@plumloom/cli 0.1.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/LICENSE +202 -0
- package/NOTICE +2 -0
- package/README.md +71 -0
- package/dist/actions/configured-runs.d.ts +25 -0
- package/dist/actions/configured-runs.d.ts.map +1 -0
- package/dist/actions/configured-runs.js +35 -0
- package/dist/actions/configured-runs.js.map +1 -0
- package/dist/actions/context.d.ts +5 -0
- package/dist/actions/context.d.ts.map +1 -0
- package/dist/actions/context.js +2 -0
- package/dist/actions/context.js.map +1 -0
- package/dist/actions/doctor.d.ts +46 -0
- package/dist/actions/doctor.d.ts.map +1 -0
- package/dist/actions/doctor.js +130 -0
- package/dist/actions/doctor.js.map +1 -0
- package/dist/actions/evaluations.d.ts +87 -0
- package/dist/actions/evaluations.d.ts.map +1 -0
- package/dist/actions/evaluations.js +307 -0
- package/dist/actions/evaluations.js.map +1 -0
- package/dist/actions/gate.d.ts +63 -0
- package/dist/actions/gate.d.ts.map +1 -0
- package/dist/actions/gate.js +272 -0
- package/dist/actions/gate.js.map +1 -0
- package/dist/actions/identity.d.ts +4 -0
- package/dist/actions/identity.d.ts.map +1 -0
- package/dist/actions/identity.js +4 -0
- package/dist/actions/identity.js.map +1 -0
- package/dist/actions/models.d.ts +4 -0
- package/dist/actions/models.d.ts.map +1 -0
- package/dist/actions/models.js +4 -0
- package/dist/actions/models.js.map +1 -0
- package/dist/actions/quality-standards.d.ts +30 -0
- package/dist/actions/quality-standards.d.ts.map +1 -0
- package/dist/actions/quality-standards.js +86 -0
- package/dist/actions/quality-standards.js.map +1 -0
- package/dist/actions/results.d.ts +12 -0
- package/dist/actions/results.d.ts.map +1 -0
- package/dist/actions/results.js +65 -0
- package/dist/actions/results.js.map +1 -0
- package/dist/actions/runs.d.ts +22 -0
- package/dist/actions/runs.d.ts.map +1 -0
- package/dist/actions/runs.js +59 -0
- package/dist/actions/runs.js.map +1 -0
- package/dist/actions/suite-gate.d.ts +24 -0
- package/dist/actions/suite-gate.d.ts.map +1 -0
- package/dist/actions/suite-gate.js +40 -0
- package/dist/actions/suite-gate.js.map +1 -0
- package/dist/actions/suite.d.ts +49 -0
- package/dist/actions/suite.d.ts.map +1 -0
- package/dist/actions/suite.js +68 -0
- package/dist/actions/suite.js.map +1 -0
- package/dist/actions/validation.d.ts +3 -0
- package/dist/actions/validation.d.ts.map +1 -0
- package/dist/actions/validation.js +26 -0
- package/dist/actions/validation.js.map +1 -0
- package/dist/actions/workspaces.d.ts +13 -0
- package/dist/actions/workspaces.d.ts.map +1 -0
- package/dist/actions/workspaces.js +18 -0
- package/dist/actions/workspaces.js.map +1 -0
- package/dist/api/api.d.ts +237 -0
- package/dist/api/api.d.ts.map +1 -0
- package/dist/api/api.js +396 -0
- package/dist/api/api.js.map +1 -0
- package/dist/api/client.d.ts +38 -0
- package/dist/api/client.d.ts.map +1 -0
- package/dist/api/client.js +295 -0
- package/dist/api/client.js.map +1 -0
- package/dist/api/endpoints.d.ts +31 -0
- package/dist/api/endpoints.d.ts.map +1 -0
- package/dist/api/endpoints.js +32 -0
- package/dist/api/endpoints.js.map +1 -0
- package/dist/api/errors.d.ts +22 -0
- package/dist/api/errors.d.ts.map +1 -0
- package/dist/api/errors.js +84 -0
