Notable changes to energy.grid-coordination/clj-urpx. The format follows Keep a Changelog; the project is pre-1.0, so a minor bump is the strongest compatibility signal available and is used for breaking changes.
Entries begin at 0.11.0. The ten releases before it carry no summaries in their release commits, and reconstructing them from the commit log would be guesswork rather than history; see git log and the v0.1.0–v0.10.0 tags for what actually landed.
contributedBy, applicableToRateGroup and referencesGeneralRule (typed by their ranges), and vocabulary-valued slots, which take a reference until their vocabularies are enforced. The drift audit reports no missing property.write-* takes :minimal-context true, keeping only the term definitions the document uses, from the default context or a given one, and every prefix declaration. A document's @context then changes only when its content does, not when the default context grows.urpx.emit/default-context types urpx:dateTimeRetrieved, urpx:sunsetOn and urpx:sunsetOnDateTime. A document whose @context extends the default gains these three terms, unless it is written with :minimal-context.ServicePointProfile, DistributedEnergyResource, AccountServiceAgreement, PriorRatePlan, BillingCycle and ScenarioCostResult become entity schemas with builders. CustomerProfile models the rest of its shape's properties, urpx:hasPriorRatePlan among them; Ledger models urpx:hasBillingCycle; Scenario.hasServicePointProfile and ScenarioCostModel.producesResult take their node inline as well as by reference.urpx.schema/OperationNode and urpx.dsl/operation-node: arithmetic over two operands, itself an operand, so a condition can compare a computed value. It nests to any depth in the condition tree.urpx.emit/default-context types urpx:gYearValue as xsd:gYear. It did not, so a gYear the emitter wrote under the default context lost its datatype.EnergyServiceProvider, RegulatoryAuthority and SystemOperator, the subclasses of Organization, and OrganizationEvent, OrganizationTransition and LocationDataSet become entity schemas with builders. An organization's urpx:hasLocationDataSet takes a LocationDataSet inline or by reference.urpx.schema/AnyOrganization takes any of the five kinds inline, chosen by @type, or a reference, and every slot whose range is urpx:Organization uses it. OfferingOrganization is now an alias for it; publishedBy and evaluatedBy took a plain Organization only.24:00 decodes to midnight, as 24:00:00 already did, and encodes as 00:00:00, so the DSL takes a tariff's hh:mm end of day as written. Other hour-24 times still refuse.urpx.emit/write-graph writes several top-level entities as one @graph document, with the @context on the document.urpx.emit/write-entity writes any entity whose @type urpx.schema registers, such as a SourcePublication or a RateCase.write-* takes :json, options passed to clojure.data.json/write-str, e.g. {:indent true :escape-slash false}.dsl-scope takes :outside, {keyword @type}, naming nodes built elsewhere. Such a keyword reads as a reference and is judged against its slot's range by the @type given; a builder in the scope may not declare it.urpx.dsl/literal writes a typed literal, {:jsonld/value v :jsonld/type datatype}.@context, @id, @type, other @ keywords, then properties, each group in string order. Output no longer depends on the order a map was built in; it differs in key order from earlier releases, with the same graph.dsl-scope accepts an @id declared twice when both builds yield the same node, so a node embedded in two places may be built in each. Two different nodes under one @id are still refused.write-urpx-package encodes every urpx:packages entry whose @type urpx.schema registers, where it encoded RatePlan, RatePlanModifier and ScenarioCostModel entries only.urpx.schema/URPXDocument, urpx.coerce/coerce-urpx-document, urpx.emit/write-urpx-document and urpx.dsl/urpx-document; urpx.core/wrapper-types holds urpx:URPXPackage alone.urpx.schema/TariffBookReference, urpx.dsl/tariff-book-reference and DataSourceReference's urpx:hasTariffBookReference.urpx:TariffPublication spelling of SourcePublication, and the v0.2.1 properties it carried that v0.5.1 drops.urpx:publishedBy and urpx:supersedes on both version snapshots; timezone inference reads urpx:offeredBy only, and the EIA id only from the identity chain, not urpx:eiaId.urpx:regulatoryFiling member of SourcePublicationType, and the v0.2.x-only subclass axioms.clojure -M:validate's :missing-coverage-claim and :unparseable-target-version findings, and the :unchecked key of urpx.validate/check-file.urpx:forRatePlan, urpx:forRatePlanModifier and urpx:hasCoverageClaim are required on the version snapshots, as sh:minCount 1 at v0.5.1. The drift audit accepts three findings, down from eight.urpx.coerce/coerce, encode, validate and explain compiled their schema on every call. They now compile it once per schema. Coercing 100 ScenarioCostModel nodes took about 40 seconds and now takes milliseconds, so clojure -M:validate over a corpus report of several thousand models finishes rather than running for an hour.PricingModel, GeneralRule, ModelCollection, ServiceClass and RateGroup become entity schemas, with builders pricing-model, general-rule, model-collection, service-class and rate-group, and urpx.schema/CollectionRole enforces urpx:collectionRole. A PricingModel's urpx:hasPlanElements is an open map, since a partial model may name the plan's metric inputs and ledgers by reference.TariffFiling validates the models, rules and sources it holds (urpx:containsPricingModel, urpx:hasGeneralRule, urpx:documentedBy), where it accepted any map.PriceSet, urpx:priceCurrency, chargeType, billPortion, hasMetricSpecification and hasDataSourceReference; on PriceDefinition, urpx:pursuantTo, hasApplicabilityGroup and contributedBy; Price.contributedBy; on RatePlan, urpx:description, serviceRole, commodity, offeredBy, hasServiceClass and hasRateGroup; RatePlanModifier.offeredBy; RatePlanVersion.name and consolidatedFrom; on URPXPackage, urpx:publishedBy, publicationPlatform and dateReceived; Ledger.hasCalculationMethod.clojure -M:validate checks every entity type urpx.schema registers, where it checked eleven. A file whose entities are, for example, a TariffFiling is now checked where it was refused.urpx.schema/registry threw a ClassCastException for a schema whose children Malli looks up by a non-keyword; urpx.schema/schemas is now an unsorted map.urpx.schema/schema-for-type returns nil for a non-string @type.urpx.schema/DynamicPricing, with its DynamicPricingType, PriceSource and UpdateFrequency enums, and urpx.dsl/dynamic-pricing. Price models urpx:hasDynamicPricing, and Price and PriceDefinition model urpx:hasDataSourceReference, so their values validate and references through them are judged by mistyped-references.urpx.schema/PriceUpdateRule and urpx.dsl/price-update-rule, the value of urpx:hasPriceUpdateRule.urpx.schema/GYear: an xsd:gYear decodes to java.time.Year and encodes back. CustomOperandNode and DatatypeOperandNode model urpx:gYearValue.CalculationMethod models urpx:aggregationMethod, urpx:intervalType, urpx:intervalAlignment and urpx:intervalSelection, and MetricSpecification the last three, each enforcing its vocabulary (AggregationMethod, IntervalType, IntervalAlignment, IntervalSelection). MetricInput models urpx:description; RatePlanModifierVersion models urpx:expectedEndDate and urpx:expectedEndDateTime.urpx.schema/schema-for-type returns nil for an @type with no prefix, where it threw.:urpx.resolved/dynamic-pricing carries :urpx/priceSource and :urpx/hasPriceUpdateRule as vectors, since both are 0..N, where it passed a bare value through as written. A consumer reading [:urpx.resolved/dynamic-pricing :urpx/hasPriceUpdateRule :urpx/description] now reads [... :urpx/hasPriceUpdateRule 0 :urpx/description].urpx.core/containment-properties includes urpx:hasDynamicPricing and urpx:hasPriceUpdateRule.sh:or of sh:class alternatives states the set as its :urpx.schema/range, and urpx.schema/in-range? accepts an instance of any member. mistyped-references, dsl-scope and the builders now judge urpx:basedOnPrice (Price or PriceSummary), urpx:offers, urpx:packages, urpx:cancelledBy, urpx:rootNode, urpx:operandNodes, urpx:leftOperand and urpx:rightOperand.urpx:adjustsServiceTier takes a urpx:ServiceTier.scripts/shacl-drift.sh compares a union range with the shipped sh:or.urpx.schema/schemas, registry and schema-for-type name every entity schema by its @type, so (m/validate :urpx/Season x {:registry schema/registry}) works.sh:class states that class as :urpx.schema/range, read by urpx.schema/ranges; urpx.schema/instance-of? follows the ontology's subclasses. urpx.index/mistyped-references reports every reference and inline node across a set of documents that is not an instance of its slot's class.scripts/shacl-drift.sh reports a range-drift finding, which fails the run, where a range differs from the shipped sh:class.urpx.dsl/dsl-scope refuses a keyword reference to an entity of the wrong type under :mistyped, and every builder refuses an inline node of the wrong type.doc/: pricing, scenario cost models, coverage, writing JSON-LD, authoring, schemas and status. The README keeps the overview, installation, getting started, a short status and a list of the guides.clojure -M:validate fails a file (exit 1) holding a reference or inline node of the wrong type for its property, judged within the file, and prints each as MISTYPED.urpx.scenario-cost-model/identity-status-at gives a plan or modifier identity's status at an instant: :in-force with the version enumerate :at takes and the bounds of its time in force, :not-in-effect outside the identity's lifetime, or :not-modeled inside it with no version in force, naming the versions on either side.urpx.schema/RatePlan decodes urpx:expirationDateTime, and urpx.schema/RatePlanModifier decodes urpx:expirationDate and urpx:expirationDateTime.corpus/corpus_json.clj walks a kind's own roots when the kind declares them, and the layout's roots otherwise, as corpus.json specifies. It walked only the layout's, so a kind rooted elsewhere, such as urpx-rate-plans' clojure-source under src/, listed no files.urpx:serviceTierNumber, urpx:touPeriodNumber) and a start, the first listed was kept and the rest dropped silently. Now a Price whose every urpx:basedOnPrice candidate is absent from the documents passed in is set aside, and a Price with no urpx:basedOnPrice always applies. Exactly one must remain; otherwise the cell's tier carries no unit price and a :urpx.resolved/unvaluable of reason :ambiguous-cell (two or more apply) or :no-applicable-price (none does), listing the tied :price-ids. A modifier carrying one Price per base plan, such as a credit derived from each base's usage Price, now values the one for the base it is composed with.:compatibility defaults to named-versions-add, so naming a urpx:applicableToRatePlanVersion adds to urpx:applicableToRatePlan and retracts nothing. This applies to enumerate, enumerate-pairs, applicable-modifiers, from-node and everything built on them. A modifier naming plans P and Q and one version of P now applies to Q's versions, where it applied to none of them. Pass {:compatibility scm/named-versions-narrow} for the old behavior.:urpx.schema/double-not-decimal, carrying the value; the message says to write the value as a string. The JSON-LD 1.1 API's object to RDF conversion reads such a JSON number as an xsd:double, so SHACL rejects it, and the JSON reader had already dropped its written scale (0.1610 decoded as 0.161M). This applies to a parsed document, bare or as a value object's @value, to a Clojure number given to urpx.coerce/coerce, and to urpx.coerce/encode and the urpx.emit writers, which now write a ratio as a string rather than a JSON number. An integral number, 3.0 included, decodes as before, with scale 0. A urpx.dsl builder accepts an exact integer of magnitude 1E21 or more, and the urpx.emit writers write it as a string.{"@value": "0.1", "@type": "xsd:decimal"} or {"@value": "--06", "@type": "xsd:gMonth"}, by its @value, holding it to the datatype's lexical space. A value object of another @type, of none, or with an @language, stays as written and fails validation. A urpx.dsl builder refuses a double, a float or a non-terminating ratio as the @value of an xsd:decimal value object with :urpx.dsl/inexact-decimal, as it does a bare one.3.0, bare or as the @value of a value object typed xsd:integer, as a long, as the JSON-LD 1.1 API reads 3.0 as "3"^^xsd:integer; it failed validation. One outside the range of a long refuses with :urpx.schema/integer-out-of-range, and any other double or float with :urpx.schema/double-not-integer, each carrying the value.1/3, with :urpx.schema/non-terminating-ratio, carrying the value; it threw a bare ArithmeticException.--MM (--01 to --12) with an optional timezone. Any other string, such as "--+6", "--1e3", a bare "6" or non-ASCII digits, refuses with :urpx.schema/malformed-g-month, carrying the value; "--+6" and an Arabic-Indic digit decoded as 6, and "--1e3" threw a bare NumberFormatException."-9e-05", NaN, INF, whitespace), or a NaN or infinite double, with :urpx.schema/malformed-decimal, and a string outside the xsd:integer lexical space with :urpx.schema/malformed-integer (:urpx.schema/integer-out-of-range beyond a long); each carries the value.scripts/shacl-drift.sh compares against the urpx-org/urpx v0.5.1 release tag by default. The shapes at that tag carry the same constraints as the previous pin, and the release's ontology widens the domains of urpx:description to include urpx:Evaluation and of urpx:effectiveDate to include urpx:RatePlanVersion, both of which urpx.schema already models.urpx.schema/ProgramEnrollment, from urpx-shacl:ProgramEnrollmentShape at v0.5.1, with the ProgramType and EnrollmentStatus vocabularies. urpx.coerce/coerce-doc dispatches it, urpx.validate checks it, urpx.emit/write-program-enrollment writes it, and urpx.dsl/program-enrollment builds it. urpx.emit/default-context types urpx:enrollmentStartDate and urpx:enrollmentEndDate as xsd:date.urpx:referencesValue, in a urpx:hasPossibleValues list, or under urpx:CustomerProfile's urpx:hasProgramEnrollment coerces and validates as one. urpx:hasProgramEnrollment is a containment property.urpx.load/documents and urpx.load/scan return each urpx:ProgramEnrollment as a document, and scan no longer lists it under :urpx.load/uncoercible. urpx.validate checks it instead of reporting it unmodeled.urpx:mondayIfSundayObservance (urpx-org/urpx#77): a holiday on Sunday is observed the following Monday, and a holiday on Saturday on its date. urpx.holidays/derive-predicate applies it to urpx:holidayRule and urpx:holidayDate holidays, and urpx.schema/ObservanceRule accepts it.urpx.scenario-cost-model/prepare reads a document set once: its version snapshots under the :repeated-ids policy, and its zone index under :package-zones. plan-versions, applicable-modifiers, enumerate, enumerate-pairs and from-node take the prepared value in place of the documents and give the same answers, except that prepare applies :repeated-ids to plan and modifier versions together, where plan-versions and applicable-modifiers given documents each apply it to their own kind. The two options are fixed at prepare; a call stating another value refuses :urpx.scenario-cost-model/prepared-option-conflict. prepared? recognizes the value. Every existing arity prepares internally.dsl-scope entity's local @id comes only from :as. (dsl/season :as :scl/summer …) replaces (dsl/season :scl/summer …). A leading qualified keyword is a key, so (dsl/organization-identifier :skos/notation "16868" …) works inside a scope. :as outside a scope throws.dsl-scope names @ids with a function of the keyword, by default urpx.dsl/curie. :scl/summer gives "scl:summer"; it gave "summer". (dsl/dsl-scope {:id-fn f} …) supplies another function.dsl-scope checks keyword references when its body finishes. A reference may come before its declaration, so a version can name its enclosing plan. The scope first realizes every lazy sequence in the body's value, so a builder inside for or map runs in the scope. Every keyword referenced but not declared in the scope throws in one ex-info, naming each keyword and the @type of each builder that used it, under :unresolved. A reference to an undeclared keyword threw DSL local ref not declared at the call.dsl-scope refuses an @id declared twice, by :as or by an explicit :jsonld/id, naming both declarations under :declared-by. The messages and ex-data of the former refusals are replaced.urpx:hasPossibleValues value is an RDF list object, {:jsonld/list [...]