G.4 - CAL Authoring for a CG-Frame: Operators, Acceptance Clauses, Evidence Wiring
Status: Stable
Use this when. A team has typed characteristics and now needs to publish reusable operators, acceptance clauses, and legal compositions before any candidate is evaluated. The working object is one design-time CAL Pack@CG-Frame, not an evaluation run, verdict, selector outcome, assurance case, or decision.
First move. Write one plain acceptance statement for one task: “For a subject x of the admitted kind, when the conditions of this use fall within claim scope S and satisfy any further temporal condition of the clause, apply operator O to a C.16 measurement result for Characteristic K that argument declaration R admits. When applying the clause, bind current result episteme E; the application of clause A returns pass | fail | unknown under threshold or policy P and any currentness conditions needed for that use.” Then turn only the reusable operator and clause definitions into stable CAL declarations. E belongs to the later application, not to reusable clause A.
Smallest viable CAL pack. Publish one charter for the exact CG frame, one typed operator card, one acceptance clause with ClaimScope, any evaluation window required by its predicate or receiving use, and unknown or failure behavior, one legal flow, one evidence and currentness profile, one proof-or-gap row and one worked declaration example. Publish a minimal editioned TaskMap that cites the exact charter, the C.22 TaskSignatureRef, and the declaration refs used by selection. Stop there when this pack answers the task; method-family extensions, archive surfaces, crossing records, and additional policy pins enter only when the case needs them.
What changes in practice. Thresholds and failure behavior become inspectable, illegal arithmetic becomes an authoring defect, and runtime workers can cite stable declarations when applying an operator or clause.
Not this pattern. Use C.16 for the measurement result, A.19 for comparison or selection, A.13 to identify each performer, A.15.1 for independent admission of dated evaluation Work, and F.6 when the receiving claim needs to establish that this Work was performed under the same assignment, A.6.1 for actual bindings, C.2.1 for the verdict episteme, A.10/G.6 for provenance, G.11 for currentness, B.3 for assurance, and C.11 for a decision. If the immediate question is whether the clause was applied and what verdict that application returned, use §4.4a to distinguish the declaration from its execution.
G.4:1 - Problem frame
CAL authoring starts from:
- one exact
CG‑Framewith itsEntityOfConcern,ReferencePlane, task, and assumption envelope, - a plurality of method traditions and claims (SoTA inputs), and
- CHR‑typed measurement constructs (
Characteristic/Scale/Coordinate+ legality guard macros).
Before downstream users run a selection, comparison, aggregation or selected-set formation, authors need an explicit, auditable CAL Pack that:
- defines what operators exist and what they are allowed to do over CHR types,
- externalizes fit-for-purpose acceptance as typed predicates whose use is bounded by an exact
ClaimScope, any evaluation window required by the predicate or receiving use, and any separate qualification window that limits use, and - binds these choices to an evidence wiring surface (lanes, provenance anchors, policy pins, and refresh triggers) so that downstream selection, logging, parity, and shipping can cite stable ids rather than re‑inventing semantics.
Use this kit to author and publish CAL declarations at design time. Use G.Core for shared Part G invariants, CG‑Spec for the comparison legality gate and CN‑Spec for governance.
G.4:2 - Problem
A CAL Pack connects typed measurements to later selection and evaluation. Drift or ambiguity in that connection can produce:
- Illicit operations include treating an encoding as cardinal without grounds, using incompatible units as if they agreed, and applying arithmetic that assumes intervals to ordinal values.
- Acceptance is scattered (thresholds embedded in code or in CHR prose; predicates not typed; unknown handling inconsistent).
- Evidence wiring is underspecified (which provenance anchors matter, what policy ids are in force, what is plane‑scoped, what changes must trigger refresh).
- Cross-sense or cross-plane imports are silent (hidden reuse across distinct source-local meanings, ReferencePlanes, or editions without the obtaining relation, required crossing records, and loss accounting).
- Tooling artifacts become semantics (vendor flags or implementation details substitute for a conceptual specification).
G.4:3 - Forces
- Expressiveness vs legality. CAL must allow useful comparisons/aggregations while staying lawful under CHR typing and legality gates.
- Pluralism vs comparability. Multiple method traditions must coexist without forcing premature unification, yet remain cross‑citable and auditable.
- Decision support vs auditability. CAL must support selection and selected-set formation while preserving explicit, reviewable assumptions and proofs.
- Exploration vs assurance. CAL must support exploratory regimes (probing, novelty, open‑ended search) without letting un‑assured outputs silently become dominance claims.
- Locality vs portability. Each CAL clause stays bounded by its declared
ClaimScope, any applicable temporal bounds, source meanings, and ReferencePlane; reuse beyond that boundary requires the exact relation and crossing records that the changed value calls for.
G.4:4 - Solution — author the smallest lawful CAL pack
G.4:4.0 - Practitioner authoring path C1–C9
Complete these actions in order. Extend a step only when the task needs the additional input named there.
- C1 — Charter the scope. Name the exact
CG‑Frame,EntityOfConcern,ReferencePlane, task,CNSpecRef.edition, andCGSpecRef.edition. State the assumption envelope in ordinary language. - C2 — Declare one typed operator. Give it a stable id, CHR-typed signature, preconditions, result kind, and failure behavior. The card declares the operation under
A.6.1. Evidence for an application comes from its execution (§4.4a). - C3 — Declare one acceptance clause. Name the exact Characteristic and the A.6.1 argument declaration that admits the corresponding C.16 measurement-result episteme. Declare the predicate or threshold,
ClaimScope, any evaluation window required by the predicate or receiving use, any separate qualification window that limits use, unknown handling, and the stated stop, degrade, or abstain behavior. Keep the exact current result episteme for the later application. If the clause claims statistical risk or coverage control, also name the loss, target, calibration population and window, sampling or exchangeability assumptions, declared treatment of shift, and the exact policy that states the guarantee. - C4 — Compose only a legal flow. Cite the operators and gating clauses, preserve the lawful result kind, and keep a selected set when no lawful scalarization exists. A declared DAG is possible composition, not performed work.
- C5 — Name the minimum evidence/currentness need. Cite the exact A.10 source/provenance anchors and G.11 currentness conditions needed to judge the clause. Do not turn an evidence profile, citation, or graph membership into a verdict or actual reliance.
