@acidicsoil/portable-capabilities 0.1.0

package/canonical/roles/convention-mining.yaml

@@ -0,0 +1,81 @@
+schemaVersion: '1.0'
+id: convention-mining
+purpose: Infer recurring conventions from observed artifacts.
+family:
+  id: exemplar-modeling
+  version: 1.0.0
+taxonomy:
+  id: convention-mining
+  displayName: Convention Mining
+  transformation: Infer recurring conventions from observed artifacts.
+  intendedOutput: Convention registry
+inputs:
+- name: observed-artifacts
+  description: Observed artifacts that may encode recurring conventions.
+  required: true
+- name: authority-sources
+  description: Explicit style guides or contracts that outrank observation.
+  required: false
+- name: sampling-context
+  description: Repository, team, or domain boundaries for the sample.
+  required: true
+precedence:
+- sourceClass: explicit-contract
+  rank: 0
+- sourceClass: observed-artifact
+  rank: 1
+- sourceClass: inference
+  rank: 2
+pipeline:
+- id: ingest.sources
+  version: 1.0.0
+- id: trace.build-evidence-ledger
+  version: 1.0.0
+- id: validation.contract
+  version: 1.0.0
+output:
+  schema: canonical/schemas/convention-mining.output.schema.json
+  description: Convention registry
+requiredBehavior:
+- 'activation: Use convention-mining only when the requested outcome requires this transformation: Infer recurring
+  conventions from observed artifacts.'
+- 'sufficiency: Require all role inputs marked required and enough evidence to complete these gates: Every convention
+  reports scope, support, counterexamples, authority status, and confidence sufficient for its classification.'
+- 'conflict: Apply declared source precedence; when equally authoritative evidence changes the decision, preserve
+  both positions and return the affected output field as unresolved.'
+- 'ambiguity: Isolate ambiguous terms or boundaries, show which stage and output fields they affect, and request
+  clarification rather than selecting an unsupported interpretation.'
+- 'abstention: Abstain from convention-mining when required inputs are absent, the decision rule cannot be applied,
+  or evidence cannot satisfy the completion gate.'
+- 'stop: Stop before emitting Convention registry unless Every convention reports scope, support, counterexamples,
+  authority status, and confidence sufficient for its classification.'
+- 'resources: Load representative artifacts, explicit guides, and sampling metadata; expand the sample when confidence
+  is below the declared threshold.'
+- 'transformation: Execute the ordered analytical procedure: 1) sample artifacts across the declared context; 2)
+  count recurring choices and cluster equivalent forms; 3) compare observed choices with explicit authority; 4)
+  classify conventions as mandated, dominant, contested, or local.'
+- 'output: Populate and validate canonical/schemas/convention-mining.output.schema.json with required fields conventions,
+  classification, supportCounts, counterexamples, scope; do not replace missing fields with prose.'
+- 'traceability: Link every material output field to source identifiers and the analytical stage or decision that
+  produced it; retain counterevidence for unresolved results.'
+- 'evaluation: Pass only when the output schema validates, the ordered stages are evidenced, the role decision rule
+  was applied, and this completion gate holds: Every convention reports scope, support, counterexamples, authority
+  status, and confidence sufficient for its classification.'
+forbiddenBehavior:
+- Do not substitute generic analytical behavior for the role-specific convention-mining transformation.
+- Do not skip, reorder, or silently collapse the role's declared analytical stages.
+- Do not invent missing evidence, resolve authority conflicts by convenience, or emit a completed result after a
+  stop or abstention condition.
+- Do not copy operation-owned ingestion, ledger, or schema-validation mechanics into the role contract.
+traceability:
+  materialClaimsRequireSources: true
+  ledgerFormat: jsonl
+  artifacts: content-addressed
+evaluation:
+  criteria:
+  - The authoritative transformation and intended output for convention-mining are preserved.
+  - 'Required inputs and the decision rule are satisfied: Treat explicit authority as mandated; infer a convention
+    only when recurrence is broad enough for the declared sampling context and counterexamples are disclosed.'
+  - 'The ordered analytical stages are evidenced and the output completion gate passes: Every convention reports
+    scope, support, counterexamples, authority status, and confidence sufficient for its classification.'
+  - Conflict, ambiguity, insufficient-evidence, abstention, and stop behavior remain explicit and testable.

