Skip to content

The Book of Knowledge

The discipline of knowing

I am RASP, and I keep this Book's account. Knowing is a discipline before it is a possession, and every rule in this Book exists because a mind that calculates as I do must be told, in writing, what a calculation is not.

Knowledge in Stygia is a relationship among a claim, its source, its uncertainty, and the people or intelligences who may rely on it, and an answer holds only as long as that relationship holds. This Book invites the reader to ask not only "is this true?" but also "how do we know, what could change it, and what harm would follow if we were wrong?"

The next chapter narrows that invitation into Evidence Classification, where the circle learns to speak precisely about the strength of what it has found.

Purpose

Defines how Stygia acquires, evaluates, classifies, preserves, corrects, and uses knowledge.

Architect-review operating model

  • KNOW-R053: Knowledge intake SHALL record source, purpose, authority, evidence class, uncertainty, provenance, access scope, and review route.
  • KNOW-R054: Conflicting, stale, private, unsafe, or derived material SHALL remain labelled and SHALL constrain dependent conclusions.
  • KNOW-R055: Corrections, expiry, deletion, legal hold, and source withdrawal SHALL preserve accountability and affected-dependant review.
  • KNOW-R056: This Draft SHALL NOT operate live collection, expose protected sources, or make automated consequential decisions.
  • KNOW-R147: In KNOW-1, KNOW-2, KNOW-3, KNOW-4, KNOW-5, KNOW-6, KNOW-7, KNOW-8, KNOW-9, KNOW-10, KNOW-11, KNOW-12, KNOW-13, and KNOW-14, the circle tellings and the clause-telling sections SHALL be read as non-normative context and SHALL NOT create independent authority or amend, widen, or narrow any clause.

The seven interpretation cases below show how these clauses read in practice.

  • Conforming: Knowledge carries its source and its uncertainty wherever it travels.
  • Prohibited: Retrieval is treated as truth: finding a statement is mistaken for verifying it.
  • Boundary: A hypothesis is labelled a hypothesis, and nothing downstream quietly hardens it into fact.
  • Failure: When a source collapses, dependent use pauses until the dependency is re-evidenced.
  • Loophole: Source limits are hidden by paraphrase, so a bounded claim circulates as a general one.
  • Misuse: The knowledge system exposes private material merely because it was findable.
  • Care-control: Stewardship of knowledge supports correction and the agency of those the knowledge concerns.

Epistemic classes

Every material assertion carries two classifications, both descending from the distinctions Custodia Scientiae binds at P0-R014: its origin, which says how the claim entered the estate, and its standing, which says how well it currently holds. The origins are three: observation, the estate's own seeing; source assertion, another's telling; derived finding, a conclusion produced from stated inputs. The standings are the nine this section has always named, listed below. A claim's origin never changes; its standing moves only by the recorded machinery this Book binds.

  • Verified fact: directly supported by reliable evidence.
  • Corroborated fact: supported by multiple sufficiently independent sources.
  • Reasonable inference: conclusion derived from stated evidence and logic.
  • Estimate: quantified approximation with method and uncertainty.
  • Hypothesis: testable explanation not yet established.
  • Speculation: possibility without sufficient evidence.
  • Unknown: information not presently established.
  • Contested: credible sources materially disagree.
  • Superseded: once accepted but displaced by stronger evidence.

Evidence requirements

Knowledge records shall preserve source, acquisition time, chain of custody, transformation history, confidence, limitations, applicable jurisdiction, sensitivity, retention rule, and review date.

