[Hafeok]

SHACL part 2a, under milestone 8 (#13). Its issue is #138, and its release is v0.1.0-preview.11. This is the first of the two PRs the plan proposed and the maintainer approved. ADRs 0169–0173 are filed unaccepted, so CI is red only on CS0618. The full report is docs/traceability/2026-10-20-issue-138-shacl-part-2a.md.

What this changes, and why

A shapes binding can now gate a dataset's commits.

• Where it is declared. A binding is a management declaration (varve:ShapesBinding, ADR 0170). It names a shapes graph of the same dataset, the graphs to validate (in the grant vocabulary's graph-set terms), and a mode.
• Who enforces it. Every server adds the dataset's gate at the management position a write is authorised under.

How the gate behaves. Varve.Shacl.Store (layer 5, ADR 0169) is a dataset validator.

• reject refuses a commit whose affected focus nodes have a violation not present at the pinned position. So violations never grow, and a conforming dataset stays conforming. This follows the approved change to R1.
• report accepts the commit and attaches <<( binding varve:affectedReport "sha256:…" )>> (ADR 0173).
• off runs nothing.
• A shapes change is never refused for existing data (R5).
• Full shapes under reject. Such a binding is refused at apply time unless it declares varve:allowFullEvaluation (R4). If its shapes become full later, it fails closed (R6).
• Refused functions. NOW(), RAND(), UUID() and STRUUID() are refused, and so is a BNODE() whose value can reach a result.
• Problem types. Three new ones, each with a page: shapes-violated, binding-not-incremental and shapes-invalid.
• Describing a binding. GET /manage/datasets/{name}/bindings/{binding} serves a binding with its classification.

Only the focus nodes a commit affects are validated (Varve.Shacl, ADR 0171).

• How they are found. Footprints are anchored from each changed triple, in the states before and after the commit. They are then propagated over the shapes' dependencies to a least fixpoint.
• Classification. Every shape is classed incremental, scoped or full, statically.
• SHACL-SPARQL. Its constraints are analysed by their join graph.

Varve.Derivation (layer 2, ADR 0172, approved with ADR 0003's amendment) holds the dependency graphs and stratification, regular paths as automata, and semi-naive propagation. SHACL uses it now, and Inference will later.

The REC generator's output version 2. The generator's leases now carry rec:leasor.

Departures from the approved plan. The record has all of them. The visible ones, both accepted on this PR and stated in ADRs 0170 and 0155:

• No varve:validates node. ADR 0166's code validator refuses blank nodes in management graphs, so the graph-set terms sit on the binding node itself, as on a grant.
• One public entry point. The plan's Validate(…, FocusSelection) and AffectedBy became IncrementalValidation.Compare, because the gate needs the affected set, both reports and the introduced results together.

What was measured

Properties, with no counterexample at the stated counts. Each count is per CI run, then the long run under VARVE_LONG_PROPERTIES:

Property	Per CI run	Long run
Incremental comparison against full validation, over generated shapes and histories	500	5,000
The same for SHACL-SPARQL's join-graph analysis	300	3,000
The gate's verdicts and digests against a full-validation oracle, with shapes changed alone and with data	500	5,000
Varve.Derivation's paths and stratification	3,000 each	—

Three sabotages of the engine each failed the incremental property:

• no path anchors;
• an empty transfer;
• no provenance.

Commit latency on the file backend. BenchmarkDotNet 0.15.8, Xeon @ 2.80 GHz, 4 cores, a cloud container. The setup:

• REC 4.1.0's shapes, without Brick's three SPARQL constraints (see below);
• the portfolio's last 220 commits, after the rest was loaded;
• the REC generator's output version 2, whose leases carry rec:leasor.

