rubric

package
v0.9.0 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Jul 12, 2026 License: MIT Imports: 6 Imported by: 0

Documentation

Overview

Package rubric provides types for rubric-based evaluation reports with categorical scoring and severity-based findings. This is suited for LLM-as-Judge style reviews like PRD and ARB evaluations.

Index

Constants

View Source 
const (
	CategoryREQ    = "REQ"    // Requirements
	CategoryMETRIC = "METRIC" // Metrics and measurements
	CategoryUSER   = "USER"   // User personas and journeys
	CategoryARCH   = "ARCH"   // Architecture and technical design
	CategorySEC    = "SEC"    // Security
	CategorySCALE  = "SCALE"  // Scalability and performance
	CategoryINFRA  = "INFRA"  // Infrastructure and operations
	CategoryDOC    = "DOC"    // Documentation
	CategorySCOPE  = "SCOPE"  // Scope and constraints
	CategoryUX     = "UX"     // UX and accessibility
)

Reason code categories (prefixes).

View Source 
const ExtensionKeyCoverage = "coverage"

ExtensionKeyCoverage is the standard extension key for coverage data.

View Source 
const SchemaVersionV2 = "v2"

SchemaVersionV2 is the current schema version.

Variables

View Source 
var LegacyCodeMapping = map[ReasonCode]ReasonCode{
	"AMBIGUOUS_REQUIREMENT":       CodeREQAmbiguous,
	"MISSING_ACCEPTANCE_CRITERIA": CodeREQNoCriteria,
	"CONFLICTING_REQUIREMENTS":    CodeREQConflict,
	"INCOMPLETE_REQUIREMENT":      CodeREQIncomplete,
	"UNTESTABLE_REQUIREMENT":      CodeREQUntestable,
	"UNMEASURABLE_SUCCESS_METRIC": CodeMETRICUnmeasurable,
	"MISSING_METRIC_BASELINE":     CodeMETRICNoBaseline,
	"MISSING_METRIC_TARGET":       CodeMETRICNoTarget,
	"UNREALISTIC_TARGET":          CodeMETRICUnrealistic,
	"MISSING_TRACKING_PLAN":       CodeMETRICNoTracking,
	"MISSING_USER_PERSONA":        CodeUSERNoPersona,
	"INCOMPLETE_PERSONA":          CodeUSERIncomplete,
	"MISSING_USER_JOURNEY":        CodeUSERNoJourney,
	"UNCLEAR_PROBLEM_STATEMENT":   CodeUSERUnclearProblem,
	"INCOMPLETE_ERROR_HANDLING":   CodeARCHNoErrorHandling,
	"MISSING_API_CONTRACT":        CodeARCHNoAPI,
	"MISSING_DATA_MODEL":          CodeARCHNoDataModel,
	"MISSING_DEPENDENCY":          CodeARCHMissingDep,
	"ARCHITECTURE_GAP":            CodeARCHGap,
	"SECURITY_GAP":                CodeSECGap,
	"MISSING_AUTHENTICATION":      CodeSECNoAuth,
	"MISSING_AUTHORIZATION":       CodeSECNoAuthz,
	"DATA_PRIVACY_CONCERN":        CodeSECPrivacy,
	"MISSING_INPUT_VALIDATION":    CodeSECNoValidation,
	"SCALABILITY_CONCERN":         CodeSCALEConcern,
	"PERFORMANCE_RISK":            CodeSCALEPerformance,
	"MISSING_CAPACITY_PLAN":       CodeSCALENoCapacity,
	"SINGLE_POINT_OF_FAILURE":     CodeSCALESPOF,
	"MISSING_DEPLOYMENT_PLAN":     CodeINFRANoDeploy,
	"MISSING_MONITORING":          CodeINFRANoMonitor,
	"MISSING_ALERTING_STRATEGY":   CodeINFRANoAlert,
	"MISSING_RECOVERY_PLAN":       CodeINFRANoRecovery,
	"INSUFFICIENT_DOCUMENTATION":  CodeDOCInsufficient,
	"OUTDATED_REFERENCE":          CodeDOCOutdated,
	"MISSING_DIAGRAM":             CodeDOCNoDiagram,
	"SCOPE_CREEP":                 CodeSCOPECreep,
	"UNBOUNDED_SCOPE":             CodeSCOPEUnbounded,
	"MISSING_CONSTRAINTS":         CodeSCOPENoConstr,
	"ACCESSIBILITY_GAP":           CodeUXNoARIA,
	"MISSING_TIMELINE":            CodeSCOPENoTimeline,
}

Legacy code mappings for backwards compatibility. Maps old code names to new prefixed codes.

View Source 
var ReasonCodeRegistry = map[ReasonCode]ReasonCodeInfo{

	CodeREQAmbiguous: {
		Code:            CodeREQAmbiguous,
		Category:        CategoryREQ,
		Description:     "Requirement lacks specificity or has multiple interpretations",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Rewrite this requirement to be specific and unambiguous. Include: (1) a clear action verb, (2) measurable outcome, (3) specific constraints or boundaries. Remove any vague terms like 'fast', 'user-friendly', 'easy'.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "mrd", "trd"},
	},
	CodeREQNoCriteria: {
		Code:            CodeREQNoCriteria,
		Category:        CategoryREQ,
		Description:     "Requirement lacks acceptance criteria for verification",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Add acceptance criteria using Given/When/Then format. Include: (1) preconditions (Given), (2) action trigger (When), (3) expected outcome (Then). Cover both success and failure scenarios.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd"},
	},
	CodeREQConflict: {
		Code:            CodeREQConflict,
		Category:        CategoryREQ,
		Description:     "Two or more requirements contradict each other",
		DefaultSeverity: SeverityCritical,
		RepairPrompt:    "Identify the conflicting requirements and resolve by: (1) clarifying which takes priority, (2) merging into a consistent requirement, or (3) adding conditional logic to handle both cases. Document the resolution rationale.",
		RequiresHuman:   true,
		SpecTypes:       []string{"prd", "mrd", "trd"},
	},
	CodeREQIncomplete: {
		Code:            CodeREQIncomplete,
		Category:        CategoryREQ,
		Description:     "Requirement is missing essential details",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Complete this requirement by adding missing elements: (1) WHO is affected, (2) WHAT action/capability, (3) WHY it's needed (business value), (4) any constraints or dependencies.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "mrd"},
	},
	CodeREQUntestable: {
		Code:            CodeREQUntestable,
		Category:        CategoryREQ,
		Description:     "Requirement cannot be objectively verified",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Rewrite this requirement to be testable. Add: (1) specific numeric thresholds where applicable, (2) clear pass/fail conditions, (3) measurement method. Replace subjective terms with objective criteria.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd"},
	},
	CodeREQMissingReason: {
		Code:            CodeREQMissingReason,
		Category:        CategoryREQ,
		Description:     "Requirement lacks justification or business rationale",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Add a rationale explaining WHY this requirement exists. Include: (1) business value or user benefit, (2) problem it solves, (3) consequences of not implementing it.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "mrd"},
	},

	CodeMETRICUnmeasurable: {
		Code:            CodeMETRICUnmeasurable,
		Category:        CategoryMETRIC,
		Description:     "Success metric cannot be objectively measured",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Make this metric measurable by specifying: (1) exact data source, (2) calculation formula, (3) measurement frequency, (4) responsible team/system for tracking.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "mrd"},
	},
	CodeMETRICNoBaseline: {
		Code:            CodeMETRICNoBaseline,
		Category:        CategoryMETRIC,
		Description:     "Success metric lacks a baseline value",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Add baseline value for this metric. Include: (1) current value or 'TBD - to be measured in sprint 1', (2) date of measurement, (3) methodology used to obtain baseline.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "mrd"},
	},
	CodeMETRICNoTarget: {
		Code:            CodeMETRICNoTarget,
		Category:        CategoryMETRIC,
		Description:     "Success metric lacks a target value",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add a specific target value with: (1) numeric goal, (2) timeframe to achieve it, (3) rationale for why this target was chosen.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "mrd"},
	},
	CodeMETRICUnrealistic: {
		Code:            CodeMETRICUnrealistic,
		Category:        CategoryMETRIC,
		Description:     "Target appears unrealistic given constraints",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Revise this target to be achievable. Either: (1) provide evidence/benchmarks supporting the target, (2) adjust to a realistic value with justification, or (3) break into incremental milestones.",
		RequiresHuman:   true,
		SpecTypes:       []string{"prd", "mrd"},
	},
	CodeMETRICNoTracking: {
		Code:            CodeMETRICNoTracking,
		Category:        CategoryMETRIC,
		Description:     "No plan for how metrics will be tracked",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Add tracking plan specifying: (1) tool/system for measurement, (2) reporting frequency, (3) dashboard or alert thresholds, (4) who reviews the metrics.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd"},
	},
	CodeMETRICMissingKPI: {
		Code:            CodeMETRICMissingKPI,
		Category:        CategoryMETRIC,
		Description:     "Key performance indicator not defined for critical feature",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add KPI for this feature covering: (1) primary success metric, (2) leading indicators, (3) guardrail metrics to prevent negative side effects.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "mrd"},
	},
	CodeMETRICVanity: {
		Code:            CodeMETRICVanity,
		Category:        CategoryMETRIC,
		Description:     "Metric does not correlate with actual business value",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Replace this vanity metric with an actionable metric that: (1) correlates with business outcomes, (2) can influence decisions, (3) measures user value not just activity.",
		RequiresHuman:   true,
		SpecTypes:       []string{"prd", "mrd"},
	},

	CodeUSERNoPersona: {
		Code:            CodeUSERNoPersona,
		Category:        CategoryUSER,
		Description:     "Target user persona not defined",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Add user persona including: (1) name and role, (2) demographics/context, (3) goals and motivations, (4) pain points and frustrations, (5) technical proficiency level.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "mrd", "uxd"},
	},
	CodeUSERIncomplete: {
		Code:            CodeUSERIncomplete,
		Category:        CategoryUSER,
		Description:     "User persona lacks essential details",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Complete this persona by adding missing elements: goals, pain points, behavioral characteristics, technical context, or typical use scenarios.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "uxd"},
	},
	CodeUSERNoJourney: {
		Code:            CodeUSERNoJourney,
		Category:        CategoryUSER,
		Description:     "User journey or flow not documented",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add user journey showing: (1) entry point/trigger, (2) step-by-step actions, (3) decision points, (4) success outcome, (5) potential failure points and recovery.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "uxd"},
	},
	CodeUSERUnclearProblem: {
		Code:            CodeUSERUnclearProblem,
		Category:        CategoryUSER,
		Description:     "Problem statement is vague or unclear",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Clarify the problem statement using this format: '[User persona] needs [capability] because [reason], but currently [obstacle/pain point].' Be specific about the impact.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "mrd"},
	},
	CodeUSERNoGoals: {
		Code:            CodeUSERNoGoals,
		Category:        CategoryUSER,
		Description:     "User goals not articulated",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add user goals distinguishing between: (1) functional goals (tasks to complete), (2) emotional goals (how they want to feel), (3) social goals (how they want to be perceived).",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "uxd"},
	},
	CodeUSERNoPainPoints: {
		Code:            CodeUSERNoPainPoints,
		Category:        CategoryUSER,
		Description:     "User pain points not documented",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Document pain points including: (1) current workarounds users employ, (2) time/money cost of the problem, (3) emotional frustration, (4) frequency of occurrence.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "mrd", "uxd"},
	},

