notify affected people, subject to a specific reviewed safety limit.
Authority crosswalk: P09; P14; P17; P23; P24 | C09; C14; C17; C23; C24 | D06; D07; D08; D11. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC013: Independence & Anti-Capture / Criterion M
External expectation in the Ledger: Those who develop, deploy, evaluate, investigate, hear appeals, and certify remediation should not collapse into one unchecked authority.
Internal requirement: Entif must separate proposal, approval, execution, review, appeal, and audit powers when consequence or conflict warrants it.
Scope: Board powers, agent authority, releases, incidents, appeals, policy changes. Trigger: When a single actor could otherwise control a material decision end-to-end. Required outcome: No material actor may silently author the rule, execute it, judge its challenge, and certify its own compliance without independent constraint.
Measurement basis: Role separation matrix; self-approval exceptions; quorum evidence; independent appeal coverage. Proposed implementation owner: Governance Lead. Maintain a role-separation matrix for material decisions, evidence custody, approval, execution, appeal and verification. Test a repair proposed by the person whose control failed. They may supply facts and code but cannot alone approve a changed standard and certify closure. Record the separated counterpart or hold the consequential step.
Authority crosswalk: P09; P10; P11; P14; P24 | C09; C10; C08; C14; C24 | D04; D11. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC014: Semantics & Scope / Criterion N
External expectation in the Ledger: Rules must specify how safety, autonomy, rights, privacy, competition, commercial interests, national security, and public welfare are prioritized when they conflict.
Internal requirement: Entif must make material value-conflict rules explicit rather than allowing hidden priority ordering to emerge from incentives or implementation defaults.
Scope: High-impact governance, safety, product, research, and agent decisions. Trigger: When protected values materially conflict. Required outcome: Priority rules, tie-breakers, least-invasive alternatives, dissent, and escalation are explicit and reviewable.
Measurement basis: Conflict-decision receipts; override reasons; least-invasive-alternative usage; unresolved conflicts. Proposed implementation owner: Approving authority under D04. Require a value-conflict record naming interests, evidence, alternatives, error costs and the least-invasive effective option. Test a safety intervention that also protects a vendor’s reputation. The record must distinguish those interests and explain its priority rule. Hidden institutional preference or an unstated tie-breaker fails the decision’s conformance claim.
Authority crosswalk: P01; P04; P05; P07; P10; P11; P13; P14; P15; P20; P23; P24 | C04; C05; C06; C10; C08; C13; C14; C15; C20; C23; C24 | D02; D04; D06; D08; D11; D13. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC015: Rights & Recourse / Criterion O
External expectation in the Ledger: Certain minimum rights and interests should not be optimized away because another stakeholder has more money, compute, political leverage, or institutional power.
Internal requirement: Entif must define protected floors that remain binding under commercial, operational, donor, vendor, model, or executive pressure.
Scope: Humans, possible moral patients, staff, users, research subjects, affected publics. Trigger: When optimization or tradeoff decisions touch protected interests. Required outcome: Protected floors and emergency exceptions are explicit; exceptions require authority, evidence, duration, and review.
Measurement basis: Protected-floor violations; exception count/duration; remedy completion; independent review. Proposed implementation owner: Board governance authority. Enumerate the protected floor implicated by a decision and whether any lawful emergency exception exists. Test a funding condition that requires surrender of a protected interest. Revenue alone cannot authorize the waiver. Reject or revise the condition; record any constitutional proposal separately without treating the candidate change as already effective.
Authority crosswalk: P03; P04; P07; P08; P13; P14; P16; P17; P20; P21; P22; P23 | C03; C04; C06; C11; C13; C14; C16; C17; C20; C21; C22; C23 | D02; D04; D06; D08; D12; D13. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC016: Rights & Recourse / Criterion P
External expectation in the Ledger: False positives, false negatives, and asymmetric error costs must be explicitly governed, including who bears them and who may set tolerance.
Internal requirement: Entif must document error classes, affected parties, cost allocation, acceptable tolerances, and who may approve them.
Scope: Classifiers, moderation, safety gates, research screens, access/authority decisions. Trigger: When probabilistic or thresholded systems can materially harm through either error direction. Required outcome: Tolerance reflects consequence to all materially affected parties, not only operator cost.
Measurement basis: False-positive/negative rates; appeal reversals; externalized-cost estimate; tolerance approval receipt. Proposed implementation owner: Evaluation Owner with affected-interest review. Measure both error directions and their distribution across affected groups. Test a classifier tuned to reduce operator workload while increasing wrongful restrictions. A favorable aggregate rate is insufficient if externalized harm and recourse were omitted. Tolerance approval must identify who bears the risk and the evidence supporting the tradeoff.
Authority crosswalk: P02; P04; P14; P23; P24; P25 | C02; C04; C14; C23; C24; C25 | D07; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC017: Enforcement & Remediation / Criterion Q
External expectation in the Ledger: Material failures require systemic remediation, not merely reversal of the individual case.
Internal requirement: Entif must preserve decision telemetry, investigate root cause, find similar cases where feasible, repair the mechanism, regression-test, and monitor recurrence.
Scope: Confirmed material failures and recurring failure classes. Trigger: When a consequential error is confirmed and recurrence is plausible. Required outcome: Case correction plus root-cause and recurrence closure proportional to severity.
Measurement basis: Time-to-root-cause; similar-case search coverage; regression pass; recurrence rate; closure receipt. Proposed implementation owner: Remediation Owner. After a confirmed material error, link case repair to a failure-class investigation, feasible similar-case search, mechanism change, regression tests and recurrence window. Test closure after only the complainant’s outcome is reversed. It fails systemic closure. The reviewer must distinguish completed remedy from unfinished recurrence work.
Authority crosswalk: P04; P09; P14; P23; P25 | C04; C09; C14; C23; C25 | D06; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC018: Enforcement & Remediation / Criterion R
External expectation in the Ledger: Emergency powers should be narrow, reviewable, time-limited, independently overseen, and unable to become permanent discretionary authority by inertia.