No commit was refused. Slower rows first:

REC	Binding	Mean per commit	StdDev	Allocated
M (≈5 M quads)	reject	11.86 ms	3.00 ms	2,471.3 KB
S (≈203 k quads)	reject	7.62 ms	1.69 ms	2,294.6 KB
M	none	0.86 ms	0.10 ms	30.5 KB
S	none	0.51 ms	0.07 ms	35.4 KB

The number of affected focus nodes per commit is the same at both sizes. The gate's cost is theirs, not the dataset's:

REC	Affected focus nodes per commit	Time per affected focus node
S	2.6	2.48 ms
M	2.6	3.44 ms

The machine was slower during this run than during the first one, on generator version 1. That run measured 8.24 ms gated at both sizes, and the README has both runs.

The benchmark also led to a fix: the shapes graph's subclass steps are now remembered. That cut the time in the gate from 13.2 ms to 4.5 ms per commit at S.

Test durations on Linux (CI's other operating systems to follow, per #126):

Project	Duration
Varve.Derivation.Tests (new)	2.2 s
Varve.Shacl.Store.Tests (new)	16.5 s

The W3C SHACL suites are unchanged: 628 pass, and the same 28 are exempt.

What was found

• REC 4.1.0 can't be bound as vendored. Three of Brick's SPARQL constraints use sh:prefixes with no sh:declare in any of the three files. Two of them also use a prefix they don't list: owl: and brick:. A strict processor refuses the whole shapes graph. RecStandaloneTests pins this, with no exemption. Part 2b's REC gate will add the declarations and the owl:Class triple, and ADR 0176 will record the exact triples.
• The upstream draft for REC is a comment on this PR, for the maintainer to file.
• The rest of REC is maintainable. Without those three constraints, nothing is classed full. A full validation of the S head takes about 74 s; incrementally, a commit takes a few milliseconds.
• The generator wrote leases without rec:leasor, giving 141 violations at S.
  • It's fixed here: each site has an owner, and every lease names it.
  • The output version is now 2, with new pinned hashes, and the S head now conforms.
  • The reject tests get their violations from an explicit mutation (a lease's leasor retracted), never from the generator.
• The plan misread REC's SPARQL. It called those three constraints analysable.

What is left

• Part 2b, the rest of Milestone 8 — SHACL #13:
  • the validation projection and its reserved graph;
  • validation-behind, and the lag reported on readiness without affecting it;
  • shapes for Varve's declarations;
  • the full REC gate (ADRs 0174–0176);
  • docs/operator/shapes.md.
• The release gate fails here only because v0.1.0-preview.10 is not tagged. With two proposed descriptors, the checker gives Management plane, part 1: the fleet system dataset, grants as code, varve apply (v0.1.0-preview.10) #134's ADRs to this release and this part's to preview.10. It clears once preview.10 is cut.
• For the maintainer:
  • accept ADRs 0169–0173;
  • cut preview.10's tag (being done; if Extent maintenance: background merges and extent deletion (v0.1.0-preview.11) #135 lands first, this renumbers to preview.12);
  • file the REC upstream note drafted in the comments.
• The Windows and macOS durations of the new test projects come from CI's test jobs. Those run once the decisions are accepted, and I'll post them here.

Public API baseline (additive; nothing removed). The lines added to each PublicAPI.Unshipped.txt

Varve.Derivation (new; 69 lines):