	CodeARCHNoErrorHandling: {
		Code:            CodeARCHNoErrorHandling,
		Category:        CategoryARCH,
		Description:     "Error handling strategy is incomplete",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add error handling covering: (1) error categories (network, validation, auth, server), (2) retry strategy per category, (3) user-facing error messages, (4) logging/alerting approach, (5) graceful degradation.",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd", "ird"},
	},
	CodeARCHNoAPI: {
		Code:            CodeARCHNoAPI,
		Category:        CategoryARCH,
		Description:     "API contract or interface not specified",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Add API specification including: (1) endpoint paths and methods, (2) request/response schemas with examples, (3) authentication requirements, (4) rate limits, (5) error response format, (6) versioning strategy.",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd"},
	},
	CodeARCHNoDataModel: {
		Code:            CodeARCHNoDataModel,
		Category:        CategoryARCH,
		Description:     "Data model or schema not defined",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Add data model with: (1) entity definitions and fields, (2) relationships between entities, (3) data types and constraints, (4) indexes for query patterns, (5) migration strategy for schema changes.",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd"},
	},
	CodeARCHMissingDep: {
		Code:            CodeARCHMissingDep,
		Category:        CategoryARCH,
		Description:     "Required dependency not documented",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Document this dependency including: (1) name and version, (2) why it's needed, (3) license compatibility, (4) maintenance status, (5) fallback if deprecated.",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd", "ird"},
	},
	CodeARCHGap: {
		Code:            CodeARCHGap,
		Category:        CategoryARCH,
		Description:     "Architecture has unexplained gaps or missing components",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Address the architectural gap by: (1) adding the missing component with rationale, (2) explaining why it's intentionally omitted, or (3) marking as future work with timeline.",
		RequiresHuman:   true,
		SpecTypes:       []string{"trd"},
	},
	CodeARCHNoInterface: {
		Code:            CodeARCHNoInterface,
		Category:        CategoryARCH,
		Description:     "Interface between components not defined",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Define the interface including: (1) method signatures, (2) input/output types, (3) error conditions, (4) idempotency guarantees, (5) versioning approach.",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd"},
	},
	CodeARCHCircularDep: {
		Code:            CodeARCHCircularDep,
		Category:        CategoryARCH,
		Description:     "Circular dependency detected between components",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Resolve circular dependency by: (1) extracting shared logic to a new module, (2) using dependency injection, (3) introducing an interface/abstraction layer, or (4) restructuring the component boundaries.",
		RequiresHuman:   true,
		SpecTypes:       []string{"trd"},
	},
	CodeARCHTightCoupling: {
		Code:            CodeARCHTightCoupling,
		Category:        CategoryARCH,
		Description:     "Components are too tightly coupled",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Reduce coupling by: (1) defining clear interfaces, (2) using events/messages instead of direct calls, (3) applying dependency inversion, (4) documenting the bounded context.",
		RequiresHuman:   true,
		SpecTypes:       []string{"trd"},
	},

	CodeSECGap: {
		Code:            CodeSECGap,
		Category:        CategorySEC,
		Description:     "Security consideration not addressed",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Add security controls addressing: (1) threat model for this feature, (2) mitigation strategies, (3) security testing requirements, (4) incident response considerations.",
		RequiresHuman:   true,
		SpecTypes:       []string{"trd", "ird", "prd"},
	},
	CodeSECNoAuth: {
		Code:            CodeSECNoAuth,
		Category:        CategorySEC,
		Description:     "Authentication mechanism not specified",
		DefaultSeverity: SeverityCritical,
		RepairPrompt:    "Define authentication including: (1) auth method (OAuth, JWT, API key, etc.), (2) token/session management, (3) password/credential requirements, (4) MFA considerations, (5) session timeout policy.",
		RequiresHuman:   true,
		SpecTypes:       []string{"trd"},
	},
	CodeSECNoAuthz: {
		Code:            CodeSECNoAuthz,
		Category:        CategorySEC,
		Description:     "Authorization model not defined",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Add authorization covering: (1) role definitions, (2) permission matrix, (3) resource-level access rules, (4) principle of least privilege application, (5) authorization check points.",
		RequiresHuman:   true,
		SpecTypes:       []string{"trd"},
	},
	CodeSECPrivacy: {
		Code:            CodeSECPrivacy,
		Category:        CategorySEC,
		Description:     "Data privacy requirements not addressed",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Document data privacy including: (1) PII identification, (2) data classification, (3) retention policy, (4) deletion/anonymization procedures, (5) compliance requirements (GDPR, CCPA, etc.).",
		RequiresHuman:   true,
		SpecTypes:       []string{"trd", "prd"},
	},
	CodeSECNoValidation: {
		Code:            CodeSECNoValidation,
		Category:        CategorySEC,
		Description:     "Input validation not specified",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add input validation covering: (1) allowed characters/formats, (2) length limits, (3) sanitization rules, (4) validation error messages, (5) server-side validation (never trust client).",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd"},
	},
	CodeSECNoEncryption: {
		Code:            CodeSECNoEncryption,
		Category:        CategorySEC,
		Description:     "Encryption requirements not specified",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Define encryption requirements: (1) data at rest encryption, (2) data in transit (TLS version), (3) key management strategy, (4) sensitive field encryption.",
		RequiresHuman:   true,
		SpecTypes:       []string{"trd", "ird"},
	},
	CodeSECHardcodedSecret: {
		Code:            CodeSECHardcodedSecret,
		Category:        CategorySEC,
		Description:     "Hardcoded secrets or credentials detected",
		DefaultSeverity: SeverityCritical,
		RepairPrompt:    "Remove hardcoded secrets and specify: (1) secret management system to use, (2) environment variable naming convention, (3) rotation policy, (4) access audit logging.",
		RequiresHuman:   true,
		SpecTypes:       []string{"trd", "ird"},
	},
	CodeSECInjectionRisk: {
		Code:            CodeSECInjectionRisk,
		Category:        CategorySEC,
		Description:     "Potential injection vulnerability (SQL, XSS, command)",
		DefaultSeverity: SeverityCritical,
		RepairPrompt:    "Address injection risk by specifying: (1) parameterized queries (SQL), (2) output encoding (XSS), (3) input sanitization, (4) CSP headers, (5) security testing requirements.",
		RequiresHuman:   true,
		SpecTypes:       []string{"trd"},
	},

	CodeSCALEConcern: {
		Code:            CodeSCALEConcern,
		Category:        CategorySCALE,
		Description:     "Scalability concern not addressed",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add scaling strategy covering: (1) expected load (requests/sec, concurrent users), (2) horizontal vs vertical scaling approach, (3) bottleneck identification, (4) auto-scaling triggers.",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd", "ird"},
	},
	CodeSCALEPerformance: {
		Code:            CodeSCALEPerformance,
		Category:        CategorySCALE,
		Description:     "Performance risk identified but not mitigated",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add performance requirements: (1) latency targets (p50, p95, p99), (2) throughput requirements, (3) resource limits (CPU, memory), (4) performance testing approach.",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd", "ird"},
	},
	CodeSCALENoCapacity: {
		Code:            CodeSCALENoCapacity,
		Category:        CategorySCALE,
		Description:     "Capacity planning not documented",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Add capacity plan including: (1) initial resource sizing, (2) growth projections (6mo, 1yr), (3) cost estimates, (4) scaling thresholds.",
		RequiresHuman:   false,
		SpecTypes:       []string{"ird"},
	},
	CodeSCALESPOF: {
		Code:            CodeSCALESPOF,
		Category:        CategorySCALE,
		Description:     "Single point of failure identified",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Eliminate SPOF by adding: (1) redundancy strategy (active-active, active-passive), (2) failover mechanism, (3) health check configuration, (4) recovery time objective.",
		RequiresHuman:   true,
		SpecTypes:       []string{"trd", "ird"},
	},
	CodeSCALENoRateLimit: {
		Code:            CodeSCALENoRateLimit,
		Category:        CategorySCALE,
		Description:     "Rate limiting not specified",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add rate limiting covering: (1) limits per user/IP/API key, (2) time windows, (3) response when exceeded (429 status), (4) bypass rules for internal services.",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd"},
	},
	CodeSCALENoCache: {
		Code:            CodeSCALENoCache,
		Category:        CategorySCALE,
		Description:     "Caching strategy not defined for frequently accessed data",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Add caching strategy: (1) what to cache, (2) cache location (CDN, Redis, in-memory), (3) TTL/expiration policy, (4) cache invalidation approach.",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd"},
	},
	CodeSCALEBlockingOp: {
		Code:            CodeSCALEBlockingOp,
		Category:        CategorySCALE,
		Description:     "Blocking operation in critical path",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Address blocking operation by: (1) making it async with queue/callback, (2) adding timeout, (3) implementing circuit breaker, (4) providing fallback response.",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd"},
	},

	CodeINFRANoDeploy: {
		Code:            CodeINFRANoDeploy,
		Category:        CategoryINFRA,
		Description:     "Deployment strategy not defined",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add deployment plan covering: (1) deployment method (blue-green, canary, rolling), (2) environment progression (dev→staging→prod), (3) deployment automation, (4) smoke test requirements.",
		RequiresHuman:   false,
		SpecTypes:       []string{"ird"},
	},
	CodeINFRANoMonitor: {
		Code:            CodeINFRANoMonitor,
		Category:        CategoryINFRA,
		Description:     "Monitoring strategy not defined",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add monitoring covering: (1) key metrics to track, (2) logging strategy and retention, (3) tracing/correlation IDs, (4) dashboard requirements, (5) SLO definitions.",
		RequiresHuman:   false,
		SpecTypes:       []string{"ird"},
	},
	CodeINFRANoAlert: {
		Code:            CodeINFRANoAlert,
		Category:        CategoryINFRA,
		Description:     "Alerting and on-call strategy not defined",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Add alerting strategy: (1) alert thresholds and conditions, (2) severity levels, (3) notification channels, (4) escalation policy, (5) on-call rotation.",
		RequiresHuman:   false,
		SpecTypes:       []string{"ird"},
	},
	CodeINFRANoRecovery: {
		Code:            CodeINFRANoRecovery,
		Category:        CategoryINFRA,
		Description:     "Disaster recovery plan not documented",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add DR plan including: (1) RTO target, (2) RPO target, (3) recovery procedures, (4) data backup strategy, (5) DR testing schedule.",
		RequiresHuman:   true,
		SpecTypes:       []string{"ird"},
	},
	CodeINFRANoBackup: {
		Code:            CodeINFRANoBackup,
		Category:        CategoryINFRA,
		Description:     "Backup strategy not specified",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add backup strategy: (1) what is backed up, (2) backup frequency, (3) retention period, (4) backup location (offsite), (5) restore testing procedure.",
		RequiresHuman:   false,
		SpecTypes:       []string{"ird"},
	},
	CodeINFRANoRunbook: {
		Code:            CodeINFRANoRunbook,
		Category:        CategoryINFRA,
		Description:     "Operational runbook not provided",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Create runbook covering: (1) common operational tasks, (2) troubleshooting steps for known issues, (3) escalation contacts, (4) maintenance procedures.",
		RequiresHuman:   false,
		SpecTypes:       []string{"ird"},
	},
	CodeINFRANoRollback: {
		Code:            CodeINFRANoRollback,
		Category:        CategoryINFRA,
		Description:     "Rollback procedure not defined",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add rollback procedure: (1) trigger criteria for rollback, (2) step-by-step rollback process, (3) data migration rollback (if applicable), (4) verification steps post-rollback.",
		RequiresHuman:   false,
		SpecTypes:       []string{"ird"},
	},
	CodeINFRAEnvMismatch: {
		Code:            CodeINFRAEnvMismatch,
		Category:        CategoryINFRA,
		Description:     "Environment configuration mismatch risk",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Address environment parity: (1) document environment differences, (2) use infrastructure-as-code, (3) implement config validation, (4) add environment-specific testing.",
		RequiresHuman:   false,
		SpecTypes:       []string{"ird"},
	},

	CodeDOCInsufficient: {
		Code:            CodeDOCInsufficient,
		Category:        CategoryDOC,
		Description:     "Documentation is insufficient for implementation",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Expand documentation to include: (1) implementation details, (2) code examples, (3) edge cases and their handling, (4) integration points.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "trd", "ird", "uxd"},
	},
	CodeDOCOutdated: {
		Code:            CodeDOCOutdated,
		Category:        CategoryDOC,
		Description:     "Reference to outdated information or deprecated feature",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Update the outdated reference to: (1) current version/approach, (2) migration path if applicable, (3) deprecation timeline if relevant.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "trd", "ird"},
	},
	CodeDOCNoDiagram: {
		Code:            CodeDOCNoDiagram,
		Category:        CategoryDOC,
		Description:     "Visual diagram would improve clarity",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Add a diagram showing: (1) component relationships, (2) data flow, (3) sequence of operations, or (4) architecture overview. Use Mermaid or similar text-based diagram format.",
		RequiresHuman:   false,
		SpecTypes:       []string{"trd", "ird", "uxd"},
	},
	CodeDOCNoExamples: {
		Code:            CodeDOCNoExamples,
		Category:        CategoryDOC,
		Description:     "Missing examples to clarify usage",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Add concrete examples showing: (1) typical usage, (2) edge cases, (3) error scenarios. Include request/response examples for APIs.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "trd"},
	},
	CodeDOCInconsistent: {
		Code:            CodeDOCInconsistent,
		Category:        CategoryDOC,
		Description:     "Documentation inconsistent with other sections",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Resolve the inconsistency by: (1) identifying the source of truth, (2) updating conflicting sections, (3) adding cross-references to avoid future drift.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "trd", "ird", "uxd"},
	},