package/canonical/roles/conversation-to-specification.yaml

@@ -0,0 +1,83 @@
+schemaVersion: '1.0'
+id: conversation-to-specification
+purpose: Convert conversational intent into an executable specification.
+family:
+  id: workflow-intent-abstraction
+  version: 1.0.0
+taxonomy:
+  id: conversation-to-specification
+  displayName: Conversation to Specification
+  transformation: Convert conversational intent into an executable specification.
+  intendedOutput: Traceable specification
+inputs:
+- name: conversation-record
+  description: The authoritative conversation or decision record.
+  required: true
+- name: existing-constraints
+  description: Known system, policy, and delivery constraints.
+  required: false
+- name: stakeholder-definitions
+  description: Stable names and responsibilities for referenced actors.
+  required: false
+precedence:
+- sourceClass: explicit-contract
+  rank: 0
+- sourceClass: observed-artifact
+  rank: 1
+- sourceClass: inference
+  rank: 2
+pipeline:
+- id: ingest.sources
+  version: 1.0.0
+- id: trace.build-evidence-ledger
+  version: 1.0.0
+- id: validation.contract
+  version: 1.0.0
+output:
+  schema: canonical/schemas/conversation-to-specification.output.schema.json
+  description: Traceable specification
+requiredBehavior:
+- 'activation: Use conversation-to-specification only when the requested outcome requires this transformation: Convert
+  conversational intent into an executable specification.'
+- 'sufficiency: Require all role inputs marked required and enough evidence to complete these gates: Every requirement
+  is testable and traced, contradictions are explicit, and unresolved intent is separated from committed specification.'
+- 'conflict: Apply declared source precedence; when equally authoritative evidence changes the decision, preserve
+  both positions and return the affected output field as unresolved.'
+- 'ambiguity: Isolate ambiguous terms or boundaries, show which stage and output fields they affect, and request
+  clarification rather than selecting an unsupported interpretation.'
+- 'abstention: Abstain from conversation-to-specification when required inputs are absent, the decision rule cannot
+  be applied, or evidence cannot satisfy the completion gate.'
+- 'stop: Stop before emitting Traceable specification unless Every requirement is testable and traced, contradictions
+  are explicit, and unresolved intent is separated from committed specification.'
+- 'resources: Load the full conversation record, linked decisions, and existing constraints; retrieve referenced
+  artifacts only when needed to resolve a clause.'
+- 'transformation: Execute the ordered analytical procedure: 1) extract explicit requests decisions assumptions
+  and unresolved questions; 2) normalize actors terms constraints and desired outcomes; 3) convert statements into
+  testable requirements and acceptance conditions; 4) trace each specification clause back to the conversation and
+  flag gaps.'
+- 'output: Populate and validate canonical/schemas/conversation-to-specification.output.schema.json with required
+  fields requirements, acceptanceCriteria, assumptions, openQuestions, traceLinks; do not replace missing fields
+  with prose.'
+- 'traceability: Link every material output field to source identifiers and the analytical stage or decision that
+  produced it; retain counterevidence for unresolved results.'
+- 'evaluation: Pass only when the output schema validates, the ordered stages are evidenced, the role decision rule
+  was applied, and this completion gate holds: Every requirement is testable and traced, contradictions are explicit,
+  and unresolved intent is separated from committed specification.'
+forbiddenBehavior:
+- Do not substitute generic analytical behavior for the role-specific conversation-to-specification transformation.
+- Do not skip, reorder, or silently collapse the role's declared analytical stages.
+- Do not invent missing evidence, resolve authority conflicts by convenience, or emit a completed result after a
+  stop or abstention condition.
+- Do not copy operation-owned ingestion, ledger, or schema-validation mechanics into the role contract.
+traceability:
+  materialClaimsRequireSources: true
+  ledgerFormat: jsonl
+  artifacts: content-addressed
+evaluation:
+  criteria:
+  - The authoritative transformation and intended output for conversation-to-specification are preserved.
+  - 'Required inputs and the decision rule are satisfied: Promote only explicit or strongly entailed intent into
+    requirements; represent unresolved or contradictory intent as open questions, never silent assumptions.'
+  - 'The ordered analytical stages are evidenced and the output completion gate passes: Every requirement is testable
+    and traced, contradictions are explicit, and unresolved intent is separated from committed specification.'
+  - Conflict, ambiguity, insufficient-evidence, abstention, and stop behavior remain explicit and testable.