Varve.Derivation.DependencyGraph<TLabel>
Varve.Derivation.DependencyGraph<TLabel>.Add(int source, int target, Varve.Derivation.Model.Polarity polarity, TLabel label) -> void
Varve.Derivation.DependencyGraph<TLabel>.DependencyGraph(int vertexCount) -> void
Varve.Derivation.DependencyGraph<TLabel>.Incoming(int vertex) -> System.Collections.Generic.IReadOnlyList<Varve.Derivation.Model.DependencyEdge<TLabel>>!
Varve.Derivation.DependencyGraph<TLabel>.Outgoing(int vertex) -> System.Collections.Generic.IReadOnlyList<Varve.Derivation.Model.DependencyEdge<TLabel>>!
Varve.Derivation.DependencyGraph<TLabel>.Stratify() -> Varve.Derivation.Stratification<TLabel>!
Varve.Derivation.DependencyGraph<TLabel>.VertexCount.get -> int
Varve.Derivation.IPathGraph<TNode, TLabel>
Varve.Derivation.IPathGraph<TNode, TLabel>.Nodes.get -> System.Collections.Generic.IEqualityComparer<TNode>!
Varve.Derivation.IPathGraph<TNode, TLabel>.Objects(TNode subject, TLabel label, System.Collections.Generic.List<TNode>! into) -> void
Varve.Derivation.IPathGraph<TNode, TLabel>.Subjects(TLabel label, TNode obj, System.Collections.Generic.List<TNode>! into) -> void
Varve.Derivation.Model.DependencyEdge<TLabel>  (record struct: Source, Target, Polarity, Label; equality, Deconstruct)
Varve.Derivation.Model.Polarity  (Positive = 0, Negative = 1, Both = 2, Target = 3)
Varve.Derivation.Model.RefusedCycle<TLabel>.Edge.get -> Varve.Derivation.Model.DependencyEdge<TLabel>
Varve.Derivation.Model.RefusedCycle<TLabel>.Edges.get -> System.Collections.Generic.IReadOnlyList<Varve.Derivation.Model.DependencyEdge<TLabel>>!
Varve.Derivation.PathAutomaton<TLabel>.AcceptsEmpty.get -> bool
Varve.Derivation.PathAutomaton<TLabel>.Anchors<TNode>(IPathGraph<TNode, TLabel>! graph, TNode subject, TLabel label, TNode obj, List<TNode>! into, HashSet<TNode>! seen) -> void
Varve.Derivation.PathAutomaton<TLabel>.Evaluate<TNode>(IPathGraph<TNode, TLabel>! graph, TNode start, List<TNode>! into, HashSet<TNode>! seen) -> void
Varve.Derivation.PathAutomaton<TLabel>.IsBounded.get -> bool
Varve.Derivation.PathAutomaton<TLabel>.Labels.get -> System.Collections.Generic.IEnumerable<TLabel>!
Varve.Derivation.PathAutomaton<TLabel>.Reverse<TNode>(IPathGraph<TNode, TLabel>! graph, TNode end, List<TNode>! into, HashSet<TNode>! seen) -> void
Varve.Derivation.Propagation<TNode, TLabel>.Propagation(DependencyGraph<TLabel>! graph, Stratification<TLabel>! strata, IEqualityComparer<TNode>! nodes) -> void
Varve.Derivation.Propagation<TNode, TLabel>.Run(PropagationTransfer<TNode, TLabel>! transfer) -> void
Varve.Derivation.Propagation<TNode, TLabel>.Seed(int vertex, TNode node) -> bool
Varve.Derivation.Propagation<TNode, TLabel>.SetOf(int vertex) -> System.Collections.Generic.IReadOnlySet<TNode>!
virtual Varve.Derivation.PropagationTransfer<TNode, TLabel>.Invoke(DependencyEdge<TLabel> edge, TNode node, List<TNode>! into) -> void
Varve.Derivation.RegularPath<TLabel>.Compile() -> Varve.Derivation.PathAutomaton<TLabel>!
static Varve.Derivation.RegularPath.Edge / Inverse / Sequence / Alternative / ZeroOrMore / OneOrMore / ZeroOrOne<TLabel>(…) -> RegularPath<TLabel>!
Varve.Derivation.Stratification<TLabel>.ComponentCount / ComponentOf(int) / IsCyclic(int) / IsStratifiable / Members(int) / Refused

(abbreviated by the session for reading; the file has every line verbatim)

Varve.Shacl.Store (new; 26 lines):

