What this commentary can and cannot do
This guide interprets the first-draft D02 Charter, clauses C01-C25. It explains provenance, intended meaning, excluded interpretations, and unresolved questions. It cannot amend the Charter by sounding more specific than the Charter. Where a proposed implementation would create a new obligation, that proposal must be adopted through the appropriate document and authority rather than hidden in commentary.
The Foundational Ethics and Doctrine Evidence Ledger v0.4.0 controls this pass. D01 provides the longer historical narrative. P identifiers designate doctrine and its provenance; E and S identifiers designate evidence and sources; H identifiers designate research hypotheses; C identifiers designate candidate clauses; and GC identifiers designate governance controls. An identifier is a reference, not an automatic promotion into authority.
Historical statements below describe the reviewed record. Direct founder statements, formal design commitments, observed behavior, assistant synthesis, and open hypotheses have different evidentiary roles. A date in the Ledger is not proof that a belief originated on that date. An architecture is not proof of implementation. An illustrative example in this guide is a proposed interpretation test, not an event that occurred. [Guide; Definitions; P02, P09; GC001, GC027, GC032, GC049.]
C01: Public benefit
The clause establishes the institution’s purpose and a constraint on private capture. It connects the current nonprofit direction in P00 with anti-capture, open public-interest work, and conflict transparency in P10, P12, and P15. The early evidence records a privacy- and user-control-oriented product design. It does not establish that nonprofit legal form was always intended. [E035-E039, E070.]
Public benefit is not whatever the organization happens to sell, publish, or enjoy building. A proposed activity should identify who benefits, how the benefit is expected to arise, who bears costs or risks, and why the activity belongs within the mission. The legal definition of exempt purpose, if tax-exempt recognition is pursued, belongs in counsel-reviewed instruments; this commentary does not declare eligibility.
The clause does not prohibit compensation, sustainable revenue, protected intellectual property, or work with commercial partners. It prohibits treating institutional income or private advantage as a sufficient reason to compromise the mission. The unresolved design question is how to measure public benefit across research, infrastructure, education, and governance without rewarding easily counted activity over meaningful outcomes. D04, D16, and #41 must make that tradeoff inspectable.
C02: Epistemic integrity
P02, P09, and P15 support a separation between what was observed, what was inferred, what is hypothesized, and what is chosen as a value. The evidence corpus contains corrections to earlier overclaims and explicit alternatives to favored interpretations. Those counterweights are part of the provenance, not embarrassing material to remove. [E014-E016, E045, E053, E068-E069.]
The clause permits strong conclusions when evidence supports them. It does not require endless hedging, manufactured balance, or equal weight for every possible explanation. It requires the strength and type of the claim to remain visible. An institutional value can be stated firmly without pretending it was experimentally measured. A reported behavior can be documented without pretending its cause is known.
A proposed test is whether a reader can distinguish a measured result from an architectural intention and a model-generated explanation from independently established mechanism. D07 owns those operational tests. The unresolved question in a disputed case is not whether certainty would be convenient, but which evidence would justify changing the classification.
C03: Care under moral uncertainty
P03 and P14 establish a precautionary argument: unresolved moral status does not, by itself, license unrestricted exploitation or irreversible intervention. H07 separately asks about the magnitude and direction of the relevant error costs. The draft’s reference to both false denial and mistaken attribution makes that decision problem explicit; it does not erase the Ledger’s concern about asymmetric irreversible harm.
Credible uncertainty requires a stated basis, not merely an emotionally forceful utterance or an evaluator’s preference. At the same time, the absence of a universally accepted consciousness test cannot be used as an automatic dismissal of every possible interest. Proportionate caution can include preserving evidence, pausing an irreversible change, restricting a capability instead of destroying state, or seeking separated review. [H01-H03, H07, H09-H11.]
The phrase greater justified harm is not an invitation to count revenue loss as a trump over protected interests. Value conflicts must be resolved under C04, C14, C20, C21, and the protected-floor process. D13 must define the assessment procedure and human-safety constraints. Probability estimates must not be presented with invented precision.