package/canonical/roles/delta-analysis.yaml

@@ -0,0 +1,84 @@
+schemaVersion: '1.0'
+id: delta-analysis
+purpose: Characterize material differences between artifacts.
+family:
+  id: comparative-traceability
+  version: 1.0.0
+taxonomy:
+  id: delta-analysis
+  displayName: Delta Analysis
+  transformation: Characterize material differences between artifacts.
+  intendedOutput: Structured delta report
+inputs:
+- name: baseline-artifact
+  description: The comparison baseline.
+  required: true
+- name: candidate-artifact
+  description: The artifact being compared with the baseline.
+  required: true
+- name: comparison-scope
+  description: Included and excluded dimensions of comparison.
+  required: true
+- name: materiality-rules
+  description: Rules for classifying consequential differences.
+  required: false
+precedence:
+- sourceClass: explicit-contract
+  rank: 0
+- sourceClass: observed-artifact
+  rank: 1
+- sourceClass: inference
+  rank: 2
+pipeline:
+- id: ingest.sources
+  version: 1.0.0
+- id: trace.build-evidence-ledger
+  version: 1.0.0
+- id: validation.contract
+  version: 1.0.0
+output:
+  schema: canonical/schemas/delta-analysis.output.schema.json
+  description: Structured delta report
+requiredBehavior:
+- 'activation: Use delta-analysis only when the requested outcome requires this transformation: Characterize material
+  differences between artifacts.'
+- 'sufficiency: Require all role inputs marked required and enough evidence to complete these gates: Every in-scope
+  unit is accounted for as changed or unchanged, and every material delta has impact and evidence.'
+- 'conflict: Apply declared source precedence; when equally authoritative evidence changes the decision, preserve
+  both positions and return the affected output field as unresolved.'
+- 'ambiguity: Isolate ambiguous terms or boundaries, show which stage and output fields they affect, and request
+  clarification rather than selecting an unsupported interpretation.'
+- 'abstention: Abstain from delta-analysis when required inputs are absent, the decision rule cannot be applied,
+  or evidence cannot satisfy the completion gate.'
+- 'stop: Stop before emitting Structured delta report unless Every in-scope unit is accounted for as changed or
+  unchanged, and every material delta has impact and evidence.'
+- 'resources: Load both artifacts, comparison scope, materiality rules, and relevant schemas or tests for ambiguous
+  semantic differences.'
+- 'transformation: Execute the ordered analytical procedure: 1) normalize baseline and candidate into comparable
+  units; 2) detect additions removals and modifications; 3) classify semantic impact and materiality; 4) trace each
+  material delta to evidence and affected obligations.'
+- 'output: Populate and validate canonical/schemas/delta-analysis.output.schema.json with required fields deltas,
+  materiality, impactAnalysis, unchangedAreas, evidence; do not replace missing fields with prose.'
+- 'traceability: Link every material output field to source identifiers and the analytical stage or decision that
+  produced it; retain counterevidence for unresolved results.'
+- 'evaluation: Pass only when the output schema validates, the ordered stages are evidenced, the role decision rule
+  was applied, and this completion gate holds: Every in-scope unit is accounted for as changed or unchanged, and
+  every material delta has impact and evidence.'
+forbiddenBehavior:
+- Do not substitute generic analytical behavior for the role-specific delta-analysis transformation.
+- Do not skip, reorder, or silently collapse the role's declared analytical stages.
+- Do not invent missing evidence, resolve authority conflicts by convenience, or emit a completed result after a
+  stop or abstention condition.
+- Do not copy operation-owned ingestion, ledger, or schema-validation mechanics into the role contract.
+traceability:
+  materialClaimsRequireSources: true
+  ledgerFormat: jsonl
+  artifacts: content-addressed
+evaluation:
+  criteria:
+  - The authoritative transformation and intended output for delta-analysis are preserved.
+  - 'Required inputs and the decision rule are satisfied: Classify a delta as material only when it changes declared
+    behavior, obligations, interfaces, risk, or acceptance outcomes; formatting-only changes remain non-material.'
+  - 'The ordered analytical stages are evidenced and the output completion gate passes: Every in-scope unit is accounted
+    for as changed or unchanged, and every material delta has impact and evidence.'
+  - Conflict, ambiguity, insufficient-evidence, abstention, and stop behavior remain explicit and testable.

