Jump to text
On this page
Copy and source
Published source confirmed at last check

Source changed 2026-11-18 21:11:39 UTC · snapshot created 2026-11-18 21:12:35 UTC · last check 2026-11-18 21:55:20 UTC

B.3 - Is This Claim Supported for This Use? — Trust and Assurance Calculus

Type: Foundational (B) Status: Stable Normativity: Normative when an FPF use makes an assurance claim about one target claim.

Plain-English headline. B.3 helps a practitioner state what an assurance claim is about, which argument and results support it, what use they support, what remains unsupported, and what would reopen the conclusion. A visible label or artifact contributes only through the claim and basis the argument uses.

Use this when. Use B.3 when a named assurance claim is current: for example, a claim that a model claim is credible for one decision, or that a safety claim is adequately supported for one release use.

First useful move. State the target claim and the assurance use. Then ask which results and argument make that use supportable. If there is no assurance claim, stop and use the pattern that defines or tests the evidence, status, gate, permission, safety, release, work, or domain-result claim.

What goes wrong if missed. A visible label or a score starts raising trust without a target, argument, basis, limitation, and use. At the other extreme, a modest assurance question is forced through a universal score and a large record whose fields do not affect the decision.

What this buys. The user gets the smallest assurance result sufficient for the named use, with enough basis and limits to inspect or reopen it. Domain-specific characteristics and calculations remain usable without pretending that unlike measures share one scale.

Not this pattern when. Stay with A.2.4 for the classification of an episteme as evidence, A.10 and G.6 for source recovery and bounded reliance, G.11 for currentness, F.10 for a status value and its use, A.21 for a gate decision, and the domain pattern for safety, permission, access, responsibility, release, compliance, or controlled action. Consequence alone does not create an assurance claim. A domain rule may require one, but the claim must be stated before B.3 is applied.

First output. Produce either one bounded AssuranceResult claim or a plain statement that the available argument does not support the attempted assurance use. Do not create a B.3 result merely to record that another pattern is relevant.

B.3:1 - Problem frame

Assurance concerns a claim, not the world-side subject in isolation. Begin with one target claim, identified by its C.2.1 episteme or ClaimAddress as specified in §4.2, and one named use of an assurance conclusion. The claim’s EntityOfConcern remains the system, episteme, method, work occurrence, relation occurrence, or other subject identified by its pattern.

For example:

  • a battery-pack safety fact and its test result remain under their safety, measurement, and test patterns;
  • the episteme that states the safety claim remains under C.2.1;
  • an assurance result states whether a named argument carries that claim for a named release use;
  • a later gate, permission, or release decision remains a separate result.

The word calculus here means a disciplined way to select, combine, and interpret the inputs that the assurance argument consumes. B.3 defines no universal arithmetic across systems and epistemes.

B.3:2 - Problem

Five failures recur:

  1. Assurance by appearance. A badge, dashboard, attestation, card, status, or publication is treated as the assurance conclusion.
  2. One scale for unlike properties. Formality, reliability, coverage, congruence, and evidence quality are placed in one tuple even though they have different bearers and scales.
  3. Unsupported aggregation. min, an average, a penalty, or another fold is called conservative without a dependency model and calibrated quantities.
  4. Process burden by default. Every result is required to name dated Work, Method, assignment, bindings, and a reusable record even when the claim’s basis closes the use.
  5. Domain obligations absorbed into assurance. Safety, rights, access, responsibility, contest, redress, status, or controlled-action rules are replaced by a generic assurance record.

B.3:3 - Forces

ForceTension
Small result vs inspectabilityA practitioner needs a quick result, while a consequential or reusable claim may need a replayable argument.
Shared discipline vs domain meaningAssurance needs common boundaries, while each characteristic and aggregation rule gets its meaning from its subject and domain.
Conservatism vs useful synthesisThe result must avoid overstatement without discarding justified combination or independent lines of support.
Formal clarity vs warranted beliefFormalization can improve inspectability and checking while leaving truth or empirical adequacy unchanged.
Current use vs change over timeAn assurance conclusion must be usable now and reopen when its basis, scope, or conditions change.