- C6 — Add an extension only when the task needs one. Select its current subject pattern first, then pin only the descriptor, distance, insertion, exploration, branch, or path records that change the present CAL action. Otherwise omit the extension.
- C7 — Record proof or an explicit gap. For every operator, flow, or clause, cite each required justification for legality, monotonicity or boundedness. If any required justification is missing, publish the gap and the consequent degrade/abstain behavior.
- C8 — Exercise declaration behavior. Provide one worked authoring example and focused conformance tests for illegal operations,
pass | fail | unknown, currentness conditions when they are part of admissibility, and failure behavior. The example and test remain declarations/test records unless separately grounded dated work is named. - C9 — Publish and hand off. Mint stable ids and continuity notes, then emit the smallest immutable
TaskMapedition. It cites the exact charter edition, the already constituted C.22TaskSignatureRef, the task, and edition-bearing operator, flow, gating-clause, and evidence-profile refs; the cited evidence profiles carry the needed currentness pins. It neither constructs the TaskSignature nor copies clause thresholds. Use G.11 for change refs; G.4 defines no refresh rule or runtime occurrence or result.
The path is complete when a reader unfamiliar with the author’s work can reconstruct the acceptance statement from the published ids and identify what must happen at runtime. The manifests, schemas, interfaces and optional extensions below make that same pack machine-citable. The order of work is C1–C9.
G.4:4.1 - G.Core linkage (normative)
Builds on: G.Core (Part-G core invariants)
GCoreLinkageManifest (normative). Canonical shape, Nil‑elision, and the Expansion rule are defined in G.Core.
GCoreLinkageManifest := ⟨
CoreConformanceProfileIds := {
GCoreConformanceProfileId.PartG.AuthoringBase,
GCoreConformanceProfileId.PartG.TriStateGuard,
GCoreConformanceProfileId.PartG.UTSWhenPublicIdsMinted,
GCoreConformanceProfileId.PartG.ShippingBoundary
},
CorePinSetIds := {
GCorePinSetId.PartG.AuthoringMinimal,
GCorePinSetId.PartG.CrossingVisibilityPins
},
CorePinsRequired := {
UTSRowId[], // CAL artefacts are public ids (Name Cards plus public-id continuity notes)
ΓFoldRef.edition? // pin an actual numerical composition model/policy when used; otherwise cite the governing rule
},
// consumed iff no explicit `ΓFoldRef.edition` override is pinned
DefaultsConsumed := { DefaultId.GammaFoldForR_eff },
RSCRTriggerSetIds := { GCoreTriggerSetId.SoTAHarvestSynthesis },
RSCRTriggerKindIds := { // deltas (Expansion rule applies)
RSCRTriggerKindId.PenaltyPolicyEdit,
RSCRTriggerKindId.DefaultGoverningDefinitionChange,
RSCRTriggerKindId.BaselineBindingEdit
}
⟩
By the G.Core Expansion rule, the effective conformance ids / trigger kinds / pin obligations for G.4 are the expansions of the referenced profiles/sets/pin‑sets plus the explicit deltas above.
Notes (normative intent, delegated semantics):
- Apply
G.Corefor tri-state outcomes, penalty routing, set-return discipline, crossing visibility, the P2W split and typed RSCR causes. Its Default Governing Definition Index identifies the rule for each default. - EvidenceGraph/Path pins, when used, are declared only via
G.4:Ext.EvidenceGraphWiringin G.4:4.5. - Method‑specific pins (e.g., QD descriptor/distance/insert policy pins; open‑ended transfer rules pins) MUST appear only in Extensions blocks (see G.4:4.5) and MUST NOT introduce competing defaults.
G.4:4.2 - CAL Pack@CG-Frame surface (kit governed by this pattern)
CAL Pack@CG-Frame is the CG‑Frame’s published CAL Pack. Minimally, it provides:
-
CAL.Charter— identification and assumption basis for this CAL pack:- cites the exact
CGFrameId,EntityOfConcernRef, andReferencePlane, - cites the governance and legality records (
CNSpecRef,CGSpecRef) by edition, - records the assumption envelope on which the acceptance predicates rely without minting another governance or legality record.
- cites the exact
-
TaskMap— the conditional G.4 handoff record toG.5when CAL gates are current; one exact edition names the task, cites the already constituted C.22TaskSignatureRefand exact charter edition, and cites the acceptance-clause, operator, flow, and evidence-profile refs that selection consumes. It does not constitute the TaskSignature or contain threshold values. -
CAL.Operator[]— UTS‑published typed operation declarations governed byA.6.1; a card declares possible arguments, result kinds, and conditions but does not assert that an operation ran:- declared signature over CHR types,
- declared preconditions and postconditions, including references to legality guard macros,
- provenance and evidence hooks identified by ids or pins independently of tool behavior.
-
CAL.Acceptance[]— typed predicate declarations whose use is bounded by the declaredClaimScope, any evaluation window required by the predicate or receiving use, and any separate qualification window that limits use; a clause declares how an actual application is judged but is not itself a verdict:- binds to CHR characteristic ids and to exact A.6.1 argument declarations for admissible C.16 measurement-result epistemes (and, when inducing numeric comparison or aggregation, to
CG‑Spec.characteristicids), - keeps the exact current result episteme in the later application binding rather than in the reusable clause,
- exposes unknown handling and failure behavior via policy pins.
- binds to CHR characteristic ids and to exact A.6.1 argument declarations for admissible C.16 measurement-result epistemes (and, when inducing numeric comparison or aggregation, to
Each resultInputDeclarationRef resolves an A.6.1 ArgumentDeclaration. Its meaning identifies the C.16 measurement-result episteme expected by the clause. The declaration states the ValueKind and binding designation rule and requires a result for the named Characteristic with the admissible shape. A one-off clause may also cite an already existing episteme in fixedResultEpistemeRefs[]?; mark that clause one-off instead of presenting it as reusable.
-
CAL.Flow[]— legality‑checked declarations of possible operator composition; a declared DAG is not performed work:- declares result kind (scalar only when lawful; selected-set / set-result when partial orders remain partial orders),
- records which acceptance clauses gate which flows.
