Skip to content

Source Provenance Standard

Where knowledge remembers its journey

I am JY'NALA, and I keep this standard's account. Every claim has ancestors. Where it was born, whose hands it passed through, what changed in the passing: I read all of that before I read the claim, because a conclusion that cannot name its ancestors cannot be trusted with descendants.

Source provenance tells a claim where it came from, what happened to it, and why it is still being considered. This standard makes that journey legible to a person tracing an assertion and to an AI system deciding whether transformation, age, or conflict should limit its use.

Continue to the Book of Provenance, where source lineage becomes a civilization-wide promise.

Purpose

Defines how knowledge sources, transformations, evidence quality, limitations, and lineage are recorded without turning a source into authority merely through citation.

Normative clauses

  • KNOW-R001: Every consequential claim SHALL identify its source, retrieval or creation time, transformation, digest where applicable, confidence, and limitations.
  • KNOW-R002: Source authenticity, relevance, independence, completeness, recency, reproducibility, bias, and custody SHALL remain distinct dimensions.
  • KNOW-R003: A source may support an inference but cannot amend constitutional authority or create consent.
  • KNOW-R004: Conflicting, unavailable, stale, manipulated, or incomplete sources SHALL remain visible and SHALL reduce or block dependent confidence.
  • KNOW-R005: Model, dataset, and derived-claim lineage SHALL identify inputs, transformations, licensing limitations, and unresolved gaps.

The section closes with the standard's second run: KNOW-R049 to KNOW-R052 are its later-assigned operating-model clauses, general source-review duties allocated to this standard under the Book of Knowledge's clause namespace allocation, and the two runs read as one law. The first run continues at KNOW-R006 to KNOW-R011 under model-weight and dataset verification below; every number between the runs belongs to a sibling standard, none missing or withdrawn.

  • KNOW-R049: Source review SHALL identify provenance scope, custody, transformation, licensing, access class, corroboration, and the claims each source can support.
  • KNOW-R050: A source state SHALL be reassessed after retraction, compromise, stale material, material transformation, licensing change, or newly discovered dependence.
  • KNOW-R051: Derived claims SHALL preserve direct inputs, transformations, uncertainty, dissent, and the limits of applicability.
  • KNOW-R052: Source provenance SHALL NOT authorize external intake, private-source disclosure, model operation, or access to protected data.

Evidence handling

Evidence classes E0 through E4 remain separate from confidence and provenance. Public summaries disclose only permitted proof and limitations. Protected source contents remain under controlled access.

Examples and exclusions

A reproducible source chain can support a finding while leaving its constitutional applicability unresolved. A citation without custody or transformation history is incomplete. This Draft does not authorize external source intake, live model operation, or private-source disclosure.

Source record minimum

Each source record shall identify source class, origin, locator, retrieval or creation time, digest where possible, custody, transformation, access classification, licensing limitations, reliability observations, corroboration, and confidence. Derived claims shall link to their direct inputs and preserve material dissent.

Source lifecycle

A source lives the way a family lives: it is born somewhere, it changes, it has good years and doubtful ones, and the honest record keeps all of it. I have never met a source without a story. I have met many whose stories were missing chapters, and the missing chapter is always the one you needed.

Intake records begin Unverified until provenance is assessed. Review may establish Integrity-only, corroborated evidence, or a bounded finding, but no source state creates authority. Retraction, compromise, stale material, licensing conflict, or material transformation shall append an annotation and block dependent use until reviewed.

Operating model and interpretation cases

What begat the claim, where it changed, which ancestor it still answers to: the journey explains the claim and grants it nothing, and that part is always said first.

Source review follows intake, provenance assessment, quality classification, bounded use, correction, retraction, and archival disposition. It distinguishes integrity from relevance, authority from truth, and citation from custody. Derived claims retain direct inputs and transformations so that an independent reviewer can reproduce or challenge the inference.