B.3:4 - Solution

B.3:4.1 - Start from one assurance question

State these three things before choosing measures or a record:

  1. the target claim;
  2. the named assurance use;
  3. the conclusion that must be supported, narrowed, or refused for that use.

Keep the following objects separate:

  1. world-side facts and domain results;
  2. the target-claim episteme;
  3. evidence-use relations and source-provenance paths;
  4. any assessment Work and the System that performed it;
  5. formal, empirical, causal, measurement, conformance, or comparison input results;
  6. the assurance-result episteme;
  7. calculation traces, witnesses, and an optional note or publication that cites the result;
  8. later reliance, status use, gate, permission, release, or action.

Evidence can support or challenge a claim. It does not make the target fact true. A favorable assurance result does not pass a gate, grant permission, or prove that later work relied on it.

B.3:4.2 - Use the smallest sufficient result

The compact result contains only facts every B.3 use needs:

AssuranceResult:
  targetClaimRef:
  assuranceUse:
  basisRefs:
  disposition: supported-for-use | narrowed | abstain | evidence-needed | reopen | blocked
  limitationsAndNotCarried:
  reopenCondition:

targetClaimRef identifies the C.2.1 episteme or one C.2.1 ClaimAddress when the use concerns one addressed claim inside a larger episteme. basisRefs cite the results, evidence-use relations, provenance paths, argument claims, or domain rules used. A compact result is complete when these fields decide the named use and its basis and limits are recoverable.

Keep the assurance argument readable beside its result or through references that let the recipient inspect it. Explain why the cited basis supports the target claim for this use, including assumptions, limitations and contrary results. This written account is an assurance case.

Develop it while reasoning: start with the claim and use, state the proposed support, and revise the argument when a premise or inference proves inadequate. A short coherent paragraph can suffice; add detail when review, transfer or later reconsideration needs it.

Use C.2.1 to identify the epistemes whose claims make up the case. A claim about a system retains that system as its subject. For an assurance-result episteme, the EntityOfConcern is the target claim content or episteme whose support is being judged; targetClaimRef locates that subject.

A case can bring several such epistemes together, preserving their identities and argument relations. The text or diagram expresses their claims; E.24.PUB governs the expression and its carrier.

An unfinished argument remains useful working material: identify the missing support and return the appropriate unresolved disposition. A positive assurance result requires an argument sufficient for the named use under the applicable domain rules.

Add a claim scope, condition set, interpretation scheme, audience, or time window only when changing it could change the conclusion. Keep design and run conclusions separate whenever their inputs or conditions differ.

Add assessment Work, performer, Method, application bindings, witnesses, or a reusable record only when the receiving use depends on competence, conflict of interest, timing, reproducibility, contest, redress, or later replay. These identities are never mandatory merely because B.3 is used.

An optional assurance note may cite the result and its basis. Safety authority, contest, and redress remain governed by their domain rules, as described in §4.7.

B.3:4.3 - Name each characteristic by its bearer and scale

Include a characteristic result only when the assurance argument consumes it. State:

AssuranceCharacteristicResult:
  bearerRef:
  characteristic:
  scaleAndUnit:
  valueOrInterval:
  interpretationForThisUse:
  basisRef:

One characteristic name must not silently change meaning between subjects. System reliability, replication quality, evidential support, proof inspectability, and relation congruence are different characteristics even when a local source labels several of them R or CL.

The letters F, G, R, and CL may appear inside a declared local scheme, but B.3 assigns them no universal cross-domain meaning:

  • Formality or inspectability. Formal structure can make assumptions and inference steps easier to check. It raises assurance only when the named argument explains which uncertainty or verification need it closes. Making a wrong model proof-grade does not improve truth or empirical adequacy.
  • Claim scope. U.ClaimScope stays an A.2.6 value. It is not a quality coordinate. Widening or narrowing scope changes the claim and its applicable use under the declared scope rules.
  • Reliability-like characteristics. Use the domain definition, bearer and scale, with any applicable unit and the population or trials, conditions and qualification window required by that definition or the receiving use. A system reliability measure and an evidence-quality judgment are not interchangeable.
  • Relation congruence. Characterize one mapping, calibration, interface, or other relation occurrence only under a declared scale and interpretation. The value neither changes the participants nor supplies a universal penalty.