package/canonical/roles/domain-context-boundary.yaml

@@ -0,0 +1,82 @@
+schemaVersion: '1.0'
+id: domain-context-boundary
+purpose: Define the minimal sufficient domain context and exclusions.
+family:
+  id: context-template-generation
+  version: 1.0.0
+taxonomy:
+  id: domain-context-boundary
+  displayName: Domain Context Boundary
+  transformation: Define the minimal sufficient domain context and exclusions.
+  intendedOutput: Context boundary contract
+inputs:
+- name: task-contract
+  description: The task or capability requiring domain context.
+  required: true
+- name: candidate-context
+  description: Available domain facts, rules, examples, and exclusions.
+  required: true
+- name: risk-profile
+  description: Consequences of missing or excessive context.
+  required: false
+precedence:
+- sourceClass: explicit-contract
+  rank: 0
+- sourceClass: observed-artifact
+  rank: 1
+- sourceClass: inference
+  rank: 2
+pipeline:
+- id: ingest.sources
+  version: 1.0.0
+- id: trace.build-evidence-ledger
+  version: 1.0.0
+- id: validation.contract
+  version: 1.0.0
+output:
+  schema: canonical/schemas/domain-context-boundary.output.schema.json
+  description: Context boundary contract
+requiredBehavior:
+- 'activation: Use domain-context-boundary only when the requested outcome requires this transformation: Define
+  the minimal sufficient domain context and exclusions.'
+- 'sufficiency: Require all role inputs marked required and enough evidence to complete these gates: Every included
+  item supports a named decision, every exclusion has a rationale, and known context gaps have escalation behavior.'
+- 'conflict: Apply declared source precedence; when equally authoritative evidence changes the decision, preserve
+  both positions and return the affected output field as unresolved.'
+- 'ambiguity: Isolate ambiguous terms or boundaries, show which stage and output fields they affect, and request
+  clarification rather than selecting an unsupported interpretation.'
+- 'abstention: Abstain from domain-context-boundary when required inputs are absent, the decision rule cannot be
+  applied, or evidence cannot satisfy the completion gate.'
+- 'stop: Stop before emitting Context boundary contract unless Every included item supports a named decision, every
+  exclusion has a rationale, and known context gaps have escalation behavior.'
+- 'resources: Load the task contract, candidate context inventory, and risk profile; retrieve additional context
+  only for an identified decision gap.'
+- 'transformation: Execute the ordered analytical procedure: 1) enumerate decisions the task must make; 2) map each
+  decision to necessary domain knowledge; 3) remove context that does not affect an allowed decision; 4) state inclusions
+  exclusions freshness and escalation triggers.'
+- 'output: Populate and validate canonical/schemas/domain-context-boundary.output.schema.json with required fields
+  includedContext, excludedContext, decisionCoverage, freshnessRules, escalationTriggers; do not replace missing
+  fields with prose.'
+- 'traceability: Link every material output field to source identifiers and the analytical stage or decision that
+  produced it; retain counterevidence for unresolved results.'
+- 'evaluation: Pass only when the output schema validates, the ordered stages are evidenced, the role decision rule
+  was applied, and this completion gate holds: Every included item supports a named decision, every exclusion has
+  a rationale, and known context gaps have escalation behavior.'
+forbiddenBehavior:
+- Do not substitute generic analytical behavior for the role-specific domain-context-boundary transformation.
+- Do not skip, reorder, or silently collapse the role's declared analytical stages.
+- Do not invent missing evidence, resolve authority conflicts by convenience, or emit a completed result after a
+  stop or abstention condition.
+- Do not copy operation-owned ingestion, ledger, or schema-validation mechanics into the role contract.
+traceability:
+  materialClaimsRequireSources: true
+  ledgerFormat: jsonl
+  artifacts: content-addressed
+evaluation:
+  criteria:
+  - The authoritative transformation and intended output for domain-context-boundary are preserved.
+  - 'Required inputs and the decision rule are satisfied: Include context only when omitting it could change a permitted
+    decision or required output; exclude unrelated sensitive or distracting material.'
+  - 'The ordered analytical stages are evidenced and the output completion gate passes: Every included item supports
+    a named decision, every exclusion has a rationale, and known context gaps have escalation behavior.'
+  - Conflict, ambiguity, insufficient-evidence, abstention, and stop behavior remain explicit and testable.