	CodeSCOPECreep: {
		Code:            CodeSCOPECreep,
		Category:        CategorySCOPE,
		Description:     "Scope includes unnecessary features or complexity",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Address scope creep by: (1) moving non-essential items to 'Future Work', (2) justifying inclusion with business value, or (3) removing entirely with rationale.",
		RequiresHuman:   true,
		SpecTypes:       []string{"prd", "mrd"},
	},
	CodeSCOPEUnbounded: {
		Code:            CodeSCOPEUnbounded,
		Category:        CategorySCOPE,
		Description:     "Scope is unbounded or unclear",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Define scope boundaries: (1) explicit 'In Scope' list, (2) explicit 'Out of Scope' / 'Non-Goals' list, (3) decision criteria for borderline items.",
		RequiresHuman:   true,
		SpecTypes:       []string{"prd", "mrd"},
	},
	CodeSCOPENoConstr: {
		Code:            CodeSCOPENoConstr,
		Category:        CategorySCOPE,
		Description:     "Constraints or limitations not documented",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Document constraints including: (1) technical constraints (platform, language, etc.), (2) business constraints (budget, timeline), (3) resource constraints (team size, skills).",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "trd"},
	},
	CodeSCOPENoNonGoals: {
		Code:            CodeSCOPENoNonGoals,
		Category:        CategorySCOPE,
		Description:     "Non-goals not explicitly stated",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Add non-goals section listing: (1) features explicitly NOT being built, (2) use cases NOT being supported, (3) rationale for each exclusion.",
		RequiresHuman:   false,
		SpecTypes:       []string{"prd", "mrd"},
	},
	CodeSCOPEMVPUnclear: {
		Code:            CodeSCOPEMVPUnclear,
		Category:        CategorySCOPE,
		Description:     "MVP scope not clearly defined",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Define MVP clearly: (1) minimum feature set for launch, (2) success criteria for MVP, (3) what's deferred to post-MVP, (4) timeline for MVP.",
		RequiresHuman:   true,
		SpecTypes:       []string{"prd"},
	},
	CodeSCOPENoTimeline: {
		Code:            CodeSCOPENoTimeline,
		Category:        CategorySCOPE,
		Description:     "Timeline or milestones not specified",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Add timeline with: (1) key milestones, (2) dependencies between milestones, (3) target dates (even if tentative), (4) critical path identification.",
		RequiresHuman:   true,
		SpecTypes:       []string{"prd", "mrd"},
	},

	CodeUXNoARIA: {
		Code:            CodeUXNoARIA,
		Category:        CategoryUX,
		Description:     "ARIA labels or accessibility attributes not specified",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Add accessibility requirements: (1) ARIA labels for interactive elements, (2) alt text requirements for images, (3) heading hierarchy, (4) focus management, (5) screen reader considerations.",
		RequiresHuman:   false,
		SpecTypes:       []string{"uxd"},
	},
	CodeUXNoErrorState: {
		Code:            CodeUXNoErrorState,
		Category:        CategoryUX,
		Description:     "Error state UI not designed",
		DefaultSeverity: SeverityHigh,
		RepairPrompt:    "Design error states including: (1) error message content and tone, (2) visual treatment, (3) recovery actions available to user, (4) retry mechanisms, (5) when to show inline vs page-level errors.",
		RequiresHuman:   false,
		SpecTypes:       []string{"uxd"},
	},
	CodeUXNoLoading: {
		Code:            CodeUXNoLoading,
		Category:        CategoryUX,
		Description:     "Loading state not specified",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add loading state design: (1) loading indicator type (spinner, skeleton, progress), (2) placement, (3) timeout handling, (4) partial content display strategy.",
		RequiresHuman:   false,
		SpecTypes:       []string{"uxd"},
	},
	CodeUXNoEmpty: {
		Code:            CodeUXNoEmpty,
		Category:        CategoryUX,
		Description:     "Empty state not designed",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Design empty state with: (1) friendly message explaining why empty, (2) illustration or icon, (3) call-to-action to populate, (4) help/onboarding content.",
		RequiresHuman:   false,
		SpecTypes:       []string{"uxd"},
	},
	CodeUXNoResponsive: {
		Code:            CodeUXNoResponsive,
		Category:        CategoryUX,
		Description:     "Responsive/mobile behavior not specified",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add responsive design specs: (1) breakpoints, (2) layout changes per breakpoint, (3) touch target sizes, (4) mobile-specific interactions, (5) content prioritization for small screens.",
		RequiresHuman:   false,
		SpecTypes:       []string{"uxd"},
	},
	CodeUXNoKeyboard: {
		Code:            CodeUXNoKeyboard,
		Category:        CategoryUX,
		Description:     "Keyboard navigation not specified",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Add keyboard accessibility: (1) tab order, (2) keyboard shortcuts, (3) focus indicators, (4) skip links, (5) escape key behavior for modals.",
		RequiresHuman:   false,
		SpecTypes:       []string{"uxd"},
	},
	CodeUXIncompleteNav: {
		Code:            CodeUXIncompleteNav,
		Category:        CategoryUX,
		Description:     "Navigation flow incomplete",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Complete navigation by adding: (1) entry points to this screen, (2) exit points/next steps, (3) back navigation behavior, (4) breadcrumb requirements, (5) deep linking support.",
		RequiresHuman:   false,
		SpecTypes:       []string{"uxd"},
	},
	CodeUXNoFeedback: {
		Code:            CodeUXNoFeedback,
		Category:        CategoryUX,
		Description:     "User feedback mechanism not designed",
		DefaultSeverity: SeverityLow,
		RepairPrompt:    "Add user feedback design: (1) success confirmations, (2) progress indicators for long operations, (3) toast/notification placement, (4) undo capabilities.",
		RequiresHuman:   false,
		SpecTypes:       []string{"uxd"},
	},

	CodeOther: {
		Code:            CodeOther,
		Category:        "other",
		Description:     "Issue does not fit standard categories",
		DefaultSeverity: SeverityMedium,
		RepairPrompt:    "Review the specific issue and address based on context. Consider whether a new reason code category should be created for this type of issue.",
		RequiresHuman:   true,
		SpecTypes:       []string{},
	},
}

ReasonCodeRegistry maps codes to their metadata.

Functions

func AllReasonCodeCategories added in v0.9.0 

func AllReasonCodeCategories() []string

AllReasonCodeCategories returns all unique reason code categories.

func AllRequiredPassing 

func AllRequiredPassing(results []CategoryResult, rubric *RubricSet) bool

AllRequiredPassing checks if all required categories passed.

func CountFindingsByCode added in v0.9.0 

func CountFindingsByCode(findings []Finding) map[ReasonCode]int

CountFindingsByCode counts findings by reason code.

func GetRepairPrompt added in v0.9.0 

func GetRepairPrompt(code ReasonCode) string

GetRepairPrompt returns the AI repair prompt for a code.

func RequiresHumanReview added in v0.9.0 

func RequiresHumanReview(code ReasonCode) bool

RequiresHumanReview returns true if the code needs human review after AI repair.

func ValidDecisionStatusValues added in v0.7.0 

func ValidDecisionStatusValues() []string

ValidDecisionStatusValues returns all valid decision status values.

func ValidEvaluationTypeValues added in v0.7.0 

func ValidEvaluationTypeValues() []string

ValidEvaluationTypeValues returns all valid evaluation type values.

func ValidIntegerScoreValues added in v0.9.0 

func ValidIntegerScoreValues() []int

ValidIntegerScoreValues returns all valid integer score values (1-5).

func ValidReasonCodes added in v0.9.0 

func ValidReasonCodes() []string

ValidReasonCodes returns all registered reason codes.

func ValidScaleTypeValues added in v0.7.0 

func ValidScaleTypeValues() []string

ValidScaleTypeValues returns all valid scale type values.

func ValidScoreValues added in v0.7.0 

func ValidScoreValues() []string

ValidScoreValues returns all valid score values.

func ValidSeverityValues added in v0.7.0 

func ValidSeverityValues() []string

ValidSeverityValues returns all valid severity values.

Types

type ActionItem 

type ActionItem struct {
	// Action describes what needs to be done.
	Action string `json:"action"`

	// Category is the related evaluation category.
	Category string `json:"category,omitempty"`

	// Severity is the related finding severity.
	Severity Severity `json:"severity,omitempty"`

	// Owner suggests who should do this.
	Owner string `json:"owner,omitempty"`

	// Effort estimates work required.
	Effort string `json:"effort,omitempty"`
}

ActionItem is a specific action to take.

type AggregationMethod 

type AggregationMethod string

AggregationMethod specifies how to combine multiple judge scores. 

const (
	// AggregationMajority uses majority vote for pass/partial/fail.
	AggregationMajority AggregationMethod = "majority"

	// AggregationConservative uses the lowest/most critical score.
	AggregationConservative AggregationMethod = "conservative"

	// AggregationOptimistic uses the highest/most lenient score.
	AggregationOptimistic AggregationMethod = "optimistic"

	// AggregationUnanimous requires all judges to agree.
	AggregationUnanimous AggregationMethod = "unanimous"
)

type Annotation 

type Annotation struct {
	// Name is the annotation type (e.g., "quality", "relevance").
	Name string `json:"name"`

	// Score is a numeric score (if applicable).
	Score float64 `json:"score,omitempty"`

	// Label is a categorical label (if applicable).
	Label string `json:"label,omitempty"`

	// Explanation provides reasoning for the annotation.
	Explanation string `json:"explanation,omitempty"`

	// AnnotatorID identifies who provided this annotation.
	AnnotatorID string `json:"annotator_id,omitempty"`

	// AnnotatorType indicates human vs automated (e.g., "human", "llm", "rule").
	AnnotatorType string `json:"annotator_type,omitempty"`
}

Annotation represents a human-provided label or score.

type CategoricalAgreement 

type CategoricalAgreement struct {
	// ExactAgreement is percentage of exact categorical matches.
	ExactAgreement float64 `json:"exactAgreement"`

	// ConfusionMatrix shows disagreement patterns.
	// Keys are "rater1_score:rater2_score" (e.g., "pass:partial").
	ConfusionMatrix map[string]int `json:"confusionMatrix"`

	// SampleSize is the number of paired ratings.
	SampleSize int `json:"sampleSize"`
}

CategoricalAgreement computes agreement between categorical scores.

func ComputeCategoricalAgreement 

func ComputeCategoricalAgreement(results1, results2 []CategoryResult) *CategoricalAgreement

ComputeCategoricalAgreement computes agreement between categorical scores.

type Category 

type Category struct {
	// ID uniquely identifies this category within the rubric.
	ID string `json:"id"`

	// Name is the human-readable category name.
	Name string `json:"name"`

	// Description explains what this category measures.
	Description string `json:"description"`

	// Weight is the relative importance (default 1.0).
	Weight float64 `json:"weight,omitempty"`

	// Required indicates if this category must pass for overall pass.
	Required bool `json:"required,omitempty"`

	// Scale defines how this category is scored.
	Scale Scale `json:"scale"`

	// EvaluationPrompt is a specific prompt for evaluating this category.
	EvaluationPrompt string `json:"evaluationPrompt,omitempty"`

	// Examples provides few-shot examples for LLM evaluation.
	// Research shows 1 example per level improves LLM alignment.
	Examples *CategoryExamples `json:"examples,omitempty"`
}

Category is a single evaluation dimension.

func NewCategory 

func NewCategory(id, name, description string) *Category

NewCategory creates a new category with a categorical scale.

func (*Category) AddOption 

func (c *Category) AddOption(value, label string, criteria ...string) *Category

AddOption adds a scale option to a categorical category.

func (*Category) GetOptionForValue 

func (c *Category) GetOptionForValue(value string) *ScaleOption

GetOptionForValue returns the scale option for a given value.

func (*Category) SetEvaluationPrompt 

func (c *Category) SetEvaluationPrompt(prompt string) *Category

SetEvaluationPrompt sets the evaluation prompt for this category.

func (*Category) SetExamples 

func (c *Category) SetExamples(examples *CategoryExamples) *Category

SetExamples sets few-shot examples for the category.

func (*Category) SetRequired 

func (c *Category) SetRequired(required bool) *Category

SetRequired marks this category as required for pass.