Never average ordinal values. Do not subtract an ordinal value from a ratio quantity. Thresholds and order comparisons are valid only under the scale that defines them.

B.3:4.4 - Aggregate only under an applicable model

B.3 supplies no default fold. When the assurance argument combines quantitative or ordered inputs, cite:

  1. the result claims being combined;
  2. the dependency or alternative-path structure;
  3. the domain aggregation rule or model;
  4. the assumptions required by the rule or model, including any independence, dependence, calibration or unit assumptions;
  5. the calculation or ordered comparison;
  6. any rival rule that is plausible at comparable effort and could change the result or its warranted use.

Use min only when the cited domain rule makes the weakest input a lower bound or bottleneck for the quantity. It is not universally conservative.

If no applicable aggregation rule is available, report the inputs separately and return a bounded, non-positive, or unresolved disposition. Do not manufacture one score to make the result look complete.

Several independent evidence lines may strengthen an argument only through the rule that states how their dependence and coverage are handled. Claim-scope intersection or union follows A.2.6 and the evidence model; it is not an assurance arithmetic shortcut.

B.3:4.5 - Add the proof detail the assurance use needs

Add calculated or model-bearing details, replay details, or both to the compact result when the named assurance use needs them.

Compact path. Use the six-field result in 4.2. Stop when it decides the named use.

Calculated or model-bearing path. Add the characteristic results, dependency structure, assumptions, aggregation rule, rival, calculation trace, and sensitivity or failure condition used.

Replay path. Add Work, performer, Method, application bindings, witnesses, and a reusable note only when those identities change the named assurance use.

For any assessment Work, use A.13 to identify the performer and A.15.1 to admit the dated occurrence independently. Add F.6 only if the replay must also say under which assignment the Work was performed. The Work, performer, Method, optional assignment check, result, witness, note, and publication remain separate.

Do not select the replay path merely because the use is important. Importance may make more basis necessary, but every added field must change inspectability, contestability, or the decision.

B.3:4.6 - Keep visible authority outside the result

A visible artifact, such as an attestation or dashboard tile, contributes to assurance only through a claim and basis relation used by the argument. Establish whether the artifact supplies a cue, source, evidence item, or representation for that use.

If the question concerns only a status, gate decision, permission, warning, or source location, use its pattern and produce no B.3 result.

If an assurance claim is current, cite the item only for the property it establishes. A valid signature or provenance chain can establish origin and integrity without establishing safety, truth, compliance, or readiness.

B.3:4.7 - Leave domain obligations with their patterns

B.3 evaluates an assurance claim. It does not define safety duties, access rules, responsibility, affected-party disclosure, contest, redress, people or team status, resource allocation, release authority, or controlled action. Cite each applicable rule as a premise or limitation.

When a domain rule says a consequential use requires an assurance claim, state that claim and then apply B.3. When the rule requires a decision, permission, review, contest route, or redress relation instead, use that result. A display that affects behavior does not by itself open B.3.

B.3:4.8 - Preserve time, currentness, and design/run distinctions

State the window only when time changes the assurance conclusion. Monitoring, drift, incidents, evidence refresh, version change, policy change, gate change, or a newly discovered defeater can narrow, reopen, or withdraw a result while the target fact and target-claim identity remain unchanged.

Design evidence and run evidence may support different claims. Produce separate results when target use, conditions, scope, or evidence window differs; compare them instead of merging them into one score.

B.3:4.9 - Use causal-use and method-structure results

When an assurance argument depends on a causal-use claim, consume the C.28 result and its stated supported and unsupported uses. B.3 does not re-run causal identification. An unsupported causal-use result narrows, blocks, or leaves the assurance claim unresolved; it does not become a low universal reliability coordinate.

When Method composition, fallback, selection or family organization matters to the assurance argument, first use B.1.5, G.5 or the predicate for the relied-on fact.

Use A.22 only when a separately selected organization changes the assurance question; then recover its selection basis and all four identity discriminators. MethodRelationStructure is a local designator for that selected structure, not a universal kind or a required record for every composition claim.