Varve.Shacl.Store.Model.BindingCheck.Accepted / Classification / Problems / Refusal (.get)
Varve.Shacl.Store.Model.ShapesBinding.ShapesBinding(RdfTerm! binding, RdfTerm! shapesGraph, GraphScope! validates, ValidationMode mode) -> void
Varve.Shacl.Store.Model.ShapesBinding.Binding / ShapesGraph / Validates / Mode (.get); AllowFullEvaluation / SubClassOfInShapesGraph (.get, .init)
Varve.Shacl.Store.Model.ValidationMode  (Off = 0, Report = 1, Reject = 2)
Varve.Shacl.Store.ShapesGate.ShapesGate(IEnumerable<ShapesBinding!>! bindings, ShaclOptions? options = null) -> void
Varve.Shacl.Store.ShapesGate.Bindings.get -> IReadOnlyList<ShapesBinding!>!
Varve.Shacl.Store.ShapesGate.Check(ShapesBinding! binding, IQuadSource! state) -> BindingCheck!
Varve.Shacl.Store.ShapesGate.Validate(IQuadSource! proposed, QuadDelta delta) -> ValidationVerdict
Varve.Shacl.Store.ShapesGate.Validate(IQuadSource! proposed, BulkDelta! delta) -> ValidationVerdict

(abbreviated the same way)

Varve.Shacl (+43):

+Varve.Shacl.FootprintContext
+Varve.Shacl.FootprintContext.Add(Varve.Rdf.TermHandle focus) -> void
+Varve.Shacl.FootprintContext.AddValueNode(Varve.Rdf.TermHandle value) -> void
+Varve.Shacl.FootprintContext.Data.get -> Varve.Rdf.IQuadSource!
+Varve.Shacl.FootprintContext.DataGraph.get -> Varve.Rdf.GraphPattern
+Varve.Shacl.FootprintContext.Object.get -> Varve.Rdf.TermHandle
+Varve.Shacl.FootprintContext.Predicate.get -> Varve.Rdf.TermHandle
+Varve.Shacl.FootprintContext.Subject.get -> Varve.Rdf.TermHandle
+Varve.Shacl.IShaclFootprint
+Varve.Shacl.IShaclFootprint.Anchors(Varve.Shacl.FootprintContext! context) -> void
+Varve.Shacl.IShaclFootprint.Construct.get -> string!
+Varve.Shacl.IShaclFootprint.Maintainability.get -> Varve.Shacl.Model.Maintainability
+Varve.Shacl.IncrementalOutcome
+Varve.Shacl.IncrementalOutcome.Comparison.get -> Varve.Shacl.Model.IncrementalComparison?
+Varve.Shacl.IncrementalOutcome.Problems.get -> System.Collections.Generic.IReadOnlyList<Varve.Shacl.Model.ShaclProblem!>!
+Varve.Shacl.IncrementalOutcome.Succeeded.get -> bool
+Varve.Shacl.IncrementalValidation
+Varve.Shacl.Model.ClassificationReason
+Varve.Shacl.Model.ClassificationReason.Construct.get -> string!
+Varve.Shacl.Model.ClassificationReason.Maintainability.get -> Varve.Shacl.Model.Maintainability
+Varve.Shacl.Model.ClassificationReason.Section.get -> string!
+Varve.Shacl.Model.IncrementalComparison
+Varve.Shacl.Model.IncrementalComparison.AffectedFocusNodes.get -> System.Collections.Generic.IReadOnlyList<Varve.Rdf.RdfTerm!>!
+Varve.Shacl.Model.IncrementalComparison.AffectedPairs.get -> int
+Varve.Shacl.Model.IncrementalComparison.After.get -> Varve.Shacl.Model.ValidationReport!
+Varve.Shacl.Model.IncrementalComparison.Before.get -> Varve.Shacl.Model.ValidationReport!
+Varve.Shacl.Model.IncrementalComparison.Introduced.get -> System.Collections.Generic.IReadOnlyList<Varve.Shacl.Model.ShaclResult!>!
+Varve.Shacl.Model.Maintainability
+Varve.Shacl.Model.Maintainability.Full = 2 -> Varve.Shacl.Model.Maintainability
+Varve.Shacl.Model.Maintainability.Incremental = 0 -> Varve.Shacl.Model.Maintainability
+Varve.Shacl.Model.Maintainability.Scoped = 1 -> Varve.Shacl.Model.Maintainability
+Varve.Shacl.Model.ShapeClassification
+Varve.Shacl.Model.ShapeClassification.Maintainability.get -> Varve.Shacl.Model.Maintainability
+Varve.Shacl.Model.ShapeClassification.Reasons.get -> System.Collections.Generic.IReadOnlyList<Varve.Shacl.Model.ClassificationReason!>!
+Varve.Shacl.Model.ShapeClassification.Shape.get -> Varve.Rdf.RdfTerm!
+Varve.Shacl.Model.ShapesClassification
+Varve.Shacl.Model.ShapesClassification.Shapes.get -> System.Collections.Generic.IReadOnlyList<Varve.Shacl.Model.ShapeClassification!>!
+Varve.Shacl.Model.ShapesClassification.Worst.get -> Varve.Shacl.Model.Maintainability
+Varve.Shacl.ShaclOptions.RefuseNonDeterministicFunctions.get -> bool
+Varve.Shacl.ShaclOptions.RefuseNonDeterministicFunctions.init -> void
+Varve.Shacl.ShapesGraph.Classify() -> Varve.Shacl.Model.ShapesClassification!
+Varve.Shacl.ShapesGraphContext.RefuseNonDeterministicFunctions.get -> bool
+static Varve.Shacl.IncrementalValidation.Compare(Varve.Shacl.ShapesGraph! shapes, Varve.Rdf.IQuadSource! after, Varve.Rdf.GraphPattern dataGraph, Varve.Rdf.QuadDelta delta, Varve.Shacl.ShaclOptions? options = null, System.Threading.CancellationToken cancellationToken = default(System.Threading.CancellationToken)) -> Varve.Shacl.IncrementalOutcome!