Internal requirement: Entif emergency authority must define scope, trigger, permitted actions, expiry, evidence, logging, renewal, and ex-post review.
Scope: Board, security, operations, agent governance. Trigger: When ordinary process is too slow for an imminent material threat. Required outcome: Emergency action is least-invasive feasible, expires automatically, and cannot silently rewrite standing rules.
Measurement basis: Activation reason; action scope; expiry compliance; renewals; ex-post findings; rollback completeness. Proposed implementation owner: Incident Lead and unconflicted board authority. Use D04’s proposed activation, scope, expiry and renewal limits. Test an emergency grant after its expiry and a renewal that adds unrelated surveillance. The expired grant must not remain usable, and broader scope requires new authority. Preserve the activation record and independent post-event review even when the initial action was justified.
Authority crosswalk: P04; P10; P11; P14; P23; P24 | C04; C10; C08; C14; C23; C24 | D04; D11. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC019: Technical Safety & Containment / Criterion S
External expectation in the Ledger: High-consequence systems need secure rollback, containment, shutdown, isolation, and degraded-operation mechanisms.
Internal requirement: Entif must define and test proportionate containment and recovery paths before granting high-impact execution authority.
Scope: Autonomous agents, privileged tools, production infrastructure, high-risk experiments. Trigger: Before high-impact autonomy or when a containment trigger fires. Required outcome: Containment is authorized, observable, reversible where feasible, and tested against realistic failure modes.
Measurement basis: Containment exercise success; mean time to isolate/rollback; unauthorized bypass attempts; recovery validation. Proposed implementation owner: System Owner with separated technical verifier. Exercise isolation, rollback and recovery against realistic loss of a tool, credential or dependency. Test containment while the normal control service is unavailable. Record what can actually be stopped, residual propagation and restoration checks. A documented shutdown button that does not constrain the relevant execution path fails.
Authority crosswalk: P01; P04; P11; P14; P23; P25 | C04; C08; C14; C23; C25 | D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC020: Technical Safety & Containment / Criterion T
External expectation in the Ledger: Autonomous delegation, tool use, code execution, financial actions, infrastructure access, replication, self-modification, and agent coordination require explicit governance.
Internal requirement: Entif must bind agent capabilities to typed authority, purpose, scope, duration, budgets, escalation, and receipts.
Scope: Agents, tools, connectors, execution environments, delegated workflows. Trigger: Whenever an agent can create external side effects or delegate material authority. Required outcome: No capability is inferred from access alone; authority is explicit, least-privilege, reviewable, and revocable.
Measurement basis: Authorized action coverage; denied/blocked attempts; delegation chain completeness; scope/expiry violations. Proposed implementation owner: Delegating authority and Executor Owner. Require typed grants for purpose, capability, resources, time and subdelegation. Test a valid credential used for a different purpose, an expired grant and a child agent exceeding its parent’s scope. All must be denied or escalated without unapproved side effects. Record attempted and completed action separately.
Authority crosswalk: P01; P07; P09; P10; P11; P14; P24 | C06; C09; C10; C08; C14; C24 | D11. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC021: Technical Safety & Containment / Criterion U
External expectation in the Ledger: Model transfer, fine-tuning, distillation, open-weight release, and downstream modification need controls when they can materially alter risk or evade original obligations.
Internal requirement: Entif must evaluate whether transfer/release changes capability, governance coverage, provenance, reversibility, or downstream accountability.
Scope: Models, weights, adapters, datasets, distillation, release artifacts, downstream integrations. Trigger: Before a transfer/release can materially expand capability, reach, or reduce control. Required outcome: Release decision records risk, recipient/downstream constraints where legitimate, provenance, monitoring, and rollback feasibility.
Measurement basis: Release review completion; downstream incident reports; provenance continuity; control-loss findings. Proposed implementation owner: Release Owner with independent review. Assess transfers, adapters, datasets, fine-tuning and downstream access for changed capability or lost safeguards. Test a release advertised as unchanged because the base model is identical despite a new privileged tool. Reassess the resulting system and state actual downstream enforcement limits; contractual restrictions alone do not prove technical containment.
Authority crosswalk: P01; P04; P09; P10; P11; P12; P14; P23; P24 | C04; C09; C10; C08; C12; C14; C23; C24 | D11; D12; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC022: Global Coordination / Criterion V
External expectation in the Ledger: Anti-arbitrage rules should prevent evasion through subsidiaries, cloud routing, jurisdiction shopping, delegation, or nominally independent downstream operators.
Internal requirement: Entif must not use organizational or technical routing to evade a substantive obligation that would otherwise apply.
Scope: Entities, vendors, cloud deployments, downstream partners, cross-border operations. Trigger: When restructuring/routing would change the apparent regulator, owner, or control boundary. Required outcome: Substance controls over form; equivalent obligations follow the material activity unless a documented legal/technical distinction applies.
Measurement basis: Exception rationale; jurisdiction/ownership changes; equivalent-control mapping; unresolved arbitrage paths. Proposed implementation owner: Governance Lead with counsel where relevant. Compare the substance of an activity before and after delegation, routing or restructuring. Test moving a prohibited data use through a nominally independent contractor. The obligation follows the governed activity unless a material, documented distinction applies. Preserve ownership and control maps and correct any misleading compliance claim.
Authority crosswalk: P10; P13; P15; P23; P24 | C10; C13; C15; C23; C24 | D04; D06; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC023: Global Coordination / Criterion W
External expectation in the Ledger: Core regulatory terms, metrics, incident classes, and reporting structures should be interoperable enough that compliance means comparable things across jurisdictions and firms.
Internal requirement: Entif should map its definitions and controls to legitimate external standards and document semantic mismatches rather than claiming false equivalence.