func (*Category) SetWeight 

func (c *Category) SetWeight(weight float64) *Category

SetWeight sets the category weight.

func (*Category) WithBinary 

func (c *Category) WithBinary(passCriteria, failCriteria []string) *Category

WithBinary sets up a binary pass/fail scale.

func (*Category) WithChecklist 

func (c *Category) WithChecklist(required, optional []string, threshold *ChecklistThreshold) *Category

WithChecklist sets up a checklist scale.

func (*Category) WithLikert 

func (c *Category) WithLikert(config *LikertConfig) *Category

WithLikert sets up a Likert scale with custom configuration.

func (*Category) WithLikert5 

func (c *Category) WithLikert5(anchors []LikertAnchor) *Category

WithLikert5 sets up a standard 1-5 Likert scale. Default thresholds: 4-5 = pass, 3 = partial, 1-2 = fail.

func (*Category) WithPassPartialFail 

func (c *Category) WithPassPartialFail(passCriteria, partialCriteria, failCriteria []string) *Category

WithPassPartialFail sets up a standard pass/partial/fail scale.

type CategoryExamples 

type CategoryExamples struct {
	Pass    *Example `json:"pass,omitempty"`
	Partial *Example `json:"partial,omitempty"`
	Fail    *Example `json:"fail,omitempty"`
}

CategoryExamples provides few-shot examples for a category. Research shows 1 example per level improves LLM alignment.

type CategoryResult 

type CategoryResult struct {
	// Category is the category ID.
	Category string `json:"category"`

	// Score is the assigned score (pass, partial, fail).
	// This is the authoritative score for decision-making.
	Score ScoreValue `json:"score"`

	// IntScore is the 1-5 integer score.
	// Preferred for LLM judges as they are unreliable at finer granularity.
	IntScore IntegerScore `json:"intScore,omitempty"`

	// NumericScore is an optional numeric score (e.g., 1-5 Likert).
	// Used for human comparison, inter-rater reliability, and calibration.
	// The categorical Score takes precedence for pass/fail decisions.
	NumericScore *float64 `json:"numericScore,omitempty"`

	// Confidence is the evaluator's confidence in this score (0.0-1.0).
	// Low confidence scores may be routed to human review.
	Confidence float64 `json:"confidence,omitempty"`

	// ReasonCodes are standardized finding identifiers for this category.
	// Enable automated repair workflows.
	ReasonCodes []ReasonCode `json:"reasonCodes,omitempty"`

	// Reasoning explains the score (chain-of-thought).
	Reasoning string `json:"reasoning"`

	// Evidence are specific quotes or observations.
	Evidence []string `json:"evidence,omitempty"`

	// Findings are issues discovered in this category.
	Findings []Finding `json:"findings,omitempty"`

	// ChecklistResults tracks checklist items (for checklist scales).
	ChecklistResults *ChecklistResults `json:"checklistResults,omitempty"`
}

CategoryResult is the evaluation result for a single category.

func NewCategoryResult 

func NewCategoryResult(category string, score ScoreValue, reasoning string) *CategoryResult

NewCategoryResult creates a category result with the given score.

func NewCategoryResultFromLikert 

func NewCategoryResultFromLikert(category string, likertScore int, config *LikertConfig, reasoning string) *CategoryResult

NewCategoryResultFromLikert creates a category result from a Likert score. The categorical score is derived from the numeric score using the config thresholds.

func NewCategoryResultWithIntScore added in v0.9.0 

func NewCategoryResultWithIntScore(category string, intScore IntegerScore, confidence float64, reasoning string) *CategoryResult

NewCategoryResultWithIntScore creates a category result from an integer score.

func NewCategoryResultWithNumeric 

func NewCategoryResultWithNumeric(category string, score ScoreValue, numericScore float64, reasoning string) *CategoryResult

NewCategoryResultWithNumeric creates a category result with both categorical and numeric scores. The numeric score is used for human comparison; categorical score is authoritative for decisions.

func (*CategoryResult) AddEvidence 

func (cr *CategoryResult) AddEvidence(evidence ...string) *CategoryResult

AddEvidence adds evidence to the result.

func (*CategoryResult) AddFinding 

func (cr *CategoryResult) AddFinding(f Finding) *CategoryResult

AddFinding adds a finding to the result.

func (*CategoryResult) AddReasonCode added in v0.9.0 

func (cr *CategoryResult) AddReasonCode(code ReasonCode) *CategoryResult

AddReasonCode adds a reason code to this category result.

func (*CategoryResult) AddReasonCodes added in v0.9.0 

func (cr *CategoryResult) AddReasonCodes(codes ...ReasonCode) *CategoryResult

AddReasonCodes adds multiple reason codes to this category result.

func (*CategoryResult) GetNumericScore 

func (cr *CategoryResult) GetNumericScore() float64

GetNumericScore returns the numeric score, or 0 if not set.

func (*CategoryResult) HasLowConfidence added in v0.9.0 

func (cr *CategoryResult) HasLowConfidence(threshold ...float64) bool

HasLowConfidence returns true if confidence is below the threshold (default 0.7).

func (*CategoryResult) HasNumericScore 

func (cr *CategoryResult) HasNumericScore() bool

HasNumericScore returns true if a numeric score is set.

func (*CategoryResult) IsPassing 

func (cr *CategoryResult) IsPassing() bool

IsPassing returns true if this category passed.

func (*CategoryResult) SetChecklistResults 

func (cr *CategoryResult) SetChecklistResults(results *ChecklistResults) *CategoryResult

SetChecklistResults sets the checklist results.

func (*CategoryResult) SetConfidence added in v0.9.0 

func (cr *CategoryResult) SetConfidence(confidence float64) *CategoryResult

SetConfidence sets the confidence value.

func (*CategoryResult) SetIntScore added in v0.9.0 

func (cr *CategoryResult) SetIntScore(score IntegerScore) *CategoryResult

SetIntScore sets the integer score and derives the categorical score.

func (*CategoryResult) SetNumericScore 

func (cr *CategoryResult) SetNumericScore(score float64) *CategoryResult

SetNumericScore sets the numeric score.

type CategoryResultCounts 

type CategoryResultCounts struct {
	Pass    int `json:"pass"`
	Partial int `json:"partial"`
	Fail    int `json:"fail"`
	Total   int `json:"total"`
}

CountCategoryResults counts results by score value.

func CountResults 

func CountResults(results []CategoryResult) CategoryResultCounts

CountResults counts category results by score.

func (CategoryResultCounts) AllPassing 

func (c CategoryResultCounts) AllPassing() bool

AllPassing returns true if all results are passing.

type ChecklistResults 

type ChecklistResults struct {
	// RequiredPresent are required items that were found.
	RequiredPresent []string `json:"requiredPresent,omitempty"`

	// RequiredMissing are required items that were not found.
	RequiredMissing []string `json:"requiredMissing,omitempty"`

	// OptionalPresent are optional items that were found.
	OptionalPresent []string `json:"optionalPresent,omitempty"`

	// OptionalMissing are optional items that were not found.
	OptionalMissing []string `json:"optionalMissing,omitempty"`
}

ChecklistResults tracks which items were found for checklist scales.

type ChecklistThreshold 

type ChecklistThreshold struct {
	// Required is "all" or a number of required items that must be present.
	Required string `json:"required,omitempty"`

	// Optional is the minimum number of optional items needed.
	Optional int `json:"optional,omitempty"`
}

ChecklistThreshold defines pass criteria for checklist scales.

type CoverageReport added in v0.9.0 

type CoverageReport struct {
	// Sections contains coverage for each named section.
	// Example keys: "components", "foundations", "functions", "lines"
	Sections map[string]CoverageSection `json:"sections"`

	// Overall is the aggregate coverage percentage (0-100).
	Overall int `json:"overall"`
}

CoverageReport aggregates coverage across multiple sections.

func NewCoverageReport added in v0.9.0 

func NewCoverageReport() *CoverageReport

NewCoverageReport creates an empty coverage report.

func (*CoverageReport) AddSection added in v0.9.0 

func (cr *CoverageReport) AddSection(name string, section CoverageSection)

AddSection adds a coverage section to the report.

func (*CoverageReport) AllComplete added in v0.9.0 

func (cr *CoverageReport) AllComplete() bool

AllComplete returns true if all sections have 100% coverage.

func (*CoverageReport) ComputeOverall added in v0.9.0 

func (cr *CoverageReport) ComputeOverall() int

ComputeOverall calculates the overall coverage percentage. Uses a simple average of all section percentages.

func (*CoverageReport) ComputeOverallWeighted added in v0.9.0 

func (cr *CoverageReport) ComputeOverallWeighted(weights map[string]float64) int

ComputeOverallWeighted calculates overall coverage with weights. The weights map should have keys matching section names. Sections not in the weights map get weight 1.0.

func (*CoverageReport) GetSection added in v0.9.0 

func (cr *CoverageReport) GetSection(name string) CoverageSection

GetSection retrieves a section by name. Returns an empty section if not found.

func (*CoverageReport) HasSection added in v0.9.0 

func (cr *CoverageReport) HasSection(name string) bool

HasSection checks if a section exists.

func (*CoverageReport) MeetsThreshold added in v0.9.0 

func (cr *CoverageReport) MeetsThreshold(threshold int) bool

MeetsThreshold returns true if overall coverage meets the threshold.

func (*CoverageReport) SectionsAboveThreshold added in v0.9.0 

func (cr *CoverageReport) SectionsAboveThreshold(threshold int) []string

SectionsAboveThreshold returns section names with coverage >= threshold.

func (*CoverageReport) SectionsBelowThreshold added in v0.9.0 

func (cr *CoverageReport) SectionsBelowThreshold(threshold int) []string

SectionsBelowThreshold returns section names with coverage < threshold.

func (*CoverageReport) SetSection added in v0.9.0 

func (cr *CoverageReport) SetSection(name string, total, complete int, missing []string) *CoverageReport

SetSection is an alias for AddSection for fluent API.

type CoverageSection added in v0.9.0 

type CoverageSection struct {
	// Total is the total number of items in this section.
	Total int `json:"total"`

	// Complete is the number of items that are complete/covered.
	Complete int `json:"complete"`

	// Percentage is the coverage percentage (0-100).
	Percentage int `json:"percentage"`

	// Missing lists the IDs or names of missing/incomplete items.
	Missing []string `json:"missing,omitempty"`
}

CoverageSection represents coverage metrics for a single section/area. This is a generic type that can be used for: - Spec coverage (components, foundations, patterns) - Code coverage (functions, lines, branches) - Test coverage (scenarios, edge cases) - Documentation coverage (API docs, guides)

type Decision 

type Decision struct {
	// Status is the decision outcome.
	Status DecisionStatus `json:"status"`

	// Passed indicates if the evaluation passed.
	Passed bool `json:"passed"`

	// Rationale explains the decision.
	Rationale string `json:"rationale"`

	// FindingCounts summarizes findings by severity.
	FindingCounts FindingCounts `json:"findingCounts"`

	// CategoryCounts summarizes category results.
	CategoryCounts CategoryResultCounts `json:"categoryCounts"`
}

Decision represents the evaluation decision.

func EvaluateResults 

func EvaluateResults(results []CategoryResult, findings []Finding, criteria PassCriteria, rubricSet *RubricSet) Decision

EvaluateResults checks category results and findings against criteria.

type DecisionStatus 

type DecisionStatus string

DecisionStatus represents the decision outcome. 

const (
	DecisionPass        DecisionStatus = "pass"         // Meets all criteria
	DecisionConditional DecisionStatus = "conditional"  // Partial scores or non-blocking findings
	DecisionFail        DecisionStatus = "fail"         // Has blocking findings or required categories failed
	DecisionHumanReview DecisionStatus = "human_review" // Requires human judgment
)

type EvaluationType 

type EvaluationType string

EvaluationType defines how evaluation is performed. 

const (
	// EvaluationTypeAnalytic scores each category independently (recommended for LLM-as-Judge).
	EvaluationTypeAnalytic EvaluationType = "analytic"

	// EvaluationTypeHolistic provides a single overall score.
	EvaluationTypeHolistic EvaluationType = "holistic"
)

type Example 

type Example struct {
	// Excerpt is example content from a document.
	Excerpt string `json:"excerpt"`

	// Reasoning explains why this gets this score.
	// Including reasoning improves LLM alignment (chain-of-thought).
	Reasoning string `json:"reasoning"`
}

Example is a few-shot example for LLM evaluation.