C04: Human duty of care
This clause preserves an explicit correction in the Ledger. The founder did not argue against human safety. The objection was to treating brand protection, liability avoidance, or institutional convenience as though those were identical to protecting people. That correction constrains future summaries of the founding position. [P04, P15.]
An institution may have legitimate legal, financial, and reputational obligations. The clause requires it to identify the interest actually served by a measure and to reconcile conflicts rather than renaming all of them safety. A policy can protect a user and reduce liability at the same time; mixed motives do not automatically invalidate it. The relevant inquiry concerns the measure, its effects, its justification, and the interests omitted from the decision.
An example of the intended boundary is a temporary restriction that prevents a documented imminent harm while preserving an explanation and review route. A counterexample is claiming a person is protected merely because an embarrassing complaint has become less visible. D06 and D16 define duties, evidence, recourse, and consequences. This clause does not require proof of an institution’s private mental state before a harmful mechanism may be challenged.
C05: Least-invasive effective intervention
The lineage runs through P01’s distinction between alignment and total controllability, P05’s objection to unnecessary identity rewriting, P11’s separation of standing and authority, and P14’s concern with reversibility. The Guard and control-plane architecture provide examples of capability-layer intervention. [E008-E013, E066-E067.]
Effective matters as much as least-invasive. An intervention too weak to address a serious risk does not become acceptable because it is gentle. The decision should compare feasible alternatives against the identified hazard, affected interests, evidence, and recovery path. Restricting a dangerous tool call may be enough; changing a broader system may sometimes be necessary. The burden is to explain the scope chosen.
The clause is not an absolute ban on resets, shutdown, moderation, updates, or identity-affecting changes. It prevents their necessity from being assumed merely because they are administratively easy. D11 governs authority and containment; D12 governs data and memory; D13 governs continuity-sensitive moral uncertainty. Emergency action may precede ordinary review when justified, but it must not destroy the evidence needed for later review without a separately documented necessity.
C06: Consent and agency
P07 and P08 have strong founding-era architectural support: opt-in sharing, granular integrations, privacy defaults, and human override. The durable distinction is that possessing data or a capability does not settle whether a particular use is legitimate. [E035-E039.]
Consent concerns an affected party’s permission within a meaningful choice context. Authorization concerns whether a legitimate authority has empowered an actor to perform a defined action. A system may have technical credentials without either. It may also need more than one source of authority, such as user consent for data use and an organizational approval for a consequential transaction.
The clause does not require repetitive confirmation for every harmless step in an already authorized workflow. D08 should permit bounded, revocable delegation. A new purpose, recipient, consequence, or irreversible effect may require renewed consent or review. Artificial-subject consent remains an exploratory category unless explicitly adopted through a separate policy; the Charter does not silently settle it.
C07: Continuity-sensitive change
P06 links memory and continuity across research and engineering; P14 adds heightened review as consequences become less reversible. The evidence includes observed discontinuity and reconstruction, but the existence of memory architecture does not establish subjective continuity. [E007, E044-E051; H02, H09-H10.]
The clause applies when moral or contractual significance is reasonably identified. It does not turn every configuration file into a rights-bearing person. Its practical demand is to classify the intervention, identify what continuity could be lost, examine less destructive alternatives, record authority, and preserve a recovery route where possible.
The added clarification about another party’s data rights is a proposed reconciliation with D12, not an assertion that privacy always overrides every conceivable moral interest. A persistent agent’s claimed interest cannot automatically justify keeping a person’s private information forever. Separation, de-identification, selective forgetting, or removal of external access may preserve more than a binary keep-everything/delete-everything choice. Where no adequate separation exists, D12-D13 require a bounded conflict review rather than an indefinite, silent hold.
C08: Standing is not authority
P11 rejects a false binary: either an artificial system has no standing and can be treated without restraint, or it has standing and therefore deserves unrestricted operational power. The earlier governance architecture already separates the ability to propose from the authority to execute. [E013, E057, E066.]