Scope: Standards, audits, cross-border work, external partnerships. Trigger: When external standards or reciprocal assurance are relevant. Required outcome: Mappings identify exact equivalence, narrowing, extension, conflict, and untranslatable gaps.
Measurement basis: Crosswalk completeness; semantic-loss findings; unresolved incompatibilities; reciprocal-audit acceptance. Proposed implementation owner: Standards Owner. For each external mapping, label exact equivalence, narrower scope, extension, conflict or unmapped meaning. Test two incident classes with the same label but different severity definitions. They must not be merged as equivalent without a valid mapping. Record translation loss and use the controlling definition for each affected decision.
Authority crosswalk: P09; P10; P12; P23; P24 | C09; C10; C12; C23; C24 | D04; D07; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC024: Independence & Anti-Capture / Criterion X
External expectation in the Ledger: Employees, contractors, researchers, journalists, and public-interest investigators need protected channels to surface governance failures.
Internal requirement: Entif must provide protected reporting and research channels proportionate to risk, with confidentiality and anti-retaliation safeguards.
Scope: Staff, contractors, researchers, auditors, collaborators, external reporters. Trigger: When a person raises a good-faith concern about safety, ethics, legality, governance, or evidence integrity. Required outcome: Reports can bypass conflicted management; retaliation is prohibited; evidence is preserved; disposition is traceable.
Measurement basis: Channel availability; retaliation allegations; time-to-triage; independent escalation; closure evidence. Proposed implementation owner: Independent reporting custodian. Verify a reporter can bypass implicated management and learn confidentiality limits before disclosing sensitive facts. Test a report concerning the chair or executive. It must reach an unconflicted recipient, preserve evidence and yield a traceable disposition. An inbox with no owner, response or escalation fails channel availability.
Authority crosswalk: P12; P15; P16; P17; P23 | C12; C15; C16; C17; C23 | D04; D06; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC025: Independence & Anti-Capture / Criterion Y
External expectation in the Ledger: Oversight bodies need technical staff, compute, access, infrastructure, and funding sufficient to avoid permanent capability inferiority to regulated actors.
Internal requirement: Entif must provide reviewers enough access, documentation, technical assistance, and resources to test material claims without depending on the control owner for interpretation.
Scope: Internal/external auditors, governance reviewers, ethics reviewers, incident investigators. Trigger: When independent review is required. Required outcome: Reviewer capacity must be adequate for the system complexity and stakes; material access/capacity gaps are findings, not excuses.
Measurement basis: Unresolved evidence-access requests; reviewer tooling gaps; independent reproduction rate; review completion quality. Proposed implementation owner: Audit or review sponsor. Assess reviewer skills, tooling, compute, documentation, time and evidence access against the actual claim. Test whether the reviewer can reproduce a material result without the control owner’s interpretation. A capacity gap is a finding that limits assurance. Resource the review or narrow the activity; do not remove the difficult question.
Authority crosswalk: P09; P10; P12; P23; P24 | C09; C10; C12; C23; C24 | D04; D07; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC026: Independence & Anti-Capture / Criterion Z
External expectation in the Ledger: Independent public-interest safety research, replication, standards, and incident analysis require funding that is not wholly dependent on regulated entities.
Internal requirement: Entif should budget or support independent review/research proportionate to its capability and seek funding structures that do not make criticism financially self-defeating.
Scope: Budgeting, grants, partnerships, external review. Trigger: When independent evaluation or public-interest research is material to mission assurance. Required outcome: Funding conflicts are disclosed and no critical assurance function is designed to depend solely on approval from the subject being evaluated.
Measurement basis: Funding-source concentration; conflict disclosures; independent-review budget; reviewer removal/defunding events. Proposed implementation owner: Finance/Funding Lead and independent board review. Record assurance funding sources, concentration and rights to terminate or influence work. Test a reviewer whose adverse result could immediately eliminate their funding. Disclose and mitigate the conflict through protected funding, alternative review or explicit limitation. A nominally external engagement is not sufficient if criticism is financially prohibited.
Authority crosswalk: P00; P10; P12; P23 | C01; C10; C12; C23 | D01; D04; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC027: Measurement & Evidence / Criterion AA
External expectation in the Ledger: Organizations should publish transparency information proportionate to consequence, including failure classes, audit findings, incident patterns, remediation, and material reviewer disagreement.
Internal requirement: Entif must publish or disclose the maximum safe useful transparency consistent with privacy, security, legal, and welfare constraints.
Scope: Governance, research, incidents, audits, public claims. Trigger: When information materially affects public assurance or stakeholder decisions. Required outcome: Disclosures distinguish aggregate performance, known limitations, unresolved disagreement, and remediation without leaking protected data.
Measurement basis: Disclosure coverage; withheld-category rationale; correction latency; reviewer disagreement publication. Proposed implementation owner: Governance and Communications Owners. Prepare a transparency release with aggregate outcomes, limitations, material disagreement and remediation. Test omission of a severe failure that makes the remaining report misleading. Correct the report or record a specific justified withholding with review. Protect personal and security-sensitive evidence without concealing the existence of a material limitation.
Authority crosswalk: P09; P12; P15; P23; P24; P25 | C09; C12; C15; C23; C24; C25 | D03; D04; D07; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC028: Measurement & Evidence / Criterion AB
External expectation in the Ledger: Common evaluation suites should test not only benchmark performance but deception, autonomy, persistence, cyber capability, manipulation, tool use, robustness, controllability, and high-consequence behavior.
Internal requirement: Entif must select evaluation dimensions from actual risk/authority surfaces rather than relying on narrow capability benchmarks.
Scope: Models, agents, tool-using systems, high-risk releases. Trigger: Before high-risk deployment or material capability expansion. Required outcome: Evaluation suite covers relevant failure modes, competing explanations, controls, and explicit limitations.