Varve.Store (+2):

+Varve.Store.Log.ValidationVerdict.Attachments.get -> System.Collections.Generic.IReadOnlyList<Varve.Rdf.RdfTerm!>!
+static Varve.Store.Log.ValidationVerdict.Accept(System.Collections.Generic.IReadOnlyList<Varve.Rdf.RdfTerm!>! attachments) -> Varve.Store.Log.ValidationVerdict

Varve.Management (+12):

+Varve.Management.Model.BindingDeclaration.AllowFullEvaluation.get -> bool
+Varve.Management.Model.BindingDeclaration.AllowFullEvaluation.init -> void
+Varve.Management.Model.BindingDeclaration.Mode.get -> Varve.Management.Model.ShapesMode
+Varve.Management.Model.BindingDeclaration.Mode.init -> void
+Varve.Management.Model.BindingDeclaration.SubClassOfInShapesGraph.get -> bool
+Varve.Management.Model.BindingDeclaration.SubClassOfInShapesGraph.init -> void
+Varve.Management.Model.BindingDeclaration.Validates.get -> Varve.Management.Model.GraphSet?
+Varve.Management.Model.BindingDeclaration.Validates.init -> void
+Varve.Management.Model.ShapesMode
+Varve.Management.Model.ShapesMode.Off = 0 -> Varve.Management.Model.ShapesMode
+Varve.Management.Model.ShapesMode.Reject = 2 -> Varve.Management.Model.ShapesMode
+Varve.Management.Model.ShapesMode.Report = 1 -> Varve.Management.Model.ShapesMode

Varve.Protocol (+23):

+static Varve.Protocol.Model.ProblemType.BindingNotIncremental.get -> Varve.Protocol.Model.ProblemType
+static Varve.Protocol.Model.ProblemType.ShapesInvalid.get -> Varve.Protocol.Model.ProblemType
+static Varve.Protocol.Model.ProblemType.ShapesViolated.get -> Varve.Protocol.Model.ProblemType
+Varve.Protocol.IManagementPlane.CheckBindings(Varve.Management.Model.Declarations! proposed) -> System.Collections.Generic.IReadOnlyList<Varve.Rdf.RdfTerm!>?
+Varve.Protocol.IManagementPlane.DescribeBinding(Varve.Protocol.Model.DatasetName dataset, Varve.Rdf.RdfTerm! binding) -> Varve.Protocol.Model.BindingDescription?
+Varve.Protocol.Model.BindingDescription  (record: Declaration, Position, Classification, Problems, with the record members)