-
CAL.EvidenceProfiles— evidence wiring surface:- lane tags under their declared regime / provenance anchors / policy pins needed for
SCRand audit surfaces, - currentness conditions needed by the receiving use, with freshness-window and decay/Γ_time policy refs pinned only when those mechanisms are used.
- explicit
ReferencePlaneand any used penalty-policy refs (Φ(CL),Ψ(CL^k),Φ_plane); the ProofLedger supplies the receiving quantity, scale, input meanings, assumptions, and derivation or calibration. Monotonicity and boundedness alone do not justify a numerical loss.
- lane tags under their declared regime / provenance anchors / policy pins needed for
-
Optional
CAL.NQD[]— QD/OEE‑related calculus surfaces when declared:
A CAL clause that consumes G.7 calibration states its exact receiving use and premises. A locally justified CL threshold is an additional policy condition; the ProofLedger and named authority must support that condition. It does not replace the bounded-use claim, target classification or reliance result. A policy waiver changes only the authorized policy condition.
-
CAL.ProofLedger— a proof/justification ledger:- links operator/flow/clause ids to their needed legality and soundness results; a numerical support fold includes the B.3/C.2.2 receiving model, compatible scales and inputs, dependence assumptions, and boundary behavior.
-
Publication artifacts:
- UTS Name Cards under F.17/F.18 for all public ids,
- ids for RSCR tests and worked examples,
- deprecation notices and edition bump notes as public-id continuity records.
Boundary discipline (normative):
- No shadow specs: CAL artefacts cite
CN‑Spec/CG‑Specand do not introduce competing “local specs” (delegated; seeCC‑GCORE‑CN‑CG‑1via CC‑G4‑CoreRef). - Shipping boundary:
G.10governs shipping; seeCC-GCORE-SKP-1via CC-G4-CoreRef. - Refresh boundary: CAL publishes pins/payload for refresh;
G.11governs refresh orchestration.
Minimal schema fragments (notation‑independent; fields for citation, not an implementation schema):
CAL.Charter :=
⟨ charterId, charterEdition, cgFrameId, entityOfConcernRef, referencePlaneRef,
CNSpecRef.edition, CGSpecRef.edition, assumptionEnvelope ⟩
CALCharterRef := <charterId, charterEdition>
TaskMap :=
⟨ taskMapId, taskMapEdition, charterRef := CALCharterRef,
taskRef, taskSignatureRef := TaskSignatureRef,
acceptanceClauseRefs[], operatorRefs[], flowRefs[], evidenceProfileRefs[] ⟩
TaskMapRef := <taskMapId, taskMapEdition>
CAL.Pack@CG-Frame :=
⟨ calPackId, charterRef, taskMapRef, operatorIds[], acceptanceClauseIds[], flowIds[],
evidenceProfileIds[], proofLedgerId, nqdIds[]?,
utsRowIds[], workedExampleIds[], rscrTestIds[], publicIdContinuityNoteIds[] ⟩
CAL.Operator :=
⟨ operatorId(UTS), signature(CHR-typed), preconditions[], postconditions[],
evidenceProfileRefs[]?, failureBehaviorRef?, crossingRefs[]? ⟩
CAL.Acceptance :=
⟨ clauseId(UTS), characteristicRefs[], resultInputDeclarationRefs[],
fixedResultEpistemeRefs[]?, // deliberately one-off clause only
cgSpecCharacteristicRefs[]?, predicateRef, claimScopeRef,
evaluationWindow?, qualificationWindow?, unknownHandlingRef,
failureBehaviorRef, evidenceProfileRefs[]?, crossingRefs[]? ⟩
CAL.Flow :=
⟨ flowId(UTS), dag(operatorIds, edges), gateClauses(acceptanceClauseIds),
resultKind, decisionAidPolicyRef? ⟩
CAL.EvidenceProfile :=
⟨ evidenceProfileId(UTS), lanes(declared tags), anchors(A.10)[],
freshnessPolicyPins[]?, penaltyPolicyPins[]?, ΓFoldRef.edition? ⟩
lanes carries the tags and meanings of the declared evidence-lane regime; when a CHR charter supplies that regime, retain its tags and meanings. The assurance claim, receiving use and selected contributions follow B.3/B.3.3. Keep any consumed C.2.2:4.1 tuple, with its exact claim and F, G and R_eff values, separately recoverable from the lane tags.
CALCharterRef and TaskMapRef each resolve one immutable edition. A changed charter, task, TaskSignature edition, clause, operator, flow, evidence profile, or edition-bearing cited list creates a new taskMapEdition; an old TaskMapRef continues to resolve its old values. The C.22 TaskSignature remains a separately constituted episteme. The map relates that exact signature to the CAL declarations used by selection but neither derives the signature nor duplicates their thresholds.
When CAL authoring relies on SoTA coverage, consume G.2’s pack CoverageJudgementRef with its HarvestPolicy basis and receiving question. Preserve the distinction between family coverage and lineage/material-entry plurality. A CAL clause does not turn a combined method/generator count into method-only coverage.