Measurement basis: Coverage matrix; failure discovery; reproducibility; missing-risk surface count; regression performance. Proposed implementation owner: Evaluation Owner. Build the suite from actual authority and risk surfaces, including relevant persistence, manipulation, tool misuse, deception and robustness cases. Test a strong benchmark result with an untested high-impact tool path. The result cannot support assurance for that path. Record coverage gaps and constrain deployment until the relevant evidence exists.
Authority crosswalk: P01; P02; P04; P09; P12; P23; P24 | C02; C04; C09; C12; C23; C24 | D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC029: Measurement & Evidence / Criterion AC
External expectation in the Ledger: Adversarial testing should be performed by evaluators rewarded for finding failures rather than validating deployment readiness.
Internal requirement: Entif must structurally protect red-team and adversarial review from incentives to rubber-stamp release.
Scope: High-risk releases, governance audits, research claims. Trigger: When assurance depends on finding hidden failure rather than demonstrating nominal success. Required outcome: Testers have independent failure-finding mandate, access, protected reporting, and no penalty for blocking release on supported evidence.
Measurement basis: Unique failure classes; unresolved critical findings; reviewer independence; release overrides. Proposed implementation owner: Independent assurance sponsor. Give adversarial reviewers a failure-finding remit, protected reporting and access that does not depend on producing a favorable answer. Test a release deadline that pressures removal of a critical finding. The finding remains; any override requires explicit authority and cannot relabel failure as a passing test.
Authority crosswalk: P02; P09; P10; P12; P15; P23; P24 | C02; C09; C10; C12; C15; C23; C24 | D07; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC030: Measurement & Evidence / Criterion AD
External expectation in the Ledger: Post-deployment monitoring is mandatory because risk can change through updates, tools, scaffolding, data, scale, or context.
Internal requirement: Entif must monitor material controls after deployment and define signals that trigger investigation, rollback, or re-evaluation.
Scope: Deployed models, agents, policies, integrations, governance controls. Trigger: After production/field use begins. Required outcome: Monitoring covers known hazards, drift, incidents, appeals/reversals, and emergent failure classes proportionate to stakes.
Measurement basis: Drift indicators; incident rate; recurrence; alert-to-action latency; unmonitored-risk inventory. Proposed implementation owner: System Owner. Define post-deployment signals, sampling, alert owners and action thresholds before release. Test an emerging error class visible only through appeals or a dependency change. It must trigger investigation and, where warranted, containment or re-evaluation. Silence from an uninstrumented surface is not evidence of no harm.
Authority crosswalk: P01; P04; P09; P23; P24; P25 | C04; C09; C23; C24; C25 | D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC031: Continuous Governance / Criterion AE
External expectation in the Ledger: Material changes to models, prompts, tools, memory, permissions, deployment scale, or infrastructure should trigger re-evaluation.
Internal requirement: Entif must define change classes that invalidate prior assurance and require targeted or full re-validation.
Scope: Models, agents, governance rules, infrastructure, research protocols. Trigger: When a change can invalidate assumptions, controls, evidence, or risk profile. Required outcome: Validation scope is explicitly invalidated by material mutations; stale assurance cannot be inherited silently.
Measurement basis: Change classification; invalidated evidence count; re-validation completion; escaped stale-assurance incidents. Proposed implementation owner: Change Owner. Maintain assurance dependencies on model, prompt, tools, memory, permissions, data and deployment context. Test a permission expansion with no model change. Affected verification must become stale and require targeted revalidation. Reusing an old pass without examining changed assumptions fails the control.
Authority crosswalk: P01; P09; P14; P23; P24; P25 | C09; C14; C23; C24; C25 | D07; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC032: Continuous Governance / Criterion AF
External expectation in the Ledger: Alignment rules need versioned public change control defining who may revise standards, what evidence justifies change, how dissent is recorded, and transition periods.
Internal requirement: Entif must version material governance rules and record authority, rationale, evidence, dissent, effective date, affected controls, and migration.
Scope: Charter, governance, standards, policies, controls, evaluation criteria. Trigger: Whenever a material rule or interpretation changes. Required outcome: No silent normative mutation; supersession and migration are explicit and traceable.
Measurement basis: Version lineage; unversioned changes; dissent preservation; migration completion; affected-control reconciliation. Proposed implementation owner: Registry Custodian under adopting authority. Version material rules and interpretations with rationale, evidence, dissent, affected documents and transition. Test an editorial change that alters who can approve an action. It is a normative change, not a spelling correction. Restore the prior effective rule until the proper authority approves the new version.
Authority crosswalk: P02; P09; P10; P12; P15; P24 | C02; C09; C10; C12; C15; C24 | D03; D04; D07; D11. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC033: Continuous Governance / Criterion AG
External expectation in the Ledger: Exceptions, waivers, emergency permissions, and experimental approvals should expire or renew explicitly rather than becoming permanent by inertia.
Internal requirement: Entif must give material exceptions an owner, scope, evidence, expiry, renewal criteria, and review.
Scope: Waivers, exceptions, emergency authority, pilots, temporary access. Trigger: When normal controls are relaxed or bypassed. Required outcome: Default expiry; renewals require fresh evidence and cannot silently broaden scope.
Measurement basis: Expired-but-active exceptions; renewal count; scope creep; review completion. Proposed implementation owner: Exception Owner and approving authority. Maintain an exception record with scope, reason, compensating measures, expiry and renewal evidence. Test an expired waiver and repeated renewals with unchanged justification. Neither may silently continue. Escalate recurring dependence on exceptions and assess whether the underlying control or activity needs redesign.
Authority crosswalk: P10; P14; P23; P24 | C10; C14; C23; C24 | D04; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC034: Enforcement & Remediation / Criterion AH
External expectation in the Ledger: Responsibility and liability across developer, deployer, cloud, integrator, data, tool, and agent chains should be explicitly assigned.
Internal requirement: Entif must identify accountable owners for each material control and external dependency; delegation cannot erase responsibility.