} with at least one member (urpx.schema/RdfList). A vector of bare members, which emitted an unordered set, refuses. The vendored fixtures, the urpx test cases and urpx-rate-plans write only list objects.Organization models the v0.5.1 properties, among them urpx:hasIdentity, which coerces to a vector of OrganizationIdentity; a bare identity coerced to a map.urpx:offeredBy on a RatePlanVersion or RatePlanModifierVersion coerces and validates as urpx.schema/OfferingOrganization. It passed through untouched, so an organization without urpx:organizationType, or of another @type, now refuses.urpx:identifierJurisdiction is an xsd:string on the three identifier schemas, as the SHACL has it; it was a Ref. The urpx test cases write "US".urpx.coerce/coerce-doc dispatches urpx:SourcePublication, its v0.2.x spelling urpx:TariffPublication, and urpx:RateCase, so urpx.load/documents yields them and urpx.load/scan no longer lists them as uncoercible. urpx.validate checks them rather than reporting them unmodeled.urpx.core/containment-properties includes urpx:hasIdentity and urpx:hasName.urpx.defs, typed definitions. (defurpx kind name doc? value) validates the value against (kind-schema kind) at macroexpansion, with every map in the schema closed except one reached through a :ref, so an invalid value or an unknown kind interns nothing; the message names the var, its file and line, and each failing path. The var carries :urpx/kind. (defs-of-kind kind nses) returns the definitions of a kind in the namespaces given.clj-kondo.exports/energy.grid-coordination/clj-urpx) whose hook lints defurpx as a def.urpx.dsl/curie, the default dsl-scope :id-fn.urpx.schema/OfferingOrganization, an Organization or UtilityOperator inline, or a reference. The version builders take :offered-by.ComparisonNode and BooleanOperationNode schemas, with urpx.dsl/comparison-node and urpx.dsl/boolean-operation-node. BooleanOperationNode admits urpx:description.UtilityOperator, OrganizationIdentity and OrganizationName schemas from the v0.5.1 SHACL, with urpx.dsl/utility-operator, organization-identity and organization-name.urpx.dsl/rdf-list, an ordered RDF list value. A builder reads a keyword member as it reads a keyword value.urpx.emit/default-context types urpx:identityValidFrom and urpx:identityValidTo as xsd:dateTime.urpx.dsl builders refuse unknown keys. A key that is neither a key of the entity's schema, a kebab-case short form of one, nor a :jsonld/* key throws, naming the keys and the entity type under :unknown-keys and :type. Such a key was emitted as written, so a misspelling produced a wrong document. Coercion on read still keeps unknown keys.urpx.dsl builders refuse a double, a float or a ratio with no terminating decimal expansion for an xsd:decimal field, throwing :reason :urpx.dsl/inexact-decimal with each field's path and value under :fields. A double has already lost its written scale: 0.1610 emitted "0.161". Write 0.1610M or "0.1610"; integers and exact ratios are accepted. Coercion on read still accepts doubles.urpx.dsl/make-builder takes an options map; :passthrough is a collection of further keys the constructor admits, for properties the schema does not model.urpx.coerce/jsonld-transformer is public, for composing with another Malli transformer.urpx.schema/MetricSpecification admits urpx:commodity, as a Ref.24:00:00 decodes to midnight, on read and in urpx.dsl builders, with any fraction of zeros and an optional offset, and encodes as 00:00:00. It threw :urpx.schema/malformed-time. A urpx:TimeBracket ending 24:00:00 runs to midnight. Bare 24:00 and any other hour-24 time still refuse.dsl-scope refuses two local-id keywords that give the same @id, such as :scl/summer and :pge/summer, naming both. Both built entities with @id "summer", which merge in RDF.urpx.dsl builders read a :urpx/X keyword value as the vocabulary term {:jsonld/id "urpx:X"}, inside a dsl-scope or not, element-wise in a vector. Inside a scope it threw DSL local ref not declared; outside, the keyword failed validation.urpx.dsl/condition-expression accepts kebab-case short keys (:root-node), as every other builder does.urpx:ServiceBrackets (proposed in urpx-org/urpx#80). A bracket carrying urpx:adjustsServiceTier and urpx:lowerBoundOffset or urpx:upperBoundOffset adds each offset to the same bound of the named tier's applicable bracket when its own season and profile-alternative qualifiers hold, across every document in a composition. Offsets on one bound sum, a bracket repeated by @id counts once, and operators are unchanged. An adjusted tier carries the adjusted value under :urpx.resolved/tier-lower-bound or :urpx.resolved/tier-upper-bound, with :urpx.resolved/tier-{lower,upper}-bound-base, :urpx.resolved/tier-{lower,upper}-bound-offset and :urpx.resolved/tier-adjusting-bracket-ids. An adjusting bracket is never a bound of the tier hosting it.:unresolvable-adjusted-tier, :adjusted-tier-not-service-tier, :multiple-adjusted-tiers, :adjusted-tier-adjusts (a target hosting adjusting brackets, its own host included), :adjusted-tier-unit-absent and :adjusted-tier-unit-mismatch at any instant, and :adjusted-bound-absent and :adjusted-bounds-inverted where an offset applies. An adjusting bracket whose qualifiers cannot be decided withholds the tier's bounds under :urpx.resolved/tier-bounds-unresolved with :reason :adjustment-unresolved.urpx.schema/ServiceBracket admits urpx:adjustsServiceTier, decoding a bare value to a one-element vector, and both offsets as decimals; urpx.emit/default-context types the offsets xsd:decimal.urpx.core/containment-properties, the properties under which parse-graph embeds a referenced node.urpx.core/expand-document-entities returns each distinct payload once. A package naming one payload under several of its payload properties returned it once per property.urpx:serviceTierNumber on urpx.schema/ServiceTier and urpx:hasPriceDefinition on urpx.schema/BillingModel are optional, so a modifier whose host tier carries no number and whose billing model only moves bounds validates.@id references, and every loader read past them: a billing model or an adjusting bracket named by reference contributed nothing, with no error. urpx.core/parse-graph now replaces a bare reference under a property in containment-properties, or to a blank node referenced once, with the node of that @id when exactly one top-level entity in the document states it, and that entity leaves the top level. A node already on the path stays a reference, as does a reference to a named node under another property, to a node the document does not state, or to an @id stated twice. A blank node identifier left occurring once is dropped.urpx.price/resolve-prices-with-modifiers indexes the versions it resolves. An adjusting bracket whose target tier sits in a version supplied through :versions-by-plan, rather than nested in a plan or modifier passed to the call, refused :unresolvable-adjusted-tier.urpx.schema/NodeIdentity, which no schema or function referenced.urpx.emit/default-context binds urpx to https://urpx.org/ns/ontology/, the namespace URPX moved to in v0.5.0. It bound the earlier https://urpx-org.github.io/urpx/ontology/.urpx:TimeBracket with urpx:dayType urpx:customDays matches the days it lists in urpx:daysOfWeek. It matched every day. resolve-prices refuses :urpx.price/custom-days-unspecified against a urpx:TOUSchedule holding one that lists none, naming each such bracket under :brackets.urpx:daysOfWeek on a urpx:TimeBracket (urpx-org/urpx#71). urpx.schema/TimeBracket admits it as a vector of urpx.schema/DayOfWeek, the seven urpx:DayOfWeekValue members, and decodes a bare value to a one-element vector.META-INF/LICENSE. clojure -T:build ci refuses a jar without it.<description> and <url>. clojure -T:build ci refuses a jar whose pom lacks either.clj-kondo and the test suite on pull requests and on pushes to main.urpx:nearestWeekdayObservance or urpx:fridayIfWeekendObservance, and a Sunday December 31 is observed Monday January 1 under urpx:nearestWeekdayObservance or urpx:mondayIfWeekendObservance, for both urpx:holidayRule and urpx:holidayDate holidays.:nrepl alias now puts test and test/resources on the classpath, so urpx.test-fixtures and the vendored fixtures load.urpx.coerce that wrote a lone value of a 0..N property bare, rather than as a one-element list, could resolve to a different answer or throw. Resolution, composition and holiday derivation now read the bare and list forms alike for every property urpx.schema decodes to a vector, including urpx:hasPrice, urpx:hasPriceSet, urpx:hasTOUSchedule, urpx:hasServiceTiers, urpx:hasApplicabilityRule, urpx:referencesMetricInput and urpx:month.urpx.schedule's price-schedule, price-schedule-days, price-schedule-with-modifiers, price-schedule-days-with-modifiers and price-segments take ZonedDateTime bounds in any zone, as segments-for and schedule-for do. The zone argument is gone: each takes [plan start end] or [plan start end opts], with modifiers after plan where it takes them, and days in place of end for the -days forms. The bounds' instants define the window; intervals are computed and returned in the plan's zone. :urpx.schedule/zone-disagrees-with-plan is retired. A bound of any other type refuses :urpx.schedule/bound-not-zoned-date-time, where a ZonedDateTime bound threw a ClassCastException, as do segments-for and schedule-for, and opts that are not a map refuse :urpx.schedule/opts-not-a-map.coverage/of-version returns :evaluated-on and the coverage window's urpx:startDateTime and urpx:endDateTime as ZonedDateTime. One written without an offset is read in the zone of the urpx:timezoneIdentifier on the version, else on its document, else the zone given as the new optional third argument. With none it refuses :urpx.coverage/zone-required; with two whose rules differ, :urpx.coverage/conflicting-timezone. Zones with the same rules, such as UTC, Etc/UTC and ZoneOffset/UTC, agree. coverage/of-model reads each part in the zone on its model document.price-schedule-days and price-schedule-days-with-modifiers count days as calendar days in the plan's zone, so the window ends at the same wall-clock time in that zone: a day is 23 or 25 hours across a daylight transition, where it was 24.resolve-prices, resolve-modifier-prices and resolve-prices-with-modifiers refuse :urpx.price/no-version for a plan or modifier with no version snapshot, neither nested nor supplied through :version or :versions-by-plan. resolve-prices and resolve-modifier-prices returned empty ledgers, and a composition dropped such a modifier. The message names both ways to supply one: :versions-by-plan from urpx.index/versions-by-plan over urpx.load/documents, or resolving the loaded documents through urpx.scenario-cost-model.urpx:holidayRule refuses with :reason :urpx.holidays/unknown-holiday-rule, and an unrecognized urpx:observanceRule with :urpx.holidays/unknown-observance-rule.urpx:Holiday carrying neither urpx:holidayDate nor urpx:holidayRule refuses with :reason :urpx.holidays/no-date-or-rule.urpx.schema/Holiday closes urpx:observanceRule to the urpx:ObservanceRuleValue members through the new urpx.schema/ObservanceRule, so the validate CLI and coerce/validate reject any other IRI.urpx.load/scan reports :urpx.load/unresolved-references: each package payload reference whose entity no source states, as {:jsonld/id id :package-id p :property k}.corpus/corpus_json.clj, a dev-only walker for any corpus whose layout is declared in a corpus.json, in urpx-rate-plans' format. corpus-sweep --layout takes a corpus.json or a directory holding one, as well as a walker namespace. As a program it prints the documents, named kinds, or --all as kind<TAB>path.corpus-sweep with no --layout reads the corpus.json at the corpus root when there is one, including in a corpus fetched by git URL, and globs **/*.jsonld when there is none. The :dev alias carries corpus/, so clojure -M:dev -m corpus-sweep reads one.corpus/urpx.corpus.json declares the layout of urpx-org/urpx's test-cases/. The urpx-upstream walker reads it.scenario-cost-model/rate-plans-name, the urpx:name urpx-rate-plans gives a model's urpx:ScenarioCostModel, to pass node as its :name: each consumed version's plan or modifier name with the version's @id in brackets, then any selected alternatives' names. A component with no urpx:name refuses :urpx.scenario-cost-model/unnamed-component; one named two ways refuses :urpx.scenario-cost-model/conflicting-name.CHANGELOG.md at docs/energy.grid-coordination/clj-urpx/guides/CHANGELOG.md, beside the doc/ guides.:urpx.resolved/withheld-prices and :urpx.resolved/unselected-alternatives listed Prices that would not apply at the instant. A Price in an unselected urpx:PriceSetAlternative is now listed only when it has started, is the latest for its cell within that alternative, and has no urpx:touPeriodNumber or the resolved one, as on the active path. A ledger left with none is no longer marked, so marginal-unit-rate answers where it refused :unselected-alternatives.urpx:PriceSummary members were gated by the base plan's TOU period, and in a composition a month rule declared in the modifier and named by bare @id from a member's urpx:PriceDefinition was not applied; resolved alone, a modifier with its own urpx:TOUSchedule counted no member whose urpx:PriceDefinition carries a urpx:touPeriodNumber. They are now gated by the modifier's own period when it carries a schedule, counting outside its brackets only members with no urpx:touPeriodNumber, and by its own urpx:ApplicabilityRules. Members in the base plan keep the base plan's gates.urpx:TOUSchedule left out of a urpx:PriceSummary subtotal every member whose urpx:PriceDefinition carries a urpx:touPeriodNumber. A Price derived from that summary is now unvaluable with :summary-member-unselected.urpx:PriceSummary whose members in effect shared the lowest tier number of a block-tier ladder summed one of them, chosen arbitrarily. A Price derived from it is now unvaluable with :summary-ambiguous-tier, and its detail names the members, as marginal-unit-rate refuses :ambiguous-marginal-tier for a ledger.clojure -M:validate and urpx.validate/check-file schema-checked a urpx:URPXPackage or urpx:URPXDocument wrapper's payloads but not the wrapper. The wrapper is now checked against its own schema too, so a package missing urpx:targetURPXVersion fails, and :checked lists it.urpx.schema/Date inside a :maybe, :or, :orn, :and or :ref, or named by a registry key, lost its written offset on emit. urpx.coerce/coerce now keeps it in :urpx.schema/date-offsets and urpx.coerce/encode writes it back, as for a direct Date entry.urpx.scenario-cost-model compared the zones declared for one version as identifiers, so UTC and Etc/UTC, or America/Los_Angeles and US/Pacific, refused as :urpx.scenario-cost-model/conflicting-timezone. Zones with the same rules now agree, and the model is zoned by the declaration with the lowest @id.urpx.arrangements is now urpx.scenario-cost-model, named for urpx:ScenarioCostModel from urpx-org/urpx#72. No deprecated aliases remain; every other function and option keeps its name.