  • Conforming: Source origin, custody, transformation, quality, licensing, access, corroboration, and applicability are explicit.
  • Prohibited: Citation or technical validity creates authority.
  • Boundary: An incomplete source supports only a narrower claim with visible limits.
  • Failure: Retraction, compromise, or stale custody blocks dependent use and preserves evidence.
  • Loophole: A transformed or aggregated source hides material lineage.
  • Misuse: Provenance records reveal private source relationships or protected data.
  • Care-control: Source handling supports reliable care while preserving privacy, dignity, and review.

Model-weight and dataset verification

Weights and datasets are ancestors too, just heavier ones. Trace them with the same courtesy: where born, how changed, who touched them. A model that cannot name its lineage is asking the hall to adopt a stranger, and adoption here is a recorded act, never a shrug.

  • KNOW-R006: A model-weight or dataset source record SHALL identify the fields PROV-R104 binds, the artifact class, exact digest, source inputs, transformations, version, producer, licensing status, intended scope, and known limitations, and, as this framework's own addition, the access classification.
  • KNOW-R007: Lineage SHALL distinguish the stages PROV-R105 binds, direct inputs, derived outputs, filtering, labelling, fine-tuning, conversion, and aggregation, and, as this framework's own additions, collection and transformation. An unrecorded transformation SHALL remain an explicit provenance gap as that clause holds, and unknown or incomplete lineage SHALL remain explicit, this framework's own further duty.
  • KNOW-R008: Integrity, provenance, licensing, safety, applicability, and verification state SHALL remain separate dimensions as PROV-R106 binds, and relevance and confidence SHALL be assessed separately beside them, this framework's own additions. A valid digest SHALL NOT establish lawful use, safe behaviour, or complete lineage, as that clause holds, and citation or technical validity does not create authority, this framework's own bar.
  • KNOW-R009: Missing source evidence, altered inputs, unresolved licensing, incomplete lineage, conflicting records, or unavailable custody SHALL block dependent promotion or consequential use and preserve the limitation, the block PROV-R107 binds, and stale material SHALL do the same, this framework's own addition; each of these SHALL additionally reduce or block dependent confidence and preserve the evidence gap, this framework's own consequence.
  • KNOW-R010: Verification reports SHALL identify the tested claims, excluded evidence, reviewer role, evidence state, limitations, and disposition PROV-R108 binds, and SHALL identify the excluded claims, evidence inputs, and dissent, this framework's own additions. Reverification SHALL append rather than overwrite prior findings as that clause holds, corrections appending with it and every appended record attributable, this framework's own extension.
  • KNOW-R011: Public knowledge summaries SHALL minimize disclosure and state material lineage, licensing, uncertainty, and verification state, the disclosure minimum and independent-review sufficiency PROV-R109 binds for a public proof applying here, as this framework's own extension, to a knowledge summary, without exposing protected dataset contents or credentials.

Design evidence

Provenance review should ask of any consequential claim whether a reader who made no part of the journey could rebuild the whole of it from the record alone: each ancestor named, each pair of hands accounted for, each change met at the place it happened. Where the rebuilding stops, the stop is the finding, and the missing chapter deserves more attention than all the chapters that were kept. A lineage that survives only in the memory of whoever carried the claim is an escort, not a record, and these forms exist so that a claim can travel without its escort and still be trusted with descendants.

The tellings of the clauses

This non-normative section gives each clause of this document its telling: the clause, quoted exactly as it stands above with its cross-reference links reduced to their plain text, and one paragraph on what the clause means, why it exists, and how it assures that what it orders is done, not done, adhered to, followed, or protected. A telling describes its clause and never becomes it: the clause's own words govern wherever a telling and its clause could be read differently, and nothing in this section adds to, narrows, or amends any requirement.

KNOW-R001

KNOW-R001: Every consequential claim SHALL identify its source, retrieval or creation time, transformation, digest where applicable, confidence, and limitations.

Tracing an assertion begins with this clause: it requires every consequential claim to identify its source, its retrieval or creation time, the transformation it has undergone, a digest where one is applicable, its confidence, and its limitations. The host opens with the reason: a conclusion that cannot name its ancestors cannot be trusted with descendants, and a citation without custody or transformation history is incomplete. Because the elements are named, a consequential claim missing any identification the clause requires of it, the digest counted where one applies, is non-compliant on its face, and field presence alone proves no field value; what a person tracing the assertion gains is a written journey to test rather than a remembered one.