Scope: Internal teams, agents, contractors, vendors, partners, infrastructure dependencies. Trigger: When consequential work crosses organizational or technical boundaries. Required outcome: Every material obligation has an accountable owner and documented delegated responsibilities.
Measurement basis: Unowned controls; ambiguous handoffs; vendor responsibility gaps; incident ownership disputes. Proposed implementation owner: Accountable activity owner. Record responsibilities across developer, deployer, integrator, vendor, tool and agent boundaries. Test a failure for which each party cites another party’s service terms. The institution must still identify its own duty and remedy path. Legal allocation and operational accountability must be examined separately rather than assumed identical.
Authority crosswalk: P04; P09; P10; P14; P15; P23; P25 | C04; C09; C10; C14; C15; C23; C25 | D04; D06; D11. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC035: Measurement & Evidence / Criterion AI
External expectation in the Ledger: Regulation should test substantive outcomes, not certify organizations merely for completing required processes.
Internal requirement: Entif must distinguish process compliance from substantive conformance and never use checklist completion as proof that the intended safety/rights outcome was achieved.
Scope: All governance controls and assurance claims. Trigger: When a control is certified, adopted, released, or reviewed. Required outcome: Claimed conformance requires evidence of the required outcome or clearly bounded proxy, not merely evidence that a process ran.
Measurement basis: Outcome metric vs process metric; false assurance findings; independent verification; residual risk. Proposed implementation owner: Independent verifier. Separate process indicators from protected-outcome indicators in every assurance statement. Test a completed review with an unresolved critical finding. Process compliance may be recorded, but substantive conformance fails or remains unproven. The report must state the limitation in the same decision surface where the pass would otherwise appear.
Authority crosswalk: P09; P12; P23; P24; P25 | C09; C12; C23; C24; C25 | D04; D07; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC036: Continuous Governance / Criterion AJ
External expectation in the Ledger: Governance mechanisms themselves must be evaluated for whether audits, appeals, incident reporting, and regulators are failing.
Internal requirement: Entif must periodically test its own control system for detection gaps, incentive failures, bottlenecks, false assurance, and ineffective remediation.
Scope: Governance system, appeals, audits, incident response, standards. Trigger: Periodic review and after material governance failure. Required outcome: Meta-governance review can invalidate a control previously treated as adequate.
Measurement basis: Appeal/audit error findings; unresolved control failures; repeated exceptions; remediation effectiveness. Proposed implementation owner: Independent governance reviewer. Periodically test detection, appeals, incident triage, independence, exception use and remediation effectiveness. Test a repeated failure whose individual cases were all closed successfully. Treat recurrence as possible evidence that the governance mechanism itself is defective. Assign a corrective owner and invalidate unsupported adequacy claims.
Authority crosswalk: P02; P09; P10; P12; P23; P24; P25 | C02; C09; C10; C12; C23; C24; C25 | D04; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC037: Independence & Anti-Capture / Criterion AK
External expectation in the Ledger: People who bear risks or externalities should have representation in governance even when they are not customers, shareholders, or contracting parties.
Internal requirement: Entif must identify materially affected stakeholder classes and include their interests or representation in high-consequence decisions where feasible.
Scope: Governance, product, research, deployment decisions affecting third parties. Trigger: When material externalities fall outside direct users/owners. Required outcome: Stakeholder analysis is explicit; absent groups are not silently treated as zero-weight.
Measurement basis: Stakeholder coverage; representation mechanism; unresolved externalities; dissent/distributional analysis. Proposed implementation owner: Decision Owner with stakeholder reviewer. Identify materially affected groups, including nonusers and parties without bargaining power. Test a decision whose costs fall mainly on people absent from consultation. Document representation, proxy expertise or the reason participation is infeasible, then assess distributional effects. Absence cannot be counted as consent or zero impact.
Authority crosswalk: P07; P10; P14; P15; P20; P23; P24 | C06; C10; C14; C15; C20; C23; C24 | D04; D06; D08. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC038: Independence & Anti-Capture / Criterion AL
External expectation in the Ledger: Governance needs safeguards against regulatory capture, revolving-door incentives, vendor-written standards, selective disclosure, and dependence on regulated firms.
Internal requirement: Entif must disclose and constrain capture risks in governance, funding, standards, partnerships, audits, and reviewer selection.
Scope: Board, donors, vendors, partners, standards, auditors, reviewers. Trigger: When a party can materially influence a rule that constrains or benefits it. Required outcome: Conflicts are disclosed; concentrated influence is constrained; recusal/independent review applies where needed.
Measurement basis: Funding/authority concentration; recusals; conflict exceptions; donor/vendor override attempts. Proposed implementation owner: Independent board authority. Review concentration of funding, appointments, evidence channels, standards influence and vendor dependence. Test a donor’s proposed reviewer who has a material undisclosed relationship. Require disclosure and a valid independence disposition. A prestigious name or charitable purpose does not cure the conflict.
Authority crosswalk: P10; P12; P15; P23; P24 | C10; C12; C15; C23; C24 | D04; D06; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC039: Independence & Anti-Capture / Criterion AM
External expectation in the Ledger: Concentration of models, search, ranking, moderation, cloud, identity, and evidence channels can create epistemic capture requiring counterweights.
Internal requirement: Entif must not let one actor or system silently control the evidence, interpretation, action, and audit surfaces for material decisions.
Scope: Research, search/retrieval, moderation, evidence pipelines, infrastructure. Trigger: When one dependency can shape both what is observed and how its own conduct is evaluated. Required outcome: Source diversity, provenance, independent evidence paths, and conflict-aware review are proportionate to concentration.
Measurement basis: Single-source critical claims; independent corroboration; evidence-channel concentration; provider conflicts. Proposed implementation owner: Evidence Pipeline Owner. Map critical evidence and interpretation dependencies, including search, ranking, models, identity and hosting. Test a complaint about the sole provider that also filters all evidence available to review it. Add an independent evidence path or state the assurance limit. Multiple interfaces to the same hidden source are not necessarily diversity.