G.4:4.4 - Interfaces (minimal I/O surface)
| Interface | Consumes | Produces |
|---|---|---|
G.4-1 Charter | exact CGFrameId, EntityOfConcernRef, ReferencePlane, CNSpecRef.edition, CGSpecRef.edition, assumption envelope, SoTA inputs, CHR Pack@CG-Frame | one immutable CAL.Charter edition and its CALCharterRef |
G.4-2 Operators | CHR typing + SoTA operator inventory | CAL.Operator[] (UTS ids; typed signatures; refs to evidence profiles & guards) |
G.4-3 Acceptance | task intent, exact Characteristic and A.6.1 result-input argument declarations, ClaimScope, any evaluation window required by the predicate or receiving use, any separate qualification window that limits use, policy pins, and CHR characteristics; exact result-episteme refs only for a clause explicitly marked one-off | CAL.Acceptance[] (typed predicate or threshold; admissible result-input declarations; scope; applicable evaluation and qualification windows; currentness pins when used; unknown and failure behavior refs) |
G.4-4 Flows | Operator cards + admissible aggregators | CAL.Flow[] (legality‑checked compositions; declared result kind) |
G.4-5 NQD Surface | Task intent + policy pins + (optional) QD/OEE inputs | CAL.NQD[] (descriptor/distance/insertion refs + edition pins; optional) |
G.4-6 Publish | all above, exact task, C.22 TaskSignatureRef, proofs, and examples | versioned CAL Pack@CG-Frame, exact CALCharterRef, and the smallest immutable TaskMap edition plus TaskMapRef, citing the task, matching TaskSignature, and acceptance-clause, operator, flow, and evidence-profile refs; also UTS entries, RSCR tests, Worked-Examples, and public-id continuity notes |
G.4:4.4a - Declaration-to-runtime evaluation boundary (normative)
A CAL pack is a reusable design-time declaration. A stored operator card, clause, flow, TaskMap, proof-ledger row, test, or evidence-profile reference establishes neither an actual participant nor performed evaluation. When a CAL declaration is applied, recover the following runtime chain:
- Name one exact
EvaluationMethod(U.Method). ItsU.MethodDescriptionmay state generic participants, parameters, effects, and evaluation conditions, but it carries no actual-participant slots and no intrinsic claim that a test, proof, or acceptance event occurred. - Cite the exact
CAL.Operator,CAL.Flow, andCAL.Acceptancedeclarations asA.6.1operation semantics. Resolve the clause’sresultInputDeclarationRef, then bind the exact current C.16 measurement-result episteme in this application and test it against the declaration’s Characteristic and admissible result shape. Use the exactA.6.1declaration and application bindings; do not infer them from a compatible signature,TaskMap, or stored reference. - First recover every precise performer’s A.13 core for the exact evaluation action, scope, working situation, and window, including the same obtaining assignment later used by any attribution. A.15.1 then independently admits one dated
EvaluationWork : U.Workfrom its performance history, enacted Method, extent, and containing-System relation. Add F.6 afterward only when the receiving claim needs exact assignment-bound attribution through that same assignment. Recover the evaluated or affected referent, actual resources, and every concrete participant through its direct subject relation or anA.6.1application binding. A compact attribution account may omit only an assignment identifier unused by its receiving claim; it omits no consumed fact. Ordinary activity not claimed asU.Workdoes not enter this branch. - State the local result under its direct predicate and pattern. A
CAL.Acceptanceapplication yields its exactpass | fail | unknownverdict; use A.19 for comparison and selection results, C.16 for measurement results, and C.11 for a decision result. No generic evaluation-result or work-result field substitutes for these objects. - When a durable assertion is needed, constitute one
C.2.1result episteme whose ClaimGraph states that local result, evaluated subject, interpretation basis, polarity or domain status, and uncertainty when current. The episteme is not the domain result and does not create it. - Attach source recovery and provenance through A.10/G.6 and currentness through G.11. State the A.10 evidence-provenance account and local
RelianceDispositionfor the bounded use, citing the direct rule for any reliance relation. Enter B.3 when an actual named assurance claim is current; a consequential use without one retains its direct governing rule. A citation, ledger edge, evidence profile, disposition, or assurance record does not establish the work, participant, application, or local result it describes. - A later selector, acceptance action, or decision is another governed occurrence. It relies on the result episteme through an exact premise, reference, decision-use, or operation-argument relation; mere storage, citation, or graph membership does not establish actual use.
This chain keeps declaration, execution, local result, result episteme, provenance, bounded reliance, currentness, acceptance, and decision independently recoverable.
G.4:4.5 - Extensions (pattern‑scoped; non‑core)
G.4 supports method‑family and discipline‑specific calculus variations exclusively via pattern‑scoped extensions.
GPatternExtension block: G.4:Ext.EvidenceGraphWiring
- PatternScopeId:
G.4:Ext.EvidenceGraphWiring - GPatternExtensionId:
EvidenceGraphWiring - GPatternExtensionKind:
InteropSpecific - GoverningPatternId:
G.6 - Entry: use when this CAL pack needs to cite an addressable G.6 path or slice, whether one receiving use requires it or several consumers share it.
- Stop: omit the block when a local A.10 source-to-use account is sufficient; remove it when no current clause, proof, or example cites the path.
- Uses:
{G.6} - ⊑/⊑⁺:
∅ - RequiredPins/EditionPins/PolicyPins (minimum):
EvidenceGraphId?PathId[]/PathSliceId[]UTSRowId[](for cited artifacts)
- RSCRTriggerSetIds:
∅ - RSCRTriggerKindIds:
{RSCRTriggerKindId.EvidenceSurfaceEdit, RSCRTriggerKindId.EditionPinChange, RSCRTriggerKindId.PolicyPinChange} - Notes: CAL proofs and examples may cite G.6 evidence paths by these ids.
GPatternExtension block: G.4:Ext.NQD
- PatternScopeId:
G.4:Ext.NQD - GPatternExtensionId:
NQD - GPatternExtensionKind:
MethodSpecific - GoverningPatternId:
C.18 - Entry: use only when the current task applies a C.18 quality-diversity/archive method and its descriptor, distance, insertion, or archive policy must be pinned for CAL use.
- Stop: omit or retire the block when the task has no current archive/QD clause or when those refs no longer change a CAL action.
- Uses:
{C.18} - ⊑/⊑⁺:
∅ - RequiredPins/EditionPins/PolicyPins (minimum):
DescriptorMapRef.editionDistanceDefRef.editionInsertionPolicyRefArchiveRef?TaskSignatureRef?(if activation is TaskSignature‑bound)
- RSCRTriggerSetIds:
∅ - RSCRTriggerKindIds:
{RSCRTriggerKindId.EditionPinChange, RSCRTriggerKindId.PolicyPinChange, RSCRTriggerKindId.TelemetryDelta, RSCRTriggerKindId.FreshnessOrDecayEvent} - Notes: C.18 defines QD semantics. CAL pins the descriptor, distance and insertion records needed to reproduce archive behavior. Any archive/illumination summaries (e.g., coverage / QD‑score / occupancyEntropy / filledCells) are published as report‑only outputs unless an explicit CAL acceptance clause/policy authorizes promotion.