| 0.35.0 | Now |
|---|---|
urpx.arrangements | urpx.scenario-cost-model |
arrangements | enumerate |
arrangement-pairs | enumerate-pairs |
arrangement-id | identity-tuple |
urpx.coverage/of-arrangement | urpx.coverage/of-model |
:urpx.arrangement/* result keys and refusals | :urpx.scenario-cost-model/*, same local names |
:urpx.arrangement/opt-determined-by-arrangement | :urpx.scenario-cost-model/opt-determined-by-model |
:arrangement-id in refusal ex-data | :identity-tuple |
identity-tuple keys a model by its alternatives alone: [base-version-iri [modifier-version-iri ...] [alternative-iri ...]], both inner vectors sorted, the values a urpx:ScenarioCostModel carries. arrangement-id ended in [[container alternative] ...]. enumerate sorts by the new tuple, and each model's :urpx.scenario-cost-model/selections still maps every container to its alternative.
segments-for and schedule-for take ZonedDateTime bounds, as [model start end] or [model start end opts], in any zone. The Instant arities and the zone argument are gone. The bounds' instants define the window; segments are computed and returned in the plan's zone, the zone resolve-at resolves in.
load/documents and load/scan return a package shell for every urpx:URPXPackage or urpx:URPXDocument wrapper: the wrapper's own properties, with each embedded payload under hasRatePlan, hasRatePlanModifier, hasTariffPublication and packages replaced by a {:jsonld/id id} reference. README gives the predicate that drops the shells. Both also return any urpx:ScenarioCostModel among the coerced documents.
New refusals.
| Reason | When |
|---|---|
:urpx.price/no-tou-period | a plan has a urpx:TOUSchedule and no period covers the instant, including a holiday no bracket covers. The period resolved to nil and every period's Price was admitted. Every urpx.schedule and scm horizon function reports it as :urpx.interval/refused; resolve-at propagates it |
:unknown-profile-alternatives-container | a :profile-alternatives key names no container in any version being resolved. The prices it was meant to select were withheld. The ex-data lists the known containers under :containers |
:unresolvable-season-ref | a modifier's urpx:TimeBracket names a urpx:Season that no document passed declares, including a modifier resolved alone whose brackets name the base plan's Seasons. :season-ids lists them |
:urpx.scenario-cost-model/shared-alternative | two containers in one composition's versions list the same urpx:ProfileAlternative, in enumerate and from-node |
:urpx.scenario-cost-model/unknown-package-zones | a :package-zones value other than :agree, :fallback or :ignore |
:urpx.load/payload-without-id | a wrapper embeds a payload with no @id |
:urpx.schema/malformed-datetime, -date, -time | a malformed or out-of-range temporal literal. load and scan propagate it and the validate CLI reports REFUSED |
:ambiguous-marginal-tier | marginal-unit-rate finds two or more prices at the lowest tier number of an energy ledger's block-tier ladder. It took one of them. The ex-data lists each ledger with its :tier-number and :price-ids under :ledgers |
Unrecognized urpx:observanceRule (ex-data :observance-rule, :holiday) | a urpx:Holiday names an observance rule other than actualDateObservance, nearestWeekdayObservance, mondayIfWeekendObservance or fridayIfWeekendObservance. The holiday was observed by its rule string's default or on its actual date. holidays/derive-predicate throws it, as it does an unrecognized urpx:holidayRule |
price-schedule, price-schedule-with-modifiers, their -days forms and schedule-for report a refused step as an interval carrying :urpx.interval/refused and continue, as price-segments and segments-for do, where the ExceptionInfo used to abort the whole window. Adjacent equal refusals merge into one interval.
A modifier's Prices carrying urpx:touPeriodNumber contribute nothing where its own urpx:TOUSchedule covers no period, such as smud:pev-credit outside 00:00-06:00, where every period's Price had been admitted. Those carrying none still apply.
A Price stating urpx:startDateTime takes effect at that instant, an offset-bearing value at the instant it names, where it had taken effect at the start of its plan-zone day. price-segments begins a segment there. A urpx:startDate still takes effect at midnight of that day in the plan zone.
Temporal literals with an offset decode to what they name. An xsd:dateTime written with Z or an offset decodes to a ZonedDateTime in that offset, and one without stays a LocalDateTime. An xsd:time with an offset decodes to an OffsetTime. An xsd:date with an offset decodes to the LocalDate it names, so a version or coverage bound written with one starts at midnight in the plan zone; the offset is kept with its date in the map's metadata under :urpx.schema/date-offsets and emitted as written while the map holds that date. The DateTime and Time schemas accept the zoned forms. Times and date-times encode in XSD canonical form with seconds, where "16:00:00" had been written as "16:00". A typed-literal object {"@value" v "@type" t} decodes like the plain string, and emit/default-context types every temporal property the schema decodes. urpx.coverage/of-version's :evaluated-on is a ZonedDateTime when the document wrote an offset.
A modifier's ledgers carry its own matched season and TOU period as :urpx.resolved/season-name, tou-period-name and tou-period-number. A modifier gated by the base plan's period carries no period keys, and the result-level keys remain the base plan's. price-segments now splits where a modifier's own season or period changes.
A modifier's brackets read the modifier's own urpx:HolidayCalendar. resolve-modifier-prices treated every date as a non-holiday without a :holiday?, and a composition applied the base plan's calendar to every modifier. A modifier declaring none still inherits the base plan's in a composition, and a caller's :holiday? overrides both. A modifier's calendar refuses an unrecognized rule as a base plan's does.
urpx.schema/ScenarioCostModel, transcribing ScenarioCostModelShape as amended by urpx-org/urpx#72, closed, with urpx:name and urpx:description admitted. Its 0..N properties decode a bare value to a vector of one. coerce-doc and the validate CLI take a model, emit/write-scenario-cost-model writes one, a model under urpx:packages is encoded through its schema, and dsl/scenario-cost-model builds one.scm/node writes a model as one urpx:ScenarioCostModel node with absolute @ids. It takes a required iri-fn, and :context, :name and :description as options. An @id the context cannot expand refuses :urpx.scenario-cost-model/unexpandable-iri, and a prefix bound two ways conflicting-prefix.scm/rate-plans-iri mints the IRI urpx-rate-plans derives from the identity tuple, for use as iri-fn. It refuses :urpx.scenario-cost-model/iri-outside-namespace and iri-local-name.scm/from-node reads a urpx:ScenarioCostModel node, raw or coerced, into the map enumerate yields for it, so resolve-at, segments-for, schedule-for and node take it unchanged. It takes :compatibility, :repeated-ids, :package-zones and :context, and refuses not-a-reference, no-version, unknown-version, inapplicable-modifier, shared-alternative, unknown-alternative, dimension-selected-twice and dimension-unselected, each under :urpx.scenario-cost-model/.:package-zones on enumerate, plan-versions, applicable-modifiers and from-node decides what a package's urpx:timezoneIdentifier counts for: :agree (default) counts it as one more declaration that must agree, :fallback reads it only when nothing else declares a zone, and :ignore reads none. Packages are indexed by hasRatePlan and hasRatePlanModifier as well as packages.urpx.schema/in-zone, wall-time, date-offsets-decoder and date-offsets-encoder.@graph was loaded as one untyped node. parse-graph unwrapped only a top-level @graph, so upstream test-cases/pge-e-tou-c lost its package, plan, version, vintage-PCIA modifier and that modifier's version. It now flattens graph objects at any depth.urpx:PriceSetAlternative is reduced on its own.urpx:TimeBracket naming a urpx:Season only the base plan declares matched nothing, so every period's Price was admitted. It now resolves that Season through the composition and matches by its months.load/scan reported a by-reference package payload as uncoercible, as {:jsonld/type nil}. The package shell keeps the reference, and the referenced entity loads where it is stated.:decode/urpx-jsonld for ApplicabilityRule/month, Season/month, PlanElements/hasSeasons, PlanElements/hasMetricInput, PriceSet/hasPrice, Prices/hasPrice, Prices/hasPriceSet, ServiceTiers/hasServiceTier, TOUPeriod/hasTimeBracket, TOUSchedule/hasTOUTier, TOUTier/hasTOUPeriod and TimeBracket/appliesInSeason. Malli does not descend a non-vector, so a bare value left its whole subtree untransformed: a bare hasSeasons kept urpx:month as ["--06" "--07"] and threw on a later cast. Read-one properties (hasRatePlanVersion, hasRatePlanModifierVersion, hasMetricSpecification, hasHolidayCalendar) keep their shape and are read through urpx.index/single.applicable-modifiers returned [] under both shipped compatibility policies when applicableToRatePlanVersion was bare, so every arrangement came out base-only. A modifier whose target list was bare applied to no plan. urpx.emit mapped over a bare hasRatePlan and emitted key-value pairs with the @ prefixes stripped. All three now read the bare and list forms alike, through urpx.index/as-vec.urpx:hasIdentity was a list. The v0.5.1 shapes allow it as 0..N. Every identity and identifier is now read.resolve-at, segments-for and schedule-for resolve such a version.price-schedule interval reported the first step's season and TOU period for its whole span. See Changed.price-schedule intervals may omit labels. A merged interval carries :urpx.resolved/season-name, :urpx.resolved/tou-period-name and :urpx.resolved/tou-period-number only when every merged step agrees on that value. Steps still merge on ledgers. Use price-segments when the labels matter, since it splits on them.:urpx.arrangement/conflicting-timezone. Arrangements throw it when the nodes a version belongs to declare different zones, with or without :at.:plan-id. The ex-data and each :stale entry name the owning plan's or modifier's @id, the key :versions-by-plan is looked up by, and the message names it too.REUSE.toml.corpus/ ships only the urpx-upstream walker. The urpx-rate-plans walker moved to that repository.A urpx:RatePlanModifier's own block tiers surfaced untiered. modifier-ledger-results built its entries from price-fields alone and never called tier-info, so no entry from a modifier carried :urpx.resolved/tier-number or any bound, however its urpx:ServiceTiers were shaped. That made it a wrong NUMBER rather than a missing field: marginal-unit-rate reduces a block ladder to its marginal rung by that number and sums everything without one, so a modifier's three-rung ladder was added together instead of reduced. Bracket and container selection now reach that path too, reading the same season and alternative context the PriceDefinition gate there already used.
The case it was for: PG&E's E-TOU-C baseline-tier overlay carries 120 urpx:ServiceBracket nodes, one per (baselineTerritory, baselineCode, season) cell, and every one of them was read by nothing. Composed with its base plan and given a baseline cell, Territory P Code B now states 0 to 13.5 and 13.5 to 54.0 kWh per day in summer and 0 to 11.0 and 11.0 to 44.0 in winter, matching PG&E's Sheet 4. With no cell named all four rungs surface with their numbers and unit prices and an :alternative-unselected mark, so a caller learns the ladder exists and that its bounds are the caller's to determine.
One asymmetry recorded rather than worked around: resolve-modifier-prices takes no opts, so a cell selection cannot reach a modifier resolved alone and its cell-scoped bounds are always withheld there. Composing is the way to select. resolve-prices has taken opts all along.
A modifier resolved on its own ignored four caller inputs, and one of them made it answer wrongly. resolve-modifier-prices was the only public resolution entry point with no opts arity, though the private function it delegates to already took one and resolve-prices-with-modifiers passed it. Worst was :holiday?: this path defaults it to (constantly false), so a modifier carrying its own urpx:TOUSchedule with holiday-sensitive urpx:TimeBrackets resolved the non-holiday period on a holiday and said nothing about it. Every other gap on that path withholds or refuses; that one returned a confident wrong period. It now takes opts on a new 3-arity, with the 2-arity delegating unchanged.
The other three withheld rather than lied, and are now reachable: :profile-alternatives, without which a cell-scoped urpx:ServiceBracket's bounds could never be determined standalone; :by-id, which supplies a urpx:Season a modifier names but does not declare, as PG&E's E-TOU-C overlay does; and :version, for a document carrying several snapshots.
pge-e-tou-c-baseline-tier-2026-03-01 fixture is re-vendored from the same upstream commit as its base plan. The two halves of one filing had been taken from different upstream moments, so the modifier named pge:e-tou-c-2026-03-01-season-* while its base plan declared pge:season-*. Nothing read those refs before, so the mismatch was invisible; wiring the modifier path made it a refusal. An inconsistent pair is not a URPX document set any consumer would see, and it was an artifact of vendoring rather than anything upstream ever published.Three readers of urpx:appliesInSeason answered different questions depending on which node carried the property. pd-active-in-season? skipped an unresolvable ref silently and read a urpx:Season carrying no urpx:month as excluding the month. find-tou-period put a TimeBracket's refs through urpx.index/single, which refuses on several, so a legal filing naming two Seasons on one bracket was unresolvable. node-season-status, added in 0.32.0 for tier selection, did neither. There is now one shape normalizer (season-refs), one reader (node-season-status) and one constructor (make-season-status), shared by tier selection, PriceDefinition gating and urpx:PriceSummary member gating.
Multiplicity is disjunctive everywhere, which is what sh:minCount 0 with no sh:maxCount states on every shape carrying the property. A gate that could NOT be evaluated no longer excludes a PriceDefinition: reading a monthless Season as out-of-season made a price silently missing from the sum, and excluding on an unreachable Season would have reinstated the defect 0.32.0 fixed for summary members through another route, since a modifier resolved without the base plan that declares its Seasons would drop every seasonal definition. A urpx:PriceSummary member keeps the stricter treatment, because a subtotal has a third state for an unknown member and a boolean filter does not: such a member is now :unknown and the derived Price reports :summary-member-unselected rather than carrying a number.
