niceeval 0.18.0 → 0.19.0

package/dist/experiment/host/types.d.ts

@@ -1,4 +1,4 @@
-import type { ResolvedModelSlots } from "../../model-slots.js";
+import type { ResolvedModelRoles } from "../../model-roles.js";
 import { type Effect } from "effect";
 import type { Config } from "../../types.js";
 import type { FeedbackCoordinator } from "../../runner/feedback/coordinator.js";
@@ -38,7 +38,7 @@ export interface ExperimentHostEvalSummary {
     readonly evaluationKind: "pass" | "score";
 }
 export interface ExperimentHostExperimentSummary {
-    readonly models?: ResolvedModelSlots;
+    readonly models?: ResolvedModelRoles;
     readonly id: string;
     readonly description?: string;
     readonly adapter: string;
@@ -46,8 +46,8 @@ export interface ExperimentHostExperimentSummary {
     readonly attempts: number;
     readonly evalIds: readonly string[];
     readonly labels: Readonly<Record<string, string | number>>;
-    readonly hasSetup: boolean;
-    readonly hasTeardown: boolean;
+    readonly hasBefore: boolean;
+    readonly hasAfter: boolean;
     readonly sharedStateKey?: string;
 }
 export interface ExperimentHostCatalog {
@@ -79,7 +79,7 @@ export interface ExperimentHostRunOverrides {
     readonly keepSandbox?: "failed" | "all";
     readonly maxConcurrency?: number;
     readonly maxBuildConcurrency?: number;
-    readonly maxSetupPrefixConcurrency?: number;
+    readonly maxBeforeCacheConcurrency?: number;
 }
 export interface ExperimentHostCheckRequest extends ExperimentHostSelectionInput {
     readonly config: Config;
@@ -108,7 +108,7 @@ export interface ExperimentHostInvocationShape {
     readonly attempts: number;
     readonly maxConcurrency: number;
     readonly maxBuildConcurrency: number;
-    readonly maxSetupPrefixConcurrency: number;
+    readonly maxBeforeCacheConcurrency: number;
     readonly experimentConcurrency: Readonly<Record<string, number>>;
 }
 export interface ExperimentHostDrySlotTarget {
@@ -217,7 +217,7 @@ export interface ExperimentHostInvocationSummary {
     readonly inputTokens?: number;
     readonly outputTokens?: number;
     readonly estimatedCostUSD?: number;
-    readonly setupPrefixes: SetupPrefixPreparationSummary;
+    readonly beforeCaches: SetupPrefixPreparationSummary;
 }
 export interface ExperimentHostInvocationResult {
     readonly status: "finished";
@@ -313,14 +313,14 @@ export interface ExperimentHostSharedStateEvidence {
     readonly processIdentity: string;
     readonly heartbeatAt: string;
 }
-export interface ExperimentHostTeardownInspectRequest {
+export interface ExperimentHostAfterInspectRequest {
     readonly cwd: string;
     readonly experimentSelector?: string;
     readonly coordinationRoot?: string;
     readonly currentHost: string;
     readonly recoveryKey?: string;
 }
-export type ExperimentHostTeardownInspection = {
+export type ExperimentHostAfterInspection = {
     readonly status: "no-evidence";
     readonly key: string;
 } | ExperimentHostSelectionProblem | {
@@ -332,7 +332,7 @@ export type ExperimentHostTeardownInspection = {
     readonly selectedExperimentId: string;
     readonly evidence: ExperimentHostSharedStateEvidence;
 } | {
-    readonly status: "teardown-required";
+    readonly status: "after-required";
     readonly experimentId: string;
     readonly evidence: ExperimentHostSharedStateEvidence;
 } | {
@@ -341,14 +341,14 @@ export type ExperimentHostTeardownInspection = {
     readonly orphanedRegistrations: number;
     readonly evidence?: ExperimentHostSharedStateEvidence;
 };
-export interface ExperimentHostTeardownRequest extends ExperimentHostTeardownInspectRequest {
-    readonly observer?: ExperimentHostTeardownObserver;
+export interface ExperimentHostAfterRequest extends ExperimentHostAfterInspectRequest {
+    readonly observer?: ExperimentHostAfterObserver;
     readonly signal?: AbortSignal;
     readonly ownerToken?: string;
     readonly confirmOwnerTerminated?: boolean;
     readonly confirmRemoteQuiesced?: boolean;
 }
-export type ExperimentHostTeardownEvent = {
+export type ExperimentHostAfterEvent = {
     readonly type: "experiment-progress";
     readonly experimentId: string;
     readonly detail: string;
@@ -366,10 +366,10 @@ export type ExperimentHostTeardownEvent = {
     readonly durationMs?: number;
     readonly recovery?: true;
 };
-export interface ExperimentHostTeardownObserver {
-    readonly observe: (event: ExperimentHostTeardownEvent) => void;
+export interface ExperimentHostAfterObserver {
+    readonly observe: (event: ExperimentHostAfterEvent) => void;
 }
-export type ExperimentHostTeardownResult = ExperimentHostTeardownInspection | {
+export type ExperimentHostAfterResult = ExperimentHostAfterInspection | {
     readonly status: "completed";
     readonly experiments: readonly {
         readonly experimentId: string;
@@ -405,9 +405,9 @@ export interface ExperimentHostHighLevelSDK {
         readonly plan: (input: ExperimentHostRenameRequest) => Effect.Effect<ExperimentHostRenamePlan, ExperimentHostError>;
         readonly apply: (input: ExperimentHostRenameRequest) => Effect.Effect<ExperimentHostRenameResult, ExperimentHostError>;
     };
-    readonly teardown: {
-        readonly inspect: (input: ExperimentHostTeardownInspectRequest) => Effect.Effect<ExperimentHostTeardownInspection, ExperimentHostError>;
-        readonly run: (input: ExperimentHostTeardownRequest) => Effect.Effect<ExperimentHostTeardownResult, ExperimentHostError>;
+    readonly after: {
+        readonly inspect: (input: ExperimentHostAfterInspectRequest) => Effect.Effect<ExperimentHostAfterInspection, ExperimentHostError>;
+        readonly run: (input: ExperimentHostAfterRequest) => Effect.Effect<ExperimentHostAfterResult, ExperimentHostError>;
     };
     readonly accept: (input: ExperimentHostAcceptRequest) => Effect.Effect<readonly ExperimentHostAcceptedAttempt[], ExperimentHostError>;
     readonly acceptRun: {