B.3:4.10 - Use Working-Model declarations only for what they state

An E.14 Working-Model assertion may contribute its declared validation posture and grounding links. A postulate needs the empirical basis required by the assurance use; an inferential claim needs its reasoning basis; an axiomatic or constructive claim needs the construction and identity basis it relies on. The declaration, grounding link, assessment Work, assurance result, and publication remain different objects.

B.3:5 - Proof obligations

B.3:5.1 - Common obligations

Every positive or narrowed B.3 result:

  1. identifies the target claim and named assurance use;
  2. cites only basis results and relations that bear on that use;
  3. states assumptions and limitations, including any unsupported stronger use relevant to the receiving decision;
  4. keeps target fact, claim, evidence, assessment, result, record, publication, and later use distinct;
  5. names a reopen condition;
  6. uses the domain rule for every safety, permission, access, status, release, responsibility, or controlled-action premise;
  7. avoids aggregation unless its model and assumptions are explicit.

B.3:5.2 - Additional obligations for a calculated result

A calculated result also names every bearer, characteristic, scale, applicable unit, dependency, calibrated mapping, aggregation rule and calculation it uses. Explain sensitivity to assumption failures when it matters for the named assurance use. If a rival rule is plausible at comparable effort and could change the result or its warranted use, show why the selected rule fits the declared dependency structure.

B.3:5.3 - Additional obligations for replay

A replayable result adds only the Work and performance facts needed by the receiving use. Follow the §4.5 replay route for each assessment Work. Add its Method, application binding, witness, timing fact, or separate F.6 assignment check only when competence, independence, reproducibility, contest, or redress depends on that fact. No record field stands in for an obtaining relation.

B.3:6 - Worked cases

B.3:6.1 - Calculated case: two necessary independent conditions

Target claim: “The protection function succeeds on a demand.” Assurance use: a bounded reliability argument for a named design decision.

The domain model says that both independently tested conditions must succeed: sensor detection and actuator response. Each has estimated probability 0.9 under the same stated demand class and qualification window. Under the declared independence assumption, the joint probability is:

0.9 × 0.9 = 0.81

Using min(0.9, 0.9) = 0.9 would overstate this conjunction. If the conditions are dependent, the product is not justified; the result must use the applicable conditional model or remain unresolved.

The B.3 result therefore cites the two domain results, the series dependency structure, independence basis, product calculation, 0.81 conclusion, limitations, and the observation that reopens the independence assumption.

What changes in practice: the design decision is evaluated against 0.81, not a falsely “conservative” 0.9. No universal B.3 reliability formula is created.

B.3:6.2 - Routed-away case: dashboard status

Starting sentence: “The dashboard approves launch.”

Suppose the dashboard displays GateDecision GD-17, which records that a named gate passed for release candidate R. The repaired sentence is: “The dashboard shows GateDecision GD-17 for release candidate R; the decision, not the display, records that the gate passed.”

Use A.21 and the release or permission pattern that consumes the gate decision. No assurance claim is present, so B.3 stops. If the dashboard does not resolve a gate decision, it is only a cue and launch approval remains unresolved.

B.3:6.3 - Episteme credibility with a compact result

Target claim: “Model edition M predicts response Y within the declared operating region.” Assurance use: whether an engineer may use that prediction as one input to a reversible design comparison.

The engineer cites the model claim, its empirical-validation result, A.2.4 first-use classification and the direct support relation used, the A.10 provenance path, the operating region, and the expiry condition. No combination of unlike characteristics is needed. The compact disposition is supported-for-use, limited to the reversible comparison; release, safety, and operation are not carried. No dated assessment Work or reusable record is added because the use does not depend on who performed the already cited validation.

The engineer states the supporting inference: “The validation compared M’s predictions with observations over the region needed for this comparison, and its stated error bounds meet the comparison’s tolerance. Those predictions are therefore supported as one input to the reversible comparison within that region and the stated validity window.”

These are premises of this worked case. If the validation does not establish that coverage or tolerance, narrow the use or return an unresolved disposition. Keep the argument beside the compact result or accessible through basisRefs. A later correction to the validation reopens the conclusions that depended on the corrected result.

B.3:6.4 - Order-sensitive Method case

An assurance argument relies on a manufacturing sequence whose result changes when two steps are reversed. Recover the B.1.5 composition, order and join conditions and any A.15.1 Work the argument consumes. Those results may answer the assurance question.