- package/dist/api/errors.js.map +1 -0
- package/dist/api/schemas.d.ts +169 -0
- package/dist/api/schemas.d.ts.map +1 -0
- package/dist/api/schemas.js +194 -0
- package/dist/api/schemas.js.map +1 -0
- package/dist/auth/credential-store.d.ts +9 -0
- package/dist/auth/credential-store.d.ts.map +1 -0
- package/dist/auth/credential-store.js +2 -0
- package/dist/auth/credential-store.js.map +1 -0
- package/dist/auth/credentials.d.ts +27 -0
- package/dist/auth/credentials.d.ts.map +1 -0
- package/dist/auth/credentials.js +81 -0
- package/dist/auth/credentials.js.map +1 -0
- package/dist/auth/keyring-store.d.ts +9 -0
- package/dist/auth/keyring-store.d.ts.map +1 -0
- package/dist/auth/keyring-store.js +57 -0
- package/dist/auth/keyring-store.js.map +1 -0
- package/dist/auth/login.d.ts +28 -0
- package/dist/auth/login.d.ts.map +1 -0
- package/dist/auth/login.js +27 -0
- package/dist/auth/login.js.map +1 -0
- package/dist/auth/password-prompt.d.ts +5 -0
- package/dist/auth/password-prompt.d.ts.map +1 -0
- package/dist/auth/password-prompt.js +10 -0
- package/dist/auth/password-prompt.js.map +1 -0
- package/dist/auth/redact.d.ts +3 -0
- package/dist/auth/redact.d.ts.map +1 -0
- package/dist/auth/redact.js +69 -0
- package/dist/auth/redact.js.map +1 -0
- package/dist/cli.d.ts +3 -0
- package/dist/cli.d.ts.map +1 -0
- package/dist/cli.js +4 -0
- package/dist/cli.js.map +1 -0
- package/dist/commands/executor.d.ts +36 -0
- package/dist/commands/executor.d.ts.map +1 -0
- package/dist/commands/executor.js +1218 -0
- package/dist/commands/executor.js.map +1 -0
- package/dist/commands/program.d.ts +21 -0
- package/dist/commands/program.d.ts.map +1 -0
- package/dist/commands/program.js +588 -0
- package/dist/commands/program.js.map +1 -0
- package/dist/commands/types.d.ts +139 -0
- package/dist/commands/types.d.ts.map +1 -0
- package/dist/commands/types.js +2 -0
- package/dist/commands/types.js.map +1 -0
- package/dist/config.d.ts +12 -0
- package/dist/config.d.ts.map +1 -0
- package/dist/config.js +57 -0
- package/dist/config.js.map +1 -0
- package/dist/configured-run/file-inputs.d.ts +8 -0
- package/dist/configured-run/file-inputs.d.ts.map +1 -0
- package/dist/configured-run/file-inputs.js +160 -0
- package/dist/configured-run/file-inputs.js.map +1 -0
- package/dist/configured-run/validation.d.ts +103 -0
- package/dist/configured-run/validation.d.ts.map +1 -0
- package/dist/configured-run/validation.js +327 -0
- package/dist/configured-run/validation.js.map +1 -0
- package/dist/domain/common.d.ts +6 -0
- package/dist/domain/common.d.ts.map +1 -0
- package/dist/domain/common.js +10 -0
- package/dist/domain/common.js.map +1 -0
- package/dist/domain/types.d.ts +117 -0
- package/dist/domain/types.d.ts.map +1 -0
- package/dist/domain/types.js +2 -0
- package/dist/domain/types.js.map +1 -0
- package/dist/errors/autoeval-error.d.ts +13 -0
- package/dist/errors/autoeval-error.d.ts.map +1 -0
- package/dist/errors/autoeval-error.js +36 -0
- package/dist/errors/autoeval-error.js.map +1 -0
- package/dist/errors/hints.d.ts +3 -0
- package/dist/errors/hints.d.ts.map +1 -0
- package/dist/errors/hints.js +43 -0
- package/dist/errors/hints.js.map +1 -0
- package/dist/gate/decision.d.ts +97 -0
- package/dist/gate/decision.d.ts.map +1 -0