package/dist/index.cjs

@@ -1,6 +1,6 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.defineModelSlots = exports.MigrationRequiredError = exports.failureClassOf = exports.EvalFatalError = exports.ExperimentFatalError = exports.defineAdapterContract = exports.defineAdapter = exports.defineExperiment = exports.defineConfig = exports.defineEvalGroup = exports.defineScoreEval = exports.defineEval = void 0;
+exports.defineModelRoles = exports.MigrationRequiredError = exports.failureClassOf = exports.EvalFatalError = exports.ExperimentFatalError = exports.defineAdapterContract = exports.defineAdapter = exports.defineExperiment = exports.defineConfig = exports.defineEvalGroup = exports.defineScoreEval = exports.defineEval = void 0;
 // niceeval 公开导出(import { … } from "niceeval")。
 // Agent/Adapter 相关见 "niceeval/adapter";Sandbox 相关见 "niceeval/sandbox"。
 var define_ts_1 = require("./define.cjs");
@@ -20,6 +20,6 @@ Object.defineProperty(exports, "EvalFatalError", { enumerable: true, get: functi
 Object.defineProperty(exports, "failureClassOf", { enumerable: true, get: function () { return failure_class_ts_1.failureClassOf; } });
 var index_ts_1 = require("./error-assistance/index.cjs");
 Object.defineProperty(exports, "MigrationRequiredError", { enumerable: true, get: function () { return index_ts_1.MigrationRequiredError; } });
-var model_slots_ts_1 = require("./model-slots.cjs");
-Object.defineProperty(exports, "defineModelSlots", { enumerable: true, get: function () { return model_slots_ts_1.defineModelSlots; } });
+var model_roles_ts_1 = require("./model-roles.cjs");
+Object.defineProperty(exports, "defineModelRoles", { enumerable: true, get: function () { return model_roles_ts_1.defineModelRoles; } });
 //# sourceMappingURL=index.cjs.map