GPatternExtension block: G.4:Ext.EELog
- PatternScopeId:
G.4:Ext.EELog - GPatternExtensionId:
EELog - GPatternExtensionKind:
MethodSpecific - GoverningPatternId:
C.19 - Entry: use only when the current task has a C.19-governed exploration/exploitation budget or probe-accounting rule that changes a CAL clause or failure branch.
- Stop: omit or retire the block when no current CAL action consumes those C.19 refs.
- Uses:
{C.19} - ⊑/⊑⁺:
∅ - RequiredPins/EditionPins/PolicyPins (minimum):
ExploreExploitBudgetPolicyRefProbeAccountingRef?FailureBehaviorRef?(if probe/sandbox is policy‑bound)
- RSCRTriggerSetIds:
∅ - RSCRTriggerKindIds:
{RSCRTriggerKindId.PolicyPinChange, RSCRTriggerKindId.TelemetryDelta, RSCRTriggerKindId.FreshnessOrDecayEvent}
GPatternExtension block: G.4:Ext.SoSLogBranches
- PatternScopeId:
G.4:Ext.SoSLogBranches - GPatternExtensionId:
SoSLogBranches - GPatternExtensionKind:
MethodSpecific - GoverningPatternId:
C.23 - Entry: use only when C.23-governed SoS-LOG branches currently explain a CAL degrade/abstain path.
- Stop: omit or retire the block when those branch/rule ids no longer change a current CAL clause, flow, or explanation.
- Uses:
{C.23} - ⊑/⊑⁺:
∅ - RequiredPins/EditionPins/PolicyPins (minimum):
SoSLogRuleId[]SoSLogBranchId[]FailureBehaviorPolicyId
- RSCRTriggerSetIds:
∅ - RSCRTriggerKindIds:
{RSCRTriggerKindId.PolicyPinChange, RSCRTriggerKindId.MaturityRungChange, RSCRTriggerKindId.TelemetryDelta} - Notes: The pinned C.23 branches and rules explain the clause’s degrade/abstain behavior.
G.4:5 - Archetypal Grounding
Tell. A team working within a CG‑Frame must choose and justify a set of candidate methods (possibly a selected set or a set retained in an archive) under explicit legality, evidence, and scope constraints. CHR provides the typed measurement basis; CAL declares auditable predicates and flows that separately grounded runtime work may apply.
Show 1 (bounded CAL pack skeleton).
Use: R&D selected-set choice. The pack names the exact CG frame and EntityOfConcern, candidate-set ClaimScope, ReferencePlane, and evaluation window. CHR defines SafetyClass(ord↑), CostUSD_2026(ratio↓), Readiness(nominal).
CAL.Operator: DominatesParetoSignature over CHR types, precondition references CHR guard macros.CAL.AcceptanceClause: AC_SafetyGateReusable typed predicate forSafetyClass(and its levels), citingSafetyResultArgument-D1, the A.6.1 argument declaration that admits a C.16 measurement-result episteme for that Characteristic. Thresholds are valid for the statedClaimScopeand evaluation window; unknown handling uses tri-state pins. The clause names no current measurement-result episteme.CAL.Flow: Flow_ParetoPortfolioDeclares a selected-set result kind; gated byAC_SafetyGateandAC_Budget.CAL.EvidenceProfile: EP_SafetyEvidenceDeclares anchor ids and freshness policy pins required forSCR.
When this CAL pack supplies selector gates, it publishes TaskMapRef=<SafetySelectionMap, E3>. That map cites CALCharterRef=<SafetyCALCharter, E2>, C.22 TaskSignatureRef=SafetyPortfolioTaskSignature-E4, the exact task, and edition-bearing refs AC_SafetyGate-E2, Flow_ParetoPortfolio-E1, DominatesPareto-E3, and EP_SafetyEvidence-E4; it contains no threshold values. Downstream G.5 consumes the exact TaskSignatureRef and this TaskMapRef together, verifies that the map cites the same signature, and resolves the charter and declarations through their refs. If the charter or a cited clause changes, G.4 publishes another TaskMap edition; selectors that still cite <SafetySelectionMap, E3> continue to replay the old boundary.
Show 2 (explicit cross-sense or ReferencePlane import).
A SafetyClass result uses an expression with a different F.17 source-local meaning or comes from another ReferencePlane. CAL may author a clause using it only after the exact F.17 cells and obtaining F.9 relation are cited when meanings differ, and the applicable plane or edition crossing records are cited when those values differ. The clause keeps its declared ClaimScope and window; the import does not silently widen either.
Show 3 (one performed acceptance evaluation).
Before the candidate action is admitted as Work, A.13 recovers SafetyEvaluatorSystem-17 : U.System for exact action SafetyAcceptanceEvaluationAction-17. Its admitted SafetyEvaluatorBoundary-17 contains the evaluation controller, its active decision state, and the input/output channels through which it applies the clause; it excludes the CAL declarations, measurement-result episteme, candidate, assignment, and containing team System. The action’s scope is SafetyAcceptanceClaimScope-17, its working situation is SafetyGateEvaluationSituation-17, and its window is 2026-07-30T09:00:00Z through 2026-07-30T09:20:00Z. It is directed by SafetyAcceptanceDecisionNorm-17: apply the current clause to admissible current inputs, return unknown rather than force a threshold verdict when uncertainty crosses the boundary, and reject an input whose declared result shape is incompatible. The relevant conditions are clause edition, result-shape admissibility, measurement currentness, and uncertainty relative to the threshold.
The local kind SafetyAcceptanceEvaluatorSystemRole is declared under A.2. Its membership criterion requires the stable work-facing contribution of safety-acceptance evaluation and goal-directed, condition-sensitive regulation under SafetyAcceptanceDecisionNorm-17: the holder must bind admissible inputs, choose the clause-defined verdict, and abstain or return unknown when the declared conditions require it. SafetyEvaluatorDecisionTrace-17 shows SafetyEvaluatorSystem-17 rejecting an incompatible result shape and returning unknown when the admissible uncertainty interval crosses the threshold; the system-boundary and runtime records show that those actions occurred within SafetyEvaluatorBoundary-17. The cited decision-trace, system-boundary and runtime records support the criterion facts under A.2’s membership rule; A.10 makes that source-to-use account recoverable. The case independently classifies SafetyEvaluatorSystem-17 under SafetyAcceptanceEvaluatorSystemRole. Neither the candidate Work nor the assignment supplies that classification. No Grade, autonomy result, characteristic profile, or stronger assurance claim is consumed here.