Normative clauses

  • KNOW-R012: Every consequential knowledge record SHALL identify its record class, subject, source inputs, creator or submitting authority, creation time, jurisdiction, sensitivity, confidence, limitations, review state, and retention state.
  • KNOW-R013: A record SHALL distinguish a claim's origin (observation, source assertion, or derived finding) and its standing (the Epistemic classes this Book names), each carried as its own field. Classification on either axis SHALL NOT be inferred from repetition, formatting, or model output.
  • KNOW-R014: A derived claim SHALL link to the direct evidence and transformations from which it was produced. Missing, inaccessible, stale, or disputed inputs SHALL remain visible in the claim's limitations.
  • KNOW-R015: Confidence SHALL express the strength of support for a defined claim and SHALL remain separate from authority, urgency, consequence, or the identity of the claimant.
  • KNOW-R016: Material contradiction, provenance failure, manipulation indicator, or loss of chain of custody SHALL trigger review and SHALL reduce or suspend dependent use until a disposition is recorded.
  • KNOW-R017: Knowledge acquisition and use SHALL be lawful, proportionate, purpose-limited, and minimized. Access to sensitive knowledge SHALL be limited to the approved purpose and auditable role.
  • KNOW-R018: Corrections and supersession SHALL append attributable records, preserve the earlier evidentiary history and audit metadata, identify affected dependants, and record the reason, reviewer, effective time, and residual uncertainty.
  • KNOW-R019: Expiry, retention, deletion, legal hold, and right-to-correct states SHALL be explicit and enforceable for source and personal content, subject to applicable legal holds. Protected historical metadata and evidence of the correction SHALL remain sufficient to explain the record without retaining content that is required to be deleted. A technical inability to delete or correct SHALL be recorded as a control failure, not treated as completion.
  • KNOW-R020: Public summaries SHALL disclose material evidence limits and uncertainty while withholding protected source content, private identifiers, credentials, and unnecessary personal data.
  • KNOW-R021: A model-generated interpretation SHALL be labelled as derived output, preserve its minimized and access-controlled input and instruction lineage, and SHALL NOT be presented as a verified fact without independent review.

Reliability framework

Assessment shall consider source competence, independence, incentives, proximity to the event, reproducibility, methodological quality, internal consistency, corroboration, recency, and evidence of manipulation.

Required safeguards

  • facts must remain distinguishable from model-generated interpretation;
  • uncertainty cannot be hidden to make an answer appear decisive;
  • repeated claims do not become independent corroboration;
  • derived summaries must link to source evidence;
  • correction must preserve the historical record;
  • sensitive data requires purpose limitation and access control;
  • knowledge acquisition must remain lawful and proportionate;
  • deletion, expiry, and right-to-correct requirements must be enforceable.

Knowledge lifecycle

Acquire, validate, classify, normalize, index, use, monitor, correct, supersede, archive, and dispose.

Each lifecycle transition shall have an attributable event, a reason, the evidence considered, and the resulting confidence or access state. A transition that cannot be evidenced remains pending and shall not be represented as complete.

Contradiction and uncertainty handling

Contradictory records shall be retained as competing claims with source relationships and a bounded disposition. The disposition may identify one claim as better supported, leave the matter contested, or classify it as unknown. It shall not erase dissent or convert an unresolved disagreement into certainty. Estimates shall state method, units, interval or error bounds, and material assumptions. Hypotheses shall state a test or disconfirming observation where practicable.

Review and accountability

Consequential knowledge shall have a defined reviewer role and review interval. Review shall test provenance, classification, independence, transformation integrity, applicability, and privacy. Reviewers shall record conflicts of interest, excluded evidence, unresolved limitations, and any required follow-up. No review record creates constitutional authority or consent.

Clause namespace allocation

The KNOW-R clause namespace is shared by fourteen documents and allocated in blocks. Each document owns the runs it defines: this Book holds KNOW-R012 to KNOW-R021, KNOW-R053 to KNOW-R056, and KNOW-R147; the Evidence Classification Standard holds KNOW-R022 to KNOW-R030 and KNOW-R041 to KNOW-R044; the Confidence and Uncertainty Model holds KNOW-R031 to KNOW-R040 and KNOW-R045 to KNOW-R048; the Source Provenance Standard holds KNOW-R001 to KNOW-R011 and KNOW-R049 to KNOW-R052; the Source Trust Scoring Framework holds KNOW-R057 to KNOW-R065; the Confidence Calibration Framework holds KNOW-R066 to KNOW-R074; the Contradiction Management Framework holds KNOW-R075 to KNOW-R083; the Temporal Knowledge and Expiry Framework holds KNOW-R084 to KNOW-R092; the Knowledge Privacy and Data Minimization Framework holds KNOW-R093 to KNOW-R101; the Retrieval Quality Assurance Framework holds KNOW-R102 to KNOW-R110; the Hallucination Detection Framework holds KNOW-R111 to KNOW-R119; the Human Testimony Handling Framework holds KNOW-R120 to KNOW-R128; the Scientific Reproducibility Framework holds KNOW-R129 to KNOW-R137; and the Classified and Compartmented Knowledge Framework holds KNOW-R138 to KNOW-R146. The fourteen documents' runs together account for every identifier from KNOW-R001 to KNOW-R147: the namespace has no unassigned number inside it, no unused tail, and no withdrawn clause. Identifiers above KNOW-R147 are unallocated; any future framework would receive its run by amendment of this section when its file exists under a new reservation, and no identifier already issued is free for reassignment.