package/dist/index.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["index.ts"],"names":[],"mappings":";;;AAAA,+CAA+C;AAC/C,uEAAuE;AACvE,yCAA2G;AAAlG,uGAAA,UAAU,OAAA;AAAE,4GAAA,eAAe,OAAA;AAAE,4GAAA,eAAe,OAAA;AAAE,yGAAA,YAAY,OAAA;AAAE,6GAAA,gBAAgB,OAAA;AACrF,2CAAoE;AAA3D,2GAAA,aAAa,OAAA;AAAE,mHAAA,qBAAqB,OAAA;AAO7C,6CAA6C;AAC7C,6EAA6E;AAC7E,8DAAiG;AAAxF,wHAAA,oBAAoB,OAAA;AAAE,kHAAA,cAAc,OAAA;AAAE,kHAAA,cAAc,OAAA;AAO7D,wDAAiH;AAAxG,kHAAA,sBAAsB,OAAA;AAS/B,mDAAoD;AAA3C,kHAAA,gBAAgB,OAAA","sourcesContent":["// niceeval 公开导出(import { … } from \"niceeval\")。\n// Agent/Adapter 相关见 \"niceeval/adapter\";Sandbox 相关见 \"niceeval/sandbox\"。\nexport { defineEval, defineScoreEval, defineEvalGroup, defineConfig, defineExperiment } from \"./define.ts\";\nexport { defineAdapter, defineAdapterContract } from \"./adapter.ts\";\nexport type { JudgeAnchor, JudgeDefinition, JudgeOptions } from \"./assertions/judge.ts\";\nexport type { ManagedScoreMatchOptions, ScoreMatch, ScoreMatchAnchor, ScoreMatchContext, ScoreMatchLlmFailure, ScoreMatchResult } from \"./assertions/match.ts\";\nexport type { AssertionEntryId } from \"./assertions/identity.ts\";\nexport { type ScoreMatchAudit, type ScoreMatchAuditV3, type ScoreMatchAuditImageContent, type ScoreMatchAuditEnvelope, type ScoreMatchAuditReadResult, } from \"./assertions/score-match-audit.ts\";\nexport type { AgentSendWindowIdentity, AgentWorkspaceDiffEndpoint, AgentWorkspaceDiffHunks, AgentWorkspaceDiffPolicy, AgentWorkspaceDiff, AgentWorkspaceDiffWindow, AgentWorkspaceDiffWindowChange, } from \"./assertions/workspace-diff.ts\";\nexport type { AssertionCheck, ScoreFunction, PolymorphicBooleanAssertionHandle, PolymorphicMeasurementAssertionHandle, PostRunBooleanAssertionHandle, } from \"./assertions/api.ts\";\n// 执行失败分类:抛出点糖衣类(声明死因波及多远)+ 结构守卫。判据、分类链与止损语义见\n// docs/feature/error-classification/README.md;`niceeval/adapter` 复导出同一份词表类型。\nexport { ExperimentFatalError, EvalFatalError, failureClassOf } from \"./shared/failure-class.ts\";\n// 类型(eval 作者会用到;跑哪个 agent / 用哪个 sandbox 见对应子路径)\nexport type { StreamEvent, ToolName, JsonValue, FlagValue, ExperimentFlags, JsonMatch, Usage, Turn, EvidenceCoverageStatus, EvidenceCoverageEntry, EvidenceCoverage, TurnEvidenceCoverage, TurnInput, InputFile, AnswerValue, InputResponse, RespondAnswer, TurnHandle, SessionHandle, TestContext, ScoreTestContext, Verdict, EvalAuthorFields, EvalInput, ScoreEvalInput, EvalDefinition, AnyEvalDefinition, EvalGroupMember, EvalGroupInput, EvalGroupDefinition, EvaluationKind, EvalDescriptor, ExperimentAuthorFields, SharedStateConfig, ExperimentInput, ExperimentDefinition, ExperimentHook, ExperimentHookContext, Config, LocalizedText, Reporter, ReporterEvent, EvalResult, AcceptedDifference, AcceptedResult, RetryAttemptRecord, InvocationSummary, InvocationShape, DiagnosticRecord, ExperimentRunInfo, SandboxRunInfo, ScriptResult, CommandResult, InputRequest, InputRequestFilter, O11ySummary, TraceSpan, SpanKind, DerivedFacts, FailureClass, FailureScope, AttemptFailureInfo, AttemptFailureClassifier, ExternalCause, ExternalCauseFact, ExternalCauseLink, ExternalCauseScalar, } from \"./types.ts\";\nexport type { Adapter, AdapterAssertionsFactory, AdapterAssertionsFactoryContext, AdapterCleanupContext, AttemptCancellation, AttemptSignal, AdapterContract, AdapterCreateContext, AdapterDefinition, AdapterEvalDefinition, AdapterEvalInput, AdapterIdentity, AdapterImplementation, AdapterImplementationInput, AdapterRuntimeDefinition, AdapterScoreEvalInput, EvalContext, ReservedAdapterContextKey, } from \"./adapter.ts\";\nexport type { ParsedTranscript } from \"./o11y/parsers/index.ts\";\nexport type { JudgePresetOptions, FactualityMaterial, FaithfulnessMaterial, InstructionFollowingMaterial, PairwisePreferenceMaterial } from \"./assertions/judge-presets.ts\";\nexport type { JudgeProvider, JudgeSelection } from \"./judge/index.ts\";\nexport { MigrationRequiredError, type MigrationOccurrence, type ErrorSource } from \"./error-assistance/index.ts\";\nexport type { AdapterUsageInput, AdapterUsageSeal } from \"./adapter-usage.ts\";\nexport type { AdapterAttachmentInput, AdapterAttachmentReceipt } from \"./adapter-attachments.ts\";\nexport type { ExecutionDisplayBlock, ExecutionTraceActor, ExecutionTraceEvent, ExecutionTraceEvidence, ExecutionTraceInput, ExecutionTraceLimitation, ExecutionTraceLink, ExecutionTraceReceipt, ExecutionTraceScope, ExecutionTraceScopeMembership, ExecutionTraceSource, ExecutionTraceTime, TraceJson, } from \"./adapter-execution-trace.ts\";\nexport type { MaterialMatch, ContextBooleanMatch, ContextScoreMatch, MatchContext, ReadonlyMaterial, MatchFact, MaterialCollection, MaterialItem, MaterialCaptureBudget, FactCaptureBudget } from \"./assertions/context-match.ts\";\nexport type { AttemptUsageSnapshot, AdapterAttemptUsageSnapshot } from \"./o11y/adapter-usage-projection.ts\";\nexport type { EvalUsage, EvalCacheRead } from \"./o11y/eval-usage.ts\";\nexport type { AgentMatchContext } from \"./context/assert-first.ts\";\nexport type { ModelSlotSelection, ResolvedModelSlot, ResolvedModelSlots } from \"./model-slots.ts\";\nexport { defineModelSlots } from \"./model-slots.ts\";\nexport type { CacheReadReport, CacheReadSummary } from \"./o11y/cache-read.ts\";\n"]}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["index.ts"],"names":[],"mappings":";;;AAAA,+CAA+C;AAC/C,uEAAuE;AACvE,yCAA2G;AAAlG,uGAAA,UAAU,OAAA;AAAE,4GAAA,eAAe,OAAA;AAAE,4GAAA,eAAe,OAAA;AAAE,yGAAA,YAAY,OAAA;AAAE,6GAAA,gBAAgB,OAAA;AACrF,2CAAoE;AAA3D,2GAAA,aAAa,OAAA;AAAE,mHAAA,qBAAqB,OAAA;AAO7C,6CAA6C;AAC7C,6EAA6E;AAC7E,8DAAiG;AAAxF,wHAAA,oBAAoB,OAAA;AAAE,kHAAA,cAAc,OAAA;AAAE,kHAAA,cAAc,OAAA;AAO7D,wDAAiH;AAAxG,kHAAA,sBAAsB,OAAA;AAS/B,mDAAoD;AAA3C,kHAAA,gBAAgB,OAAA","sourcesContent":["// niceeval 公开导出(import { … } from \"niceeval\")。\n// Agent/Adapter 相关见 \"niceeval/adapter\";Sandbox 相关见 \"niceeval/sandbox\"。\nexport { defineEval, defineScoreEval, defineEvalGroup, defineConfig, defineExperiment } from \"./define.ts\";\nexport { defineAdapter, defineAdapterContract } from \"./adapter.ts\";\nexport type { JudgeAnchor, JudgeDefinition, JudgeOptions } from \"./assertions/judge.ts\";\nexport type { ManagedScoreMatchOptions, ScoreMatch, ScoreMatchAnchor, ScoreMatchContext, ScoreMatchLlmFailure, ScoreMatchResult } from \"./assertions/match.ts\";\nexport type { AssertionEntryId } from \"./assertions/identity.ts\";\nexport { type ScoreMatchAudit, type ScoreMatchAuditV3, type ScoreMatchAuditImageContent, type ScoreMatchAuditEnvelope, type ScoreMatchAuditReadResult, } from \"./assertions/score-match-audit.ts\";\nexport type { AgentSendWindowIdentity, AgentWorkspaceDiffEndpoint, AgentWorkspaceDiffHunks, AgentWorkspaceDiffPolicy, AgentWorkspaceDiff, AgentWorkspaceDiffWindow, AgentWorkspaceDiffWindowChange, } from \"./assertions/workspace-diff.ts\";\nexport type { AssertionCheck, ScoreFunction, PolymorphicBooleanAssertionHandle, PolymorphicMeasurementAssertionHandle, PostRunBooleanAssertionHandle, } from \"./assertions/api.ts\";\n// 执行失败分类:抛出点糖衣类(声明死因波及多远)+ 结构守卫。