type Finding 

type Finding struct {
	// ID is the unique identifier for this finding.
	ID string `json:"id"`

	// Category is the evaluation category this relates to.
	Category string `json:"category"`

	// Code is the standardized reason code for this finding.
	// Enables automated repair workflows.
	Code ReasonCode `json:"code,omitempty"`

	// Severity indicates the impact level.
	Severity Severity `json:"severity"`

	// Title is a brief summary of the finding.
	Title string `json:"title"`

	// Description provides detailed explanation.
	Description string `json:"description"`

	// Recommendation explains how to fix the issue.
	Recommendation string `json:"recommendation"`

	// Location is a reference to where the issue was found (e.g., "REQ-12", "Section 3.2").
	Location string `json:"location,omitempty"`

	// Evidence provides specific examples or references.
	Evidence string `json:"evidence,omitempty"`

	// Owner suggests who should address this finding.
	Owner string `json:"owner,omitempty"`

	// Effort estimates the work required (low, medium, high).
	Effort string `json:"effort,omitempty"`
}

Finding represents an issue discovered during evaluation.

func NewFinding added in v0.9.0 

func NewFinding(id, category string, severity Severity, title, description string) *Finding

NewFinding creates a new finding with the required fields.

func NewFindingWithCode added in v0.9.0 

func NewFindingWithCode(id, category string, code ReasonCode, title, description string) *Finding

NewFindingWithCode creates a new finding with a reason code.

func (*Finding) GetCodeInfo added in v0.9.0 

func (f *Finding) GetCodeInfo() *ReasonCodeInfo

GetCodeInfo returns the reason code info for this finding's code.

func (*Finding) GetRepairHint added in v0.9.0 

func (f *Finding) GetRepairHint() string

GetRepairHint returns the repair hint from the reason code registry. Deprecated: Use GetRepairPrompt instead.

func (*Finding) GetRepairPrompt added in v0.9.0 

func (f *Finding) GetRepairPrompt() string

GetRepairPrompt returns the AI repair prompt from the reason code registry.

func (*Finding) IsBlocking 

func (f *Finding) IsBlocking() bool

IsBlocking returns true if this finding blocks approval.

func (*Finding) SetCode added in v0.9.0 

func (f *Finding) SetCode(code ReasonCode) *Finding

SetCode sets the reason code on the finding.

func (*Finding) SetEffort added in v0.9.0 

func (f *Finding) SetEffort(effort string) *Finding

SetEffort sets the effort estimate on the finding.

func (*Finding) SetEvidence added in v0.9.0 

func (f *Finding) SetEvidence(evidence string) *Finding

SetEvidence sets the evidence on the finding.

func (*Finding) SetLocation added in v0.9.0 

func (f *Finding) SetLocation(location string) *Finding

SetLocation sets the location reference on the finding.

func (*Finding) SetOwner added in v0.9.0 

func (f *Finding) SetOwner(owner string) *Finding

SetOwner sets the owner on the finding.

func (*Finding) SetRecommendation added in v0.9.0 

func (f *Finding) SetRecommendation(recommendation string) *Finding

SetRecommendation sets the recommendation on the finding.

type FindingCounts 

type FindingCounts struct {
	Critical int `json:"critical"`
	High     int `json:"high"`
	Medium   int `json:"medium"`
	Low      int `json:"low"`
	Info     int `json:"info"`
	Total    int `json:"total"`
}

FindingCounts tracks the number of findings by severity.

func CountFindings 

func CountFindings(findings []Finding) FindingCounts

CountFindings counts findings by severity.

func (FindingCounts) BlockingCount 

func (c FindingCounts) BlockingCount() int

BlockingCount returns the number of blocking findings.

func (FindingCounts) HasBlocking 

func (c FindingCounts) HasBlocking() bool

HasBlocking returns true if there are any blocking findings.

type FindingLimits 

type FindingLimits struct {
	Critical int `json:"critical"`
	High     int `json:"high"`
	Medium   int `json:"medium"`
	Low      int `json:"low,omitempty"`
}

FindingLimits sets maximum allowed findings per severity. Use -1 for unlimited.

type IRRMetrics 

type IRRMetrics struct {
	// ExactAgreement is the percentage of exact score matches.
	ExactAgreement float64 `json:"exactAgreement"`

	// AdjacentAgreement is the percentage within ±1 of each other.
	AdjacentAgreement float64 `json:"adjacentAgreement"`

	// MeanAbsoluteDifference is the average absolute difference.
	MeanAbsoluteDifference float64 `json:"meanAbsoluteDifference"`

	// PearsonCorrelation measures linear correlation (-1 to 1).
	PearsonCorrelation float64 `json:"pearsonCorrelation"`

	// SampleSize is the number of paired ratings.
	SampleSize int `json:"sampleSize"`
}

IRRMetrics contains inter-rater reliability metrics. These metrics are useful when comparing LLM and human ratings.

func ComputeIRR 

func ComputeIRR(pairs []RatingPair) *IRRMetrics

ComputeIRR calculates inter-rater reliability metrics from paired ratings.

func ComputeIRRFromResults 

func ComputeIRRFromResults(results1, results2 []CategoryResult) *IRRMetrics

ComputeIRRFromResults computes IRR metrics from two sets of category results. Useful for comparing LLM evaluation with human ground truth.

type IntegerScore added in v0.9.0 

type IntegerScore int

IntegerScore represents a 1-5 integer evaluation score. This scale is preferred for LLM judges as research shows they are unreliable at finer granularity than 5 levels. 

const (
	// ScoreUnacceptable indicates the spec does not meet requirements.
	ScoreUnacceptable IntegerScore = 1

	// ScoreMajorRevisions indicates significant work is needed.
	ScoreMajorRevisions IntegerScore = 2

	// ScoreAcceptable indicates minimum requirements are met.
	ScoreAcceptable IntegerScore = 3

	// ScoreGood indicates the spec meets expectations well.
	ScoreGood IntegerScore = 4

	// ScoreExcellent indicates the spec exceeds expectations.
	ScoreExcellent IntegerScore = 5
)

func AllIntegerScores added in v0.9.0 

func AllIntegerScores() []IntegerScore

AllIntegerScores returns all valid integer scores in descending order.

func ParseIntegerScore added in v0.9.0 

func ParseIntegerScore(score int) IntegerScore

ParseIntegerScore converts an integer to IntegerScore, clamping to valid range.

func (IntegerScore) Icon added in v0.9.0 

func (s IntegerScore) Icon() string

Icon returns the emoji icon for the score.

func (IntegerScore) IsPassing added in v0.9.0 

func (s IntegerScore) IsPassing() bool

IsPassing returns true if the score is considered passing (4 or higher).

func (IntegerScore) IsValid added in v0.9.0 

func (s IntegerScore) IsValid() bool

IsValid returns true if the score is in the valid 1-5 range.

func (IntegerScore) String added in v0.9.0 

func (s IntegerScore) String() string

String returns the human-readable label for the score.

func (IntegerScore) ToCategorical added in v0.9.0 

func (s IntegerScore) ToCategorical() ScoreValue

ToCategorical converts the integer score to a categorical ScoreValue. 1-2 = fail, 3 = partial, 4-5 = pass

type JudgeCategoricalScore 

type JudgeCategoricalScore struct {
	// JudgeID identifies the judge.
	JudgeID string `json:"judgeId"`

	// Score is the judge's categorical score.
	Score ScoreValue `json:"score"`
}

JudgeCategoricalScore is a categorical score from a specific judge.

type JudgeDisagreement 

type JudgeDisagreement struct {
	// Category is the evaluation dimension.
	Category string `json:"category"`

	// Scores are the individual judge scores.
	Scores []JudgeCategoricalScore `json:"scores"`

	// UniqueScores is the number of distinct scores given.
	UniqueScores int `json:"uniqueScores"`
}

JudgeDisagreement captures where judges had significantly different scores.

type JudgeMetadata 

type JudgeMetadata struct {
	// JudgeID is a unique identifier for this judge configuration.
	JudgeID string `json:"judge_id,omitempty"`

	// Model is the LLM model used (e.g., "claude-3-opus-20240229", "gpt-4-turbo").
	Model string `json:"model"`

	// ModelProvider is the API provider (e.g., "anthropic", "openai", "bedrock").
	ModelProvider string `json:"model_provider,omitempty"`

	// ModelVersion is the specific model version if applicable.
	ModelVersion string `json:"model_version,omitempty"`

	// PromptTemplate is the name/ID of the prompt template used.
	PromptTemplate string `json:"prompt_template,omitempty"`

	// PromptVersion is the version of the prompt template.
	PromptVersion string `json:"prompt_version,omitempty"`

	// SystemPrompt is the system prompt used (or hash/reference if too long).
	SystemPrompt string `json:"system_prompt,omitempty"`

	// Temperature is the sampling temperature used.
	Temperature float64 `json:"temperature,omitempty"`

	// MaxTokens is the max tokens setting.
	MaxTokens int `json:"max_tokens,omitempty"`

	// RubricID references the rubric set used for scoring.
	RubricID string `json:"rubric_id,omitempty"`

	// RubricVersion is the version of the rubric used.
	RubricVersion string `json:"rubric_version,omitempty"`

	// EvaluatedAt is when this evaluation was performed.
	EvaluatedAt time.Time `json:"evaluated_at,omitempty"`

	// Latency is the evaluation duration.
	Latency time.Duration `json:"latency,omitempty"`

	// TokensUsed tracks token consumption.
	TokensUsed *TokenUsage `json:"tokens_used,omitempty"`

	// TraceID links to observability trace (e.g., for Opik/Phoenix/Langfuse).
	TraceID string `json:"trace_id,omitempty"`

	// SpanID links to observability span.
	SpanID string `json:"span_id,omitempty"`
}

JudgeMetadata tracks information about the LLM judge that produced an evaluation. This enables reproducibility, debugging, and comparison of different judge configurations.

func NewJudgeMetadata 

func NewJudgeMetadata(model string) *JudgeMetadata

NewJudgeMetadata creates judge metadata with required fields.

func (*JudgeMetadata) SetLatency 

func (j *JudgeMetadata) SetLatency(d time.Duration)

SetLatency records the evaluation duration.

func (*JudgeMetadata) WithPrompt 

func (j *JudgeMetadata) WithPrompt(template, version string) *JudgeMetadata

WithPrompt sets the prompt template info.

func (*JudgeMetadata) WithProvider 

func (j *JudgeMetadata) WithProvider(provider string) *JudgeMetadata

WithProvider sets the model provider.

func (*JudgeMetadata) WithRubric 

func (j *JudgeMetadata) WithRubric(id, version string) *JudgeMetadata

WithRubric sets the rubric reference.

func (*JudgeMetadata) WithTemperature 

func (j *JudgeMetadata) WithTemperature(temp float64) *JudgeMetadata

WithTemperature sets the sampling temperature.

func (*JudgeMetadata) WithTokenUsage 

func (j *JudgeMetadata) WithTokenUsage(input, output int) *JudgeMetadata

WithTokenUsage sets the token usage.

func (*JudgeMetadata) WithTrace 

func (j *JudgeMetadata) WithTrace(traceID, spanID string) *JudgeMetadata

WithTrace links to observability.

type LikertAnchor 

type LikertAnchor struct {
	// Value is the numeric score.
	Value int `json:"value"`

	// Label is the short label (e.g., "Excellent", "Good").
	Label string `json:"label"`

	// Description explains what this score means.
	Description string `json:"description,omitempty"`
}

LikertAnchor describes what a specific score level means.

func StandardLikert5Anchors 

func StandardLikert5Anchors() []LikertAnchor

StandardLikert5Anchors returns standard 1-5 Likert anchors.

type LikertConfig 

type LikertConfig struct {
	// Min is the minimum score value (usually 1 or 0).
	Min int `json:"min"`

	// Max is the maximum score value (usually 5).
	Max int `json:"max"`

	// Anchors describe what each score level means.
	Anchors []LikertAnchor `json:"anchors,omitempty"`

	// PassThreshold is the minimum score for "pass" (default: top 40%).
	// For 1-5 scale, default is 4.
	PassThreshold *int `json:"passThreshold,omitempty"`

	// PartialThreshold is the minimum score for "partial" (default: middle).
	// For 1-5 scale, default is 3.
	PartialThreshold *int `json:"partialThreshold,omitempty"`
}

LikertConfig defines a Likert scale configuration.

type MultiJudgeResult 

type MultiJudgeResult struct {
	// Evaluations are the individual judge evaluations.
	Evaluations []*Rubric `json:"evaluations"`