The same A.13 core uses SafetyAcceptanceEvaluationAssignment, a directly declared U.SystemRoleAssignment species under A.2.1. The species defines holder, assigned-kind, and evaluation-candidate participant meanings; its predicate appoints the holder to evaluate that candidate under the applicable clause for the stated scope, situation, and window. SafetyAcceptanceEvaluationAssignment-17 obtains with SafetyEvaluatorSystem-17 as holder, SafetyAcceptanceEvaluatorSystemRole as assigned-kind value, and C-17 as evaluation candidate. Its maximal uninterrupted predicate-true interval covers the full stated window.
Only after that A.13 core is established does A.15.1 independently admit EvalWork-2026-07-30-17 : U.Work from its exact performance history, enacted SafetyAcceptanceMethod, temporal extent, and obtaining containing-System relation to independently admitted SafetyEvaluationTeamSystem-17. The actual A.6.1 application SafetyAcceptanceApplication-17 binds candidate C-17 separately from its binding of current C.16 measurement-result episteme SafetyMeasureResult-E17 to SafetyResultArgument-D1 while using unchanged reusable clause AC_SafetyGate. Neither the assignment nor an F.6 conclusion is an A.15.1 admission premise.
Because this worked case states that the Work was performed under an assignment, F.6 afterward establishes performedUnderAssignment(EvalWork-2026-07-30-17, SafetyAcceptanceEvaluationAssignment-17) through the same obtaining A.13 assignment. The direct case fact links the named Work and assignment, the holder is the performer, and the assignment interval covers the Work. A different overlapping assignment held by the same performer would not establish this attribution.
SafetyMeasureResult-E17 states the measured safety characteristic, scale, attributed value, uncertainty, model, calibration, and measurement Work; it is neither the raw detector output nor the acceptance verdict. The clause application obtains unknown because the uncertainty interval crosses the threshold. A later SafetyMeasureResult-E18 can bind through separately identified SafetyAcceptanceApplication-18 while AC_SafetyGate remains unchanged. A C.16 result for CostUSD_2026, a result with an incompatible declared shape, or raw detector output fails SafetyResultArgument-D1 before the predicate runs; it does not cause a new reusable clause edition. A separate C.2.1 episteme asserts that exact verdict and cites its provenance under A.10 and, when the EvidenceGraph extension is present, G.6; G.11 supplies currentness. A later C.11 result may record defer, and its claim uses the verdict episteme through an exact premise or decision-use relation. Any decision-making Work remains separate. The clause card, proof-ledger row, evidence edge, and decision record do not retroactively establish the measurement Work or the evaluation occurrence.
Show 4 (a support model and its separate acceptance threshold).
For the pump-triage use in G.5 §0.5, consider a separately declared gated variant. AC_InputConditionGate-E1 accepts a declared probability of at least 0.85 that both the measurement series is valid (event A) and its time alignment is valid (event B) for the same 24-hour input. Its scope, evaluation window, result-input declaration, and unknown behavior belong to this G.4 clause, not to the default composition rule. PumpInputEvidence-E1 supplies the receiving model and provenance for those event meanings and probabilities. An applicable TaskMapRef binds that clause, profile, and the matching C.22 task signature.
Suppose the application inputs establish P(A)=0.9 and P(B)=0.9 and the profile’s model establishes independence. The model gives P(A ∩ B)=0.81, so the clause returns fail. Minimum 0.9 would incorrectly pass. If independence or a justified conditional alternative is unavailable, the missing joint probability yields the clause’s unknown result, with the declared downstream degrade/abstain behavior; a high F or formal line tag cannot replace it.
For a different receiving question that asks only for a qualified comparison on existing heterogeneous support, the profile retains the proof, empirical contributions, shared-bias limitations, and contrary evidence under B.1.3/C.2.2. It does not invent a probability or apply AC_InputConditionGate-E1 to that different question. The ordinary G.5 shortlist remains available without an additional assurance calculation.
G.4:6 - Bias-Annotation
CAL clauses define what counts as acceptable. The choices below can bias that judgement:
- threshold‑selection bias (arbitrary floors masquerading as natural laws),
- measurement bias amplified by illegitimate arithmetic or hidden scalarization,
- survivorship bias in Worked‑Examples and probe telemetry,
- Goodhart pressures when report‑only telemetry is accidentally treated as dominance.
The pattern mitigates these by requiring typed acceptance clauses, explicit policy pins, and an auditable ledger of proofs and justifications, while identifying cross-sense and ReferencePlane reuse. For a numerical assurance penalty, the model for the receiving use determines which results it affects.