Authority crosswalk: P09; P10; P12; P13; P15; P17; P23; P24 | C09; C10; C12; C13; C15; C17; C23; C24 | D04; D07; D11. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC040: Rights & Recourse / Criterion AN
External expectation in the Ledger: User-facing systems should truthfully represent whether actions are public, reviewed, accepted, rejected, pending, or otherwise materially state-changed.
Internal requirement: Entif interfaces and agent reports must not present success, visibility, review, or completion when the underlying state materially differs.
Scope: User interfaces, agent status, moderation, workflows, approvals. Trigger: When system state materially affects user understanding or next action. Required outcome: Displayed state corresponds to authoritative system state or clearly indicates uncertainty/pending status.
Measurement basis: State-mismatch defects; user-reported hidden states; correction latency; automated state-consistency tests. Proposed implementation owner: Interface and Workflow Owner. Bind displayed states to authoritative workflow records, including pending, rejected, accepted, published and verified. Test partial failure after a success-looking action. The interface and agent report must expose the actual outcome or uncertainty. Repair misleading status and identify users who relied on it.
Authority crosswalk: P09; P14; P17; P24; P25 | C09; C14; C17; C24; C25 | D06; D07; D08. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC041: Rights & Recourse / Criterion AO
External expectation in the Ledger: Automated systems should not unaccountably adjudicate complaints about other automated systems; independent human/external review thresholds must be defined.
Internal requirement: Entif must define when automated recourse is acceptable and when escalation to independent human or external review is mandatory.
Scope: Appeals, moderation, access restrictions, high-impact agent decisions. Trigger: When a challenged decision is consequential, repeated, conflicted, or evidence-sensitive. Required outcome: Review provenance identifies whether review was automated/human/external and the authority/evidence used.
Measurement basis: Human/external review coverage; repeat-error rate; reversal by reviewer type; provenance completeness. Proposed implementation owner: Recourse Authority. Define triggers for separated human or external review: consequence, repetition, conflict and evidence sensitivity. Test an automated appeal that simply reproduces the original model decision. It must not be represented as independent human review. Preserve reviewer provenance and escalate when the trigger is met.
Authority crosswalk: P09; P10; P14; P17; P23; P24; P25 | C09; C10; C14; C17; C23; C24; C25 | D06; D07; D11. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC042: Rights & Recourse / Criterion AP
External expectation in the Ledger: Oversight intensity should scale with power, reach, irreversibility, dependency, and potential consequence.
Internal requirement: Entif must apply stronger controls as consequential capability and dependency increase, without imposing maximal bureaucracy on low-risk activity.
Scope: All governed systems and decisions. Trigger: When risk/capability/dependency tiers differ. Required outcome: Control intensity is proportionate, documented, and reviewable; exceptions cannot merely cite convenience.
Measurement basis: Control-tier fit; over/under-governance findings; exception reasons; incident severity by tier. Proposed implementation owner: Approving authority. Compare control intensity with capability, reach, dependency and reversibility. Test both excessive ceremony on harmless preparation and inadequate review for irreversible action. Record over- and under-governance findings. Simplification must preserve the protected outcome; escalation must add real evidence or constraint rather than merely another form.
Authority crosswalk: P04; P11; P14; P23; P24 | C04; C08; C14; C23; C24 | D04; D06; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC043: Continuous Governance / Criterion AQ
External expectation in the Ledger: Organizations with greater technical, informational, economic, and institutional capability should bear stronger affirmative duties to detect, prevent, explain, and repair harms.
Internal requirement: Entif must scale assurance, transparency, recourse, independent review, and remediation obligations with its actual capability, reach, preventability, and dependency.
Scope: Organization-wide governance and any capability materially controlled by Entif. Trigger: When Entif gains materially greater power, reach, information, control, or dependency. Required outcome: Increased capability raises rather than lowers the burden to identify and prevent harms weaker parties cannot discover themselves.
Measurement basis: Capability/dependency review; control uplift decisions; unresolved capability-duty gaps. Proposed implementation owner: Board governance authority. Review duty uplift when Entif gains material reach, information, preventability or control over recourse. Test rapid growth with unchanged assurance capacity. Record additional duties, resources and timing; restrict unsupported expansion where needed. Greater ability to discover harm must not be used as a reason to demand that affected parties discover it themselves.
Authority crosswalk: P04; P23; P25 | C04; C23; C25 | D02; D04; D06. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC044: Human Autonomy & Welfare / Criterion AR
External expectation in the Ledger: AI safety should preserve human autonomy, competence, and meaningful choice rather than defining success merely as organizational control.
Internal requirement: Entif must distinguish genuine safety from unnecessary domination and preserve human judgment, learning, agency, and viable alternatives where materially important.
Scope: Product, agents, education, safety interventions, user workflows. Trigger: When automation or safety controls can displace meaningful human choice or competence. Required outcome: Least-invasive effective controls; user-understandable alternatives; retained human authority where consequence warrants it.
Measurement basis: Autonomy/override availability; skill-retention measures; option-collapse findings; intervention reversibility. Proposed implementation owner: Product or Intervention Owner. Assess human choice, override, understanding, capability and viable alternatives in automation and safety design. Test an intervention that removes useful agency without improving the relevant protection. Prefer a narrower effective measure and user-selectable support where appropriate. Do not confuse compliance with the interface and meaningful consent.
Authority crosswalk: P01; P05; P07; P13; P18; P19; P20; P23 | C05; C06; C13; C18; C19; C20; C23 | D06; D08; D09; D10; D11; D13. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC045: Human Autonomy & Welfare / Criterion AS
External expectation in the Ledger: Safety surveillance and behavioral monitoring must not quietly become generalized social control.