判据、分类链与止损语义见\n// docs/feature/error-classification/README.md;`niceeval/adapter` 复导出同一份词表类型。\nexport { ExperimentFatalError, EvalFatalError, failureClassOf } from \"./shared/failure-class.ts\";\n// 类型(eval 作者会用到;跑哪个 agent / 用哪个 sandbox 见对应子路径)\nexport type { StreamEvent, ToolName, JsonValue, FlagValue, ExperimentFlags, JsonMatch, Usage, Turn, EvidenceCoverageStatus, EvidenceCoverageEntry, EvidenceCoverage, TurnEvidenceCoverage, TurnInput, InputFile, AnswerValue, InputResponse, RespondAnswer, TurnHandle, SessionHandle, TestContext, ScoreTestContext, Verdict, EvalAuthorFields, EvalInput, ScoreEvalInput, EvalDefinition, AnyEvalDefinition, EvalGroupMember, EvalGroupInput, EvalGroupDefinition, EvaluationKind, EvalDescriptor, ExperimentAuthorFields, SharedStateConfig, SandboxBeforeCache, SandboxCacheConfig, ExperimentInput, ExperimentDefinition, ExperimentHook, ExperimentHookContext, Config, LocalizedText, Reporter, ReporterEvent, EvalResult, AcceptedDifference, AcceptedResult, RetryAttemptRecord, InvocationSummary, InvocationShape, DiagnosticRecord, ExperimentRunInfo, SandboxRunInfo, ScriptResult, CommandResult, InputRequest, InputRequestFilter, ObservabilitySummary, TraceSpan, SpanKind, DerivedFacts, FailureClass, FailureScope, AttemptFailureInfo, ExternalCause, ExternalCauseFact, ExternalCauseLink, ExternalCauseScalar, } from \"./types.ts\";\nexport type { Adapter, AdapterAssertionsFactory, AdapterAssertionsFactoryContext, AdapterCleanupContext, AttemptCancellation, AttemptSignal, AdapterContract, AdapterCreateContext, AdapterDefinition, AdapterEvalDefinition, AdapterEvalInput, AdapterIdentity, AdapterImplementation, AdapterImplementationInput, AdapterRuntimeDefinition, AdapterScoreEvalInput, EvalContext, ReservedAdapterContextKey, } from \"./adapter.ts\";\nexport type { ParsedTranscript } from \"./o11y/parsers/index.ts\";\nexport type { JudgePresetOptions, FactualityMaterial, FaithfulnessMaterial, InstructionFollowingMaterial, PairwisePreferenceMaterial } from \"./assertions/judge-presets.ts\";\nexport type { JudgeProvider, JudgeSelection } from \"./judge/index.ts\";\nexport { MigrationRequiredError, type MigrationOccurrence, type ErrorSource } from \"./error-assistance/index.ts\";\nexport type { AdapterUsageInput, AdapterUsageSeal } from \"./adapter-usage.ts\";\nexport type { AdapterAttachmentInput, AdapterAttachmentReceipt } from \"./adapter-attachments.ts\";\nexport type { ExecutionDisplayBlock, ExecutionTraceActor, ExecutionTraceEvent, ExecutionTraceEvidence, ExecutionTraceInput, ExecutionTraceLimitation, ExecutionTraceLink, ExecutionTraceReceipt, ExecutionTraceScope, ExecutionTraceScopeMembership, ExecutionTraceSource, ExecutionTraceTime, TraceJson, } from \"./adapter-execution-trace.ts\";\nexport type { MaterialMatch, ContextBooleanMatch, ContextScoreMatch, MatchContext, ReadonlyMaterial, MatchFact, MaterialCollection, MaterialItem, MaterialCaptureBudget, FactCaptureBudget } from \"./assertions/context-match.ts\";\nexport type { AttemptUsageSnapshot, AdapterAttemptUsageSnapshot } from \"./o11y/adapter-usage-projection.ts\";\nexport type { EvalUsage, EvalCacheRead } from \"./o11y/eval-usage.ts\";\nexport type { AgentMatchContext } from \"./context/assert-first.ts\";\nexport type { ModelRoleSelection, ResolvedModelRole, ResolvedModelRoles } from \"./model-roles.ts\";\nexport { defineModelRoles } from \"./model-roles.ts\";\nexport type { CacheReadReport, CacheReadSummary } from \"./o11y/cache-read.ts\";\n"]}