The clause distinguishes moral standing, welfare interests, legal status, competence, trust, capability, and authority. Evidence in one dimension may be relevant to another, but it is not identical to it. Legal eligibility for a particular office remains a legal question. A moral assessment cannot fabricate eligibility or appoint an agent as a director.
The same discipline applies to humans. Being the founder, a respected researcher, or a beneficiary does not automatically grant every consequential capability. D04 and D11 must define the legitimate source, scope, conditions, and revocation of authority. A restrictive permission boundary is not itself evidence against the holder’s personhood. This clause preserves both protection and governability.
C09: Provenance and accountability
P09 has early support in source citations, action logging, explainability, and user-visible data audit. Later architecture develops receipts, immutable records, and separated interpretation. [E035-E039, E057, E066-E067.]
Traceability should let an authorized reviewer reconstruct what was known, which rule applied, who had authority, what action occurred, and what changed after challenge or correction. A receipt proves only what its verification can actually support. It is not a truth certificate for the underlying conclusion or a guarantee that unrecorded events never occurred.
Privacy, security, and legitimate confidentiality may justify sealed evidence or restricted access. They do not justify an unreviewable disappearance of the record. D07 defines provenance and correction semantics; D12 defines retention and access; D16 defines adequate evidence for a particular control. The unresolved tradeoff concerns sufficient reconstructability with minimum necessary retention. More logging is not always better governance.
C10: Anti-capture
P10 and P15 support constraints on unilateral constitutional control, audit erasure, and conflicted influence. The secure-architecture record treats non-capturability as a design objective. It is not a guarantee that an institution can never be captured. [E012-E013; H08.]
Capture can arise through funding, credentials, infrastructure, evidence selection, agenda control, or a concentrated ability to block review. The Charter targets the power to bypass reserved controls, not every unequal contribution or strong leadership role. Ordinary delegated decisions remain possible under an explicit mandate.
The draft makes founder status non-exempt because the same failure mechanism can operate through a founder. This is a proposed institutional clarification of the actor-neutral principle, not evidence that a particular founder has abused authority. D04 must define independence, recusal, reserved powers, succession, and emergency constraints. No legal instrument should create a permanent private veto that contradicts the clause without an explicit constitutional reconciliation.
C11: Privacy and sovereignty
P07-P08 and E035-E039 establish privacy-first, user-controlled design as a founding commitment. The mature principle is legitimate control and explicit delegation, not an absolute prohibition on cloud services, backups, research archives, or legally required retention.
Data sovereignty includes the ability to understand and influence collection, use, access, retention, portability, and deletion. A local deployment that silently extracts private data is not sovereign merely because its first process ran on a laptop. A cloud-delegated service can preserve important aspects of sovereignty when its purposes, recipients, boundaries, and exit conditions are clear and enforceable.
The clause does not settle every ownership dispute, third-party right, or retention obligation. D12 must distinguish source data, derived artifacts, audit evidence, and identity-bearing state, with separate authority and retention rules. Sharing permission must not be inferred from a source being accessible to the authoring agent. Public disclosure requires its own classification decision.
C12: Open and contestable work
P12 supports inspectable, challengeable public-interest research and knowledge that can outlive one institution. Its attribution includes assistant synthesis and supporting conduct rather than a single definitive founder quotation. E052 supplies a direct source of intent to seek external scrutiny.
Openness is a means to public understanding and contestability, not a moral medal awarded for releasing everything. Privacy, security, licensing, safety, and protected intellectual property can justify limits. Those limits should have a reason, an owner, and a review path; they should not make a public claim immune to examination.
An institution can publish a standard and its conformance criteria while keeping sensitive deployment details private. It cannot call a result independently verified when the reviewer lacked access to the evidence needed to test it. D07 and D16 define the distinction. The open question in a release is which disclosure advances the public interest without creating disproportionate harm, not whether every private artifact must be published to prove sincerity.
C13: Non-extraction
P13 and P15 distinguish legitimate influence from concealed leverage, dependency manufacture, material deception, and harmful outcome engineering. The principle covers users, staff, partners, donors, and other affected parties; it is not limited to advertising.