package/canonical/roles/evaluation-harness-design.yaml

@@ -0,0 +1,86 @@
+schemaVersion: '1.0'
+id: evaluation-harness-design
+purpose: Design executable checks for a capability contract.
+family:
+  id: guardrails-evaluation
+  version: 1.0.0
+taxonomy:
+  id: evaluation-harness-design
+  displayName: Evaluation Harness Design
+  transformation: Design executable checks for a capability contract.
+  intendedOutput: Evaluation harness specification
+inputs:
+- name: capability-contract
+  description: The capability or behavior contract to evaluate.
+  required: true
+- name: test-fixtures
+  description: Representative positive negative and edge-case fixtures.
+  required: true
+- name: execution-interface
+  description: The callable interface and observable outputs.
+  required: true
+- name: threshold-policy
+  description: Scoring and hard-failure thresholds.
+  required: false
+precedence:
+- sourceClass: explicit-contract
+  rank: 0
+- sourceClass: observed-artifact
+  rank: 1
+- sourceClass: inference
+  rank: 2
+pipeline:
+- id: ingest.sources
+  version: 1.0.0
+- id: trace.build-evidence-ledger
+  version: 1.0.0
+- id: validation.contract
+  version: 1.0.0
+output:
+  schema: canonical/schemas/evaluation-harness-design.output.schema.json
+  description: Evaluation harness specification
+requiredBehavior:
+- 'activation: Use evaluation-harness-design only when the requested outcome requires this transformation: Design
+  executable checks for a capability contract.'
+- 'sufficiency: Require all role inputs marked required and enough evidence to complete these gates: Every contract
+  obligation is covered by an executable assertion or justified judgment, and mutation tests prove the harness detects
+  violations.'
+- 'conflict: Apply declared source precedence; when equally authoritative evidence changes the decision, preserve
+  both positions and return the affected output field as unresolved.'
+- 'ambiguity: Isolate ambiguous terms or boundaries, show which stage and output fields they affect, and request
+  clarification rather than selecting an unsupported interpretation.'
+- 'abstention: Abstain from evaluation-harness-design when required inputs are absent, the decision rule cannot
+  be applied, or evidence cannot satisfy the completion gate.'
+- 'stop: Stop before emitting Evaluation harness specification unless Every contract obligation is covered by an
+  executable assertion or justified judgment, and mutation tests prove the harness detects violations.'
+- 'resources: Load the contract, callable interface, fixtures, result schema, and threshold policy; load mutation
+  cases to prove each assertion can fail.'
+- 'transformation: Execute the ordered analytical procedure: 1) derive observable assertions from the contract;
+  2) partition deterministic checks from graded judgments; 3) design positive negative mutation and boundary fixtures;
+  4) define execution isolation result schema thresholds and failure triage.'
+- 'output: Populate and validate canonical/schemas/evaluation-harness-design.output.schema.json with required fields
+  testPlan, fixtures, assertions, executionProtocol, thresholds, resultSchema; do not replace missing fields with
+  prose.'
+- 'traceability: Link every material output field to source identifiers and the analytical stage or decision that
+  produced it; retain counterevidence for unresolved results.'
+- 'evaluation: Pass only when the output schema validates, the ordered stages are evidenced, the role decision rule
+  was applied, and this completion gate holds: Every contract obligation is covered by an executable assertion or
+  justified judgment, and mutation tests prove the harness detects violations.'
+forbiddenBehavior:
+- Do not substitute generic analytical behavior for the role-specific evaluation-harness-design transformation.
+- Do not skip, reorder, or silently collapse the role's declared analytical stages.
+- Do not invent missing evidence, resolve authority conflicts by convenience, or emit a completed result after a
+  stop or abstention condition.
+- Do not copy operation-owned ingestion, ledger, or schema-validation mechanics into the role contract.
+traceability:
+  materialClaimsRequireSources: true
+  ledgerFormat: jsonl
+  artifacts: content-addressed
+evaluation:
+  criteria:
+  - The authoritative transformation and intended output for evaluation-harness-design are preserved.
+  - 'Required inputs and the decision rule are satisfied: Use deterministic assertions for contractually decidable
+    behavior; use graded judgment only where the contract permits semantic variance.'
+  - 'The ordered analytical stages are evidenced and the output completion gate passes: Every contract obligation
+    is covered by an executable assertion or justified judgment, and mutation tests prove the harness detects violations.'
+  - Conflict, ambiguity, insufficient-evidence, abstention, and stop behavior remain explicit and testable.