package/dist/index.d.cts

@@ -7,7 +7,7 @@ export { type ScoreMatchAudit, type ScoreMatchAuditV3, type ScoreMatchAuditImage
 export type { AgentSendWindowIdentity, AgentWorkspaceDiffEndpoint, AgentWorkspaceDiffHunks, AgentWorkspaceDiffPolicy, AgentWorkspaceDiff, AgentWorkspaceDiffWindow, AgentWorkspaceDiffWindowChange, } from "./assertions/workspace-diff.cjs";
 export type { AssertionCheck, ScoreFunction, PolymorphicBooleanAssertionHandle, PolymorphicMeasurementAssertionHandle, PostRunBooleanAssertionHandle, } from "./assertions/api.cjs";
 export { ExperimentFatalError, EvalFatalError, failureClassOf } from "./shared/failure-class.cjs";
-export type { StreamEvent, ToolName, JsonValue, FlagValue, ExperimentFlags, JsonMatch, Usage, Turn, EvidenceCoverageStatus, EvidenceCoverageEntry, EvidenceCoverage, TurnEvidenceCoverage, TurnInput, InputFile, AnswerValue, InputResponse, RespondAnswer, TurnHandle, SessionHandle, TestContext, ScoreTestContext, Verdict, EvalAuthorFields, EvalInput, ScoreEvalInput, EvalDefinition, AnyEvalDefinition, EvalGroupMember, EvalGroupInput, EvalGroupDefinition, EvaluationKind, EvalDescriptor, ExperimentAuthorFields, SharedStateConfig, ExperimentInput, ExperimentDefinition, ExperimentHook, ExperimentHookContext, Config, LocalizedText, Reporter, ReporterEvent, EvalResult, AcceptedDifference, AcceptedResult, RetryAttemptRecord, InvocationSummary, InvocationShape, DiagnosticRecord, ExperimentRunInfo, SandboxRunInfo, ScriptResult, CommandResult, InputRequest, InputRequestFilter, O11ySummary, TraceSpan, SpanKind, DerivedFacts, FailureClass, FailureScope, AttemptFailureInfo, AttemptFailureClassifier, ExternalCause, ExternalCauseFact, ExternalCauseLink, ExternalCauseScalar, } from "./types.cjs";
+export type { StreamEvent, ToolName, JsonValue, FlagValue, ExperimentFlags, JsonMatch, Usage, Turn, EvidenceCoverageStatus, EvidenceCoverageEntry, EvidenceCoverage, TurnEvidenceCoverage, TurnInput, InputFile, AnswerValue, InputResponse, RespondAnswer, TurnHandle, SessionHandle, TestContext, ScoreTestContext, Verdict, EvalAuthorFields, EvalInput, ScoreEvalInput, EvalDefinition, AnyEvalDefinition, EvalGroupMember, EvalGroupInput, EvalGroupDefinition, EvaluationKind, EvalDescriptor, ExperimentAuthorFields, SharedStateConfig, SandboxBeforeCache, SandboxCacheConfig, ExperimentInput, ExperimentDefinition, ExperimentHook, ExperimentHookContext, Config, LocalizedText, Reporter, ReporterEvent, EvalResult, AcceptedDifference, AcceptedResult, RetryAttemptRecord, InvocationSummary, InvocationShape, DiagnosticRecord, ExperimentRunInfo, SandboxRunInfo, ScriptResult, CommandResult, InputRequest, InputRequestFilter, ObservabilitySummary, TraceSpan, SpanKind, DerivedFacts, FailureClass, FailureScope, AttemptFailureInfo, ExternalCause, ExternalCauseFact, ExternalCauseLink, ExternalCauseScalar, } from "./types.cjs";
 export type { Adapter, AdapterAssertionsFactory, AdapterAssertionsFactoryContext, AdapterCleanupContext, AttemptCancellation, AttemptSignal, AdapterContract, AdapterCreateContext, AdapterDefinition, AdapterEvalDefinition, AdapterEvalInput, AdapterIdentity, AdapterImplementation, AdapterImplementationInput, AdapterRuntimeDefinition, AdapterScoreEvalInput, EvalContext, ReservedAdapterContextKey, } from "./adapter.cjs";
 export type { ParsedTranscript } from "./o11y/parsers/index.cjs";
 export type { JudgePresetOptions, FactualityMaterial, FaithfulnessMaterial, InstructionFollowingMaterial, PairwisePreferenceMaterial } from "./assertions/judge-presets.cjs";
@@ -20,6 +20,6 @@ export type { MaterialMatch, ContextBooleanMatch, ContextScoreMatch, MatchContex
 export type { AttemptUsageSnapshot, AdapterAttemptUsageSnapshot } from "./o11y/adapter-usage-projection.cjs";
 export type { EvalUsage, EvalCacheRead } from "./o11y/eval-usage.cjs";
 export type { AgentMatchContext } from "./context/assert-first.cjs";
-export type { ModelSlotSelection, ResolvedModelSlot, ResolvedModelSlots } from "./model-slots.cjs";
-export { defineModelSlots } from "./model-slots.cjs";
+export type { ModelRoleSelection, ResolvedModelRole, ResolvedModelRoles } from "./model-roles.cjs";
+export { defineModelRoles } from "./model-roles.cjs";
 export type { CacheReadReport, CacheReadSummary } from "./o11y/cache-read.cjs";