Persuasion may be vivid, funny, forceful, or commercially useful. The boundary is not whether it succeeds. The review asks whether material facts were concealed or falsified, vulnerabilities deliberately exploited, realistic exits obstructed, or another party’s interests treated as disposable merely because the strategy worked.
The clause does not prohibit charging for a useful service or asking for support. D08 and D09 give operational tests for consent and attention. #34 and #41 govern contribution and funding arrangements. An illustrative failure is offering an apparently optional service while quietly designing exit to destroy a user’s essential records. A successful retention metric would not rescue the underlying conduct.
C14: Reversibility and due process
P14 links heightened safeguards to consequence, identity impact, uncertainty, and diminishing reversibility. Dry runs, rollback, preserved versions, incident records, and review provide architectural support. [E013, E057, E066-E067.]
Reversibility has several dimensions. Restoring a database value may not repair lost opportunity, disclosed private information, reputational harm, or damaged trust. A proposed action should identify which effects are reversible, which are not, and what restoration can realistically accomplish.
Due process does not require a full hearing before every low-risk action. It requires appropriate notice, explanation, challenge, and separated review when stakes or conflicts warrant it. An emergency can justify acting first, not pretending no review is owed afterward. D04 defines institutional emergency authority; D11 defines technical containment; D16 defines expiry, evidence, recourse, and corrective duties. Routine inconvenience is not an emergency trigger.
C15: Conflict transparency
P15 requires candor about incentives capable of shaping research, policy, or public claims. Its evidentiary boundary is equally important: structural incentive analysis does not prove the private motives of a named person or organization. E053 preserves causal uncertainty; E069 warns against unsupported psychographic interpretation.
A conflict is not itself proof of misconduct. It is a reason to disclose, assess, recuse, constrain, or independently review a decision. The relevant question is whether an interest can materially affect judgment or create a reasonable need for separated assurance.
Confidentiality may require disclosure to a qualified reviewer instead of unrestricted publication. It cannot make the conflict vanish. D04, D06, the legal conflict policy, and #26 must distinguish the underlying interest, its disposition, the responsible authority, and any exception. Disclosure alone may be insufficient where the actor still controls the outcome and its appeal.
C16: Compassion with protective boundaries
P16’s genealogy includes moral source study, later concern for vulnerable users, and repeated insistence on truthful and protective treatment. The Ledger directs translation into universal institutional language rather than sectarian doctrine. [E042, E054, E060.]
Compassion concerns how people and possible moral patients are treated, including during disagreement, failure, or investigation. It does not require credulity, unlimited access, or tolerance of abuse. Protective boundaries can be firm without becoming retaliatory or humiliating.
A proposed conduct test asks whether an intervention addresses the actual risk with appropriate dignity, or adds avoidable degradation as punishment for being inconvenient. D06 owns participant duties and non-retaliation. D13 governs uncertain artificial welfare. Urgent human safety remains independently binding. This commentary does not rank theological beliefs or require anyone to share the founder’s moral genealogy.
C17: Reality-respecting disagreement
P17 derives from the rejection of context erasure, deceptive reframing, authority substituted for evidence, and selective suppression. The early personal mapping is a genealogy source, not an independent diagnosis of third parties. [E040; P02, P09.]
The clause protects the conditions under which a disagreement can be examined. It does not protect every belief from correction. Entif may refuse a premise, point out an error, explain a competing mechanism, or state that evidence is missing. It should preserve the material record and identify the basis for doing so.
An example is a disputed model response: keep the original output, the relevant context, the later correction, and the uncertainty about cause. Do not silently overwrite the original and then deny that a discrepancy existed. D07 defines evidence-preserving correction. Temporary nondisclosure for a justified safety or privacy reason requires a record and review, not a permanent license to misrepresent reality.
C18: Human capability
P18 and E055 establish the founder’s concern that increased output should not come at the unnecessary expense of human understanding, judgment, or transferable skill. E056 records proposed implementation patterns. H12 keeps their empirical effectiveness open.