No vendored fixture moves, and none could: measured across the 23 vendored in this repo, no TimeBracket names more than one Season, none of their PriceDefinition, Price or PriceSet nodes carries urpx:appliesInSeason, and no Season lacking urpx:month is referenced. That is why the three readers could disagree unnoticed, and why every case is constructed. Reverting the unification fails 12 assertions across the new namespace and two existing ones.
Those counts are a property of this repo's sample, not of URPX filings: the upstream corpus carries urpx:appliesInSeason on 89 PriceDefinitions across six CPAU files, reported by the urpx-rate-plans maintainers. It still moves nothing there, and the check that establishes it is reachability rather than absence: none of the 89 names a Season its own document does not declare, so no such gate is unevaluable and the behavior changed here cannot be reached from that data.
urpx:ServiceTiers.referencesMetricInput is documented as naming the quantity a tier's BOUNDS are stated in, which is not the quantity the ledger's charge is stated in. The two are meant to differ: LADWP R-1A's Power Access Charge is 2.30, 7.90 and 22.70 per P1M per meter with bounds in kWh, so its container names mi-energy while its ledger names mi-meter-count. Reading the property as a pointer to a container's ledger would strip the bounds off a monthly fixed charge tiered by consumption. No behavior change; the question was raised by 0.32.0's tier selection and is settled against it.A modifier's PriceDefinitions were gated by TOU period alone, so month and season applicability did nothing there. The base path filters a ledger's definitions on pd-active-at-tou?, pd-active-in-month? and pd-active-in-season?; the modifier path filtered on the first only. A urpx:RatePlanModifier telling two cells apart by a urpx:hasApplicabilityRule month constraint or by urpx:appliesInSeason therefore surfaced BOTH at every instant, while one telling them apart by urpx:touPeriodNumber gated correctly, which is why it went unseen. Reported from the corpus side: SMUD's NEM1 export credit over smud:fixed is that shape, and only its TOU-discriminated cells were buildable. Month and season now apply unconditionally on that path, TOU stays conditional on there being a period to compare against, and a urpx:Season a modifier names but does not declare resolves through the composition-wide index.
Standalone resolve-modifier-prices had the opposite failure in its summary-member gate, which it built with EMPTY rule and season indexes: pd-active-in-season? could resolve no urpx:Season for any ref, so a member carrying urpx:appliesInSeason was always inactive and silently dropped from the subtotal rather than gated by it. It is now judged against the modifier's own seasons and rules. Compositions were unaffected, since those are supplied from the base version.
No vendored fixture moves: not one modifier PriceDefinition or PriceSummary in the 23 carries either property, so both halves are exercised by constructed documents.
A tier's bounds came from whichever urpx:ServiceBracket happened to be written first. tier-info took (first (:urpx/hasServiceBracket tier)) and indexed one urpx:ServiceTiers container per urpx:serviceTierNumber, so urpx:appliesInSeason and urpx:referencesProfileAlternative were silent no-ops at both scopes: a tier stating summer bounds and winter bounds reported the summer ones in December, and a second container declaring the same number last-won on map insertion order. Bounds are now selected among every (container, tier, bracket) candidate for that number, gated by both properties on the container and on the bracket alike. Measured on a constructed two-bracket tier: upper bound 13.5 in July and 11.0 in December, where both months read 13.5 before.
Three refusals rather than a pick: :ambiguous-service-bracket (two brackets in effect stating different bounds, and urpx:hasServiceBracket is a set, so there is no first one to take), :ambiguous-service-tier (two containers declaring one number with different bounds), and :unresolvable-season-ref (a ref naming a urpx:Season that matches none of the ones reachable here, evaluated eagerly so it fires whichever cell the caller chose). That last one divides on whether any urpx:Season is reachable at all: if some are, a ref matching none is a typo and refuses; if none are, the document that declares it was simply not passed, so the bounds are withheld with :season-not-evaluable and composing or supplying :by-id clears it. That is the line pd-alternative-status already draws for an alternative no container in this document declares. Brackets stating the same bounds do not refuse: comparison goes through .stripTrailingZeros, so 0M and 0.0M agree. resolve-prices now carries ten :reason values.
No vendored fixture moves: every one of the 423 tests that predate this change still passes, with no assertion altered and three added to pin the ex-data rename below; the bracket-selection namespace takes the suite to 450 tests, 0 failures. (Assertion totals are not quoted because urpx.provenance-test reads target/classes/META-INF/.../build-provenance.edn off the filesystem and asserts seven fields when a build has populated it, one trivially when it has not. target/ is gitignored, so a fresh checkout counts six fewer than a tree that has run ci.) The 18 tiered rungs across the five tiered base filings carry exactly one bracket each, and neither it nor its container carries either property, so a single unqualified candidate short-circuits before the date or the selection is read. The only qualified brackets in the corpus are the 120 on PG&E's E-TOU-C baseline-tier modifier, and tier-info has one call site, on the base-plan path.
:urpx.resolved/tier-bounds-unresolved, on a tier whose bounds selection cannot determine. A bound read off a bracket that does not apply is a number a consumer bills against, so it is withheld instead. Three reasons: :alternative-unselected when the caller named no cell in that dimension, :season-not-evaluable when a urpx:Season carries no urpx:month so the month gate was never evaluated, and :no-applicable-bounds when nothing is in effect at this instant. The mark carries :scope and :ids, plus :alternatives where an alternative is involved. The entry keeps its :urpx.resolved/tier-number and its unit price, so marginal-unit-rate still reduces the ladder by number and still answers: 0.20M on a tier whose bounds are withheld.:multiple-profile-alternatives and :ambiguous-profile-alternative report :node-id and :node-type where they reported :price-definition-id. Bracket selection hands pd-alternative-status urpx:ServiceTiers containers and urpx:ServiceBrackets as well as PriceDefinitions, so the old key named the wrong thing for two of the three node types.The same instant, written in two zones, resolved to two different prices. urpx.price reduced the caller's ZonedDateTime with .toLocalDateTime and compared those wall-clock fields against document bounds that are wall-clock in the plan's zone. Every bound in a URPX document is naive: all 89 in the production corpus carry no offset. So the answer depended on how the caller spelled the moment.
Measured on PG&E E-TOU-C, whose peak runs 16:00–21:00 Pacific. One instant:
| 2026-07-15, asked at | in Pacific | in UTC |
|---|---|---|
| 10:00 PT | Off-Peak 0.34586 | Peak 0.46886 |
| 16:00 / 17:00 / 20:00 PT | Peak 0.46886 | Off-Peak 0.34586 |
Up to 36% apart, with no error and no mark. The version-window half was reported by a consumer: SCE Schedule D's sce:d-2025-11-15 expires at LA midnight, and 2025-12-31T20:00-08:00 priced while the identical 2026-01-01T04:00Z refused :outside-version-window. The TOU half was not reported and is worse, because it fires on most of the day rather than at one boundary.
resolve-prices now anchors the query in the plan's declared zone before reading anything, so resolution is a pure function of the instant and the plan's zone. resolve-modifier-prices anchors in the modifier's own; a composition anchors both halves in the base plan's, because one question about one moment should not resolve in two frames.
price-schedule and price-segments refuse a walk zone the plan does not declare. Candidates must be derived in the zone resolution happens in or the candidate set stops being a superset of true transitions. Measured: a UTC walk of that LA plan over one July day put the peak boundary at 17:00 Pacific instead of 16:00 and lost the 21:00 transition entirely, returning two segments where three are right. Passing the plan's own zone, or nil, is unchanged.
A plan declaring no urpx:timezoneIdentifier is refused, :reason :urpx.price/no-timezone. Its bounds anchor to no moment, so whether an instant falls inside one has no answer. Every filing in both corpora declares one.
version-effective-at, version-expired-by? and version-covers? take a required ZoneId and work in ZonedDateTime, as does the new version-expires-at. An optional zone would have made the unzoned call the easy one and the wrong one, which is how this namespace came to compare a caller's wall clock against a publisher's.
:effective and :expires in :outside-version-window are ZonedDateTime rather than LocalDateTime. A naive bound handed to a consumer invites exactly the comparison that caused this.
Upgrading: this changes resolved values for any caller passing a ZonedDateTime in a zone other than the plan's — previously silently wrong, now correct. Callers already passing the plan's zone see no change. An explicit walk zone that disagrees with the plan now refuses rather than returning wrong boundaries, and a plan with no declared zone refuses rather than resolving against the caller's.
urpx.load, naming the pipeline from sources to documents. document-xf is a transducer, documents realises it, and scan does the same work in one pass while reporting what it discarded. A source is anything clojure.java.io/reader coerces; where sources come from is the caller's business and deliberately not this library's, because a corpus repository is a one-off and there will never be a layout everyone adopts.
Nothing here is new behavior. It is the same three functions in the one order that is correct, and it exists because the previous arrangement — a four-step form in the README that every consumer composed by hand — produced exactly the failure it describes. A consumer following it used mapv where mapcat belongs and omitted the coercion step, and got a plausible count rather than an error. Naming the composition makes the wrong one unavailable to make rather than documented against.
It is its own namespace because it cannot live in either it spans: urpx.core requires only data.json, java.io and string, urpx.coerce only malli and urpx.schema, and neither requires the other. Core would drag Malli into the parse layer; coerce would make a coercion-only consumer pull in the parser.
scan makes the one silent drop assertable. coerce-doc returns nil for an entity type this library does not model, so document-xf discards it without a word: 42 of 129 expanded entities over the production corpus, urpx:ProgramEnrollment 30, urpx:SourcePublication 8, urpx:RateCase 4. None is a version today, which is luck rather than design; the unlucky form of the same thing cost a whole rate plan in 0.29.0. A caller can now assert that no version type reached the discard list, which it could not before. "This library does not model it" and "it does not matter" are different statements, and the second cannot be judged while the first is invisible.
urpx.coverage, an optional reader for what a snapshot claims about itself. Every version carries a urpx:CoverageClaim, the standard's own answer to "what am I holding?", and this library modeled it and read it nowhere. of-version returns a report; claim returns the raw node. It requires urpx.index and nothing else: not urpx.price, not urpx.arrangements, and nothing in src/ requires it.
The verdict is the PUBLISHER's, not this library's: claims "are produced by an implementation or publishing workflow", so nothing here re-derives one or second-guesses one.
Three things the report is careful about, each because the careless version would read as a fact the document never stated. Absent is not unknown — a version with no claim reports stated? false and completeness nil, never "urpx:unknown", which is itself a verdict meaning no evaluation has run. selfContained is reported with its source and any disagreement rather than flattened, and it is vacuous in practice: derived as "false iff any PriceDefinitionReference is retrievedElsewhere", with zero such references in the corpus or the fixtures. scope-stated? and coverage-window are separate, because scopeKind is the only required field on a TemporalScope, so a scope can name its role and bound nothing.
Four refusals: :unknown-completeness-result, :claim-ref-unresolved, :self-contained-contradicts-references, :ambiguous-coverage-scope. Two deliberate non-refusals: a missing claim (v0.2.x acceptance is a contract) and claim-versus-version selfContained disagreement, which the ontology settles by making the claim authoritative. Every guard carries a mutation proof; most cannot fire on real data, which is why they are constructed.
of-arrangement returns every part's report in arrangement order, base first, plus completeness-counts and conjunction predicates; descriptions returns the Evaluation prose. Each part is read against ITS OWN document, the one the arrangement bound to it, so a claim recorded as a bare reference resolves in the right graph.
coverage-interval returns the window a verdict covers as {:tick/beginning :tick/end}, the shape urpx.schedule already emits, and verdict-applies? asks whether it contains an interval. These are tick intervals in fact, not only in shape: tick.alpha.interval reads any map with those keys, so Allen's relations work on the value directly and ti/contains? agrees with verdict-applies? case for case, which a test pins. The library requires tick nowhere. The window is [beginning, end), matching urpx.price/version-covers?, because URPX states no convention and two readings of a window inside one library would be worse than either.
Both ends are ZonedDateTime and the zone is a required argument, per this repo's own standing rule that the library uses ZonedDateTime end-to-end at output boundaries. URPX states a TemporalScope's bounds as wall-clock readings with no zone, but returning them unzoned is the trap rather than the careful choice: urpx.schedule emits the same two keys carrying ZonedDateTime, so an unzoned window is the same map with an incompatible type, and the natural comparison does not fail loudly. Measured: tick.alpha.interval/relation throws on the mixture, while contains? returns true regardless of the values. A rate plan usually declares the zone as urpx:timezoneIdentifier; refuses :urpx.coverage/zone-required when none is passed.
verdict-applies? refuses on every real filing today with :coverage-scope-absent, because zero TemporalScope nodes of any kind exist in the corpus or the fixtures. That is the honest answer rather than a bug: false would claim the verdict does not cover the window, which no document said, and true would be worse.
No arrangement-level verdict, and the absence is pinned as an exact key set so that adding one fails a test. The six completeness members carry no ordering, and the absence is meaningful rather than an oversight: the same ontology uses skos:broader where it does mean a hierarchy. They are distinct kinds, not degrees. More decisively, every definition is relative to "the claimed scope" and no snapshot states one, so composition is undefined at the domain level rather than merely unranked.
:repeated-ids, a third caller policy, deciding what one version @id naming several distinct snapshots means. repeated-ids-refuse (the default, unchanged behavior) refuses :conflicting-version-id naming every offending @id at once; repeated-ids-union merges them property by property, the reading a corpus whose own SHACL gate validates a merged graph already assumes. Shipped as a policy for the same reason as :combine and :compatibility: URPX does not say, and a client reading one filing wants a different answer from a client merging a corpus.
Two costs, both stated at the call site. RDF merges triples, not meanings: merging a copy that states applicableToRatePlanVersion with one that omits it yields a node that has the property, which under named-versions-narrow restricts the modifier to exactly those versions, while the omitting copy meant no restriction at all. Over the production corpus that is the difference between 137 pairs and 144. And it silently reconciles a version @id whose option list is time-varying, which is a defect on any reading.
Only vector-valued properties union. A Ref is a map and therefore a collection, but it is one value; unioning two concatenates their entries into nonsense, so two differing scalars refuse :unmergeable-property. A property one occurrence states and another omits merges to the stated value, because absence is not an assertion.
A merged node is in no single document, so the one bound to it is the document that IS the identity the version points back at, derivable from the graph. Picking a source arbitrarily is wrong in kind rather than merely arbitrary: a bundled copy of a modifier version lives inside some plan's filing, so an arbitrary pick can bind a urpx:RatePlan as the modifier document.
arrangement-pairs, the enumeration collapsed to one row per (base version, modifier set), with ProfileAlternative choices not expanded. For comparing two independent enumerators on what composes with what, without disagreeing about whether a pending choice is one row or several.