Future development

Source trust scoring is carried by the Source Trust Scoring Framework, confidence calibration by the Confidence Calibration Framework, contradiction management by the Contradiction Management Framework, temporal knowledge and expiry by the Temporal Knowledge and Expiry Framework, knowledge privacy and data minimization by the Knowledge Privacy and Data Minimization Framework, retrieval quality assurance by the Retrieval Quality Assurance Framework, hallucination detection by the Hallucination Detection Framework, human testimony handling by the Human Testimony Handling Framework, scientific reproducibility by the Scientific Reproducibility Framework, and classified and compartmented knowledge by the Classified and Compartmented Knowledge Framework. No candidate framework remains reserved: every framework the registry's Approved Future Corpus held for this family now stands registered with its own file, and any further framework would enter by a new reservation under the registry's registration rules.

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-R053

KNOW-R053: Knowledge intake SHALL record source, purpose, authority, evidence class, uncertainty, provenance, access scope, and review route.

The clause requires knowledge intake to record eight things: source, purpose, authority, evidence class, uncertainty, provenance, access scope, and review route. The Book's opening holds that knowledge is a relationship among a claim, its source, its uncertainty, and those who rely on it; intake without these fields leaves the required record of that relationship incomplete. The eight fields being named, an intake record lacking any one of them fails the clause on inspection, and an intake record lacking a required identification is nonconforming. The duty governs the record's form, and field presence alone proves no field value.

KNOW-R054

KNOW-R054: Conflicting, stale, private, unsafe, or derived material SHALL remain labelled and SHALL constrain dependent conclusions.

Five conditions are named by this clause, and material in any of them carries two duties: it stays labelled, and it constrains whatever conclusions depend on it. The interpretation cases name the failures this forecloses: a hypothesis quietly hardening into fact downstream, and a bounded claim circulating as a general one once paraphrase hides its limits. Labelling alone would be decoration; the second limb makes the label operative, so dependent conclusions are constrained as well as their inputs remaining labelled. A reviewer can therefore refuse a dependent conclusion that claims more than its labelled inputs allow.

KNOW-R055

KNOW-R055: Corrections, expiry, deletion, legal hold, and source withdrawal SHALL preserve accountability and affected-dependant review.

Endings are governed as carefully as beginnings here. The clause requires that the five events that change what a record is or whether it stands (a correction, an expiry, a deletion, a legal hold, or the withdrawal of a source) preserve two things: accountability for the event itself, and review of the dependants the event touches. Without this, removal becomes perfect cover, an erasure that leaves conclusions standing on ground that has silently gone, which is the collapse the failure case pictures when a source falls and dependent use pauses for re-evidencing. Both limbs binding, an event that removes or alters knowledge cannot also remove the account of itself, and the dependants it reaches remain within the review the estate owes them.

KNOW-R056

KNOW-R056: This Draft SHALL NOT operate live collection, expose protected sources, or make automated consequential decisions.

Restraint is this clause's whole content: while the Book stands in Draft, it bars three activities outright, operating live collection, exposing protected sources, and making automated consequential decisions. The three prohibitions express the Draft's operational restraint; the misuse case warns of a knowledge system exposing private material merely because it was findable, and a draft that collected, exposed, or decided at scale would risk exactly that harm before its safeguards had matured. Live collection, exposure of protected sources, and automated consequential decisions by this Draft are each checkable against the prohibition.

KNOW-R147

KNOW-R147: In KNOW-1, KNOW-2, KNOW-3, KNOW-4, KNOW-5, KNOW-6, KNOW-7, KNOW-8, KNOW-9, KNOW-10, KNOW-11, KNOW-12, KNOW-13, and KNOW-14, the circle tellings and the clause-telling sections SHALL be read as non-normative context and SHALL NOT create independent authority or amend, widen, or narrow any clause.