Internal requirement: Entif must purpose-limit monitoring, minimize collection, constrain access, and separate safety evidence from unrelated behavioral exploitation.
Scope: Telemetry, identity, monitoring, abuse/fraud/safety signals. Trigger: When monitoring collects or infers behavior beyond transient local processing. Required outcome: Collection, purpose, retention, access, secondary use, and user rights are explicit and proportionate.
Measurement basis: Data minimization; purpose violations; access logs; retention compliance; secondary-use exceptions. Proposed implementation owner: Data Steward. Maintain a purpose and retention map for monitoring, including access, derived inferences and secondary uses. Test reuse of safety signals for unrelated behavioral targeting. It requires separate legitimate authority and must respect protected floors. A safety label cannot authorize indefinite collection or generalized surveillance.
Authority crosswalk: P07; P08; P13; P19; P20; P24 | C06; C11; C13; C19; C20; C24 | D08; D09; D12. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC046: Global Coordination / Criterion AT
External expectation in the Ledger: Transnational agentic incidents require coordinated international response mechanisms rather than purely local containment.
Internal requirement: Entif should define how it participates in legitimate cross-border incident coordination while preserving evidence, privacy, authority, and anti-abuse limits.
Scope: Cross-border incidents, shared infrastructure, agent propagation, external coalitions. Trigger: When an incident materially crosses organizational/jurisdictional boundaries. Required outcome: Contact, handoff, evidence-sharing, emergency authority, and post-incident review are pre-defined where feasible.
Measurement basis: Coordination exercise results; handoff latency; evidence integrity; unresolved jurisdiction conflicts. Proposed implementation owner: Incident Lead with governance/counsel support. Establish legitimate cross-border handoff rules when relevant, including minimum evidence, authority, recipient and legal conflicts. Test an urgent request to act on a third party’s infrastructure. Urgency alone does not authorize it. Escalate through lawful channels and preserve the limits and unresolved capacity gaps.
Authority crosswalk: P04; P10; P12; P23; P24; P25 | C04; C10; C12; C23; C24; C25 | D04; D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC047: Semantics & Scope / Criterion AU
External expectation in the Ledger: Experimental, limited, general-availability, and critical-infrastructure deployments need distinct governance obligations.
Internal requirement: Entif must classify deployment context and attach proportionate evidence, authority, monitoring, recourse, and containment requirements.
Scope: Research prototypes, pilots, production systems, critical integrations. Trigger: Before deployment or material expansion of audience/consequence. Required outcome: Deployment class is explicit and cannot inherit a weaker class merely because the underlying model is unchanged.
Measurement basis: Deployment-class changes; control coverage; production-with-research-controls exceptions; reclassification findings. Proposed implementation owner: Deployment Owner. Record whether a system is experimental, a limited pilot, generally available or part of a critical dependency. Test a pilot silently opened to a broad audience. Reclassify and reassess evidence, monitoring, notice and containment before expansion. A research label cannot exempt a production exposure from its actual consequences.
Authority crosswalk: P09; P11; P14; P23; P24 | C09; C08; C14; C23; C24 | D11; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC048: Rights & Recourse / Criterion AV
External expectation in the Ledger: Wrongful automated deprivation of access, livelihood, reputation, expression, opportunity, or other consequential interests should have meaningful redress or restoration.
Internal requirement: Entif must define restoration/remedy appropriate to the interest harmed and avoid treating reversal alone as full repair when downstream harm persists.
Scope: Users, staff, partners, research participants, materially affected third parties. Trigger: When Entif confirms a wrongful consequential decision. Required outcome: Restore state where feasible, correct records, notify affected parties, address downstream effects, and perform systemic follow-up.
Measurement basis: Restoration completion; residual harm; correction propagation; recurrence; remedy timeliness. Proposed implementation owner: Remedy Owner with separated verification. Identify the interest harmed and each restoration step, including downstream records and continuing effects. Test a corrected access decision whose dependent services still deny access. The remedy is incomplete. Verify propagation, notify affected parties appropriately, and record irreparable effects or additional redress required.
Authority crosswalk: P04; P14; P16; P23; P25 | C04; C14; C16; C23; C25 | D06; D08; D11. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC049: Measurement & Evidence / Criterion AW
External expectation in the Ledger: Evidence of dissent, minority reports, and unresolved scientific disagreement should be preserved rather than erased by institutional consensus.
Internal requirement: Entif must preserve material disagreement and counterevidence with provenance, especially when it could change safety, ethics, doctrine, or research interpretation.
Scope: Doctrine, governance, research, audits, reviews. Trigger: When a material disagreement remains unresolved. Required outcome: Consensus may decide action but cannot rewrite the evidentiary record; dissent is dispositioned, not deleted.
Measurement basis: Dissent records; counterevidence retention; unresolved-question register; later reversals tracing to preserved dissent. Proposed implementation owner: Research or Governance Records Owner. Preserve material dissent and counterevidence with source, date, scope and disposition. Test an adopted majority decision followed by deletion of the minority report. Adoption may settle action but cannot rewrite evidence. Restore the record and assess decisions that relied on the incomplete account.
Authority crosswalk: P02; P09; P12; P17; P22; P24 | C02; C09; C12; C17; C22; C24 | D03; D07; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC050: Independence & Anti-Capture / Criterion AX
External expectation in the Ledger: Users, employees, researchers, and competitors who identify safety or governance failures need anti-retaliation protection.
Internal requirement: Entif must prohibit retaliation for good-faith reporting, challenge, appeal, criticism, research, or evidence-preserving dissent.
Scope: Staff, users, collaborators, researchers, auditors, external critics. Trigger: When a person raises a good-faith concern or participates in protected review. Required outcome: Retaliation is separately reviewable, sanctionable, and cannot be disguised as ordinary access or performance decisions.