package/dist/index.d.mts

@@ -7,7 +7,7 @@ export { type ScoreMatchAudit, type ScoreMatchAuditV3, type ScoreMatchAuditImage
 export type { AgentSendWindowIdentity, AgentWorkspaceDiffEndpoint, AgentWorkspaceDiffHunks, AgentWorkspaceDiffPolicy, AgentWorkspaceDiff, AgentWorkspaceDiffWindow, AgentWorkspaceDiffWindowChange, } from "./assertions/workspace-diff.mjs";
 export type { AssertionCheck, ScoreFunction, PolymorphicBooleanAssertionHandle, PolymorphicMeasurementAssertionHandle, PostRunBooleanAssertionHandle, } from "./assertions/api.mjs";
 export { ExperimentFatalError, EvalFatalError, failureClassOf } from "./shared/failure-class.mjs";
-export type { StreamEvent, ToolName, JsonValue, FlagValue, ExperimentFlags, JsonMatch, Usage, Turn, EvidenceCoverageStatus, EvidenceCoverageEntry, EvidenceCoverage, TurnEvidenceCoverage, TurnInput, InputFile, AnswerValue, InputResponse, RespondAnswer, TurnHandle, SessionHandle, TestContext, ScoreTestContext, Verdict, EvalAuthorFields, EvalInput, ScoreEvalInput, EvalDefinition, AnyEvalDefinition, EvalGroupMember, EvalGroupInput, EvalGroupDefinition, EvaluationKind, EvalDescriptor, ExperimentAuthorFields, SharedStateConfig, ExperimentInput, ExperimentDefinition, ExperimentHook, ExperimentHookContext, Config, LocalizedText, Reporter, ReporterEvent, EvalResult, AcceptedDifference, AcceptedResult, RetryAttemptRecord, InvocationSummary, InvocationShape, DiagnosticRecord, ExperimentRunInfo, SandboxRunInfo, ScriptResult, CommandResult, InputRequest, InputRequestFilter, O11ySummary, TraceSpan, SpanKind, DerivedFacts, FailureClass, FailureScope, AttemptFailureInfo, AttemptFailureClassifier, ExternalCause, ExternalCauseFact, ExternalCauseLink, ExternalCauseScalar, } from "./types.mjs";
+export type { StreamEvent, ToolName, JsonValue, FlagValue, ExperimentFlags, JsonMatch, Usage, Turn, EvidenceCoverageStatus, EvidenceCoverageEntry, EvidenceCoverage, TurnEvidenceCoverage, TurnInput, InputFile, AnswerValue, InputResponse, RespondAnswer, TurnHandle, SessionHandle, TestContext, ScoreTestContext, Verdict, EvalAuthorFields, EvalInput, ScoreEvalInput, EvalDefinition, AnyEvalDefinition, EvalGroupMember, EvalGroupInput, EvalGroupDefinition, EvaluationKind, EvalDescriptor, ExperimentAuthorFields, SharedStateConfig, SandboxBeforeCache, SandboxCacheConfig, ExperimentInput, ExperimentDefinition, ExperimentHook, ExperimentHookContext, Config, LocalizedText, Reporter, ReporterEvent, EvalResult, AcceptedDifference, AcceptedResult, RetryAttemptRecord, InvocationSummary, InvocationShape, DiagnosticRecord, ExperimentRunInfo, SandboxRunInfo, ScriptResult, CommandResult, InputRequest, InputRequestFilter, ObservabilitySummary, TraceSpan, SpanKind, DerivedFacts, FailureClass, FailureScope, AttemptFailureInfo, ExternalCause, ExternalCauseFact, ExternalCauseLink, ExternalCauseScalar, } from "./types.mjs";
 export type { Adapter, AdapterAssertionsFactory, AdapterAssertionsFactoryContext, AdapterCleanupContext, AttemptCancellation, AttemptSignal, AdapterContract, AdapterCreateContext, AdapterDefinition, AdapterEvalDefinition, AdapterEvalInput, AdapterIdentity, AdapterImplementation, AdapterImplementationInput, AdapterRuntimeDefinition, AdapterScoreEvalInput, EvalContext, ReservedAdapterContextKey, } from "./adapter.mjs";
 export type { ParsedTranscript } from "./o11y/parsers/index.mjs";
 export type { JudgePresetOptions, FactualityMaterial, FaithfulnessMaterial, InstructionFollowingMaterial, PairwisePreferenceMaterial } from "./assertions/judge-presets.mjs";
@@ -20,6 +20,6 @@ export type { MaterialMatch, ContextBooleanMatch, ContextScoreMatch, MatchContex
 export type { AttemptUsageSnapshot, AdapterAttemptUsageSnapshot } from "./o11y/adapter-usage-projection.mjs";
 export type { EvalUsage, EvalCacheRead } from "./o11y/eval-usage.mjs";
 export type { AgentMatchContext } from "./context/assert-first.mjs";
-export type { ModelSlotSelection, ResolvedModelSlot, ResolvedModelSlots } from "./model-slots.mjs";
-export { defineModelSlots } from "./model-slots.mjs";
+export type { ModelRoleSelection, ResolvedModelRole, ResolvedModelRoles } from "./model-roles.mjs";
+export { defineModelRoles } from "./model-roles.mjs";
 export type { CacheReadReport, CacheReadSummary } from "./o11y/cache-read.mjs";