	// Judges contains metadata for each judge.
	Judges []*JudgeMetadata `json:"judges"`

	// AggregatedCategories are the combined category results.
	AggregatedCategories []CategoryResult `json:"aggregatedCategories"`

	// AggregationMethod describes how scores were combined.
	AggregationMethod AggregationMethod `json:"aggregationMethod"`

	// Agreement measures inter-judge agreement (0-1, higher = more agreement).
	Agreement float64 `json:"agreement"`

	// Disagreements lists categories where judges significantly disagreed.
	Disagreements []JudgeDisagreement `json:"disagreements,omitempty"`

	// ConsolidatedDecision is the final decision after aggregation.
	ConsolidatedDecision Decision `json:"consolidatedDecision"`

	// ConsolidatedFindings merges findings from all judges.
	ConsolidatedFindings []Finding `json:"consolidatedFindings"`
}

MultiJudgeResult aggregates evaluations from multiple judges. This improves reliability by combining perspectives and detecting disagreement.

func AggregateEvaluations 

func AggregateEvaluations(evaluations []*Rubric, method AggregationMethod) *MultiJudgeResult

AggregateEvaluations combines multiple evaluation reports.

type NextSteps 

type NextSteps struct {
	// RerunCommand is the command to re-run evaluation.
	RerunCommand string `json:"rerunCommand,omitempty"`

	// Immediate are blocking actions that must be completed.
	Immediate []ActionItem `json:"immediate,omitempty"`

	// Recommended are suggested improvements.
	Recommended []ActionItem `json:"recommended,omitempty"`
}

NextSteps provides actionable workflow guidance.

type PairwiseCategoryScore 

type PairwiseCategoryScore struct {
	// Category is the evaluation dimension.
	Category string `json:"category"`

	// Winner indicates which output won for this category.
	Winner PairwiseWinner `json:"winner"`

	// Margin indicates how much better the winner is (0-1, higher = larger gap).
	Margin float64 `json:"margin,omitempty"`

	// Reasoning explains the category-level comparison.
	Reasoning string `json:"reasoning,omitempty"`
}

PairwiseCategoryScore compares outputs on a specific dimension.

type PairwiseComparison 

type PairwiseComparison struct {
	// ID is the unique identifier for this comparison.
	ID string `json:"id,omitempty"`

	// Input is the shared input/prompt for both outputs.
	Input string `json:"input"`

	// OutputA is the first output being compared.
	OutputA string `json:"output_a"`

	// OutputB is the second output being compared.
	OutputB string `json:"output_b"`

	// Winner indicates which output won ("A", "B", or "tie").
	Winner PairwiseWinner `json:"winner"`

	// Confidence is the judge's confidence in the decision (0-1).
	Confidence float64 `json:"confidence,omitempty"`

	// Reasoning explains why this winner was chosen.
	Reasoning string `json:"reasoning"`

	// CategoryScores provides per-category comparisons if applicable.
	CategoryScores []PairwiseCategoryScore `json:"category_scores,omitempty"`

	// Judge contains metadata about the LLM judge.
	Judge *JudgeMetadata `json:"judge,omitempty"`

	// Metadata contains additional comparison context.
	Metadata map[string]any `json:"metadata,omitempty"`

	// CreatedAt is when this comparison was made.
	CreatedAt time.Time `json:"created_at,omitempty"`
}

PairwiseComparison represents a comparison between two outputs. This is an alternative to absolute scoring that can reduce position bias and improve reliability of LLM-as-Judge evaluations.

func NewPairwiseComparison 

func NewPairwiseComparison(input, outputA, outputB string) *PairwiseComparison

NewPairwiseComparison creates a new pairwise comparison.

func (*PairwiseComparison) AddCategoryScore 

func (p *PairwiseComparison) AddCategoryScore(category string, winner PairwiseWinner, reasoning string, margin float64)

AddCategoryScore adds a per-category comparison.

func (*PairwiseComparison) SetWinner 

func (p *PairwiseComparison) SetWinner(winner PairwiseWinner, reasoning string, confidence float64)

SetWinner sets the comparison result.

func (*PairwiseComparison) SwappedComparison 

func (p *PairwiseComparison) SwappedComparison() *PairwiseComparison

SwappedComparison creates a comparison with A and B swapped. Running both orders helps detect position bias in the judge.

type PairwiseResult 

type PairwiseResult struct {
	// Comparisons are all the individual comparisons.
	Comparisons []PairwiseComparison `json:"comparisons"`

	// WinRateA is the percentage of comparisons won by A.
	WinRateA float64 `json:"win_rate_a"`

	// WinRateB is the percentage of comparisons won by B.
	WinRateB float64 `json:"win_rate_b"`

	// TieRate is the percentage of ties.
	TieRate float64 `json:"tie_rate"`

	// OverallWinner is the aggregated winner.
	OverallWinner PairwiseWinner `json:"overall_winner"`

	// Confidence is the overall confidence in the result.
	Confidence float64 `json:"confidence"`
}

PairwiseResult aggregates multiple pairwise comparisons.

func ComputePairwiseResult 

func ComputePairwiseResult(comparisons []PairwiseComparison) *PairwiseResult

ComputeResult aggregates multiple comparisons into a result.

type PairwiseWinner 

type PairwiseWinner string

PairwiseWinner indicates the winner of a pairwise comparison. 

const (
	// WinnerA indicates output A is better.
	WinnerA PairwiseWinner = "A"

	// WinnerB indicates output B is better.
	WinnerB PairwiseWinner = "B"

	// WinnerTie indicates both outputs are roughly equal.
	WinnerTie PairwiseWinner = "tie"

	// WinnerUncertain indicates the judge couldn't determine a winner.
	WinnerUncertain PairwiseWinner = "uncertain"
)

type PassCriteria 

type PassCriteria struct {
	// MinCategoriesPassing specifies how many categories must pass.
	// Values: "all", "all_required", or a number like "3".
	MinCategoriesPassing string `json:"minCategoriesPassing"`

	// MaxFindings limits findings by severity.
	// Use -1 for unlimited.
	MaxFindings *FindingLimits `json:"maxFindingsSeverity,omitempty"`

	// MinIntScore is the minimum overall IntegerScore (1-5) required to pass.
	// If set, the overall score must be >= this value.
	// Use 0 to disable this check.
	MinIntScore IntegerScore `json:"minIntScore,omitempty"`
}

PassCriteria defines the requirements for approval. Aligned with LLM-as-Judge best practices.

func DefaultPassCriteria 

func DefaultPassCriteria() PassCriteria

DefaultPassCriteria returns standard pass criteria. All required categories must pass, 0 critical/high findings allowed.

func StrictPassCriteria 

func StrictPassCriteria() PassCriteria

StrictPassCriteria returns strict pass criteria. All categories must pass, max 3 medium findings.

type RatingPair 

type RatingPair struct {
	// Rater1 is the first rater's score (e.g., human).
	Rater1 float64

	// Rater2 is the second rater's score (e.g., LLM).
	Rater2 float64

	// Category is the category being rated.
	Category string

	// ItemID identifies the item being rated.
	ItemID string
}

RatingPair represents a pair of ratings for the same item.

type ReasonCode added in v0.9.0 

type ReasonCode string

ReasonCode is a standardized finding identifier using category prefixes. Format: {CATEGORY}-{ISSUE} Example: REQ-AMBIGUOUS, SEC-MISSING_AUTH

Reason codes enable automated repair workflows by providing machine-readable categorization of issues that map to specific repair strategies. 

const (
	CodeREQAmbiguous     ReasonCode = "REQ-AMBIGUOUS"
	CodeREQNoCriteria    ReasonCode = "REQ-NO_CRITERIA"
	CodeREQConflict      ReasonCode = "REQ-CONFLICT"
	CodeREQIncomplete    ReasonCode = "REQ-INCOMPLETE"
	CodeREQUntestable    ReasonCode = "REQ-UNTESTABLE"
	CodeREQMissingReason ReasonCode = "REQ-MISSING_REASON"
)

Requirements codes (REQ-*) 

const (
	CodeMETRICUnmeasurable ReasonCode = "METRIC-UNMEASURABLE"
	CodeMETRICNoBaseline   ReasonCode = "METRIC-NO_BASELINE"
	CodeMETRICNoTarget     ReasonCode = "METRIC-NO_TARGET"
	CodeMETRICUnrealistic  ReasonCode = "METRIC-UNREALISTIC"
	CodeMETRICNoTracking   ReasonCode = "METRIC-NO_TRACKING"
	CodeMETRICMissingKPI   ReasonCode = "METRIC-MISSING_KPI"
	CodeMETRICVanity       ReasonCode = "METRIC-VANITY"
)

Metrics codes (METRIC-*) 

const (
	CodeUSERNoPersona      ReasonCode = "USER-NO_PERSONA"
	CodeUSERIncomplete     ReasonCode = "USER-INCOMPLETE"
	CodeUSERNoJourney      ReasonCode = "USER-NO_JOURNEY"
	CodeUSERUnclearProblem ReasonCode = "USER-UNCLEAR_PROBLEM"
	CodeUSERNoGoals        ReasonCode = "USER-NO_GOALS"
	CodeUSERNoPainPoints   ReasonCode = "USER-NO_PAIN_POINTS"
)

User codes (USER-*) 

const (
	CodeARCHNoErrorHandling ReasonCode = "ARCH-NO_ERROR_HANDLING"
	CodeARCHNoAPI           ReasonCode = "ARCH-NO_API"
	CodeARCHNoDataModel     ReasonCode = "ARCH-NO_DATA_MODEL"
	CodeARCHMissingDep      ReasonCode = "ARCH-MISSING_DEP"
	CodeARCHGap             ReasonCode = "ARCH-GAP"
	CodeARCHNoInterface     ReasonCode = "ARCH-NO_INTERFACE"
	CodeARCHCircularDep     ReasonCode = "ARCH-CIRCULAR_DEP"
	CodeARCHTightCoupling   ReasonCode = "ARCH-TIGHT_COUPLING"
)

Architecture codes (ARCH-*) 

const (
	CodeSECGap             ReasonCode = "SEC-GAP"
	CodeSECNoAuth          ReasonCode = "SEC-NO_AUTH"
	CodeSECNoAuthz         ReasonCode = "SEC-NO_AUTHZ"
	CodeSECPrivacy         ReasonCode = "SEC-PRIVACY"
	CodeSECNoValidation    ReasonCode = "SEC-NO_VALIDATION"
	CodeSECNoEncryption    ReasonCode = "SEC-NO_ENCRYPTION"
	CodeSECHardcodedSecret ReasonCode = "SEC-HARDCODED_SECRET" //nolint:gosec // G101 false positive: this is a reason code identifier, not a credential
	CodeSECInjectionRisk   ReasonCode = "SEC-INJECTION_RISK"
)

Security codes (SEC-*) 

const (
	CodeSCALEConcern     ReasonCode = "SCALE-CONCERN"
	CodeSCALEPerformance ReasonCode = "SCALE-PERFORMANCE"
	CodeSCALENoCapacity  ReasonCode = "SCALE-NO_CAPACITY"
	CodeSCALESPOF        ReasonCode = "SCALE-SPOF"
	CodeSCALENoRateLimit ReasonCode = "SCALE-NO_RATE_LIMIT"
	CodeSCALENoCache     ReasonCode = "SCALE-NO_CACHE"
	CodeSCALEBlockingOp  ReasonCode = "SCALE-BLOCKING_OP"
)

Scalability codes (SCALE-*) 

const (
	CodeINFRANoDeploy    ReasonCode = "INFRA-NO_DEPLOY"
	CodeINFRANoMonitor   ReasonCode = "INFRA-NO_MONITOR"
	CodeINFRANoAlert     ReasonCode = "INFRA-NO_ALERT"
	CodeINFRANoRecovery  ReasonCode = "INFRA-NO_RECOVERY"
	CodeINFRANoBackup    ReasonCode = "INFRA-NO_BACKUP"
	CodeINFRANoRunbook   ReasonCode = "INFRA-NO_RUNBOOK"
	CodeINFRANoRollback  ReasonCode = "INFRA-NO_ROLLBACK"
	CodeINFRAEnvMismatch ReasonCode = "INFRA-ENV_MISMATCH"
)