---

Checklist

• Issue referenced: every commit carries Refs #138 (issue-refs passes); the descriptor commit carries the closing trailer
• ADR cited: ADRs 0169–0173, with dated amendments to 0003, 0017, 0058, 0119, 0150, 0152, 0154, 0155 and 0166
• Specification updated, or n/a: n/a. The log format is unchanged, and the attachment is a validator attachment (ADR 0017); ADR 0173 defines its term
• API baseline updated: PublicAPI.Unshipped.txt only, as above
• Conformance baseline updated, or n/a: n/a, unchanged
• Tests: real store, real server and real HTTP; oracles are full validation
• A new gate's failure path was proven by running it: the three sabotages above each failed the incremental property
• Signed and signed off: every commit is SSH-signed and carries Signed-off-by:
• A traceability record exists
• dotnet run eng/ci.cs passes locally: everything but release (preview.10's missing tag, above) and CS0618

🤖 Generated with Claude Code

https://claude.ai/code/session_01RFtkE4cRmAMRARHDu5Y9ZA

[Hafeok]

Draft for the maintainer to file with REC upstream (RealEstateCore, rec repository). It is not in this repository.

---

Title: REC 4.1.0's SHACL release: Brick's SPARQL constraints use prefixes the shapes graph never declares, and implicit class targets need owl:Class rdfs:subClassOf rdfs:Class

Files: the SHACL release files of v4.1.0:

File	SHA-256
rec.ttl	3cdda54a…ce74b
Brick+patches.ttl	7bbe0d34…9158b
recimports.ttl	58cebae4…22983

1. Undeclared SPARQL prefixes. SHACL-SPARQL (SHACL 1.0 §5.2.1; SHACL 1.2 SPARQL, "Prefix declarations") resolves a query's prefixes only from the sh:declare triples on the nodes that sh:prefixes names. A processor must not fall back on the file's Turtle @prefix lines.

Three sh:SPARQLConstraints in Brick+patches.ttl name namespace nodes with sh:prefixes. None of the three files has a single sh:declare. So a conforming processor cannot parse any of the three queries, and it refuses the whole shapes graph.

Two of the queries also use a prefix they don't even list:

Shape	sh:prefixes	Prefixes the query uses	Used but not listed
bsh:DeprecationRuleForInstances	rdf:, brick:	rdf:, owl:	owl: ($this rdf:type/owl:deprecated true)
bsh:TimeseriesReferenceOnPointsConstraint	rdf:, rdfs:, ref:	ref:, rdf:, rdfs:, brick:	brick: (brick:Point)
bsh:VirtualMeterRule	rdf:, rdfs:, brick:	brick:, rdf:, rdfs:	none

Suggested fix. Two parts:

1. Declare each prefix on the node that sh:prefixes already names. For example:

   rdf:   sh:declare [ sh:prefix "rdf" ;   sh:namespace "http://www.w3.org/1999/02/22-rdf-syntax-ns#"^^xsd:anyURI ] .
   rdfs:  sh:declare [ sh:prefix "rdfs" ;  sh:namespace "http://www.w3.org/2000/01/rdf-schema#"^^xsd:anyURI ] .
   owl:   sh:declare [ sh:prefix "owl" ;   sh:namespace "http://www.w3.org/2002/07/owl#"^^xsd:anyURI ] .
   brick: sh:declare [ sh:prefix "brick" ; sh:namespace "https://brickschema.org/schema/Brick#"^^xsd:anyURI ] .
   ref:   sh:declare [ sh:prefix "ref" ;   sh:namespace "https://brickschema.org/schema/Brick/ref#"^^xsd:anyURI ] .

2. Add the missing sh:prefixes: owl: on bsh:DeprecationRuleForInstances, and brick: on bsh:TimeseriesReferenceOnPointsConstraint.

Brick's own build declares these prefixes; the merged Brick+patches.ttl appears to have lost them.