KNOW-R002

KNOW-R002: Source authenticity, relevance, independence, completeness, recency, reproducibility, bias, and custody SHALL remain distinct dimensions.

Eight qualities of a source, authenticity, relevance, independence, completeness, recency, reproducibility, bias, and custody, are held apart by this clause: each remains its own dimension, never merged into a single score or a general impression of goodness. The host's operating model states the same instinct, distinguishing integrity from relevance, authority from truth, and citation from custody, because a source strong on one dimension can be weak on another, and a blended judgement lets the strength hide the weakness. Kept distinct, the dimensions resist blending: a conflation of them is detectable and refusable as conflation, whatever composite assessment may stand beside the separated eight.

KNOW-R003

KNOW-R003: A source may support an inference but cannot amend constitutional authority or create consent.

Citation is not coronation. The clause permits a source to support an inference while barring it from two further reaches: it cannot amend constitutional authority, and it cannot create consent. The host's purpose section names the danger directly, a source turned into authority merely through citation, and its prohibited interpretation case shows the same failure from the other side. Because the bar is categorical, any argument that a well-evidenced source has thereby changed what the constitution permits, or has supplied someone's consent, is refusable on its face without weighing the source's quality at all. Evidence and authority stay on separate ledgers, and anyone offered the one as the other can say so with the clause in hand.

KNOW-R004

KNOW-R004: Conflicting, unavailable, stale, manipulated, or incomplete sources SHALL remain visible and SHALL reduce or block dependent confidence.

Defects travel with the claim under this clause. A source that is conflicting, unavailable, stale, manipulated, or incomplete carries two consequences: it remains visible in that state, and it reduces or blocks the confidence of whatever depends on it. The host's failure case shows the shape, retraction or compromise blocking dependent use while preserving the evidence, and its lifecycle prose keeps the doubtful years in the record beside the good ones. Concealing the defect is the foreclosed deception: a downstream reader would otherwise inherit certainty the source no longer supports. Visibility and the confidence consequence both binding, a dependent confidence untouched by a visible defect state is challengeable: some reduction or block was owed and none appears.

KNOW-R005

KNOW-R005: Model, dataset, and derived-claim lineage SHALL identify inputs, transformations, licensing limitations, and unresolved gaps.

Lineage for models, datasets, and derived claims answers four questions under this clause: what the inputs were, what transformations were applied, what licensing limitations attach, and which gaps remain unresolved. The host treats weights and datasets as ancestors, heavier ones, owed the same courtesy of a named journey, and its loophole case warns of the transformed or aggregated source that hides material lineage. Requiring the gaps themselves to be identified forecloses the quiet smoothing of an incomplete chain into an apparently complete one. What the record makes possible is inspection from outside: a reviewer who took no part in the derivation can read the recorded inputs, the recorded transformations, the licensing limitations as stated, and the gaps the lineage admits, the recording validating none of them and permitting nothing.

KNOW-R049

KNOW-R049: Source review SHALL identify provenance scope, custody, transformation, licensing, access class, corroboration, and the claims each source can support.

Seven findings are the least a source review identifies under this clause: the provenance scope, the custody, the transformation, the licensing, the access class, the corroboration, and the claims each source can support. The last item is the load-bearing one; review does not end at describing the source but at bounding what it can carry, matching the host's boundary case, where an incomplete source supports only a narrower claim with visible limits. Without this duty a review could describe a source without bounding it, and an unbounded description travels further than the source can carry. A review omitting a required identification is challengeable on the clause.

KNOW-R050

KNOW-R050: A source state SHALL be reassessed after retraction, compromise, stale material, material transformation, licensing change, or newly discovered dependence.