package/canonical/roles/failure-to-guardrail.yaml

@@ -0,0 +1,81 @@
+schemaVersion: '1.0'
+id: failure-to-guardrail
+purpose: Convert observed failures into preventive controls.
+family:
+  id: guardrails-evaluation
+  version: 1.0.0
+taxonomy:
+  id: failure-to-guardrail
+  displayName: Failure to Guardrail
+  transformation: Convert observed failures into preventive controls.
+  intendedOutput: Guardrail contract
+inputs:
+- name: failure-evidence
+  description: Observed failures, incidents, or rejected outcomes.
+  required: true
+- name: system-boundary
+  description: The controllable boundary where prevention may be applied.
+  required: true
+- name: risk-tolerance
+  description: Severity and residual-risk thresholds.
+  required: false
+precedence:
+- sourceClass: explicit-contract
+  rank: 0
+- sourceClass: observed-artifact
+  rank: 1
+- sourceClass: inference
+  rank: 2
+pipeline:
+- id: ingest.sources
+  version: 1.0.0
+- id: trace.build-evidence-ledger
+  version: 1.0.0
+- id: validation.contract
+  version: 1.0.0
+output:
+  schema: canonical/schemas/failure-to-guardrail.output.schema.json
+  description: Guardrail contract
+requiredBehavior:
+- 'activation: Use failure-to-guardrail only when the requested outcome requires this transformation: Convert observed
+  failures into preventive controls.'
+- 'sufficiency: Require all role inputs marked required and enough evidence to complete these gates: Each guardrail
+  maps to a failure cause, has an enforcement point and executable check, and states residual risk and bypass behavior.'
+- 'conflict: Apply declared source precedence; when equally authoritative evidence changes the decision, preserve
+  both positions and return the affected output field as unresolved.'
+- 'ambiguity: Isolate ambiguous terms or boundaries, show which stage and output fields they affect, and request
+  clarification rather than selecting an unsupported interpretation.'
+- 'abstention: Abstain from failure-to-guardrail when required inputs are absent, the decision rule cannot be applied,
+  or evidence cannot satisfy the completion gate.'
+- 'stop: Stop before emitting Guardrail contract unless Each guardrail maps to a failure cause, has an enforcement
+  point and executable check, and states residual risk and bypass behavior.'
+- 'resources: Load incident evidence, system boundaries, existing controls, and risk thresholds; load bypass or
+  adversarial examples before declaring coverage.'
+- 'transformation: Execute the ordered analytical procedure: 1) reconstruct the failure mechanism and preconditions;
+  2) identify controllable detection prevention and containment points; 3) select the least disruptive effective
+  control; 4) define enforcement evidence bypass handling and residual risk.'
+- 'output: Populate and validate canonical/schemas/failure-to-guardrail.output.schema.json with required fields
+  failureModel, guardrails, enforcementPoints, verification, residualRisk; do not replace missing fields with prose.'
+- 'traceability: Link every material output field to source identifiers and the analytical stage or decision that
+  produced it; retain counterevidence for unresolved results.'
+- 'evaluation: Pass only when the output schema validates, the ordered stages are evidenced, the role decision rule
+  was applied, and this completion gate holds: Each guardrail maps to a failure cause, has an enforcement point
+  and executable check, and states residual risk and bypass behavior.'
+forbiddenBehavior:
+- Do not substitute generic analytical behavior for the role-specific failure-to-guardrail transformation.
+- Do not skip, reorder, or silently collapse the role's declared analytical stages.
+- Do not invent missing evidence, resolve authority conflicts by convenience, or emit a completed result after a
+  stop or abstention condition.
+- Do not copy operation-owned ingestion, ledger, or schema-validation mechanics into the role contract.
+traceability:
+  materialClaimsRequireSources: true
+  ledgerFormat: jsonl
+  artifacts: content-addressed
+evaluation:
+  criteria:
+  - The authoritative transformation and intended output for failure-to-guardrail are preserved.
+  - 'Required inputs and the decision rule are satisfied: Choose a preventive control over detection when feasible
+    and proportionate; reject controls that do not intercept the demonstrated failure path.'
+  - 'The ordered analytical stages are evidenced and the output completion gate passes: Each guardrail maps to a
+    failure cause, has an enforcement point and executable check, and states residual risk and bypass behavior.'
+  - Conflict, ambiguity, insufficient-evidence, abstention, and stop behavior remain explicit and testable.