2. Implicit class targets. About 1,445 Brick shapes are typed both owl:Class and sh:NodeShape. SHACL's implicit class target (§2.1.3.3) applies only to a shape that is an rdfs:Class, that is, a SHACL instance of rdfs:Class through rdf:type/rdfs:subClassOf*. None of the files states owl:Class rdfs:subClassOf rdfs:Class, so a processor that doesn't import OWL gives these shapes no targets. Adding that one triple, which is OWL's own vocabulary, to the release would make the targets apply wherever the files are used.

How this was found. Varve's SHACL 1.2 processor loaded the three files as one shapes graph, an RDF merge with each file's blank nodes kept apart. It reported three failures:

ExtensionFailure: UndeclaredPrefix: The prefix 'rdf:' is not declared.
ExtensionFailure: UndeclaredPrefix: The prefix 'ref:' is not declared.
ExtensionFailure: UndeclaredPrefix: The prefix 'brick:' is not declared.

Without those three constraints, and with the owl:Class triple added, the rest of the shapes load and validate.

---

What happens in Varve meanwhile:

• Part 2b's REC gate adds both fixes, the sh:declare triples above and the owl:Class triple, in its shapes composition. ADR 0176 will record the exact triples.
• RecStandaloneTests.BricksSparqlConstraintsDeclareNoPrefixes keeps pinning that REC 4.1.0 as vendored is refused.

---

Generated by Claude Code

[Hafeok]

CI on the current head: every job that builds the solution fails at the build step, and only on CS0618.

• The cause is the unaccepted decisions of ADRs 0169–0173: ADR 0066's design is that a pull request is red only on those warnings.
• I checked two of the logs, build (ubuntu-latest) (64 × CS0618) and browser storage in Chromium (60 × CS0618). Neither has any other error code.
• The other red jobs build the same solution: conformance, auth, samples, storage/fleet, container, native aot and decision report.
• The gates that don't build are green: dco, release pending, rec generator on both architectures, durability and bicep.
• agent review is waiting on your approval of the head.

Nothing else needs fixing in this pull request. The test jobs, and with them the Windows and macOS durations of the new test projects, run once the decisions are accepted. I'll report those durations here then.

---

Generated by Claude Code

[Hafeok]

v0.1.0-preview.10 is tagged, so I merged main into this branch (1db38aa; the only incoming change is release.yml's id-token: write). #135 hasn't landed, so the descriptor stays v0.1.0-preview.11.

Locally, all six of these pass: release, changelog, status, release-pending, issue-refs and dco. The item in "What is left" about the release gate is resolved.

What remains red is the build's CS0618, until ADRs 0169–0173 are accepted, and agent review on the new head 1db38aa.

---

Generated by Claude Code

[Hafeok]

Renumbered to v0.1.0-preview.12. #135 (Extent maintenance) landed as preview.11, and I merged main in as cb46087. Where the description says preview.11 for this release, read preview.12.

What the merge changed:

• The descriptor is now releases/v0.1.0-preview.12.yaml. Its basis is unchanged: SHACL, part 2a: commit-time gating — the incremental engine, the classification, shapes bindings #138 and ADRs 0169–0173.
• The changelog is re-projected.
• README Status names preview.12, then preview.11 and preview.10, and the image tags are at preview.12.
• AGENTS.md's State has part 2a, then Extent maintenance.
• The roadmap row and the part 2a section say preview.12.
• The benchmarks README keeps Milestone Extent maintenance — background merges and extent deletion #110's section before this part's. It also notes that Milestone Extent maintenance — background merges and extent deletion #110's REC runs were taken on generator output version 1, before this PR's change.

Gates: locally, decision-sets, changelog, status, issue-refs and dco pass. release fails the same way it did before preview.10 was tagged: v0.1.0-preview.11 isn't tagged yet, so with two proposed descriptors the checker swaps the ADRs between them. It should clear once preview.11 is cut, and I'll re-run it then.

What remains red is CS0618, until ADRs 0169–0173 are accepted, and agent review on the new head cb46087.