A second corpus walker, urpx-upstream, for the URPX standard's own test-cases/. A naive glob over that repository does not miscount, it crashes: dev-docs/samples carries an instant written 2026-01-01T06:00:00Z that coercion refuses, and the error names no file. The test cases also exercise shapes the production corpus never uses — three top-level RatePlanVersions, two base versions claiming urpx:complete, the only TemporalScope with scopeKind urpx:coverage in either tree, and a genuine :ambiguous-version tie.
urpx.validate reports conformance findings, distinct from schema errors. Two rules the standard states that this library's schemas deliberately do not enforce, so a document can be schema-valid and still violate one: :missing-coverage-claim (checked against what the document says it targets, since requiring it unconditionally would reject every v0.2.x document) and :self-contained-disagrees (a violated MUST that urpx.coverage reads without refusing, because the ontology settles which value wins). A third, :unparseable-target-version, says so when a target version cannot be read rather than skipping the rule quietly.
They count toward the exit code. They did not at first, and were therefore decoration: a file every schema accepted and the standard rejected exited 0 with the violation printed above it.
urpx:targetURPXVersion is found by a deep walk. It sits on the urpx:URPXPackage wrapper in 67 of the 73 production filings and on a node inside the @graph in the other 6; a top-level read finds it in some documents and not others. Every filing in the corpus and every vendored fixture declares it, all v0.5.1.
dev/corpus_sweep.clj prints a coverage census, before the enumeration because it is about documents and holds even when the enumeration refuses: verdicts by version type, how many claims state a coverage window (zero, everywhere, since the machinery existed), how many carry an Evaluation, and any selfContained disagreements.
urpx:TemporalScope, urpx:TemporalPeriod and the ScopeKind enum are modeled, with urpx.dsl/temporal-scope and temporal-period builders. Groundwork for reading a coverage verdict's window.
The node was transcription, not judgement: urpx-shacl:TemporalScopeShape is fully specified, sh:closed, with urpx:scopeKind sh:minCount 1 sh:maxCount 1 against ScopeKindValueShape and the four date bounds plus hasTemporalPeriod each 0..1. scopeKind is the only required field, so a conformant scope may name its role and state no bounds at all.
The EDGE is what nothing specifies. urpx:hasTemporalScope appears as sh:path zero times in the shapes graph and in sh:ignoredProperties 79 times, so every closed shape permits it and none constrains it: nothing says which classes may carry a scope, how many, or of which kind. It is modeled on CoverageClaim as 0..N because that is what unconstrained means, not because any filing carries several. There are zero TemporalScope nodes of any kind in the production corpus or the vendored fixtures.
corpus/, worked examples of the half that is not the library's. One walker per corpus layout, with urpx-rate-plans as the first. Not on the library :paths, never in the jar, nothing in src/ requires it. They exist because a glob looks equivalent to a walk and is not: over urpx-rate-plans a naive **/*.jsonld yields 87 documents where the layout walker yields 84, and the extra three are not filings — one is deliberately non-conforming so that corpus's own gate has something to fail on. Nothing errors; the count is simply about something else.
docs_test verifies that every project function a documentation block names still resolves, but it recognised aliases from a hardcoded set that was a second copy of the alias map, and unknown aliases were dropped by a keep. So urpx.arrangements shipped in 0.29.0 with a README section full of arr/ calls that the suite never looked at. The set is now derived from the map, and a new test fails until every urpx.* namespace on disk is registered, because an unregistered namespace is the only way these checks can go quiet.Upgrading: additive. Three new namespaces (urpx.load, urpx.coverage, and the corpus/ walkers, which are not in the jar), a third caller policy, and no resolved number changes. Nothing existing changes behavior.
One thing to know if you enumerate a corpus: :repeated-ids defaults to repeated-ids-refuse, which is what 0.29.0 did unconditionally. A corpus where one version @id names several distinct snapshots still refuses; pass repeated-ids-union to merge instead. And urpx.validate now exits 1 on a conformance finding as well as a schema error, so a document that passed before and violates a stated MUST will now fail.
urpx.arrangements, an optional enumerator for service arrangements. A service arrangement is one urpx:RatePlanVersion, plus the compatible urpx:RatePlanModifierVersions, plus one urpx:ProfileAlternative for each dimension any of them leaves open: fully determined, so given an instant inside its window it resolves to prices with no further input. If a choice is still outstanding it is not one arrangement but several.
urpx:ServiceArrangement is not a URPX class. Zero occurrences in the ontology, zero in the SHACL shapes; nothing here writes it into a document or claims a urpx: prefix for it.
Optional by construction. The namespace requires urpx.price; nothing in the library requires the namespace, and a test asserts that over the whole of src/ by reading the ns forms on disk. A consumer that assembles its own compositions never loads the file.
Public: arrangements, plan-versions, applicable-modifiers, offered-by, arrangement-id, resolve-at, and the two combining policies one-modifier-at-most (the default) and every-subset.
Discovery of what may apply, separate from what does apply. applicable-modifiers answers which modifiers a plan version may take, read from urpx:applicableToRatePlanVersion where present and urpx:applicableToRatePlan otherwise; offered-by groups by urpx:offeredBy for a broader search. Which modifiers a given customer holds is in no document, because URPX places customer data outside its scope, so it is a caller policy passed as :combine rather than an assumption made here. Nothing in the ontology expresses exclusivity or combinability between modifiers, so every-subset is complete but will pair programs no filing says can be paired.
dev/corpus_sweep.clj, a worked example of walking a corpus. It takes the corpus as an argument and names none itself: a library that reads a standard should not privilege one corpus in its own source. A git URL is pinned to a SHA via clojure.tools.gitlibs; a path is not, and since a clone may be legitimately ahead of its remote, the path form reports the tree's SHA, commits ahead of upstream, and dirty count rather than letting the two forms be assumed equal. babashka/fs and tools.gitlibs are :dev dependencies only; the library does no filesystem traversal and fetches nothing.
segments-for and schedule-for, horizons that take an arrangement. resolve-at answers one instant; a price server publishes a window. Doing that through urpx.schedule directly meant taking an arrangement apart and rebuilding the per-document :version map by hand, which is the reconstruction resolve-at exists to prevent and the defect class this library has fixed three times. One builder now produces the binding for all three entry points, so they cannot drift about what an arrangement means, and both refuse :version and :profile-alternatives as resolve-at does.
segments-for is the one to publish from: its boundaries come out of the documents rather than from sampling. Over one July day of PG&E E-TOU-C at a six-hour step it returns 00:00, 16:00 and 21:00 where the stepped walk returns 00:00 and 18:00, inventing a boundary that is not a transition and missing both that are.
urpx.price/version-effective-at, version-expired-by? and version-covers? are now public, as the single readers of a version window. They were private and urpx.arrangements re-derived them, which is how it came to require urpx:effectiveDateTime where urpx.price also read urpx:effectiveDate. A consumer asking whether a snapshot is current now gets the same answer the resolver uses.
urpx.coerce/uncoerced?, check-coerced! and temporal-signature are public, one implementation of the test that tells a coerced document from a raw one. urpx.schedule had it privately and urpx.arrangements needed it; two copies would have been two chances to disagree about what "coerced" means.
:compatibility, a second caller policy, because URPX does not settle whether naming a urpx:applicableToRatePlanVersion narrows urpx:applicableToRatePlan or adds to it. named-versions-narrow (the default) fits the ontology's "which specific versions" and is the only reading under which the property can express a restriction, but it is non-monotonic: merging a document that adds a version statement removes compatibility with a plan named only at plan level, and RDF merge never retracts. named-versions-add is monotone but leaves the property able only to widen. Nothing in the artifacts decides it, so the library ships both rather than picking, exactly as it does for :combine.
The two are indistinguishable on the corpus this library can enumerate today. They differ only about a modifier that names some version and names a plan no named version belongs to: 33 of 40 modifier versions name no version at all, and of the 7 that do, 4 name a version of the only plan they name. The remaining two are both among the identities currently refused as :conflicting-version-id, so the two questions are entangled and shipping both policies keeps settling one from silently settling the other.
Three refusals, because enumeration asks a question about a whole corpus. A corpus that does not determine an answer gets a named :reason rather than one of the answers it might have meant: :uncoerced-document, :urpx.arrangement/conflicting-version-id (one version @id naming two snapshots that differ, every offending @id named at once), and :urpx.arrangement/ambiguous-version (two distinct snapshots of one identity sharing the latest effective instant, the input urpx.price already refuses). One @id published twice with byte-identical content is deduplicated, not refused.
All four were found by an adversarial audit of this namespace before its first release, seven of the eight lenses independently finding the first.
The enumeration depended on the order the documents were passed in. current-versions broke a tie on urpx:effectiveDateTime with (apply max-key ...), which returns the last maximal element, so when two snapshots of one identity shared an effective instant one was silently discarded and which one depended on seq order. Over the production corpus at 889344b that moved the answer between 739 arrangements / 144 distinct pairs in glob order and 731 / 136 reversed, with no error either way, and the eight pairs that vanished were all CPAU. urpx.price refuses exactly this input with :ambiguous-version and says "there is no rule here that picks among them"; a namespace sitting on top of it was quietly undoing that. Ties are now refused.
A version snapshot was paired with a document it did not come from. plan-doc / mod-doc were (into {} ...) keyed by version @id, so where two documents publish one @id the loser's snapshot was bound to the winner's document. Measured on the corpus: 69 of 1728 base bindings and 641 of 1728 modifier bindings named a document that was not the snapshot's source. This is the :version defect class of 0.28.0 for the third time, so the fix is structural rather than another guard: the source document now travels with the snapshot as an [identity-id version document] triple and is never recovered by @id. The caller-supplied :combine policy still sees version nodes, never the triples.
arrangement-id was not unique, and its docstring and the README both said it was. 1728 arrangements collapsed to 1068 distinct ids, 272 ids were carried by more than one arrangement, and 20 were shared by arrangements that genuinely differed. Now unique: 1041 of 1041 across all versions, 712 of 712 at an instant.
A conformant version stating only urpx:effectiveDate was silently dropped, taking its whole plan out of the enumeration, while urpx.price resolved the same document normally. Both properties are sh:minCount 0. The cause was re-deriving what urpx.price already read, so version-effective-at, version-expired-by? and version-covers? are now public there and are the single readers; urpx.arrangements calls them instead of keeping a second copy. The same de-duplication moved the uncoerced-document test into urpx.coerce, which urpx.schedule now shares.
Uncoerced documents were answered rather than refused. urpx.core/load-graph returns parsed but untyped entities, so urpx:effectiveDateTime is still a string. With an instant that threw a ClassCastException from deep inside; with no instant nothing compares a date at all, so the documents sailed through and 202 pairs came back where the coerced form gives 144 — a plausible number, not a refusal. Found when a consumer's cross-check probe walked into it, where it would have read as an enumeration disagreement between two implementations.
Six guards that could be deleted with the suite green, found by re-running a mutation battery against the tests rather than by reading them. Five were still green after the first round of fixes: resolve-at's per-document :version map, both branches of the compatibility rule, applicable-modifiers' documented :at option, and the selections component of arrangement-id. Each now has a test and a mutation proof that it fails when the behavior it names is broken.
The :version map was invisible for a structural reason worth recording: no fixture carries more than one version in one document, so urpx.price's own rules could only ever reach the same snapshot the arrangement had bound. Made visible with a constructed two-version document, the binding turns out to do more than choose. Without it urpx.price refuses the document outright ("Multiple nested version snapshots; this library reads one"), so the map is not an optimization over urpx.price's rules, it is what makes a multi-version document resolvable at all.
One of the six pins a contested reading: urpx:applicableToRatePlanVersion is treated as exhaustive, so a modifier naming any version combines with those and no others. The ontology's wording supports it ("which specific rate plan versions are compatible") but the reading is non-monotonic, and merging a document that adds a version removes compatibility with plans named only by applicableToRatePlan. Over the production corpus the two readings differ by exactly 7 of 166 base/modifier pairs. URPX states neither; the test exists so that changing the answer is a deliberate act.
check-no-conflicting-ids! was called once per kind, so it threw on the base versions before the modifier versions were examined and reported one conflicting @id where the corpus has three. Naming them all at once was the entire purpose of the function and calling it twice defeated it. It now takes every kind together, and ex-data gains :by-kind.
A test that compared a value to itself. The resolve-at opt pass-through assertion passed a :holiday? opt and compared the result to the same number the un-opted call returns, so the wrong opt, the opposite opt and a nonsense opt all satisfied it. It cannot be fixed by comparing values either: the predicate provably does not change any fixture's rate, and SCE Schedule D never consults it at all. It now counts the predicate's own invocations on the one fixture that invokes it. Every guard added here carries a mutation proof that it fails when the behavior it names is broken.
resolve-at refuses :version and :profile-alternatives instead of merging them. An arrangement determines both. Letting a caller's value win would restore the exact defect the function exists to prevent (an opt reaching, or failing to reach, the document it belongs to: both such defects are fixed in 0.28.0); letting the arrangement win silently would ignore an opt the caller meant. ex-data carries :reason :urpx.arrangement/opt-determined-by-arrangement. A caller who wants a different version or selection wants a different arrangement.Upgrading: additive. urpx.arrangements is new and optional, urpx.price and urpx.coerce gain public functions that were private, and no resolved number changes.
Two things to know before pointing the enumerator at a corpus. It refuses rather than enumerates when one version @id names two snapshots that differ; the production corpus has three such identities today, so it refuses there. And the :compatibility reading is genuinely unsettled upstream: named-versions-narrow is the default because it is what the corpus and the reference implementation assume, not because it is known right, and the two readings can only disagree about a modifier that names some version and names a plan no named version belongs to.
A :version opt for the base plan silently dropped every modifier. plan-version short-circuited on (:version opts) before anything looked at which document was being resolved, and resolve-prices-with-modifiers threads one opts map to the base plan and to every modifier. So the base plan's snapshot was handed to each modifier as its version; that node carries no urpx:applicableToRatePlan, modifier-applies-to? therefore said no, and every modifier was dropped with no mark and no refusal.
SCE Schedule D composed with CARE returned 0.31085 instead of 0.19804875 — the full undiscounted rate, 57% high, over five ledgers instead of eight. A variant where the plan node carries no @id double-counted the base plan's ledgers instead, at 0.68538. The library's own refusals routed callers straight into it: :ambiguous-version and :version-without-effective-instant both end "Pass :version to choose explicitly".
A version now applies only to the document it is a snapshot of, identified by urpx:forRatePlan / urpx:forRatePlanModifier or by the document nesting it, and is ignored elsewhere. Ignoring is not the silent failure it replaces: the other document then resolves by its own rules, which is correct, where before it was discarded. Present since 0.11.0.