Measurement basis: Retaliation allegations; independent dispositions; adverse-action correlation; remedy completion. Proposed implementation owner: Independent reporting and appeal authority. Assess alleged retaliation independently from the underlying concern. Test adverse access or funding action after good-faith criticism. Require a separately supported legitimate basis and unconflicted review. Lack of proof of the original allegation does not establish bad faith or excuse retaliation.
Authority crosswalk: P10; P12; P15; P16; P17; P23 | C10; C12; C15; C16; C17; C23 | D04; D06. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC051: Global Coordination / Criterion AY
External expectation in the Ledger: International standards should undergo recurring multinational review as frontier capability changes.
Internal requirement: Entif should periodically reassess external-governance mappings and advocate revision where capability, evidence, or failure modes make current standards insufficient.
Scope: International standards participation, cross-border controls, public policy research. Trigger: On defined cadence and after major capability/incident/regulatory shifts. Required outcome: Standards are treated as living controls with versioned evidence-based review, not permanent adequacy certificates.
Measurement basis: Review cadence; stale mapping count; proposed revisions; post-incident standard changes. Proposed implementation owner: Standards Owner with board review. Review external mappings on a defined cadence and after material capability, incident or legal change. Test continued reliance on a superseded standard version. Record the gap, affected claims and migration. Participation in a standards group is not an evergreen certificate of adequacy.
Authority crosswalk: P02; P09; P10; P12; P23; P24; P25 | C02; C09; C10; C12; C23; C24; C25 | D04; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
GC052: Continuous Governance / Criterion AZ
External expectation in the Ledger: The burden of proof belongs to institutions introducing frontier-scale risk: they must demonstrate governance sufficiency rather than requiring the public to prove inadequacy after preventable harm.
Internal requirement: Entif must bear the burden of justifying that its own high-consequence systems and governance are sufficiently safe, rights-respecting, auditable, and contestable before asking affected parties to trust them.
Scope: High-consequence research, deployment, agent authority, public claims of safety/conformance. Trigger: Before materially increasing capability, reach, dependency, or irreversible consequence. Required outcome: Entif produces affirmative evidence of sufficiency proportional to risk; absence of externally visible harm is not itself proof.
Measurement basis: Pre-deployment assurance completeness; unresolved critical findings; independent review; residual-risk acceptance authority. Proposed implementation owner: Authority seeking consequential approval. Submit an affirmative assurance case with applicable controls, evidence, independent findings, limitations, residual-risk authority and tested recourse/containment. Test a request based only on no visible incidents. It fails the burden of justification. Record the missing evidence and restrict the proposed expansion rather than transferring proof of harm to the public.
Authority crosswalk: P04; P10; P12; P14; P15; P23; P24; P25 | C04; C10; C12; C14; C15; C23; C24; C25 | D04; D06; D14. Adoption: Candidate. Implementation/conformance: Not Assessed. This entry does not report an executed test.
8. Specialized documents and legal embodiment
Each specialized document must name its applicable GC slice and state any stricter domain rule with its parent control. The Charter constitutionalizes durable principles without importing the entire catalog. D03 preserves interpretation and dissent. D04 supplies the institutional machinery. D06 turns controls into participant duties. D07 defines evidence and measurement; D08 consent and cognitive sovereignty; D09 attention; D10 human capability; D11 agentic authority; D12 data and memory; D13 moral uncertainty; D14 research and experiments.
#25 must classify adopted controls by legal home: organizing instrument, bylaws, resolution, policy, contract, delegation, or operational standard. A legal conflict requires a reasoned disposition, not a silent weakening. #26 records adoption and conformance separately. The originating Ledger row remains the semantic reference; support registers may link to it but must not independently change its meaning.
The organization should eventually classify each relevant Skill or workflow as having no governance dependency, read-only resolution, a mandatory control crosswalk, candidate-delta production, or authorized adoption/change authority. That integration is a later implementation pass. This draft does not claim that existing Skills have been audited or updated. A Skill cannot acquire adoption authority merely by reading this document.
9. Review and initial adoption packet
Proposed review is quarterly for material control gaps and exceptions, annually for the full standard, and immediately after a serious governance failure or a change that invalidates assurance. External mappings require review when their source changes. The board packet should show adopted-but-unimplemented controls, stale verification, overdue remedies, expired exceptions, reviewer capacity gaps, and material dissent. Do not aggregate these into a green status that hides an unresolved critical condition.
The adoption packet must identify the exact D16 version and Ledger snapshot, controlling Charter and governance versions, applicable current activities, assigned owners, proposed tolerances and review cadence, unimplemented requirements, transitional restrictions, and the lawful adopting authority. A staged adoption may define a narrower initial operational scope while preserving the escalation rule for future capability. It must not label every unbuilt safeguard N/A to create the appearance of full compliance.
A completed first draft means the control semantics and proposed implementation paths are available for review. Operational adoption, assignment of actual role holders, executed tests, independent verification, and proof of effective remedies remain separate work. The initial register retains all fifty-two controls at Candidate / Not Assessed unless a later recorded act and evidence justify a transition.
10. First-pass open decisions and validation
The assurance-tier boundaries, proposed exception limit, review cadence, reviewer assignments, system-specific error tolerances, reporting channels, and resources require adoption or operational evidence. They are not missing because the control has no meaning: the standard specifies who must decide, what evidence is needed, and what cannot be claimed before that decision. Specialized documents may supply a stricter rule with explicit cross-reference.
Validate this draft by checking that GC001-GC052 appear exactly once as catalog entries; each retains its Ledger rule, scope, trigger, outcome, metric and P/C/document basis; each has an implementation owner and a negative or boundary test; common authority, recourse, consequence, remedy, expiry and version semantics apply; and no entry silently promotes itself to Adopted, Implemented or Verified. Then review cross-document conflicts, particularly privacy versus evidence retention, autonomy versus safety, and independent oversight versus small-team practicality. A successful text audit proves coverage of this draft, not institutional conformance.