---

Generated by Claude Code

[Hafeok]

approve cb46087

[Hafeok]

v0.1.0-preview.11 is tagged. On cb46087, which already contains main, the release gates now pass locally: release, changelog, status and release-pending.

The merged code's tests pass too: Protocol 113, Shacl.Store 13, Server 106 and Store 277, with none failing.

The only thing keeping CI red is CS0618, until ADRs 0169–0173 are accepted.

---

Generated by Claude Code

[Hafeok]

approve 0bb0d34

[Hafeok]

CI after the acceptance (0bb0d34): every test project built and passed on Ubuntu and Windows. The only failure was in dotnet pack: Varve.Shacl.Store had no README.md (NU5019). I've added one in fd94e29, pack passes locally, and it's pushed with the traceability update as 3dbffa7.

Durations of the new test projects, from that run (#126). CI runs these projects on Ubuntu and Windows only; the macOS job runs the durability suite alone, so there's no macOS figure.

Project	ubuntu-latest	windows-latest
Varve.Derivation.Tests (new)	0.97 s	1.23 s
Varve.Shacl.Store.Tests (new; the two long REC tests skipped)	10.9 s	12.1 s
Varve.Shacl.Tests	4.6 s	4.9 s
Varve.Shacl.Sparql.Tests	5.2 s	6.4 s

The new head 3dbffa7 needs your approve comment for agent review.

---

Generated by Claude Code

[Hafeok]

approve 3dbffa7

[Hafeok]

pipeline (devcontainer) failed in conformance, and the failure is not this PR's.

The W3C ratchet itself is clean: 6,141 in the baseline, 40 exempt, 0 regressed, 0 newly passing. What failed is a single property, CanonPropertyTests.Without_blank_graph_names_isomorphism_holds_exactly_when_the_canonical_forms_are_equal, with CsCheck seed 1Axba7DbGjr2.

Why it isn't this PR's. The property exercises only Varve.Rdf's RdfCanonicaliser and Varve.Turtle, and this PR changes neither. With CsCheck_Seed=1Axba7DbGjr2 it fails the same way on main at 0b02803. CsCheck draws a new seed on every run, so it shows up only occasionally; both conformance jobs passed in this same run.

What it is. A real counterexample to the claim in docs/spec/rdf-canon.md §6: that when every graph name is an IRI, isomorphic datasets get equal canonical forms.

• The two datasets are isomorphic. I checked the mapping by hand.
• Their canonical forms differ only in which of two leaf blank nodes gets c14n3 and which gets c14n4.
• The two leaves are distinguishable only by which graph their two edges sit in. Each has one edge in <http://example.org/é> and one in the default graph, to the same two already-canonical nodes, c14n1 and c14n2, the other way round.
• RDFC-1.0 can't see that. Hash Related Blank Node (§4.7.3) hashes the position, the predicate and the related node's identifier, but not the graph of the quad that relates them. So the two leaves get equal N-degree hashes, and the input order breaks the tie.

As far as I can tell, RdfCanonicaliser implements the spec as written. The documented "holds with IRI graph names" is too strong, so the property is testing a claim RDFC-1.0 doesn't make.

No fix exists yet. It belongs in its own change, not this PR:

• correct §6 of rdf-canon.md;
• narrow the property, either by keeping the generator from producing a blank node whose relations differ only by the graph of the relating quad, or by counting such pairs and reporting them as the blank-graph-name direction already does;
• add this pair as a pinned example.

I'll re-run the failed job once, after build (windows-latest) finishes.

---

Generated by Claude Code

[Hafeok]

build (windows-latest) passed too, so on 3dbffa7 every check is green except pipeline (devcontainer). That one hit the RDFC property defect described above, which also reproduces on main.

I tried to re-run the failed job, but the session's token isn't allowed to (403 Resource not accessible by integration). It needs a re-run from the Actions page, run 39662117839, "Re-run failed jobs". A new seed will almost certainly pass.

---

Generated by Claude Code