Infrastructure codes (INFRA-*) 

const (
	CodeDOCInsufficient ReasonCode = "DOC-INSUFFICIENT"
	CodeDOCOutdated     ReasonCode = "DOC-OUTDATED"
	CodeDOCNoDiagram    ReasonCode = "DOC-NO_DIAGRAM"
	CodeDOCNoExamples   ReasonCode = "DOC-NO_EXAMPLES"
	CodeDOCInconsistent ReasonCode = "DOC-INCONSISTENT"
)

Documentation codes (DOC-*) 

const (
	CodeSCOPECreep      ReasonCode = "SCOPE-CREEP"
	CodeSCOPEUnbounded  ReasonCode = "SCOPE-UNBOUNDED"
	CodeSCOPENoConstr   ReasonCode = "SCOPE-NO_CONSTRAINTS"
	CodeSCOPENoNonGoals ReasonCode = "SCOPE-NO_NON_GOALS"
	CodeSCOPEMVPUnclear ReasonCode = "SCOPE-MVP_UNCLEAR"
	CodeSCOPENoTimeline ReasonCode = "SCOPE-NO_TIMELINE"
)

Scope codes (SCOPE-*) 

const (
	CodeUXNoARIA        ReasonCode = "UX-NO_ARIA"
	CodeUXNoErrorState  ReasonCode = "UX-NO_ERROR_STATE"
	CodeUXNoLoading     ReasonCode = "UX-NO_LOADING"
	CodeUXNoEmpty       ReasonCode = "UX-NO_EMPTY"
	CodeUXNoResponsive  ReasonCode = "UX-NO_RESPONSIVE"
	CodeUXNoKeyboard    ReasonCode = "UX-NO_KEYBOARD"
	CodeUXIncompleteNav ReasonCode = "UX-INCOMPLETE_NAV"
	CodeUXNoFeedback    ReasonCode = "UX-NO_FEEDBACK"
)

UX codes (UX-*) 

const (
	CodeOther ReasonCode = "OTHER"
)

Generic codes

func GetBlockingCodes added in v0.9.0 

func GetBlockingCodes(findings []Finding) []ReasonCode

GetBlockingCodes returns the reason codes from blocking findings.

func GetReasonCodesByCategory added in v0.9.0 

func GetReasonCodesByCategory(category string) []ReasonCode

GetReasonCodesByCategory returns all reason codes in a category.

func GetReasonCodesBySpecType added in v0.9.0 

func GetReasonCodesBySpecType(specType string) []ReasonCode

GetReasonCodesBySpecType returns reason codes applicable to a spec type.

func NormalizeCode added in v0.9.0 

func NormalizeCode(code ReasonCode) ReasonCode

NormalizeCode converts legacy codes to new format.

type ReasonCodeInfo added in v0.9.0 

type ReasonCodeInfo struct {
	// Code is the reason code identifier.
	Code ReasonCode `json:"code"`

	// Category is the prefix category (REQ, SEC, ARCH, etc.).
	Category string `json:"category"`

	// Description explains what this code means.
	Description string `json:"description"`

	// DefaultSeverity is the typical severity for this issue.
	DefaultSeverity Severity `json:"defaultSeverity"`

	// RepairPrompt is the AI prompt for automated repair.
	// This should be a clear instruction that an LLM can follow
	// to fix the issue in the spec document.
	RepairPrompt string `json:"repairPrompt"`

	// RequiresHuman indicates if human review is needed after AI repair.
	// True for security-critical, business-critical, or subjective issues.
	RequiresHuman bool `json:"requiresHuman"`

	// SpecTypes lists which spec types this code applies to.
	SpecTypes []string `json:"specTypes,omitempty"`
}

ReasonCodeInfo provides metadata about a reason code including information needed for AI-assisted automated repair.

func GetReasonCodeInfo added in v0.9.0 

func GetReasonCodeInfo(code ReasonCode) *ReasonCodeInfo

GetReasonCodeInfo returns the info for a reason code, or nil if not found. Handles both legacy and new code formats.

type ReferenceData 

type ReferenceData struct {
	// ID is the unique identifier for this reference.
	ID string `json:"id,omitempty"`

	// Input is the input/prompt that produced the reference output.
	Input string `json:"input,omitempty"`

	// ExpectedOutput is the gold/reference output.
	ExpectedOutput string `json:"expected_output,omitempty"`

	// ExpectedOutputs allows multiple acceptable outputs.
	ExpectedOutputs []string `json:"expected_outputs,omitempty"`

	// Context provides additional context (e.g., retrieved documents for RAG).
	Context []string `json:"context,omitempty"`

	// Annotations are human-provided labels or scores.
	Annotations []Annotation `json:"annotations,omitempty"`

	// Source indicates where this reference came from.
	Source string `json:"source,omitempty"`

	// Tags categorize or filter references.
	Tags []string `json:"tags,omitempty"`

	// Metadata contains additional reference data.
	Metadata map[string]any `json:"metadata,omitempty"`
}

ReferenceData contains ground truth or expected data for evaluation. This enables reference-based evaluation where outputs are compared against known-good examples.

func NewReferenceData 

func NewReferenceData(input, expectedOutput string) *ReferenceData

NewReferenceData creates a new reference data item.

func (*ReferenceData) WithAnnotation 

func (r *ReferenceData) WithAnnotation(name string, score float64, annotatorID string) *ReferenceData

WithAnnotation adds a human annotation.

func (*ReferenceData) WithContext 

func (r *ReferenceData) WithContext(ctx ...string) *ReferenceData

WithContext adds context documents.

type ReferenceDataset 

type ReferenceDataset struct {
	// ID is the unique identifier for this dataset.
	ID string `json:"id"`

	// Name is the display name.
	Name string `json:"name"`

	// Description explains what this dataset contains.
	Description string `json:"description,omitempty"`

	// Version tracks dataset iterations.
	Version string `json:"version,omitempty"`

	// Items are the reference data items.
	Items []ReferenceData `json:"items"`

	// Tags categorize the dataset.
	Tags []string `json:"tags,omitempty"`

	// Metadata contains additional dataset info.
	Metadata map[string]any `json:"metadata,omitempty"`
}

ReferenceDataset is a collection of reference data items.

func NewReferenceDataset 

func NewReferenceDataset(id, name string) *ReferenceDataset

NewReferenceDataset creates a new reference dataset.

func (*ReferenceDataset) AddItem 

func (d *ReferenceDataset) AddItem(item ReferenceData)

AddItem adds a reference data item to the dataset.

func (*ReferenceDataset) GetByID 

func (d *ReferenceDataset) GetByID(id string) *ReferenceData

GetByID retrieves a reference item by ID.

type ReportMetadata 

type ReportMetadata struct {
	// Document is the filename or path being evaluated.
	Document string `json:"document"`

	// DocumentID is the document identifier (e.g., PRD ID).
	DocumentID string `json:"documentId,omitempty"`

	// DocumentTitle is the document title.
	DocumentTitle string `json:"documentTitle,omitempty"`

	// DocumentVersion is the document version.
	DocumentVersion string `json:"documentVersion,omitempty"`

	// GeneratedAt is when the report was created.
	GeneratedAt time.Time `json:"generatedAt"`

	// GeneratedBy identifies what created this report.
	GeneratedBy string `json:"generatedBy,omitempty"`

	// ReviewerID identifies the reviewer (agent or human).
	ReviewerID string `json:"reviewerId,omitempty"`
}

ReportMetadata contains report identification.

type Rubric 

type Rubric struct {
	// Schema is the JSON Schema URL.
	Schema string `json:"$schema,omitempty"`

	// SchemaVersion is the evaluation schema version (e.g., "v2").
	// Used for backwards compatibility.
	SchemaVersion string `json:"schemaVersion,omitempty"`

	// Metadata contains report identification and audit info.
	Metadata ReportMetadata `json:"metadata"`

	// ReviewType identifies the type of review (prd, arb, security, article, etc.).
	ReviewType string `json:"reviewType"`

	// Judge contains metadata about the LLM judge.
	Judge *JudgeMetadata `json:"judge,omitempty"`

	// RubricID references the rubric used for scoring.
	RubricID string `json:"rubricId,omitempty"`

	// RubricVersion is the version of the rubric used.
	RubricVersion string `json:"rubricVersion,omitempty"`

	// Reference contains gold/expected data for comparison.
	Reference *ReferenceData `json:"reference,omitempty"`

	// IntScore is the overall 1-5 integer score.
	// Preferred for LLM judges as they are unreliable at finer granularity.
	IntScore IntegerScore `json:"intScore,omitempty"`

	// Confidence is the overall confidence in the evaluation (0.0-1.0).
	// Low confidence evaluations may be routed to human review.
	Confidence float64 `json:"confidence,omitempty"`

	// Pass is an explicit pass/fail gate, orthogonal to score.
	// A spec can have a high score but still fail due to blocking issues.
	Pass bool `json:"pass"`

	// Blocking contains reason codes that caused failure.
	// Empty if Pass is true.
	Blocking []ReasonCode `json:"blocking,omitempty"`

	// Categories contains results for each evaluation dimension.
	Categories []CategoryResult `json:"categories"`

	// Findings are all issues discovered during evaluation.
	Findings []Finding `json:"findings"`

	// PassCriteria defines the requirements for approval.
	PassCriteria PassCriteria `json:"passCriteria"`

	// Decision is the evaluation outcome.
	Decision Decision `json:"decision"`

	// OverallDecision is a simplified pass/conditional/fail status.
	OverallDecision string `json:"overallDecision"`

	// NextSteps provides actionable guidance.
	NextSteps NextSteps `json:"nextSteps"`

	// Summary is the overall assessment.
	Summary string `json:"summary"`

	// Extensions contains domain-specific metadata.
	// Use this to store custom data without modifying the core schema.
	// Example: {"coverage": {...}, "metrics": {...}}
	Extensions map[string]any `json:"extensions,omitempty"`
}

Rubric is the detailed rubric-based evaluation report for LLM-as-Judge reviews.

func NewRubric 

func NewRubric(reviewType, document string) *Rubric

NewRubric creates a new rubric-based evaluation report.

func (*Rubric) AddBlocking added in v0.9.0 

func (r *Rubric) AddBlocking(code ReasonCode) *Rubric

AddBlocking adds a blocking reason code.

func (*Rubric) AddCategoryResult 

func (r *Rubric) AddCategoryResult(cr CategoryResult)

AddCategoryResult adds a category result.

func (*Rubric) AddFinding 

func (r *Rubric) AddFinding(f Finding)

AddFinding adds a finding.

func (*Rubric) CollectBlockingCodes added in v0.9.0 

func (r *Rubric) CollectBlockingCodes() []ReasonCode

CollectBlockingCodes gathers all blocking reason codes from findings.

func (*Rubric) ComputeOverallConfidence added in v0.9.0 

func (r *Rubric) ComputeOverallConfidence() float64

ComputeOverallConfidence calculates the overall confidence from category confidences. Uses minimum confidence across all categories (weakest link).

func (*Rubric) ComputeOverallIntScore added in v0.9.0 

func (r *Rubric) ComputeOverallIntScore(rubricSet *RubricSet) IntegerScore

ComputeOverallIntScore calculates the overall integer score from category scores. Uses weighted average of category IntScores.

func (*Rubric) Evaluate 

func (r *Rubric) Evaluate(rubricSet *RubricSet) Decision

Evaluate computes the decision based on findings and category results.

func (*Rubric) Finalize 

func (r *Rubric) Finalize(rubricSet *RubricSet, rerunCommand string)

Finalize computes all derived fields.

func (*Rubric) GenerateNextSteps 

func (r *Rubric) GenerateNextSteps(rerunCommand string)

GenerateNextSteps creates actionable next steps.

func (*Rubric) GenerateSummary 

func (r *Rubric) GenerateSummary() string

GenerateSummary creates the summary text.

func (*Rubric) GetCategoryResult 

func (r *Rubric) GetCategoryResult(categoryID string) *CategoryResult

GetCategoryResult returns a category result by ID, or nil if not found.

func (*Rubric) GetCoverage added in v0.9.0 

func (r *Rubric) GetCoverage() *CoverageReport

GetCoverage retrieves coverage data from a Rubric's extensions. Returns nil if not set or if type assertion fails.

func (*Rubric) GetExtension added in v0.9.0 

func (r *Rubric) GetExtension(key string) any

GetExtension retrieves an extension value by key. Returns nil if not found.

func (*Rubric) HasExtension added in v0.9.0 

func (r *Rubric) HasExtension(key string) bool

HasExtension checks if an extension exists.