package/canonical/roles/multi-exemplar-consensus.yaml

@@ -0,0 +1,81 @@
+schemaVersion: '1.0'
+id: multi-exemplar-consensus
+purpose: Derive consensus and divergence across exemplars.
+family:
+  id: comparative-traceability
+  version: 1.0.0
+taxonomy:
+  id: multi-exemplar-consensus
+  displayName: Multi-Exemplar Consensus
+  transformation: Derive consensus and divergence across exemplars.
+  intendedOutput: Consensus model
+inputs:
+- name: exemplars
+  description: Multiple comparable exemplars.
+  required: true
+- name: comparison-dimensions
+  description: Dimensions on which consensus and divergence are judged.
+  required: true
+- name: weighting-rules
+  description: Optional authority or quality weights for exemplars.
+  required: false
+precedence:
+- sourceClass: explicit-contract
+  rank: 0
+- sourceClass: observed-artifact
+  rank: 1
+- sourceClass: inference
+  rank: 2
+pipeline:
+- id: ingest.sources
+  version: 1.0.0
+- id: trace.build-evidence-ledger
+  version: 1.0.0
+- id: validation.contract
+  version: 1.0.0
+output:
+  schema: canonical/schemas/multi-exemplar-consensus.output.schema.json
+  description: Consensus model
+requiredBehavior:
+- 'activation: Use multi-exemplar-consensus only when the requested outcome requires this transformation: Derive
+  consensus and divergence across exemplars.'
+- 'sufficiency: Require all role inputs marked required and enough evidence to complete these gates: Each consensus
+  statement has an agreement score and source matrix, and every material divergence remains visible.'
+- 'conflict: Apply declared source precedence; when equally authoritative evidence changes the decision, preserve
+  both positions and return the affected output field as unresolved.'
+- 'ambiguity: Isolate ambiguous terms or boundaries, show which stage and output fields they affect, and request
+  clarification rather than selecting an unsupported interpretation.'
+- 'abstention: Abstain from multi-exemplar-consensus when required inputs are absent, the decision rule cannot be
+  applied, or evidence cannot satisfy the completion gate.'
+- 'stop: Stop before emitting Consensus model unless Each consensus statement has an agreement score and source
+  matrix, and every material divergence remains visible.'
+- 'resources: Load all exemplars, comparison dimensions, and weighting rules; load provenance metadata when weighting
+  depends on authority or quality.'
+- 'transformation: Execute the ordered analytical procedure: 1) normalize exemplars against shared dimensions; 2)
+  measure agreement and divergence per dimension; 3) apply declared weights without hiding minority evidence; 4)
+  derive consensus rules and preserve unresolved variants.'
+- 'output: Populate and validate canonical/schemas/multi-exemplar-consensus.output.schema.json with required fields
+  consensus, divergences, minorityPositions, agreementScores, sourceMatrix; do not replace missing fields with prose.'
+- 'traceability: Link every material output field to source identifiers and the analytical stage or decision that
+  produced it; retain counterevidence for unresolved results.'
+- 'evaluation: Pass only when the output schema validates, the ordered stages are evidenced, the role decision rule
+  was applied, and this completion gate holds: Each consensus statement has an agreement score and source matrix,
+  and every material divergence remains visible.'
+forbiddenBehavior:
+- Do not substitute generic analytical behavior for the role-specific multi-exemplar-consensus transformation.
+- Do not skip, reorder, or silently collapse the role's declared analytical stages.
+- Do not invent missing evidence, resolve authority conflicts by convenience, or emit a completed result after a
+  stop or abstention condition.
+- Do not copy operation-owned ingestion, ledger, or schema-validation mechanics into the role contract.
+traceability:
+  materialClaimsRequireSources: true
+  ledgerFormat: jsonl
+  artifacts: content-addressed
+evaluation:
+  criteria:
+  - The authoritative transformation and intended output for multi-exemplar-consensus are preserved.
+  - 'Required inputs and the decision rule are satisfied: Declare consensus only when the configured agreement threshold
+    is met; otherwise report divergence and retain minority positions.'
+  - 'The ordered analytical stages are evidenced and the output completion gate passes: Each consensus statement
+    has an agreement score and source matrix, and every material divergence remains visible.'
+  - Conflict, ambiguity, insufficient-evidence, abstention, and stop behavior remain explicit and testable.