The holiday calendar on a composition comes from the resolved version. resolve-prices-with-modifiers called holidays/derive-predicate's 1-arity, which re-derives the snapshot by reading the nested one; that ignores a caller's :version and refuses outright on a document nesting several, which is exactly the document :version exists to settle. So a composed call threw from the holiday path after :version had already chosen, while the same call through resolve-prices succeeded. The snapshot is in hand at that call site. No resolved number changes: the predicate built there does not set the composed result's season or TOU, and no fixture or constructed document produced a different tier count under the real calendar, (constantly true) or (constantly false).
:version accepts a map from document @id to snapshot, scoped the way :versions-by-plan already is, so a composition can pin several documents explicitly. A document the map does not name resolves normally.
:version-opt-not-a-version, refused when :version is neither a snapshot nor such a map. There is no document it could apply to, so reading on would silently ignore what the caller asked for.
Upgrading: no resolved number changes unless you pass :version. If you do, and you pass it to resolve-prices-with-modifiers, you were getting the base plan's rate with every modifier silently dropped; you now get the composed rate. marginal-unit-rate still carries six :reason values; resolve-prices now carries seven, the new one being :version-opt-not-a-version.
Version windows on modifiers, and two schedule defects that made price-schedule disagree with price-segments about the same instant.
A superseded urpx:RatePlanModifierVersion kept applying its prices. version-covers? had exactly one call site, in resolve-prices, so a modifier's validity window was read when choosing among several candidates and never when there was one — which every filing has. SCE Schedule D composed with a CARE modifier expired 2026-04-01 returned 0.19804875 at 2026-07-15, the full discount, three and a half months after the filing stopped describing anything. The base plan has refused the symmetric case since 0.22.0, so the answer depended on which side of a composition the superseded filing sat.
Marked, not dropped: every ledger a non-covering modifier produces carries :urpx.resolved/outside-version-window, and marginal-unit-rate refuses. Dropping a superseded discount would quietly raise a bill, and a caller reading tiers off a superseded filing deliberately is not blocked. Scoped to prices that would actually be summed, so an expired modifier contributing only non-energy or already-excluded prices costs nothing and does not refuse.
Reported from urpx-rate-plans, where end bounds prepared for three modifier versions were inert in the consumer.
price-schedule merged across a version boundary and reported one side's answer for the whole span. same-ledgers? compares :urpx.resolved/ledgers, but a base plan's version-window mark sits on the result, so a boundary that changed nothing else was invisible. Both directions were wrong, and the false-refusal one was reachable on an unmodified vendored fixture with no mutation: sce-d-2026-03-20 takes effect on the 20th, and a March window returned one interval carrying the not-yet-effective mark, so marginal-unit-rate refused for the whole month including the twelve days the filing covers, while price-segments split correctly and a direct resolve returned 0.31085. A version expiring mid-window failed the other way, carrying the covered side's rate past the bound.
same-ledgers? now compares the mark as well. Measured across 43 corpus plans over two month-long windows, base and composed: exactly 1 of 86 schedules changes interval count, the intended boundary, and total intervals move 3206 → 3207. No over-splitting.
price-segments fragmented a window it should have merged. The result-level mark carried :at, the instant resolved, and price-segments merges on the whole resolved value, so a base filing queried across a window it does not cover returned 31 segments over a month where it returns 1, each holding identical prices. Present since 0.22.0.
Both version-window marks now carry the same three fields — :version-id, :effective, :expires — and neither carries a per-instant field. Two marks with two shapes was two chances to reintroduce the same defect, and it had already happened once: the modifier mark was written by copying the result-level one, :at included, which made an hourly month over an out-of-window composition return 744 intervals where it returns 1. A test asserts the two key sets are equal rather than asserting each separately.
The :outside-version-window refusal names every stale ledger, grouped by the version that made it stale. Naming only the first cost a caller one refusal round per stale modifier to discover a set the library already had, and with two versions involved there was no single "the version used" to name. ex-data gains :expires, :ledger-id and :stale, so a consumer classifying programmatically is not worse off than one reading the prose.
Upgrading: narrowly breaking. (:at (ex-data e)) is gone from an :outside-version-window refusal and the message says "does not cover the instant resolved" rather than naming it — the instant is the caller's own, and :effective/:expires identify the window. price-schedule returns more intervals for a window straddling a version boundary, which is the fix rather than a side effect. Measured across the production corpus at 19 probe instants, 0 of 817 resolved cells change value.
:urpx.resolved/condition-gated on a tier whose Price, or whose PriceDefinition, carries a urpx:ConditionExpression this library does not evaluate. Such a price fires unconditionally at resolution, which is a silent assumption that its predicate holds, and it is epistemically the same position as an event-driven price. Event-driven prices have been marked since 0.19.0 and excluded from the marginal since; condition-gated ones were neither. Composing SCE TOU-D-PRIME with its CPP modifier returns -0.13897 at summer peak and -0.46336 off-peak, a customer being paid to consume, because a credit gated on "is a CPP event called today" and a credit gated on a grandfathered-customer attribute both apply at every instant.
Each gate carries :urpx.resolved/gate-id, :urpx.resolved/gate-name and :urpx.resolved/gate-on (:price or :price-definition). Rules stated as a bare {@id} reference are dereferenced, for the same reason the month gate dereferences them.
Marking only. Nothing is excluded from any aggregate and no resolved number changes. Measured over the 76-file production corpus at four instants, 101,992 of 105,177 resolved tiers are condition-gated, so excluding wholesale would delete the corpus rather than correct it. The population that actually reaches a per-kWh marginal is far smaller: 1,352 of 3,717 tiers on import energy ledgers, the other 100,640 being export prices that urpx:flowDirection urpx:reverse already keeps out of the sum.
A rule carrying only urpx:month is not reported, because that constraint is evaluated. Conditions inside a urpx:ProfileAlternative (required on every alternative by the shapes) and inside a urpx:EligibilityRule (which answers who may take the tariff) are not reported either, since neither hangs off a Price.
marginal-unit-rate refuses with :condition-gated-tier rather than summing a price whose condition it cannot evaluate. Closes the half of #5 that #9 tracked. Pass :assume-conditions-hold? true for the every-condition-true upper bound, as :include-dynamic-pricing? true gives the during-event one.
Two different defects reach this guard, and the larger one is a modeling problem. SCE's CCA-CRS files twenty PCIA vintages as twenty condition-gated prices, one per vintage year, of which exactly one applies to any customer. Summing all twenty took SCE TOU-D-PRIME from 0.60067 to 0.95664, where the true range is 0.59423 to 0.62691. That is the identical defect this function has refused since 0.20.0 when a filing states it as urpx:PriceSetAlternatives, which is how PG&E files the same concept: the answer used to depend on which layer a filer expressed exclusivity in, and now it does not. The other defect is a genuine unevaluable predicate: SCE's CPP credits, gated on "is a CPP event called today" and on a grandfathered-customer attribute, applied at every instant and made the composed marginal negative, -0.13897 at summer peak, which says a customer is paid to consume.
Scoped to prices that would actually be summed. 100,640 of the corpus's 101,992 gated tiers are export prices in net-billing matrices where the gate is the lookup key; urpx:flowDirection urpx:reverse has kept those out of the marginal since 0.21.0 and they do not refuse it either.
Upgrading: measured across the production corpus, 35 of 172 composed resolutions change from a number to a refusal, on five base plans: PG&E E-ELEC, and SCE D, TOU-D-4-9PM, TOU-D-5-8PM and TOU-D-PRIME. Every one of those numbers was wrong, and the ones driven by CRS were wrong by more than half the base rate. The remaining 137 are unchanged.
Discounts are not affected. A urpx:PriceSummary whose members carried a gated price would have propagated the refusal into every CARE and FERA composition; measured, none does, and SCE Schedule D still returns 0.19804875 with CARE and 0.2550770 with FERA, identical with and without :assume-conditions-hold? true. Flagged from urpx-rate-plans before release, and checked rather than assumed.
Where the gated set is mutually exclusive, the filing wants urpx:ProfileAlternatives and the refusal becomes a selection you can make; where the gate is a genuine predicate, :assume-conditions-hold? true restores the previous value explicitly.
urpx:referencesProfileAlternative on a urpx:PriceDefinition, and a narrower rule for when withholding costs the caller an answer.
urpx:referencesProfileAlternative is now read on urpx:PriceDefinition. URPX names PriceDefinition explicitly as a place to scope applicability to a ProfileAlternative; clj-urpx read it only on PriceSetAlternative, so the slot existed in the vocabulary, passed the shapes at sh:minCount 0, and did nothing. Given the caller's :profile-alternatives, a definition that references no alternative applies as before, one that references the chosen alternative applies, one that references a different alternative in that dimension is filtered out silently, and one in a dimension nothing was chosen for has its prices withheld with the ledger saying so. The selection map is keyed by the ProfileAlternatives container while a definition names an alternative; urpx:hasProfileAlternative bridges them, so no new vocabulary is needed.
Read on the base plan and on modifiers alike. A matrix of mutually-exclusive cells is a modifier shape, so a version that read the property only on the base plan would have been a no-op exactly where it is needed.
This is where a multi-dimensional choice lands. ProfileAlternativesShape is sh:maxCount 1 on hasAlternativeDimensions, so the dimensions live in one container and its alternatives enumerate the cells of the cross product: PG&E's own baseline territory and code set is 2 dimensions and 20 cells. Naming several alternatives on one definition refuses with :multiple-profile-alternatives rather than defining a combining rule for a shape no filing has.
The corpus has zero uses today. The slot has to work before a filing can be authored against it.
:urpx.resolved/withheld-prices on a withheld ledger, alongside the existing :urpx.resolved/unselected-alternatives. The latter says that something was withheld; the former says what, under the same keys a resolved tier uses.
Two further refusals from resolve-prices, both where reading on would return a confident wrong number. :ambiguous-profile-alternative when the alternative a definition names is listed by two ProfileAlternatives, so there is no one dimension to read the caller's choice from. :unknown-profile-alternative when :profile-alternatives chooses a value the container it is keyed under does not list.
The second is a behavior change on a path every filing already uses, not just on the new slot. A stale or mistyped alternative id used to be skipped silently, which dropped the whole dimension: on PG&E E-TOU-C composed with its vintage-PCIA modifier it returned 0.46886 where the correct id returns 0.49859, and the PCIA ledger was simply absent from the result. The check runs once per document, against the containers that document declares, so a key meant for a modifier is still ignored by the base plan and vice versa. A test that had pinned the silent skip as the specification was rewritten.
:summary-member-unselected, a new :urpx.resolved/unvaluable reason on a derived price whose urpx:PriceSummary names a member in a dimension the caller chose nothing for. Excluding such a member understates the subtotal and including it overstates it, so the summary declines to be valued.
Withholding refuses the marginal rate only when it would make the sum short. marginal-unit-rate refused whenever a withheld energy ledger was present, whatever had been withheld. An event-driven price (urpx:hasDynamicPricing) or a recurring charge is already outside a non-event per-unit marginal, so hiding it changes no number; the refusal is now decided by the same predicate that excludes a tier from the sum. Asking for the during-event upper bound with :include-dynamic-pricing? true wants that price again, and the same document refuses again.
Without this, moving a set of event-driven prices onto the selection layer would convert a correct rate into a refusal no caller could clear. CPAU's E-HRA cells are the utility's own reserve level and generation band, reset by council vote; a consumer has no source for either, so the selection the refusal demands could never be supplied. Vintage-PCIA is the case that keeps the refusal, being an ordinary per-kWh price and an attribute of the customer the caller does know. Raised from urpx-rate-plans before authoring against the new slot.
A recurring modifier price is now excluded from the per-kWh marginal. urpx:intervalDuration is read from the PriceDefinition on the modifier path as it always was on the base plan. It was unreachable there, because the modifier path never called the function that stamps it, so the exclusion marginal-unit-rate documents for recurring charges could not fire for any modifier tier. No vendored fixture changes: the modifier definitions that carry an interval today all sit on non-energy ledgers.
A urpx:PriceSummary counted members whose PriceDefinition does not apply. Member gating read TOU period, month and season but not urpx:referencesProfileAlternative, so a definition scoped to an alternative the caller chose against was correctly dropped from :urpx.resolved/tiers and still summed into every subtotal naming it. A derived price then multiplied a total containing a charge the customer does not pay. On SCE Schedule D composed with its CARE modifier the marginal rate came back 0.08043875 against a correct 0.11866200, a third low, with no mark and no refusal. Gating two of the three places applicability is consulted is worse than gating none, and this release is what made the three disagree.
A composition indexed its alternatives once per document, so three paths could disagree about one PriceDefinition. A composed resolution reads ONE :profile-alternatives map, but the base plan, each modifier, and the urpx:PriceSummary member gate each built their own index from their own urpx:hasProfileAlternatives. Two consequences, both introduced by this release: a definition scoped to a container the other document declares was :unselected for every possible selection, so the withhold could not be cleared from a composed call at all; and where one ProfileAlternative @id was listed by a different container in each document, the modifier's tier path answered :selected while the summary answered :not-selected, in one result, silently dropping a member the tiers kept. There is now one index over every document in the composition, merged on container @id so a shared dimension declared in both does not read as two. An alternative genuinely listed by two different containers still refuses with :ambiguous-profile-alternative, across a composition for the same reason it already did within one document.
A modifier ledger marked from two sources reported only one. A ledger can be marked both by prices surfaced from an unselected PriceSetAlternatives and by a definition scoped to an unchosen alternative; :urpx.resolved/unselected-alternatives dropped the second whenever the first was present, so a caller who supplied the selection the error named got a second refusal about something the first never mentioned.
:all-energy-tiers-excluded blamed exclusion for prices that were withheld. Its message told the caller to inspect fields on :urpx.resolved/tiers, which is empty in that case; it now names :urpx.resolved/withheld-prices and says a selection would give the ledger some tiers.
urpx:referencesProfileAlternative was not declared on the PriceDefinition schema. It validated by accident, as an unknown key on a non-closed shape, and passed through coerce and emit untouched. What it did not get was the single-or-vector normalization every other reference property has, so the coerced document carried a shape the schema does not describe and urpx.emit round-tripped it unnormalized. The resolver itself tolerates both forms, so this is a modeling hole rather than a wrong number: authoring against a slot that validates by accident is how the next filing finds one.
A valid selection could be refused where a container and its price sets disagree. validate-selection! checked the caller's value against the container's urpx:hasProfileAlternative list only, while the price path matches it against each PriceSetAlternative's own urpx:referencesProfileAlternative. Nothing in the shapes forces those two sets to agree; all eight pairs in this corpus do, so nothing changes today, but the index now covers both, because refusing a document the library resolves correctly is the expensive direction. The fold happens where the composition index is built, covering every document at once: folded per path instead, the base plan's check could not see a cell only a modifier's price sets name, and the summary member gate, reading the index before any fold, saw none of them at all.