Use A.22 only if a separately selected organization contributes another needed premise; then identify its constituents, obtaining relations, applied constraints and use frame. The assurance result cites only the results and structure it relies on.

B.3:6.5 - Bias annotation

RiskCountermeasure
Formal prestigeAsk what uncertainty or verification need the formalization closes.
One-number preferenceKeep unlike characteristics separate unless an applicable model combines them.
Visible-authority biasResolve the claim and property established by the artifact.
Record completion biasAdd only fields the named use consumes.
Consequence inflationUse domain rules; consequence alone creates no assurance claim.

B.3:7 - Conformance checklist

IDRequirement
CC-B3-1One target claim and one named assurance use are stated before measures or records are selected.
CC-B3-2The compact result contains target, use, basis, disposition, limits, and reopen condition; optional fields appear only when they change the conclusion or its replay.
CC-B3-3Every characteristic has one bearer, property, scale and unit where applicable, interpretation, and basis.
CC-B3-4Formality is not treated as monotone truth, empirical adequacy, or warrant.
CC-B3-5Aggregation cites the domain model, dependencies, required assumptions, applicable units, calculation or ordered comparison, and any relevant rival; no default min, mean, penalty, or ordinal arithmetic is used.
CC-B3-6Design and run results remain separate when their conditions or evidence differ.
CC-B3-7Target fact, target claim, evidence use, assessment Work, input results, assurance result, witness, note, publication, and later reliance or decision remain recoverable separately.
CC-B3-8A label, dashboard, card, provenance mark, attestation, or publication contributes only the property established through a cited relation.
CC-B3-9Safety, rights, access, responsibility, contest, redress, status, permission, release, and controlled action remain with their patterns.
CC-B3-10A causal-use premise cites the C.28 result. A Method premise cites its composition, selection or other result; add an A.22 structure only when independently selected organization changes the assurance question, with all four identity discriminators recoverable.
CC-B3-11Work, performer, Method, bindings, witnesses, reusable notes, and an optional F.6 assignment check are added only for a replay, competence, independence, timing, contest, or redress need. Every Work follows the §4.5 A.13 then independent A.15.1 route.
CC-B3-12A positive result states its reopen condition and any unsupported stronger use relevant to the receiving decision.

B.3:8 - Common anti-patterns and repairs

Anti-patternWhy it failsRepair
Universal weakest-link formulamin can be an upper bound for a conjunction and can ignore alternatives or dependence.Use the dependency model and domain rule; otherwise report inputs separately.
Ordinal penalty subtractionAn ordinal label is subtracted from a probability or ratio quantity.Use a calibrated mapping to the quantity when one exists, or keep the values separate.
Formality raises assuranceA more formal wrong model receives a better result.State the inspectability gain and the uncertainty it closes; keep truth and empirical adequacy separate.
Same letter, different propertyR names system reliability in one row and evidence quality in another.Give each characteristic its bearer and definition.
Safety-case inflationA warning, access value, or consequential display triggers a generic B.3 package.Use the domain pattern; apply B.3 only to an assurance claim.
Evidence creates truthNew evidence is said to make the target fact obtain.Revise the warrant or disposition unless the world-side facts changed.
Assessment-record collapseA checklist, trace, witness, or note is treated as the assessment Work or result.Identify each object separately and add only what the use consumes.
Design/run chimeraBlueprint evidence and runtime observations are merged into one score.Produce separate results and compare them.

B.3:9 - Consequences

Benefits

  • Assurance remains explicit without forcing one cross-domain score.
  • A small local claim can stop with the compact result and its recoverable argument.
  • Visible assumptions and dependency structure let the practitioner check whether a calculation applies.
  • Domain safety, access, responsibility, status, and decision rules retain their meaning.
  • Visible artifacts can contribute useful provenance or evidence without becoming authority.

Trade-offs

  • B.3 supplies no convenient universal number. A project must use the domain model that gives its inputs meaning.
  • Some assurance questions remain unresolved until a dependency model, calibrated mapping, or domain requirement is supplied.
  • Reusable replay records can add cost; include them when the receiving use needs them.