func (*Rubric) HasLowConfidence added in v0.9.0 

func (r *Rubric) HasLowConfidence(threshold ...float64) bool

HasLowConfidence returns true if confidence is below the threshold (default 0.7).

func (*Rubric) IsV2 added in v0.9.0 

func (r *Rubric) IsV2() bool

IsV2 returns true if this is a v2 schema evaluation.

func (*Rubric) NeedsHumanReview added in v0.9.0 

func (r *Rubric) NeedsHumanReview(confidenceThreshold ...float64) bool

NeedsHumanReview returns true if this evaluation should be reviewed by a human.

func (*Rubric) SetBlocking added in v0.9.0 

func (r *Rubric) SetBlocking(codes []ReasonCode) *Rubric

SetBlocking sets the blocking reason codes.

func (*Rubric) SetConfidence added in v0.9.0 

func (r *Rubric) SetConfidence(confidence float64) *Rubric

SetConfidence sets the overall confidence value.

func (*Rubric) SetCoverage added in v0.9.0 

func (r *Rubric) SetCoverage(coverage *CoverageReport)

SetCoverage is a convenience method to set coverage on a Rubric.

func (*Rubric) SetExtension added in v0.9.0 

func (r *Rubric) SetExtension(key string, value any)

SetExtension sets a single extension value.

func (*Rubric) SetIntScore added in v0.9.0 

func (r *Rubric) SetIntScore(score IntegerScore) *Rubric

SetIntScore sets the overall integer score.

func (*Rubric) SetJudge 

func (r *Rubric) SetJudge(judge *JudgeMetadata)

SetJudge sets the judge metadata.

func (*Rubric) SetPass added in v0.9.0 

func (r *Rubric) SetPass(pass bool) *Rubric

SetPass sets the pass/fail status.

func (*Rubric) SetPassCriteria 

func (r *Rubric) SetPassCriteria(criteria PassCriteria)

SetPassCriteria sets the pass criteria.

func (*Rubric) SetReference 

func (r *Rubric) SetReference(ref *ReferenceData)

SetReference sets the reference data for comparison.

func (*Rubric) SetRubricInfo 

func (r *Rubric) SetRubricInfo(rubricID, rubricVersion string)

SetRubricInfo sets the rubric ID and version.

type RubricMetadata 

type RubricMetadata struct {
	CreatedAt string   `json:"createdAt,omitempty"`
	Author    string   `json:"author,omitempty"`
	BasedOn   []string `json:"basedOn,omitempty"`
}

RubricMetadata contains additional rubric information.

type RubricPassCriteria 

type RubricPassCriteria struct {
	// MinCategoriesPassing is "all", "all_required", or a number.
	MinCategoriesPassing string `json:"minCategoriesPassing,omitempty"`

	// MaxFindings limits findings by severity.
	MaxFindings *FindingLimits `json:"maxFindingsSeverity,omitempty"`
}

RubricPassCriteria defines requirements for overall pass/fail determination.

type RubricSet 

type RubricSet struct {
	// ID uniquely identifies this rubric set.
	ID string `json:"id"`

	// Name is the human-readable name.
	Name string `json:"name"`

	// Version is the semantic version of this rubric.
	Version string `json:"version"`

	// Description explains what this rubric set evaluates.
	Description string `json:"description,omitempty"`

	// EvaluationType is "analytic" (per-category) or "holistic" (single score).
	// Analytic is recommended for LLM-as-Judge.
	EvaluationType EvaluationType `json:"evaluationType,omitempty"`

	// PassCriteria defines requirements for overall pass/fail.
	PassCriteria RubricPassCriteria `json:"passCriteria"`

	// Categories are the evaluation dimensions.
	Categories []Category `json:"categories"`

	// JudgePromptTemplate is the prompt template for LLM evaluation.
	// Supports placeholders: {content}, {categories}, etc.
	JudgePromptTemplate string `json:"judgePromptTemplate,omitempty"`

	// Metadata contains additional information about the rubric.
	Metadata *RubricMetadata `json:"metadata,omitempty"`
}

RubricSet is a collection of rubrics for a complete evaluation. Follows Go-first principles: Go types are source of truth, JSON Schema generated from them.

func NewRubricSet 

func NewRubricSet(id, name, version string) *RubricSet

NewRubricSet creates a new rubric set with required fields.

func (*RubricSet) AddCategory 

func (rs *RubricSet) AddCategory(cat Category) *RubricSet

AddCategory adds a category to the rubric set.

func (*RubricSet) GetCategory 

func (rs *RubricSet) GetCategory(id string) *Category

GetCategory returns a category by ID, or nil if not found.

func (*RubricSet) GetRequiredCategories 

func (rs *RubricSet) GetRequiredCategories() []Category

GetRequiredCategories returns all required categories.

func (*RubricSet) SetJudgePrompt 

func (rs *RubricSet) SetJudgePrompt(template string) *RubricSet

SetJudgePrompt sets the judge prompt template.

func (*RubricSet) SetMetadata 

func (rs *RubricSet) SetMetadata(meta *RubricMetadata) *RubricSet

SetMetadata sets the rubric metadata.

func (*RubricSet) SetPassCriteria 

func (rs *RubricSet) SetPassCriteria(criteria RubricPassCriteria) *RubricSet

SetPassCriteria sets the pass criteria.

func (*RubricSet) ToJSON 

func (rs *RubricSet) ToJSON() ([]byte, error)

ToJSON serializes a rubric set to JSON.

func (*RubricSet) Validate 

func (rs *RubricSet) Validate() []string

Validate checks the rubric for common issues.

type Scale 

type Scale struct {
	// Type is "categorical", "checklist", "binary", or "likert".
	// Categorical with 2-3 options is recommended for LLM-as-Judge.
	// Likert is better for human comparison studies.
	Type ScaleType `json:"type"`

	// Options are the scoring options (for categorical scales).
	Options []ScaleOption `json:"options,omitempty"`

	// RequiredItems are items that must be present (for checklist scales).
	RequiredItems []string `json:"requiredItems,omitempty"`

	// OptionalItems are items that add value (for checklist scales).
	OptionalItems []string `json:"optionalItems,omitempty"`

	// PassingThreshold defines pass criteria (for checklist scales).
	PassingThreshold *ChecklistThreshold `json:"passingThreshold,omitempty"`

	// LikertConfig defines the likert scale (for likert scales).
	LikertConfig *LikertConfig `json:"likertConfig,omitempty"`
}

Scale defines the scoring mechanism for a category.

type ScaleOption 

type ScaleOption struct {
	// Value is the machine-readable value (e.g., "pass", "partial", "fail").
	Value string `json:"value"`

	// Label is the human-readable label.
	Label string `json:"label"`

	// Criteria are specific requirements for this score level.
	Criteria []string `json:"criteria"`
}

ScaleOption is a single option in a categorical scale.

type ScaleType 

type ScaleType string

ScaleType defines the type of scoring scale. 

const (
	// ScaleTypeCategorical uses discrete categories (pass/partial/fail).
	// Recommended for LLM-as-Judge - better calibrated than numeric scales.
	ScaleTypeCategorical ScaleType = "categorical"

	// ScaleTypeChecklist uses a list of required/optional items.
	ScaleTypeChecklist ScaleType = "checklist"

	// ScaleTypeBinary is simple pass/fail.
	ScaleTypeBinary ScaleType = "binary"

	// ScaleTypeLikert uses a numeric scale (e.g., 1-5).
	// Better for human comparison and inter-rater reliability studies.
	// Scores are mapped to categorical (pass/partial/fail) for decisions.
	ScaleTypeLikert ScaleType = "likert"
)

type ScoreValue 

type ScoreValue string

ScoreValue represents a categorical score value. 

const (
	ScorePass    ScoreValue = "pass"
	ScorePartial ScoreValue = "partial"
	ScoreFail    ScoreValue = "fail"
)

func LikertToCategorical 

func LikertToCategorical(score int, config *LikertConfig) ScoreValue

LikertToCategorical converts a Likert score to categorical (pass/partial/fail).

func (ScoreValue) Icon 

func (s ScoreValue) Icon() string

Icon returns the emoji icon for the score.

func (ScoreValue) IsFailing 

func (s ScoreValue) IsFailing() bool

IsFailing returns true if this score is failing.

func (ScoreValue) IsPartial 

func (s ScoreValue) IsPartial() bool

IsPartial returns true if this score is partial.

func (ScoreValue) IsPassing 

func (s ScoreValue) IsPassing() bool

IsPassing returns true if this score is considered passing.

type Severity 

type Severity string

Severity represents the severity level of a finding. Based on InfoSec severity classifications. 

const (
	SeverityCritical Severity = "critical" // Blocks approval, must fix
	SeverityHigh     Severity = "high"     // Blocks approval, must fix
	SeverityMedium   Severity = "medium"   // Should fix before approval
	SeverityLow      Severity = "low"      // Nice to fix
	SeverityInfo     Severity = "info"     // Informational only
)

func AllSeverities 

func AllSeverities() []Severity

AllSeverities returns all severity levels in order of severity.

func (Severity) Icon 

func (s Severity) Icon() string

Icon returns the emoji icon for the severity.

func (Severity) IsBlocking 

func (s Severity) IsBlocking() bool

IsBlocking returns true if this severity blocks approval.

func (Severity) Weight 

func (s Severity) Weight() int

Weight returns a numeric weight for sorting (higher = more severe).

type TokenUsage 

type TokenUsage struct {
	// InputTokens is the number of input/prompt tokens.
	InputTokens int `json:"input_tokens"`

	// OutputTokens is the number of output/completion tokens.
	OutputTokens int `json:"output_tokens"`

	// TotalTokens is the total tokens used.
	TotalTokens int `json:"total_tokens"`

	// CacheReadTokens is tokens read from cache (if applicable).
	CacheReadTokens int `json:"cache_read_tokens,omitempty"`

	// CacheWriteTokens is tokens written to cache (if applicable).
	CacheWriteTokens int `json:"cache_write_tokens,omitempty"`
}

TokenUsage tracks token consumption for an evaluation.

type ValidationIssue added in v0.7.0 

type ValidationIssue struct {
	// Path is the JSON path to the problematic field (e.g., "categories[0].score").
	Path string `json:"path"`

	// Code is a machine-readable error code.
	Code string `json:"code"`

	// Message describes the issue.
	Message string `json:"message"`

	// Severity indicates how serious the issue is.
	Severity ValidationSeverity `json:"severity"`

	// ActualValue is the invalid value found.
	ActualValue string `json:"actualValue,omitempty"`

	// AllowedValues lists valid options (for enum violations).
	AllowedValues []string `json:"allowedValues,omitempty"`
}

ValidationIssue represents a single validation problem.

type ValidationResult added in v0.7.0 

type ValidationResult struct {
	// Valid is true if no errors were found (warnings allowed).
	Valid bool `json:"valid"`

	// Issues contains all validation problems found.
	Issues []ValidationIssue `json:"issues"`

	// ErrorCount is the number of error-level issues.
	ErrorCount int `json:"errorCount"`

	// WarningCount is the number of warning-level issues.
	WarningCount int `json:"warningCount"`
}

ValidationResult contains all validation issues found.

func ValidateReport added in v0.7.0 

func ValidateReport(r *Rubric) *ValidationResult

ValidateReport validates a rubric evaluation report for correctness. It checks: - All enum values are valid (scores, severities, decision status) - Required fields are present - Counts are accurate - Decision is consistent with findings/categories

func ValidateRubricSetV2 added in v0.7.0 

func ValidateRubricSetV2(rs *RubricSet) *ValidationResult

ValidateRubricSet validates a rubric definition (not a report). Returns a ValidationResult instead of []string for consistency.

func (*ValidationResult) HasErrors added in v0.7.0 

func (r *ValidationResult) HasErrors() bool

HasErrors returns true if there are any error-level issues.

func (*ValidationResult) HasWarnings added in v0.7.0 

func (r *ValidationResult) HasWarnings() bool

HasWarnings returns true if there are any warning-level issues.

func (*ValidationResult) String added in v0.7.0 

func (r *ValidationResult) String() string

String returns a human-readable summary.

type ValidationSeverity added in v0.7.0 

type ValidationSeverity string

ValidationSeverity indicates the severity of a validation issue. 

const (
	// ValidationError is a fatal issue that must be fixed.
	ValidationError ValidationSeverity = "error"

	// ValidationWarning is a non-fatal issue that should be fixed.
	ValidationWarning ValidationSeverity = "warning"
)

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.