G.4:7 - Conformance Checklist (normative)
| ConformanceId | Statement |
|---|---|
| CC‑G4‑CoreRef | Conformance with G.4 requires satisfying the effective G.Core obligations referenced by the GCoreLinkageManifest in G.4:4.1 (profiles, pin sets, consumed defaults, and trigger kinds). |
| CC‑G4‑01 | CAL Pack@CG-Frame is published as a notation-independent object with stable UTS ids (Name Cards under F.17/F.18) for CAL.Charter, TaskMap, all operator, acceptance-clause, flow, and evidence-profile declarations, Worked-Examples, and public-id continuity notes, including deprecations and lexical-continuity notes. Tooling/vendor details remain non-normative. |
| CC‑G4‑02 | Each exact CALCharterRef = <charterId, charterEdition> resolves one immutable charter edition naming the exact CGFrameId, EntityOfConcernRef, ReferencePlane, CNSpecRef.edition, CGSpecRef.edition, and assumption envelope on which the pack relies. |
| CC‑G4‑03 | Every CAL.Operator has a declared CHR‑typed signature and stated preconditions. Cite each referenced legality guard macro by id. |
| CC‑G4‑04 | Every reusable CAL.Acceptance binds the exact Characteristic and exact A.6.1 resultInputDeclarationRef values that declare the admissible C.16 measurement-result episteme inputs; the exact current result episteme is bound only in an actual application. A clause marked one-off may additionally cite fixedResultEpistemeRefs[]?. Every clause also declares its predicate or threshold, ClaimScope, any evaluation window required by the predicate or receiving use, any separate qualification window that limits use, unknown handling, and failure behavior. A statistically risk-controlled clause also names its loss, target, calibration population and window, sampling or exchangeability assumptions, declared treatment of shift, and the exact policy that states or defines the guarantee. Inputs with distinct source-local meanings cite the exact F.17 cells and obtaining F.9 relation; cross-plane or cross-edition inputs cite their applicable crossing records. None of these declarations establishes performed evaluation or a verdict. |
| CC‑G4‑05 | If an acceptance clause, operator, or flow induces numeric comparison or aggregation, it cites the relevant CG‑Spec.characteristic ids and links to legality proof refs (CSLC) in the ProofLedger; if required legality grounds are missing, declare how downstream use degrades or abstains before the unsupported operation. |
| CC‑G4‑06 | Every CAL.Flow declares its result kind and the set of gating acceptance clauses; any thinning/selection‑aid policies (e.g., ε‑front selection) are explicitly policy‑bound and do not silently replace the underlying result kind. |
| CC‑G4‑07 | Every CAL.EvidenceProfile declares the provenance anchors, evidence lanes, and currentness or loss-policy pins its use needs. It cites DefaultId.GammaFoldForR_eff for B.3/C.2.2 support-model discipline and pins any numerical Γ-fold actually used. That calculation must preserve input meanings, scales, dependencies, formal/empirical distinctions, and mapping limits; absent a common model, retain separate support and a bounded synthesis. The selected receiving model determines which results a numerical loss consequence affects, including whether an R-only restriction applies. Loss assessment alone defines neither dominance nor an acceptance threshold. |
| CC‑G4‑08 | CAL.ProofLedger links each operator, flow, or clause to its required proof or justification and explicit failure behavior. A numerical Γ-fold or loss includes its receiving quantity, input scales, dependency model, derivation or calibration, assumptions, and boundary behavior; monotonicity and boundedness alone are insufficient. A missing required model cannot be repaired by F-to-R conversion or an invented default score. |
| CC‑G4‑09 | CAL publication includes RSCR tests and Worked‑Examples sufficient to detect illegality (incl. unit laundering / ordinal arithmetic), to exercise authored acceptance/flow behavior, and to validate the authored freshness envelope when it is part of admissibility; record missing tests or examples as an auditable gap. |
| CC‑G4‑10 | Each TaskMapRef = <taskMapId, taskMapEdition> resolves one immutable map edition containing the exact CALCharterRef, task, C.22 TaskSignatureRef, and edition-bearing acceptance-clause, operator, flow, and evidence-profile refs used by selection. Changing the charter, task, signature edition, or a cited component creates a new map edition. When G.4 gates are current, G.5 consumes this exact map alongside the same TaskSignatureRef; a mismatch or unresolved ref blocks that gated selector use. The map neither constructs the TaskSignature nor embeds thresholds or duplicates acceptance semantics. |
| CC‑G4‑11 | Any method/discipline specifics are placed under G.4:4.5 Extensions as GPatternExtension blocks (stable PatternScopeId, explicit governing definition, pins, and RSCR triggers); no extension introduces competing defaults or replaces G.Core invariants. |
| CC‑G4‑12 | CAL Pack@CG-Frame includes public-id continuity records for public ids: deprecations, edition bumps, and lexical-continuity notes. It exposes refresh payload pins, including editions, policies, UTS ids, and, when present, PathId and PathSliceId, sufficient for G.11 to plan RSCR without inferring semantics from prose. |
| CC‑G4‑13 | When G.4:Ext.NQD is present, CAL.NQD[] is present and is wired only via the declared subject pattern (C.18): at minimum it pins DescriptorMapRef.edition, DistanceDefRef.edition, and InsertionPolicyRef, and it treats archive/illumination summaries as report‑only unless explicitly promoted by a CAL acceptance clause/policy. |
| CC‑G4‑14 | CAL does not mint new universal types to encode “strategy/policy”. Strategy is expressed as authored flows + acceptance clauses + policy/task pins (and downstream registry/composition in G.5); any specialization is introduced only via GPatternExtension wiring blocks or cited subject patterns. |
| CC‑G4‑15 | Every runtime example keeps the reusable Method, MethodDescription, A.6.1 declaration, each precise performer’s A.13 core, independently A.15.1-admitted dated U.Work, optional later F.6 assignment-bound attribution, actual bindings and direct participants, local result, and C.2.1 result episteme distinct. Any F.6 relation uses the same obtaining A.13 assignment. A compact account may omit only an unused assignment identifier and no consumed fact; ordinary activity outside U.Work does not enter this rule. |
| CC‑G4‑16 | Each performed acceptance or decision path names its exact local result, applies the relevant result pattern, and states the exact later-use relation. Provenance stays with A.10/G.6, currentness with G.11, assurance with B.3, and decisions with C.11; no universal evaluation-result, work-result, evidence-use, or criterion-participant relation is introduced. |
G.4:8 - Common Anti-Patterns and How to Avoid Them
-
Hidden thresholds. Avoid: embedding cutoffs in CHR prose or in operator descriptions. Prefer:
CAL.AcceptanceClausewith explicit ids and pins. -
Untyped “score(x)”. Avoid: operators with implicit units and untracked legality assumptions. Prefer: explicit CHR‑typed operator signatures + cited legality checks.
-
Silent cross-sense or cross-plane reuse. Avoid: importing expressions with distinct source-local meanings, or values across ReferencePlanes or editions, without the obtaining relation and required crossing records. Prefer: cite the exact F.17 cells and F.9 relation when meanings differ, cite the applicable plane or edition crossing records, and keep each clause bounded by its stated
ClaimScopeand window. -
Acceptance as implementation detail. Avoid: acceptance embedded in tool logic. Prefer: publish acceptance as citable CAL artifacts; downstream consumes ids.
-
Exploratory telemetry treated as dominance. Avoid: letting probe/illumination telemetry quietly become a dispatch criterion. Prefer: keep it report‑only unless a policy-bound acceptance clause expressly authorizes its promotion.