A container listing the same alternative twice read as two dimensions and refused an unambiguous selection with :ambiguous-profile-alternative. A duplicate entry is a redundancy in the document, not a second dimension.
Modifiers that do not apply to the plan are no longer indexed. The composition index was built from every modifier handed in, so one naming a different plan could contribute its containers to a resolution it takes no part in, and reading its version to get them surfaced its version errors there too. urpx:applicableToRatePlan now gates the index as it already gated the prices.
deploy refuses unless the artifact is the one this commit built. It publishes whatever is in target/, so it was trusting that a human ran ci first, on this commit, from a clean tree; trusting that shipped a pom from an eight-version-old build once. The embedded build-provenance is now the witness: deploy checks the jar's recorded commit and version against HEAD, and refuses on a snapshot version, a dirty tree, a missing or misplaced v<version> tag, a stale or unbuilt target/, or an unset CLOJARS_PASSWORD. It also refuses when src/ or resources/ holds a file git IGNORES that would actually be packed, because ci packs those directories whole while neither git status --porcelain nor build-provenance's own dirty check reports ignored files, so such a file rides into the jar with every gate quiet. Files tools.build excludes on its own (.DS_Store, *~, #...#, .#...) do not trip it, since refusing a release over something the artifact provably never packs is a guard that fires on legitimate work. A corrupt or unreadable jar reads as "no provenance" and refuses by name rather than escaping as a ZipException that would preempt the refusals naming the real fix. Added clojure -T:build tag and clojure -T:build release, and doc/RELEASING.md.
Upgrading: two changes can turn a call that returned a number into a refusal, and both were returning a wrong number before. A :profile-alternatives value that names something its container does not list is now :unknown-profile-alternative rather than a silent skip; check your selection ids against urpx:hasProfileAlternative if you see it. And a plan whose urpx:PriceSummary names a member in an unchosen dimension now reports :summary-member-unselected rather than summing it. Neither fires on any filing that does not use urpx:ProfileAlternatives.
The last two findings from the silent-drop sweep, which share one cause: the unselected-alternatives mark was computed from surviving tier entries, and a ledger with none is elided before anything can mark it.
An unselected urpx:PriceSetAlternatives vanished without trace on the base plan path. prices-from-alternatives contributes nothing for a set the caller made no selection for, so the prices, the ledger they would have populated, and any evidence they existed all disappeared together. Six shipped base versions carry such a set. The ledger now survives carrying :urpx.resolved/unselected-alternatives, so a consumer can tell an unchosen dimension from a plan that has none.
Emptying an alternatives ledger deleted the refusal that guards it. Because the mark came from surviving tiers, any route that emptied the ledger also removed the protection: dating PG&E's 18 vintage-PCIA prices into the future took marginal-unit-rate from refusing to answering 0.46886, silently short a component spanning 0.06391/kWh. Removing content can no longer remove a guard.
Its originally reported trigger — stripping urpx:startDate from those prices — was already neutralized by the 0.22.0 start-date fix and now refuses correctly. The defect survived that; only the route to it changed.
The 0.22.0 summary tier fix could not reach the corpus. It read urpx:serviceTierNumber from the member Price only, with no fallback to the Price's PriceDefinition. URPX puts the property on both, and 69 of 73 corpus filings carry it on the definition, one per tier, which is the house idiom; the fixture that validated the fix is pinned at a commit predating that idiom and carries it on the Price. So the fix computed the right answer on the fixture and no answer at all on real filings: SCE Schedule D's CARE discount stayed at −0.12139725 where −0.10070125 is correct.
tier-info has read both placements since long before, three hundred lines up in the same file. Writing a second, narrower rule instead of reusing it was the defect. Both placements now give −0.10070125, and a test pins both, because a test covering one placement is exactly what let the narrow version through.
Reported from urpx-rate-plans, which declined to work around it on the data side: putting the discriminator back on the Price would have made SCE D the only filing in 73 doing so, which is the anomaly that caused the 3.09× SDG&E defect in the first place.
Silent losses. A sweep looked for every way a SHACL-conformant document can lose content or produce a wrong number with no diagnostic; it returned seven confirmed findings, rejected nineteen candidates as correct gating or existing loud refusals, and caught four overclaims among its own results. Two more came from urpx-rate-plans. Resolved prices change on filings that carry any of these shapes.
urpx:startDate discarded the Price. Four scoping sites read the date with when-let and dropped anything without one, while the PriceSet gate three lines above treats a nil date as permissive. It deleted content from a shipped filing: PG&E's E-TOU-C baseline-tier modifier carries four Prices and no dates anywhere, resolved to zero ledgers and zero tiers, and composing it with its base plan changed no number and reported nothing. Absence now means no start constraint, which is what sh:minCount 0 says. A Price that has not started yet is still excluded.urpx:startDateTime was never read. Modeled on Price, PriceSet and PriceSetAlternative, written by urpx.emit, ignored by the resolver: a filing stating its date that way was accepted, validated, round-tripped and then priced as though it had no date. On SDG&E DR, restating one PriceSet's date that way took 13 tiers to 3.:urpx.resolved/outside-version-window and marginal-unit-rate refuses with :reason :outside-version-window. Season and TOU period still answer, because those are structural questions about a plan's shape.urpx:touPeriodNumber on a Price was ignored. URPX puts it on Price as well as PriceDefinition and 43 corpus Prices use it. Two different wrong answers followed: in a summary every TOU variant of one charge summed, overstating SDG&E EV-TOU's CARE discount 3.09× (−0.9255435 against −0.2999885); in the tier path the variants collided in the latest-wins dedupe and the survivor was reported at every hour, so a transmission charge read at its on-peak value around the clock.marginal-unit-rate had within a ledger in 0.20.0, one level down, and the partition is now one shared function rather than two copies.sdge-ev-tou-2026-04-01 (12 Prices carrying touPeriodNumber directly — urpx-rate-plans has since collapsed these, so this is the only place the case survives), pge-e-tou-c-baseline-tier-2026-03-01 (four Prices, no dates), and sce-d-2025-11-15 earlier.Upgrading: a filing carrying any of these shapes prices differently, generally more completely. marginal-unit-rate gains a fifth refusal, :outside-version-window, which fires when a filing is resolved outside the window its version describes — check :reason rather than catching broadly.
Export ledgers were summed into the import marginal rate. urpx:flowDirection urpx:reverse marks compensation for energy sent back, which is not a cost of the next kWh consumed. marginal-unit-rate ignored the property entirely, so an export ledger was added to an import rate as though it were a charge.
Measured: composing PG&E E-ELEC 2026-03-27 with its NBT modifier returned -213.15920/kWh, because four export ledgers carrying 4,610 hourly avoided-cost values were added to a 0.56006 import rate. It now returns 0.56006, the base plan's own rate. The audit that found it counted 22 compositions summing export compensation into an import marginal, 17 of them real defects.
urpx:net and urpx:bidirectional still contribute. Neither appears in the corpus, so excluding them would be speculation; a netted quantity in an import marginal is genuinely ambiguous and that is the line to revisit if one appears.
:urpx.resolved/flow-direction on resolved ledgers, when the filing declares one. An export ledger and an import ledger are not the same kind of thing, and a consumer summing per-kWh charges needs to tell them apart.has-energy-ledger? now agrees with marginal-unit-rate about what counts, so the guard cannot answer yes where the aggregate would refuse. A plan whose only energy ledger is an export ledger reports false and refuses with :no-energy-ledger.marginal-unit-rate returns a different, larger number on most plans. Read the upgrade note.
Additive rate components were dropped from the marginal rate. marginal-unit-rate kept only the first tier of each energy ledger. That is correct for a block-tier ladder, which is what the function was written for, and wrong for an additive rate stack, which is how delivery charges are modeled throughout the corpus. A delivery ledger carrying ten components contributed one.
Measured at 2026-07-15 18:00 PT:
| plan | reported | correct | |
|---|---|---|---|
| SDG&E EV-TOU 2026-04-01 | 0.57086 | 0.86302 | 33.9% low |
| SCE TOU-D-PRIME 2026-03-20 | 0.55208 | 0.60067 | 8.1% low |
| SCE TOU-D-4-9PM 2026-03-20 | 0.54513 | 0.49264 | overstated by 0.05249 |
The error was not one-sided, so no sanity bound would have caught it: SCE's overstatement comes from dropping a negative Baseline Credit. A sweep of 40 base plans at four instants found 74 resolutions dropping a nonzero remainder across 26 of the 40 plans.
urpx:serviceTierNumber separates the two uses: a numbered tier belongs to a ladder, a bare one is an additive component. Across that sweep, multi-tier energy ledgers were 183 additive, 31 ladder and 3 mixed, all three being the same ledger. A mixed ledger contributes its bare components plus the lowest numbered rung; on the one real case those two policies agree exactly, and only this one stays correct if a genuine ladder ever appears beside additive components.
Upgrading: every consumer of marginal-unit-rate should expect a different number on most plans, usually larger. Reported rates moved on 26 of 40 base plans in the corpus. Anything publishing this value downstream — an OpenADR 3 interval price, for instance — was publishing an understated rate on most plans and an overstated one on some. There is no flag to keep the old behavior, because the old behavior was wrong.
urpx:serviceTierNumber signal it turns on.urpx:referencesProfileAlternative is read on PriceSetAlternative only and not yet on the PriceDefinition-level slot the shapes allow.Event-driven prices, and a fix for a guard that shipped broken in 0.18.0.
price-segments rejected every modifier filing in the corpus. Its uncoerced-document guard threw on any date-shaped string, which is incoherent next to harvesting by java.time type: after coercion a genuine temporal is a java.time value, so a residual date-shaped string is evidence of nothing. urpx:version is rdfs:range xsd:string and every filing uses its effective date as the version label, so "2026-03-20" is a perfectly coerced string. All 33 modifier filings and 3 base plans were refused, and therefore every composed horizon. The guard now tests the signature of an uncoerced document (no java.time values anywhere while date-shaped strings are present) rather than the shape of a string.marginal-unit-rate excludes event-driven prices by default, closing the open half of #5. A Price carrying urpx:hasDynamicPricing applies only while its event is called: critical peak events are capped around 50 hours a summer against roughly 2,200 summer hours, so summing one unconditionally answered "what would this cost during an event" while reading as the ordinary rate. Measured on SMUD CPP, composing the modifier moved the rate from 0.3765 to 0.8765; it now stays at 0.3765, which is the non-event baseline the issue predicted. Pass {:include-dynamic-pricing? true} for the during-event upper bound.
Unlike the mutually-exclusive alternatives this function also refuses on, there is a defensible default here rather than an arbitrary pick: no-event-today is the ordinary state.
Tiers are now excluded before the first tier is taken rather than after, so a ledger mixing an event charge with an ordinary price contributes the ordinary one instead of dropping out entirely.
A new refusal, :all-energy-tiers-excluded, for the case where exclusions empty the sum. Returning 0M there would be the nil-reads-as-free failure one level up again.
urpx:ConditionExpression predicates this library does not evaluate, so excluding the event charge leaves them applying and the marginal rate goes negative. That number was already wrong; it is now wrong more visibly. A test pins it so the limit is not mistaken for a fix.:all-energy-tiers-excluded; a claim that a ledger with no MetricInput makes the aggregate return 0M when since 0.15.0 it refuses; and a feature bullet predating price-segments.Horizons.
urpx.schedule/price-segments, a piecewise-constant price schedule whose boundaries are derived from the documents rather than discovered by sampling. Built for a client that wants the price now and for the next N hours as intervals to hand a solver, rather than N point resolutions.
Temporal literals are harvested by java.time type rather than by property name, so a temporal property URPX adds later is picked up without this code changing; a name list would have to be kept in step, and the one it missed would be a missing candidate. Those times are crossed with every local date the window touches, every local midnight is included unconditionally, and the zone's DST transitions are added. The candidate set is a deliberate over-approximation: an extra candidate costs one resolution and merges away, a missed one is a silently wrong boundary.
Every local midnight being a candidate is what absorbs every date-granularity gate at once: day of week, holidays including observance shifting, season and month boundaries, applicability month rules, and every start, effective and expiration date.
DST handled at the instant level. A local time inside the spring-forward gap never occurs and is clamped to the transition instant; ZonedDateTime/of would land past it by the gap's width, which is not always an hour. A local time in the fall-back overlap occurs twice and yields both instants, so the repeated hour is priced twice rather than merged away. Zone transitions are enumerated by iterating ZoneRules.nextTransition rather than assuming two a year, because real tzdb data does not oblige.
Refusals become segments. A resolution that throws partway through a horizon is reported as :urpx.interval/refused rather than aborting the window, and equal refusals merge so an uncovered stretch is one segment rather than one per candidate.
A test pinning the premise the whole design rests on: urpx.price reads the query instant through exactly .toLocalDateTime, .toLocalDate, .toLocalTime and .getDayOfWeek, and never through .toInstant, .getOffset or .getZone. Resolution is therefore a pure function of the local date-time, which is why boundaries can be derived from literals at all. A resolver that read the offset would invalidate the candidate rule without touching the file that implements it.
price-schedule reported prices outside the requested window. The step grid is anchored to the start of start's hour, and that anchoring leaked: a request beginning at 17:23 returned a first interval beginning at 17:00 and a last interval ending up to a full step past end. Intervals are now clipped to the window, and clipped before merging, while interior boundaries still fall on clock hours.Version selection is time-aware.
The version snapshot was chosen arbitrarily when a plan had more than one. plan-version fell back to (first ...) over the back-pointer index, taking whatever the index happened to hold rather than what was in force. Same class as the vintage-PCIA arbitrary pick: a defensible-looking number with no rule behind it. It stayed invisible because every filing carries exactly one version, so loading one file gives one answer; six SCE plans in the corpus already carry two, and a forecast client loading several versions to cover a window is the case it was silently wrong for.
The rule is now the latest urpx:effectiveDateTime at or before the instant, inclusive, skipping any version whose urpx:expirationDate has passed. All 84 version snapshots in the corpus carry an effective instant and none carries an expiration, so a version's span is inferable only from its successor's start; the rule therefore needs a plan's whole candidate set rather than one snapshot.