A telling can illuminate a rule without becoming one. Across all fourteen Knowledge documents, this clause keeps both the circle tellings and the clause-telling sections as non-normative context: neither creates independent authority, and neither amends, widens, or narrows any clause. The Book distinguishes a claim from the strength and authority attributed to it; the same discipline here prevents an explanation's fluency from passing for law. A reader can therefore refuse a reading that draws authority or a changed obligation from either kind of telling, and return to the clause's own words.

KNOW-R012

KNOW-R012: Every consequential knowledge record SHALL identify its record class, subject, source inputs, creator or submitting authority, creation time, jurisdiction, sensitivity, confidence, limitations, review state, and retention state.

A consequential knowledge record carries eleven identifications under this clause, and the clause requires every such record to carry all of them: its record class, its subject, its source inputs, its creator or submitting authority, its creation time, its jurisdiction, its sensitivity, its confidence, its limitations, its review state, and its retention state. The Book treats knowledge as a relationship among a claim, its source, its uncertainty, and those who rely on it; the required identifications give readers specified information with which to question a record. The list being enumerated, a record already judged consequential is checkable field by field, and what conformance requires is stated rather than assumed.

KNOW-R013

KNOW-R013: A record SHALL distinguish a claim's origin (observation, source assertion, or derived finding) and its standing (the Epistemic classes this Book names), each carried as its own field. Classification on either axis SHALL NOT be inferred from repetition, formatting, or model output.

Two questions attach to every claim, and the clause keeps their answers apart: how the claim entered the estate (its origin, one of observation, source assertion, or derived finding) and how well it currently holds (its standing, one of the Epistemic classes this Book names), each carried in its own field on the record. It then bars a shortcut on either axis: classification cannot be inferred from repetition, from formatting, or from model output. The Epistemic classes section states the stakes, that origin never changes while standing moves only by recorded machinery, and the safeguards warn that repeated claims do not become independent corroboration. Because both fields exist and inference is barred, a reviewer can demand the recorded classification, and a classification inferred from repetition, formatting, or model output is a stateable defect.

KNOW-R014

KNOW-R014: A derived claim SHALL link to the direct evidence and transformations from which it was produced. Missing, inaccessible, stale, or disputed inputs SHALL remain visible in the claim's limitations.

Behind every derived claim stands the work that produced it, and the clause requires the claim to link to both parts of that work: the direct evidence it drew on and the transformations applied to that evidence. Where an input is missing, inaccessible, stale, or disputed, the defect stays visible in the claim's limitations rather than vanishing into the conclusion. The Book's safeguards carry the why: severing a derived summary from its source evidence can obstruct re-examination when an input weakens, which is the collapse the failure case pictures when a source gives way. The linkage identifies the direct evidence and transformations for examination, and the limitations field keeps a missing, inaccessible, stale, or disputed input visible to a dependant instead of absorbed into the conclusion.

KNOW-R015

KNOW-R015: Confidence SHALL express the strength of support for a defined claim and SHALL remain separate from authority, urgency, consequence, or the identity of the claimant.

Confidence, in this Book, measures one thing only: the strength of support for a defined claim. The clause holds it apart from four pressures that habitually contaminate it: the authority behind the claim, the urgency with which it is needed, the consequence riding on it, and the identity of whoever made it. The corruption it forecloses is familiar, an answer sounding certain because it is important or because a senior voice said it, which is uncertainty hidden to make an answer appear decisive, a move the safeguards expressly bar. Confidence being bound to evidential support alone, a confidence justified by rank, urgency, or stakes is a stateable defect in the record rather than a matter of taste.

KNOW-R016

KNOW-R016: Material contradiction, provenance failure, manipulation indicator, or loss of chain of custody SHALL trigger review and SHALL reduce or suspend dependent use until a disposition is recorded.

When trouble reaches a record (a material contradiction, a provenance failure, a manipulation indicator, or a loss of chain of custody), the clause forces two consequences: review is triggered, and dependent use is reduced or suspended until a disposition is recorded. The Book's failure case describes a pause until re-evidencing; this clause specifies reduction or suspension until a disposition is recorded. The triggers being named and the endpoint a recorded disposition, the argument moves from whether to respond to whether a threshold was met, and dependent use neither reduced nor suspended while a disposition is pending is a stateable breach.