B.3:10 - Rationale

Assurance-case practice supports explicit claims, arguments, evidence, and maintenance. Reliability engineering supports calculations tied to dependency structure and assumptions. Neither supports one universal F-G-R-CL score across unlike subjects. B.3 therefore standardizes the boundaries and the minimum result while leaving characteristics and aggregation with the models that define them.

B.3:10.1 - Decision-bearing SoTA account

Practical questionSourceContribution to B.3Applicability condition
What makes an assurance case inspectable?ISO/IEC/IEEE 15026-2:2022, edition 2 specifies assurance-case structure and maintenance. The GSN Community Standard v3 gives engineering-argument notation and guidance.Adopt explicit claim, argument, evidence, context, defeater, and maintenance structure. Adapt it to the compact and replay paths. Reject document or diagram appearance as assurance, approval, or permission. Sections 4.1–4.2 apply this structure to the smallest result sufficient for the use.Use the governing assurance-case edition and check that the compact result states the claim, argument, and evidence needed for the receiving use.
How may reliability-like inputs be combined?NASA’s Fault Tree Handbook with Aerospace Applications, v1.1, chapter 6, derives AND-event probability from conditional probability and uses a product only under independence.Adopt dependency- and assumption-specific calculation. Reject universal min and any claim that it is always conservative. Section 4.4 and case 6.1 apply the dependency model before selecting arithmetic.Use a dependence model warranted for the claim. A different validated domain model changes which calculation applies.
Can one trustworthiness tuple compare unlike AI or system properties?NIST AI Risk Management Framework 1.0 treats trustworthiness characteristics as context- and use-dependent and recognizes trade-offs and domain metrics.Adopt named characteristics and use-specific thresholds. Reject same-letter cross-domain comparability and monotone formality. Section 4.3 keeps the bearer and scale explicit.Keep numerical assessments of unlike properties separate unless a validated common measurement model supplies the bearers, scales, units, and interpretation needed for their comparison.
What can provenance and attestation establish?C2PA Technical Specification 2.4 distinguishes provenance validity from value judgments. in-toto Attestation Framework 1.2 defines authenticated metadata about software artifacts.Adopt subject, source, binding, and verification claims. Reject a valid credential, manifest, attestation, or display as automatic truth, safety, compliance, readiness, or release. Section 4.6 and case 6.2 distinguish the verified property from the later decision.Check the property warranted by the applicable specification and whether the domain rule makes it sufficient for the named use.

B.3:11 - Relations

  • Builds on: C.2.1 for target and assurance-result epistemes; A.2.4 for evidence-use classification; A.10 and G.6 for source-provenance paths and bounded reliance; A.2.6 for ClaimScope; C.16 and C.16.Q for characteristic and scale discipline; and the domain patterns for every input result.
  • Coordinates with: A.15.1 and A.6.1 only when assessment Work and applications matter; G.11 for currentness; F.10 for status; A.21 for gates; permission, commitment, release, access, responsibility, contest, redress, safety, and controlled-action patterns for their results; and E.17, E.24.PUB, and C.29 for publication and representation.
  • Coordinates with: C.28 for causal-use results; B.1.5, G.5 or the pattern for the relied-on Method fact; A.22 only for an independently selected organization needed by the assurance argument.
  • Used by: a pattern or project decision that consumes one assurance result. The consumer applies its decision, permission, gate, status, or work rule.

B.3:11a - Quantum-like claims

Quantum-like wording does not require assurance by itself. If the wording only prevents a local representation mistake, keep the note with C.26 and ordinary evidence. Apply B.3 only when an assurance claim about that quantum-like result and use is current. A comparative superiority claim must name the rival model, baseline, claimed mechanism, scope, evidence, and loss. Mathematical novelty or prestige supplies no assurance.

B.3:11b - Mathematical-lens use

When a C.29 mathematical-lens result is an input to an assurance claim, cite the lens-result claim, its interpretation and limits, the evidence-use and provenance relations relied on, and the named assurance use. Mathematical elegance or a structure-preserving mapping does not raise assurance by itself. Measurement construction and comparability remain with C.16.

B.3:End

Referenced in the corpus

757 literal mentions in other sections. Read their context to establish the relation.