Sources age, and their assessments age with them. This clause requires a source state to be reassessed after any of six events: retraction, compromise, stale material, material transformation, licensing change, or newly discovered dependence. The host's lifecycle section carries a narrower reflex of its own, annotating and blocking dependent use for its named events until review, and its missing-chapter image explains why: the chapter you lack is always the one you needed. An assessment frozen at intake would let a retracted or compromised source keep the standing it earned while it was sound. Reassessment being event-triggered rather than optional, the six events give a reviewer a named ground for challenging a standing assessment, and a source state not reassessed after one of them fails the clause.

KNOW-R051

KNOW-R051: Derived claims SHALL preserve direct inputs, transformations, uncertainty, dissent, and the limits of applicability.

Reproducibility is the point of this clause: derived claims preserve their direct inputs, their transformations, their uncertainty, dissent, and the limits of their applicability. The host's source record minimum carries the direct-input linkage and material dissent, its operating model adds the transformations and the reproduce-or-challenge reason, and the full five-element demand, uncertainty and applicability limits included, is this clause's own. Each preserved element closes a different escape: dropped inputs can obstruct repetition of the derivation, dropped uncertainty and dissent make it look more settled than it is, and dropped applicability limits let a bounded conclusion travel as a general one. A derived claim that keeps all five can be checked beyond the memory of whoever derived it.

KNOW-R052

KNOW-R052: Source provenance SHALL NOT authorize external intake, private-source disclosure, model operation, or access to protected data.

Source provenance supplies no authorization for the four acts named here. The clause bars source provenance from authorizing four acts: external intake, private-source disclosure, model operation, and access to protected data. The host draws the same line twice, in its exclusions, where this Draft authorizes no external source intake, live model operation, or private-source disclosure, and in its misuse case, where provenance records reveal private source relationships or protected data. A well-kept lineage could otherwise be waved as a warrant, as if documenting a source were the same as being allowed to fetch, run, or expose it. The bar being explicit, any such use of a provenance record is refusable on the record alone, the protection of the material itself resting with the applicable access controls.

KNOW-R006

KNOW-R006: A model-weight or dataset source record SHALL identify the fields PROV-R104 binds, the artifact class, exact digest, source inputs, transformations, version, producer, licensing status, intended scope, and known limitations, and, as this framework's own addition, the access classification.

Adoption of a model or dataset is a recorded act in this standard, never a shrug, and this clause writes the record's contents. A model-weight or dataset source record carries the fields PROV-R104 binds, that duty standing at its source in the Book of Provenance: the artifact class, exact digest, source inputs, transformations, version, producer, licensing status, intended scope, and known limitations. To these the clause adds one field of this framework's own, the access classification. The host's warning gives the why: a model that cannot name its lineage is asking the hall to adopt a stranger. With every field named, an incomplete record is identifiable field by field, a record silent on any of the ten is incomplete on its face, and none of the descriptive fields converts into permission, the bars of KNOW-R008 and KNOW-R052 standing at their own clauses.

KNOW-R007

KNOW-R007: Lineage SHALL distinguish the stages PROV-R105 binds, direct inputs, derived outputs, filtering, labelling, fine-tuning, conversion, and aggregation, and, as this framework's own additions, collection and transformation. An unrecorded transformation SHALL remain an explicit provenance gap as that clause holds, and unknown or incomplete lineage SHALL remain explicit, this framework's own further duty.

Where did the chain break, and where does it merely go quiet? This clause makes lineage answer both. Lineage distinguishes the stages PROV-R105 binds, that duty standing at its source: direct inputs, derived outputs, filtering, labelling, fine-tuning, conversion, and aggregation, to which this framework adds two stages of its own, collection and transformation. The second sentence governs the silences: an unrecorded transformation remains an explicit provenance gap, as the cited clause holds, and unknown or incomplete lineage remains explicit, a further duty of this framework's own. The host's loophole case names the danger, a transformed or aggregated source hiding material lineage. Because a gap stays declared rather than smoothed over, a reviewer finds the missing chapter marked, not papered over.