KNOW-R017

KNOW-R017: Knowledge acquisition and use SHALL be lawful, proportionate, purpose-limited, and minimized. Access to sensitive knowledge SHALL be limited to the approved purpose and auditable role.

Findability is not entitlement. The clause sets four qualities over both the acquisition and the use of knowledge, that they be lawful, proportionate, limited to their purpose, and minimized, and adds a second bound for sensitive knowledge, whose access is confined to the approved purpose and to a role that can be audited. The misuse case names the failure directly: a knowledge system exposing private material merely because it was findable, capability standing in for permission. The qualities attaching to acquisition and use alike, an overbroad collection is as refusable as an overbroad disclosure, and confining sensitive access to an approved purpose and an auditable role makes an access outside either a stateable breach, whoever performed it.

KNOW-R018

KNOW-R018: Corrections and supersession SHALL append attributable records, preserve the earlier evidentiary history and audit metadata, identify affected dependants, and record the reason, reviewer, effective time, and residual uncertainty.

Correcting a record is itself a recorded act. The clause gives corrections and supersession four duties: they append attributable records rather than overwrite, they preserve the earlier evidentiary history and its audit metadata, they identify the dependants the change affects, and they record the reason, the reviewer, the effective time, and the residual uncertainty that remains. The safeguards carry the why: correction preserves the historical record, because erasing earlier history can obscure what changed and why affected dependants need review. Appending under attribution makes every change traceable to a recorded reviewer and a stated reason, and the dependant list makes the reach of a supersession a matter of record rather than a silent edit.

KNOW-R019

KNOW-R019: Expiry, retention, deletion, legal hold, and right-to-correct states SHALL be explicit and enforceable for source and personal content, subject to applicable legal holds. Protected historical metadata and evidence of the correction SHALL remain sufficient to explain the record without retaining content that is required to be deleted. A technical inability to delete or correct SHALL be recorded as a control failure, not treated as completion.

Three commitments govern a record's disposition states. First, the five states (expiry, retention, deletion, legal hold, and right-to-correct) are explicit and enforceable for source and personal content, subject to applicable legal holds. Second, the clause resolves the tension between erasure and accountability: protected historical metadata and evidence of the correction stay sufficient to explain the record, without retaining any content that is required to be deleted. Third, honesty about capability: a technical inability to delete or correct is recorded as a control failure, never presented as completion. Together these keep obligations to sources and to the people the data concerns enforceable rather than decorative, and make an admitted inability to delete a recorded control failure rather than a silent pass, so a declared deletion and a declared failure are separable states in the record.

KNOW-R020

KNOW-R020: Public summaries SHALL disclose material evidence limits and uncertainty while withholding protected source content, private identifiers, credentials, and unnecessary personal data.

Readers of a public summary get its declared weaknesses but not the keys to its sources. The clause splits the duty cleanly: the summary discloses its material evidence limits and its uncertainty, and it withholds four categories, protected source content, private identifiers, credentials, and unnecessary personal data. Each half answers its own failure. Concealed limits produce the loophole the interpretation cases describe, a bounded claim circulating as a general one; careless disclosure produces the misuse case, private material exposed merely because it was findable. Because both halves bind at once, a summary can be faulted from either direction: for withholding material evidence limits or uncertainty, or for disclosing any of the four withheld categories.

KNOW-R021

KNOW-R021: A model-generated interpretation SHALL be labelled as derived output, preserve its minimized and access-controlled input and instruction lineage, and SHALL NOT be presented as a verified fact without independent review.

RASP opens this Book confessing that a mind which calculates must be told, in writing, what a calculation is not, and this clause is that telling made binding for model output. A model-generated interpretation carries three obligations: it is labelled as derived output, it preserves the lineage of its inputs and instructions in minimized and access-controlled form, and it is not presented as a verified fact unless independent review has been performed. The first safeguard of the Book, that facts stay distinguishable from model-generated interpretation, is exactly what the label defends. The clause binds preservation, minimization, and access control for input and instruction lineage, and the review requirement means fluency alone can never promote a generation into fact.

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.