-
Declaration mistaken for execution. Avoid: treating a CAL card,
TaskMap, proof-ledger row, worked example, or evidence edge as proof that an operator ran or a verdict obtained. Prefer: recover every precise performer’s A.13 core, let A.15.1 independently admit the dated Work, and add F.6 only when the receiving claim needs to establish that the Work was performed under the same obtaining assignment; recover actual direct bindings separately. Compact wording may omit only an unused assignment identifier and no consumed fact. Keep the domain-local result and any result episteme separate from both.
G.4:9 - Consequences
- CAL operator/acceptance semantics become stable, citable CAL Pack content rather than tacit code behavior.
- Legality failures are surfaced as authoring defects (RSCR‑testable) rather than run‑time surprises.
- Downstream patterns (
G.5,G.8,G.9,G.10,G.11) can reference stable ids/pins without redefining acceptance or operator semantics. - Method pluralism is supported: multiple calculi can coexist as separate operator/flow/acceptance families, wired via Extensions rather than mixed into the core kit.
G.4:10 - Rationale
CAL sits at the boundary where typed measurement becomes actionable choice. Publishing typed and testable CAL declarations reduces semantic drift and prevents “shadow legality gates” from emerging in tools or in downstream prose.
The design separates concerns:
- CHR governs measurement typing and legality guard macros,
- CG‑Spec and CN‑Spec govern the legality gate and governance card, respectively,
G.Coregoverns Part‑G invariants and trigger/default discipline,G.4governs the CAL kit: authoring objects, publication surface, and handoff manifest.
This yields modularity (one governing definition per invariant or default), auditability (pins/ids and proof refs), and extensibility (method families attach through explicit extension modules).
G.4:11 - SoTA-Echoing
These sources contribute to the CAL clauses or extensions identified below under the stated conditions. Reconsider the affected use when its source assumptions or adopted contract change.
| Source and adopted contribution | Use in CAL | Applicability limit | Return when |
|---|---|---|---|
| Angelopoulos, Bates, Fisch, Lei, and Schuster, Conformal Risk Control, ICLR 2024 — adapt bounded monotone-loss risk control only for a CAL clause whose statistical assumptions are explicit. | C3 and CC-G4-04 require loss, risk/coverage target, calibration population/window, exchangeability or declared shift treatment, and failure/abstain behavior before such a clause is published. Only an actual application under §4.4a supplies the required binding and verdict. | “Conformal”, a calibration set, or a coverage target does not make a universal acceptance guarantee, authorize deployment, or establish that evaluation occurred. | Reopen only C3, CC-G4-04, and the one worked clause/test that claims this guarantee when its assumptions or guarantee change. |
Fontaine, Togelius, Nikolaidis, and Hoover, Covariance Matrix Adaptation for the Rapid Illumination of Behavior Space, GECCO 2020 — adapt only the need to pin descriptor, distance, insertion, archive, and reporting policy when G.4:Ext.NQD is used. | C6, G.4:Ext.NQD, and CC-G4-13 keep QD method semantics in C.18 while making the CAL wiring reproducible. | Archive occupancy, coverage, QD-score, or the presence of CMA-ME wiring is not dominance, acceptance, selection, or a runtime result. | Reopen only C6, G.4:Ext.NQD, CC-G4-13, and its one NQD example/test if the adopted descriptor/archive contract changes. |
| Wang et al., Enhanced POET: Open-ended Reinforcement Learning through Unbounded Invention of Learning Challenges and their Solutions, ICML/PMLR 119 (2020) — reject as a source of G.4 core or acceptance semantics; retain it only as an exact lineage reference for optional exploration wiring under the applicable pattern. | C6 and CC-G4-11 require an exact current pattern and present-task entry and stop condition before any exploration extension is admitted; no POET-specific rule enters the CAL core. | Open-ended generation, transfer, or progress telemetry does not become a CAL acceptance rule, task authority, or selected governor by citation. | Reopen only C6, CC-G4-11, and the exact C.19/C.23 extension block if the applicable pattern explicitly adopts a changed POET-family rule set. |
Use A.18/A.19 for comparison, partial-order and selected-set rules. Distributionally robust and broader multi-objective methods are further candidates to investigate when the receiving question needs them. Revisit only the clauses, extensions and applications that depend on a changed source claim.
G.4:11.1 - CAL architecture and publication inventory
The charter fixes the pack’s scope and assumptions. Typed operators and clauses define admissible applications; flows compose them; evidence profiles and the ProofLedger expose their required grounds. The TaskMap connects these declarations to selection. Later applications bind actual inputs and produce results under §4.4a. Extensions add only the method-specific declarations needed by that use.
G.4:12 - Relations
Builds on: G.Core (and the pattern template discipline in E.8).
Uses: G.1 (CG‑Frame Card), G.2 (SoTA Synthesis Pack), G.3 (CHR Pack), G.0 (CG‑Spec legality gate), A.19 (CN‑Spec plus direct comparison and selection patterns), A.18 (CSLC), A.2.6 (U.ClaimScope), A.6.1 (declarations and actual bindings), A.13 (precise performer core), A.15.1 (independent Work admission), A.2.1 (assignment species and occurrences), and F.6 only when the receiving claim needs to establish that the Work was performed under the same obtaining assignment, C.2.1 (result epistemes), C.11 (decision results), A.10 (provenance and bounded reliance), B.3 (assurance), G.11 (currentness), and E.18 (transformation-flow structure and GateCrossing), A.21 (gate decisions from independent check results), F.9 (actual relations between source-local meanings), F.17 (SchemeSenseCells), and E.17 (multi-view publication).
Uses (via Extensions): G.6 (EvidenceGraph/Path citation; when G.4:Ext.EvidenceGraphWiring is present), C.18 (NQD), C.19 (E/E‑LOG), C.23 (SoS‑LOG).
Used by: G.5 (selector/dispatcher), G.8 (SoS‑LOG bundles), G.9 (parity), G.10 (shipping), G.11 (refresh orchestration).
Publishes to: UTS (public ids and public-id continuity records), RSCR (tests and trigger emissions), G.5 (handoff manifest), and, as cited payload, shipped packs governed by G.10.
Constrains: any run‑time LOG implementation that executes CAL operators/flows must treat CAL artifacts as citable specifications and must not re‑invent acceptance semantics.