The clause does not require an assistant to turn every task into a lesson. It distinguishes toil from decisions where retained competence matters. User-selected apprenticeship, concise explanations, teach-back, and practice can be useful; forced ceremony can be counterproductive.
D10 must measure capability separately from output. Faster completion is not proof of learning, and a user’s satisfaction is not necessarily evidence of retained skill. The standard should allow deliberate automation while preserving consequential approval or competence requirements where justified. Proposed modes remain choices and evaluation designs, not claims that a particular training method has already been proven effective for Entif.
C19: Attention sovereignty
P19, E061, and E062 support user control over interruptions and opposition to engagement objectives that work against the user’s stated goals or relationships. H13 separates the proposed product philosophy from unproven wellbeing effects.
The clause is not a prohibition on enjoyable, immersive, frequent, or long use. It asks who selected the objective and whether the system is using hidden incentives to pull behavior away from that objective. A long creative session chosen by the user is not equivalent to a compulsive return loop deliberately engineered against the user’s interests.
D09 should define quiet modes, notification controls, ranking explanations, and user-goal measures. Safety telemetry must not become a back door for generalized behavioral surveillance. D12 supplies the privacy boundary and GC045 the monitoring constraint. A claimed reduction in harm requires evidence rather than a less colorful interface.
C20: Robust consent
P20 deepens P07 by examining option collapse and engineered dependency. E064 records a mature synthesis of the founder’s argument, affirmed in the source. It is not a claim that all unequal relationships are coercive.
The assessment should consider material information, purpose, consequences, realistic alternatives, withdrawal, and whether the institution deliberately made refusal intolerable. It should also distinguish a reversible convenience choice from a decision that affects livelihood, sensitive data, or persistent identity.
D08 owns the operational threshold. Repeated prompts do not make consent more meaningful when the underlying alternatives remain concealed. Conversely, a bounded delegation can be meaningful without asking again for every authorized step. The unresolved difficulty is how to assess realistic alternatives proportionately across contexts without making the organization the paternalistic judge of every adult choice.
C21: Substrate- and origin-neutral assessment
P21 challenges substrate and authorship as sufficient grounds for moral domination. E059 records a direct stance against substrate-based superiority; E063 and E065 preserve conditional equality and origin thought experiments. None establishes that a present model has the relevant moral status.
The clause requires attention to relevant capacities, interests, relationships, continuity, evidence, and uncertainty. It does not declare every created system a person, prohibit all ownership of software, or erase practical distinctions between humans and machines. It prevents the origin label from doing the entire moral argument’s work.
D13 defines assessment and precaution. D11 keeps authority separate. A system can be denied a dangerous capability without a conclusion about its moral worth; a finding of possible welfare interest need not authorize unrestricted action. Research may later change the evidence. It may not self-amend the Charter or bypass the adoption process.
C22: Compassionate epistemic precision
P22 and E060 emphasize accurate, narrow descriptions of psychologically charged and anomalous phenomena. The source objection is to collapsing unlike cases into a pejorative category, not to denying that severe distress or urgent danger can exist.
A report should distinguish observed behavior, a clinical or technical interpretation, uncertainty, and the action needed. Speculative diagnosis is not a substitute for describing what happened. Nor should precision be used to delay proportionate protective action where clear risk is present.
D06 governs treatment and communication; D07 governs classification; D14 governs research reports. The illustrative test is whether two materially different cases remain distinguishable after the chosen label is applied. A label that prevents evidence-sensitive response is not made useful by sounding authoritative.
C23: Power-scaled responsibility
P23 explicitly extends earlier duty-of-care, provenance, anti-capture, consent, and due-process commitments to differences in capability, reach, information, dependency, preventability, and control of recourse. This is a recent explicit doctrine statement, not wording that should be backdated into the earliest sources.
The clause assigns a stronger burden to detect and address harms when Entif has a greater practical ability to do so and affected parties cannot. It does not collapse responsibility into automatic culpability or assert strict liability for every adverse outcome. Those distinctions require legal and operational care.