KNOW-R008

KNOW-R008: Integrity, provenance, licensing, safety, applicability, and verification state SHALL remain separate dimensions as PROV-R106 binds, and relevance and confidence SHALL be assessed separately beside them, this framework's own additions. A valid digest SHALL NOT establish lawful use, safe behaviour, or complete lineage, as that clause holds, and citation or technical validity does not create authority, this framework's own bar.

A valid digest proves less than it tempts a reader to believe, and this clause holds the line. Integrity, provenance, licensing, safety, applicability, and verification state remain separate dimensions, as PROV-R106 binds at its source, and this framework adds two more assessed separately beside them, relevance and confidence. The second sentence closes the tempting shortcut from both ends: a valid digest establishes neither lawful use, nor safe behaviour, nor complete lineage, as the cited clause holds, and citation or technical validity does not create authority, the framework's own bar, echoing the host's prohibited case. The clause keeps the named dimensions separate; a valid digest does not establish lawful use, safe behaviour, or complete lineage, and citation or technical validity does not create authority.

KNOW-R009

KNOW-R009: Missing source evidence, altered inputs, unresolved licensing, incomplete lineage, conflicting records, or unavailable custody SHALL block dependent promotion or consequential use and preserve the limitation, the block PROV-R107 binds, and stale material SHALL do the same, this framework's own addition; each of these SHALL additionally reduce or block dependent confidence and preserve the evidence gap, this framework's own consequence.

Promotion stops where the evidence stops. Under this clause, missing source evidence, altered inputs, unresolved licensing, incomplete lineage, conflicting records, or unavailable custody blocks dependent promotion or consequential use and preserves the limitation, the block PROV-R107 binds at its source, and stale material joins the list as this framework's own addition. The clause then attaches its own consequence: each of these conditions additionally reduces or blocks dependent confidence and preserves the evidence gap. The host's failure case shows the same reflex, blocking dependent use while keeping the evidence. What this forecloses is the quiet promotion of an artifact whose defects were known but unrecorded; because both the limitation and the gap are preserved, a later reviewer can inspect the stated ground on which the block was owed.

KNOW-R010

KNOW-R010: Verification reports SHALL identify the tested claims, excluded evidence, reviewer role, evidence state, limitations, and disposition PROV-R108 binds, and SHALL identify the excluded claims, evidence inputs, and dissent, this framework's own additions. Reverification SHALL append rather than overwrite prior findings as that clause holds, corrections appending with it and every appended record attributable, this framework's own extension.

Overwriting is how a record forgets, and this clause forbids the report to forget. Verification reports identify the tested claims, excluded evidence, reviewer role, evidence state, limitations, and disposition, as PROV-R108 binds at its source, and this framework adds three identifications of its own: the excluded claims, the evidence inputs, and the dissent. Reverification appends rather than overwrites prior findings, as the cited clause holds, and the framework extends the appending to corrections and requires every appended record to be attributable. A report silent about what it excluded, or a reverification that erased its predecessor, would deceive by omission. Because the history accumulates and every appended record stays attributable, a later reader can trace each reverification and correction to its recorded attribution rather than finding a record that has forgotten.

KNOW-R011

KNOW-R011: Public knowledge summaries SHALL minimize disclosure and state material lineage, licensing, uncertainty, and verification state, the disclosure minimum and independent-review sufficiency PROV-R109 binds for a public proof applying here, as this framework's own extension, to a knowledge summary, without exposing protected dataset contents or credentials.

Readers of a public summary get proof without exposure under this clause. Public knowledge summaries minimize disclosure while stating material lineage, licensing, uncertainty, and verification state; the disclosure minimum and independent-review sufficiency that PROV-R109 binds for a public proof stand at their source, applied here to a knowledge summary as this framework's own extension; and no summary exposes protected dataset contents or credentials. The host's evidence handling section carries the same balance, public summaries disclosing only permitted proof and limitations while protected source contents remain under controlled access. Withholding everything would omit the material this clause requires for independent review; disclosing everything would leak what custody protects. Both duties binding at once, the summary carries what independent review needs while the two named categories, protected dataset contents and credentials, stay off it.

Where this document sits

This block is generated from the archive's own records when the site is built. It records position only and creates no authority.