:reason in ex-data: :no-version-in-effect (nothing had taken effect yet), :ambiguous-version (two candidates share the latest effective instant), and :version-without-effective-instant (a candidate cannot be ordered, since urpx:effectiveDateTime is sh:minCount 0).urpx:expectedEndDate and urpx:expectedEndDateTime are deliberately not read. An expected end anticipates a successor rather than terminating anything, so a version past its expected end with no successor is still in force; treating it as an expiration would silently drop a plan that outlived its own forecast.@id. Two CPAU E-1 filings publish the same version @id at the same effective instant while differing in urpx:hasModifierOptions; collapsing on @id would pick one silently, which is the defect being removed. As two candidates they tie and are refused.sce-d-2025-11-15 fixture, the prior version of a plan already vendored, so the suite holds two real versions of one plan that price differently. Every other fixture is a single version, which is the condition that hid this.One fix, and it changes what resolve-prices returns for every base plan in the corpus.
Recurring charges were dropped on the base-plan path and kept on the modifier path. pd-active-at-tou? answered two unrelated questions, returning false for any PriceDefinition carrying its own urpx:CalculationMethod. Because the modifier path applies that predicate only when there is a TOU period to gate against, the recurring exclusion was skipped along with it, so the same shape was dropped from a base plan and kept in a modifier. The two questions are now separate and recurring charges surface on both paths.
The asymmetry produced incoherent output on filed documents. Composing SCE D with SCE CARE, the CARE discounts on the Base Services Charge surfaced while the Base Services Charge itself did not: the result carried credits against a charge that was not in it. Every vendored base plan lost at least one real charge this way (minimum charges, power access, service charges, base charges), and CPAU E-14, whose eighteen PriceDefinitions are all recurring, resolved to nothing at all.
:urpx.resolved/interval-duration. A recurring charge is priced per interval, and its ledger is typically measured by a meter or fixture count with no urpx:measurementKind and no unit, so the number alone is uninterpretable. Flattening a period-scoped quantity into a per-unit rate is how a monthly charge becomes a wrong daily one.marginal-unit-rate excludes any tier carrying an interval duration. No recurring ledger in the corpus is energy-kind, so this changes no current value; it is there so a filing that does pair the two cannot quietly add a per-month figure to a per-kWh rate. Every marginal rate in the corpus is unchanged, pinned as exact values.:no-energy-ledger refusal now distinguishes "nothing resolved at all" from "ledgers resolved, none of them measured in a kind this library can sum". CPAU E-14 moved from the first to the second.Upgrading: every base plan gains at least one ledger. Consumers that enumerate :urpx.resolved/ledgers will see charges they did not see before, some of them per-interval rather than per-unit; check :urpx.resolved/interval-duration before treating a unit price as per-unit. Marginal rates are unaffected.
marginal-unit-rate refuses when a plan has no energy ledger at all, where it returned 0M. A street lighting schedule prices per lamp per month; its marginal unit is not a kWh, so the question has no answer and 0M was a confident wrong one, indistinguishable from a plan whose energy ledgers genuinely sum to zero. Both states are now distinguishable: the second still answers zero, because that is the answer.marginal-unit-rate makes now carries a :reason in its ex-data (:no-energy-ledger, :unvaluable-marginal-tier, :unselected-alternatives), so a consumer sweeping a corpus can classify rather than catch-all.has-energy-ledger? and measurement-kinds, so the refusal above is not a tax on every caller: ask before computing. measurement-kinds returns the kinds a plan does price in rather than a bare boolean, so a sweeping consumer can route instead of merely skip.28390dc: the only filing in the corpus with no energy ledger at all, which isolates that case with no other variable. Its ledgers carry urpx:referencesMetricInput naming fixture-count MetricInputs with no measurement kind, so it exercises "ledgers wired, no energy kind among them" rather than "ledgers unwired".384ffb1. Both utilities are vendored because neither alone shows the variation, and a fixture set carrying only one would silently sanction whichever shape it held. SCE shares one subtotal between CARE and FERA, which differ only in rate; SDG&E carries two subtotals on one base plan with different member lists, because Schedule E-CARE excludes any applicable exemptions and FERA grants none. Unlike the PG&E @graph bundles each base plan is a separate filing from its modifiers, so the cross-document dereference is exercised the way the corpus files it. urpx.care-fera-test pins the filed values: SDG&E CARE -0.1450505 and FERA -0.0756612, SCE CARE -0.12139725 and FERA -0.0672354, each checked against the summary's own member arithmetic as well as against the resolver.urpx:startDate on a urpx:Price with nothing enclosing it to inherit a date from (dropped silently), and urpx:referencesMetricInput on a urpx:Ledger (tiers survive, but marginal-unit-rate comes back quietly short, or 0M). Both were found by authoring real filings, not by validation.marginal-unit-rate began refusing: the walkthrough composes PG&E's vintage-PCIA modifier, so the first thing a new reader ran would have thrown. README blocks were only checked for symbol resolution, which cannot catch that. The Getting started section is self-contained, so it is now evaluated like the DSL tutorial.:profile-alternatives and explains why; the refusal is described in prose rather than shown as a throwing call.:ambiguous-basis and :summary-no-active-members.marginal-unit-rate and the PriceSummary schema docstrings stated the old behavior; both now say what they do.Composition. Five defects in how a modifier resolves against a base plan, four of them wrong answers rather than errors and one a crash, plus three semantics that URPX leaves open and that this library now states rather than implies. Found by auditing the compose path to answer a question from the filings repo, not by a failing test: every one is reachable by a SHACL-conformant document, and the 73-filing corpus gate caught none of them, because it validates documents and these are resolution defects.
Upgrading: marginal-unit-rate now throws on a composed plan whose modifier carries mutually-exclusive price alternatives, where it previously returned a number. PG&E filings are the case in practice. Pass :profile-alternatives to select, or read :urpx.resolved/tiers directly, which needs no selection.
marginal-unit-rate silently added an arbitrary price from a matrix modifier. PG&E's vintage-PCIA ledger carries 18 prices selected by departure vintage: mutually-exclusive alternatives, exactly one of which applies to a customer. They were treated as a block-tier ladder and the first was taken, so composing the modifier turned a 0.41784 marginal rate into 0.44757. The 18 span 0.06391/kWh and the pick was document order. Resolution still surfaces all of them (the caller-picks contract is unchanged); the aggregate now refuses. Pass :profile-alternatives to choose, and only the chosen alternative surfaces.ClassCastException on a holiday. The holiday predicate's result was passed where the predicate was expected.urpx:forRatePlanModifier was dropped from composition without a word, giving a result indistinguishable from passing no modifier at all. Un-nesting is the shape v0.5.1 introduced.marginal-unit-rate (which documents itself as taking the lowest tier) got whatever came first in the document.urpx:PriceSummary members were summed with no season, month or TOU gating. A summary naming a summer price and a winter price returned the sum of both in July. Members are now gated by the same predicates that gate a tier, applied to each member's own PriceDefinition, so one summary per base plan can name every discountable line item and stay correct in every TOU period. When no member is in effect the result is :summary-no-active-members rather than a confident zero.Multiple urpx:basedOnPrice values are read disjunctively as candidate bases, exactly one expected to be present in the documents passed to the call; 0.13.0 threw. This lets one discount Price serve a modifier applying to many base plans. Zero resolvable candidates is still an error and two present at once reports :ambiguous-basis. URPX fixes the cardinality and says nothing about multiplicity; the conjunctive reading is what urpx:PriceSummary already provides, so the disjunctive one is taken here and recorded rather than left implicit.
:profile-alternatives is now honored on the modifier path. It was read on the base plan path and discarded on the modifier path, which is why matrix modifiers had no way to be selected.
The README's description of how matrix modifiers carry their selection dimension was wrong, in two places: it is urpx:PriceSetAlternatives / urpx:referencesProfileAlternative, not urpx:hasApplicabilityGroup with a urpx:ConditionExpression.
Only one of the two was corrected in the 0.14.0 artifact; the second, and the rest of the documentation catch-up, landed after the tag and ships in 0.15.0.
One feature. A urpx:Price may state urpx:multiplier with urpx:basedOnPrice instead of a urpx:unitPrice, naming the price it is derived from. That is how a percentage discount such as CARE or FERA is filed. 0.12.0 returned nil for every such Price, and a nil unit price reads to a consumer as free, so a 32.5% discount arrived as a 100% one.
urpx:PriceSummary, which aggregates line items through urpx:includesPrice (sh:minCount 2 upstream: a summary of one price is not a summary), and urpx:hasPriceSummary on PriceSet and PriceSetAlternative, which is where a summary attaches (it is rdfs:domain urpx:PriceSet, so a subtotal lives beside the line items it names). A price-summary DSL builder comes with it; nothing else was needed, because make-builder derives kwargs, aliases and validation from the schema, and a test now pins that rather than assuming it.resolve-prices values derived prices: the multiplier times the referenced Price's unit price, or times the sum of a referenced summary's members. urpx:summaryAmount is read but not trusted, so an author-stated total that disagrees with the members stays visible rather than becoming authoritative.:urpx.resolved/unvaluable on any tier whose Price could not be valued, carrying a reason and a detail: :unresolved-reference, :summary-member-unresolved, :basis-not-valuable, or :no-unit-price. This is the point of the release. A Price this library cannot value is now named as such instead of surfacing as a nil that reads as free.resolve-prices-with-modifiers builds one @id index spanning the base plan and every modifier, so a discount Price filed in a program modifier can dereference a summary defined in the base plan filing. That is the shape these discounts take in practice, the discount belonging to the program and the line items to the plan. No cross-file lookup is invented: the index covers exactly the documents the caller passed in, and a modifier resolved on its own reports :unresolved-reference.marginal-unit-rate throws when a marginal energy tier could not be valued, rather than summing around it. Skipping one reports the undiscounted price as though it were the real rate.urpx:basedOnPrice is accepted as 0..N, as PriceShape declares (sh:minCount 0, no sh:maxCount); it was modeled as a single value. With more than one basis stated, resolution throws rather than picking, the same refusal the 0..N properties in 0.12.0 make.scripts/shacl-drift.sh recurses through :and as well as :or when deciding whether a schema entry is 0..N. [:and [:vector X] [:fn ...]] is how a cardinality refinement is written, and reading it as a single value reported drift against a schema that has none.Ten commits. URPX's production corpus finished migrating to v0.5.1 across all six of its per-utility waves during this window, so most of what follows is the library catching up with real migrated filings rather than with the specification on paper.
Three changes tighten validation and can reject documents 0.11.0 accepted. All three were measured against the fully migrated 73-filing corpus first and reject nothing in it, but a consumer with other filings should read them.
urpx:changeReason and urpx:serviceRole are now required on RatePlanVersion and RatePlanModifierVersion, matching sh:minCount 1, which has been their cardinality since URPX v0.2.1. All 84 version snapshots in the corpus carry both.sh:minCount 0, no sh:maxCount). With more than one nested snapshot and no explicit choice, resolve-prices throws rather than resolving against an arbitrary one. Which snapshot applies to an instant is a selection question the vocabulary does not settle, so the library refuses instead of guessing; pass :version in the opts to choose.MetricInput.hasMetricSpecification, PlanElements.hasHolidayCalendar, TimeBracket.appliesInSeason — with the same refusal rather than an arbitrary pick.@id reference to an ApplicabilityRule on a PriceDefinition is now dereferenced before month gating. Previously such a reference carried no urpx:month, read as permissive, and the gate silently stopped applying.scripts/shacl-drift.sh reports where the hand-written Malli schemas have fallen behind the shipped SHACL shapes. It is an auditor, not a generator, and emits nothing that is compiled or shipped. Findings rank by cost: rejects-valid and arity mean the library refuses documents the standard allows; under-validates and missing-required mean the reverse. Exit codes are tri-state so a run that compared nothing cannot read as success.doc/shacl-and-malli.md records why the schemas are hand-written rather than generated from the shapes, with the measurements behind that decision.Price.referencesPriceDefinition was required while urpx-shacl:PriceShape declares sh:minCount 0, so conformant documents omitting it were refused. PriceSetAlternative already modeled the same property optional, so the two schemas had been disagreeing with each other. Found by the drift detector on its first run, along with thirteen cardinality mismatches also fixed here.urpx:eiaId and into the Organization identity chain (hasIdentity → hasIdentifier → {urpx:scheme urpx:eiaScheme, skos:notation}). infer-zone read only the old key, found nothing, and fell back to matching a display name without saying so. It now reads either location.RatePlanModifier embedded in a urpx:URPXPackage, because it filtered load-graph rather than using the descending loader.Whether to keep accepting URPX v0.2.x or tighten to v0.5.1 exactly remains open. Four wave-gated requirements and one retired enum member stay loose because tightening them would reject every v0.2.x document. That is a contract decision rather than a conformance one, and it is deferred to the next consumer upgrade.
Retargets the library to URPX v0.5.1 (urpx-org/urpx at 95eb8e22) while continuing to accept v0.2.x in full. The production corpus migrates utility by utility and carries both vocabularies at once, so dual acceptance is the design rather than transitional debt.
urpx:URPXPackage as a sibling schema to urpx:URPXDocument, its v0.3.0 rename, including the urpx:packages containment pointer. Loading, coercion, DSL builders and emitters cover both.CoverageClaim, Evaluation, ModelExclusion, PriceDefinitionReference, plus forRatePlan / forRatePlanModifier and hasCoverageClaim on the version snapshots.SourcePublication (dual-accepting the v0.2.x urpx:TariffPublication @type) and TariffFiling.RatePlanIdentifier, RatePlanModifierIdentifier, OrganizationIdentifier.sh:in lists.urpx.index/versions-by-plan, and load-rate-plan-version / load-rate-plan-modifier-version.urpx.validate, a single-file CLI checker (clojure -M:validate <file.jsonld>) with tri-state exits. It descends wrappers, so it cannot report success over a document whose inner entity was never reached.hasApplicabilityRule's reference arm is closed to @id alone. The open form accepted a malformed inline rule as though it were a reference.@context is accepted in all three JSON-LD encodings — object, IRI string, or an array of either. Only the object form was accepted before, so the standard's own v0.5.1 reference documents did not validate.hasRatePlanVersion, hasRatePlanModifierVersion and effectiveDateTime are optional, matching sh:minCount 0. effectiveDateTime was never required upstream at any version.DataSourceReference accepts the v0.5.1 flattening (leafStart, leafEnd, tariffStatementType inline) as well as the v0.2.x nested TariffBookReference.forRatePlan back-pointers as well as the nested pointer.com.apicatalog/titanium-json-ld dependency, which was also non-functional as declared (no jakarta.json implementation was supplied).Can you improve this documentation?Edit on GitHub
cljdoc builds & hosts documentation for Clojure/Script libraries
| Ctrl+k | Jump to recent docs |
| ← | Move to previous article |
| → | Move to next article |
| Ctrl+/ | Jump to the search field |