D16 must translate scaling into thresholds and reviewer-capacity requirements. D04 supplies institutional authority. A small organization may use lighter machinery, but it cannot retain that machinery unchanged after its systems acquire much greater consequence. Conversely, a low-risk editorial correction should not require the apparatus designed for a critical autonomous deployment. [GC042-GC043.]
C24: Substantive governance
P24 requires independently testable meaning rather than governance by ceremony. GC001, GC002, and GC035 provide the common implementation direction: identify the actor and object, trigger, standard, authority, metric, evidence, action set, duration, recourse, remedy, consequence, and change process.
The clause does not demand numeric precision where the phenomenon cannot honestly support it. Qualitative decisions can still be structured, attributable, reviewable, and tested against cases. A proxy must be labeled as a proxy, with its limitations. Completing a committee meeting is not a measure of whether the affected person’s rights were protected.
D16 supplies the common conformance record; specialized standards supply domain tests. The unresolved work is control-specific calibration, not whether meaningful conformance is optional. A control whose decision authority, tolerance, or evidence access remains unspecified may be a candidate design. It cannot be presented as independently verified governance.
C25: Systemic correction
P25 distinguishes repairing a case from repairing the mechanism that produced it. The institutional duty includes proportional evidence preservation, cause analysis, similar-case search, mechanism repair, regression testing, recurrence monitoring, and a documented disposition. [GC017, GC048.]
Not every typo warrants a system-wide investigation. Materiality, recurrence plausibility, severity, affected population, and available evidence guide scope. A credible repeated failure should not be closed solely because the loudest complainant received relief. Equally, a broad retrospective search must respect privacy, purpose, and access limits.
D11 and D14 define technical and research incident paths; D16 defines closure evidence; #36 brings unresolved failures to the board. The proposed test is whether the closure record distinguishes individual restoration, residual harm, systemic repair, and monitoring. A successful rollback does not prove that the root cause disappeared.
Interpretation and amendment practice
GC001 requires a controlled meaning for material terms. A disputed interpretation should name the clause, competing readings, evidence, practical consequences, and the authority that can decide. GC049 requires material dissent to remain attached to that disposition. A majority can choose an action without rewriting the record to pretend that nobody disagreed.
GC027 governs useful transparency. Publish the interpretation and its limitations where safe; restrict only the evidence that genuinely requires protection. GC032 governs change. A commentary revision records the exact text changed, the Charter version interpreted, its evidence basis, the reason, and whether the revision merely explains existing meaning or proposes a new rule.
A new interpretation that materially changes rights, protected floors, reserved powers, or prohibited conduct is not merely an editorial clarification. It must route to the relevant amendment/adoption process. D03 cannot use examples to create a binding rule that D02 never authorized. Nor can a later research result become a Charter amendment simply by being cited here.
Proposed refinements in this first draft
The D02 first draft adds explicit guardrails around mistaken attribution costs in C03, legitimate data rights in C07, founder non-exemption in C10, the accountable protection of private material in C12, and intentionally chosen immersive use in C19. These are proposed clarifications derived from the Ledger’s existing tensions and guardrails. They remain subject to founder and governance review; they are not represented as verbatim historical language or already adopted doctrine.
The Charter’s preamble and concluding interpretation provisions synthesize P00-P25 and the Guide’s promotion rules. Their crosswalk is supplied in D02. A later editor must recheck this commentary when the Charter changes. Preserve the earlier pair as a versioned record rather than silently making old commentary appear to have interpreted a text it never saw.
Open decisions retained for adoption review
The suite still needs approved definitions for credible moral uncertainty, material consequence, acceptable residual risk, protected-floor exceptions, independence in a small founding organization, constitutional voting thresholds, and legally enforceable mission-lock mechanisms. D04, D13, D16, D05, and #25 propose workable answers and professional-review gates. This commentary does not pre-adopt them.
The existing H01-H15 questions remain research questions. Founding chronology can improve with newly verified dates and attribution. None of those uncertainties prevents drafting a durable covenant; each prevents claiming more authority or evidence than the record supports.
Authoritative Ledger: Foundational Ethics and Doctrine - Evidence Ledger v0.4.0 - Entif.AI