package/dist/index.d.ts

@@ -7,7 +7,7 @@ export { type ScoreMatchAudit, type ScoreMatchAuditV3, type ScoreMatchAuditImage
 export type { AgentSendWindowIdentity, AgentWorkspaceDiffEndpoint, AgentWorkspaceDiffHunks, AgentWorkspaceDiffPolicy, AgentWorkspaceDiff, AgentWorkspaceDiffWindow, AgentWorkspaceDiffWindowChange, } from "./assertions/workspace-diff.js";
 export type { AssertionCheck, ScoreFunction, PolymorphicBooleanAssertionHandle, PolymorphicMeasurementAssertionHandle, PostRunBooleanAssertionHandle, } from "./assertions/api.js";
 export { ExperimentFatalError, EvalFatalError, failureClassOf } from "./shared/failure-class.js";
-export type { StreamEvent, ToolName, JsonValue, FlagValue, ExperimentFlags, JsonMatch, Usage, Turn, EvidenceCoverageStatus, EvidenceCoverageEntry, EvidenceCoverage, TurnEvidenceCoverage, TurnInput, InputFile, AnswerValue, InputResponse, RespondAnswer, TurnHandle, SessionHandle, TestContext, ScoreTestContext, Verdict, EvalAuthorFields, EvalInput, ScoreEvalInput, EvalDefinition, AnyEvalDefinition, EvalGroupMember, EvalGroupInput, EvalGroupDefinition, EvaluationKind, EvalDescriptor, ExperimentAuthorFields, SharedStateConfig, ExperimentInput, ExperimentDefinition, ExperimentHook, ExperimentHookContext, Config, LocalizedText, Reporter, ReporterEvent, EvalResult, AcceptedDifference, AcceptedResult, RetryAttemptRecord, InvocationSummary, InvocationShape, DiagnosticRecord, ExperimentRunInfo, SandboxRunInfo, ScriptResult, CommandResult, InputRequest, InputRequestFilter, O11ySummary, TraceSpan, SpanKind, DerivedFacts, FailureClass, FailureScope, AttemptFailureInfo, AttemptFailureClassifier, ExternalCause, ExternalCauseFact, ExternalCauseLink, ExternalCauseScalar, } from "./types.js";
+export type { StreamEvent, ToolName, JsonValue, FlagValue, ExperimentFlags, JsonMatch, Usage, Turn, EvidenceCoverageStatus, EvidenceCoverageEntry, EvidenceCoverage, TurnEvidenceCoverage, TurnInput, InputFile, AnswerValue, InputResponse, RespondAnswer, TurnHandle, SessionHandle, TestContext, ScoreTestContext, Verdict, EvalAuthorFields, EvalInput, ScoreEvalInput, EvalDefinition, AnyEvalDefinition, EvalGroupMember, EvalGroupInput, EvalGroupDefinition, EvaluationKind, EvalDescriptor, ExperimentAuthorFields, SharedStateConfig, SandboxBeforeCache, SandboxCacheConfig, ExperimentInput, ExperimentDefinition, ExperimentHook, ExperimentHookContext, Config, LocalizedText, Reporter, ReporterEvent, EvalResult, AcceptedDifference, AcceptedResult, RetryAttemptRecord, InvocationSummary, InvocationShape, DiagnosticRecord, ExperimentRunInfo, SandboxRunInfo, ScriptResult, CommandResult, InputRequest, InputRequestFilter, ObservabilitySummary, TraceSpan, SpanKind, DerivedFacts, FailureClass, FailureScope, AttemptFailureInfo, ExternalCause, ExternalCauseFact, ExternalCauseLink, ExternalCauseScalar, } from "./types.js";
 export type { Adapter, AdapterAssertionsFactory, AdapterAssertionsFactoryContext, AdapterCleanupContext, AttemptCancellation, AttemptSignal, AdapterContract, AdapterCreateContext, AdapterDefinition, AdapterEvalDefinition, AdapterEvalInput, AdapterIdentity, AdapterImplementation, AdapterImplementationInput, AdapterRuntimeDefinition, AdapterScoreEvalInput, EvalContext, ReservedAdapterContextKey, } from "./adapter.js";
 export type { ParsedTranscript } from "./o11y/parsers/index.js";
 export type { JudgePresetOptions, FactualityMaterial, FaithfulnessMaterial, InstructionFollowingMaterial, PairwisePreferenceMaterial } from "./assertions/judge-presets.js";
@@ -20,6 +20,6 @@ export type { MaterialMatch, ContextBooleanMatch, ContextScoreMatch, MatchContex
 export type { AttemptUsageSnapshot, AdapterAttemptUsageSnapshot } from "./o11y/adapter-usage-projection.js";
 export type { EvalUsage, EvalCacheRead } from "./o11y/eval-usage.js";
 export type { AgentMatchContext } from "./context/assert-first.js";
-export type { ModelSlotSelection, ResolvedModelSlot, ResolvedModelSlots } from "./model-slots.js";
-export { defineModelSlots } from "./model-slots.js";
+export type { ModelRoleSelection, ResolvedModelRole, ResolvedModelRoles } from "./model-roles.js";
+export { defineModelRoles } from "./model-roles.js";
 export type { CacheReadReport, CacheReadSummary } from "./o11y/cache-read.js";

package/dist/index.js

@@ -19,8 +19,8 @@ const __niceevalExport7 = __niceevalCanonical["defineEvalGroup"];
 export { __niceevalExport7 as defineEvalGroup };
 const __niceevalExport8 = __niceevalCanonical["defineExperiment"];
 export { __niceevalExport8 as defineExperiment };
-const __niceevalExport9 = __niceevalCanonical["defineModelSlots"];
-export { __niceevalExport9 as defineModelSlots };
+const __niceevalExport9 = __niceevalCanonical["defineModelRoles"];
+export { __niceevalExport9 as defineModelRoles };
 const __niceevalExport10 = __niceevalCanonical["defineScoreEval"];
 export { __niceevalExport10 as defineScoreEval };
 const __niceevalExport11 = __niceevalCanonical["failureClassOf"];

package/dist/index.mjs

@@ -19,8 +19,8 @@ const __niceevalExport7 = __niceevalCanonical["defineEvalGroup"];
 export { __niceevalExport7 as defineEvalGroup };
 const __niceevalExport8 = __niceevalCanonical["defineExperiment"];
 export { __niceevalExport8 as defineExperiment };
-const __niceevalExport9 = __niceevalCanonical["defineModelSlots"];
-export { __niceevalExport9 as defineModelSlots };
+const __niceevalExport9 = __niceevalCanonical["defineModelRoles"];
+export { __niceevalExport9 as defineModelRoles };
 const __niceevalExport10 = __niceevalCanonical["defineScoreEval"];
 export { __niceevalExport10 as defineScoreEval };
 const __niceevalExport11 = __niceevalCanonical["failureClassOf"];

package/dist/inspection/assertions.cjs

@@ -3,7 +3,7 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.projectAttemptAssertionIndex = projectAttemptAssertionIndex;
 exports.projectAttemptAssertionDetail = projectAttemptAssertionDetail;
 exports.projectAttemptAssertionImage = projectAttemptAssertionImage;
-// @concord-file ne-eval-inspection-assertion-projection
+// @concord-file
 // @concord-implements docs/feature/inspection/README.md
 // @concord-implements docs/feature/assertions/README.md
 const effect_1 = require("effect");