- package/dist/gate/decision.js +436 -0
- package/dist/gate/decision.js.map +1 -0
- package/dist/gate/policy.d.ts +31 -0
- package/dist/gate/policy.d.ts.map +1 -0
- package/dist/gate/policy.js +26 -0
- package/dist/gate/policy.js.map +1 -0
- package/dist/guide/eval-creation-backend.d.ts +20 -0
- package/dist/guide/eval-creation-backend.d.ts.map +1 -0
- package/dist/guide/eval-creation-backend.js +16 -0
- package/dist/guide/eval-creation-backend.js.map +1 -0
- package/dist/guide/eval-guide.d.ts +24 -0
- package/dist/guide/eval-guide.d.ts.map +1 -0
- package/dist/guide/eval-guide.js +219 -0
- package/dist/guide/eval-guide.js.map +1 -0
- package/dist/mcp/context.d.ts +21 -0
- package/dist/mcp/context.d.ts.map +1 -0
- package/dist/mcp/context.js +19 -0
- package/dist/mcp/context.js.map +1 -0
- package/dist/mcp/schemas.d.ts +206 -0
- package/dist/mcp/schemas.d.ts.map +1 -0
- package/dist/mcp/schemas.js +80 -0
- package/dist/mcp/schemas.js.map +1 -0
- package/dist/mcp/server.d.ts +5 -0
- package/dist/mcp/server.d.ts.map +1 -0
- package/dist/mcp/server.js +10 -0
- package/dist/mcp/server.js.map +1 -0
- package/dist/mcp/tools.d.ts +42 -0
- package/dist/mcp/tools.d.ts.map +1 -0
- package/dist/mcp/tools.js +287 -0
- package/dist/mcp/tools.js.map +1 -0
- package/dist/mcp.d.ts +3 -0
- package/dist/mcp.d.ts.map +1 -0
- package/dist/mcp.js +32 -0
- package/dist/mcp.js.map +1 -0
- package/dist/output/big-digits.d.ts +7 -0
- package/dist/output/big-digits.d.ts.map +1 -0
- package/dist/output/big-digits.js +43 -0
- package/dist/output/big-digits.js.map +1 -0
- package/dist/output/clipboard.d.ts +29 -0
- package/dist/output/clipboard.d.ts.map +1 -0
- package/dist/output/clipboard.js +82 -0
- package/dist/output/clipboard.js.map +1 -0
- package/dist/output/colors.d.ts +21 -0
- package/dist/output/colors.d.ts.map +1 -0
- package/dist/output/colors.js +21 -0
- package/dist/output/colors.js.map +1 -0
- package/dist/output/digest.d.ts +165 -0
- package/dist/output/digest.d.ts.map +1 -0
- package/dist/output/digest.js +771 -0
- package/dist/output/digest.js.map +1 -0
- package/dist/output/human.d.ts +95 -0
- package/dist/output/human.d.ts.map +1 -0
- package/dist/output/human.js +643 -0
- package/dist/output/human.js.map +1 -0
- package/dist/output/json.d.ts +2 -0
- package/dist/output/json.d.ts.map +1 -0
- package/dist/output/json.js +5 -0
- package/dist/output/json.js.map +1 -0
- package/dist/output/layout.d.ts +18 -0
- package/dist/output/layout.d.ts.map +1 -0
- package/dist/output/layout.js +64 -0
- package/dist/output/layout.js.map +1 -0
- package/dist/output/markdown.d.ts +16 -0
- package/dist/output/markdown.d.ts.map +1 -0
- package/dist/output/markdown.js +38 -0
- package/dist/output/markdown.js.map +1 -0
- package/dist/output/progress.d.ts +29 -0
- package/dist/output/progress.d.ts.map +1 -0
- package/dist/output/progress.js +136 -0
- package/dist/output/progress.js.map +1 -0
- package/dist/output/safe-text.d.ts +11 -0
- package/dist/output/safe-text.d.ts.map +1 -0
- package/dist/output/safe-text.js +44 -0
- package/dist/output/safe-text.js.map +1 -0
- package/dist/output/scorecard.d.ts +23 -0
- package/dist/output/scorecard.d.ts.map +1 -0
- package/dist/output/scorecard.js +506 -0
- package/dist/output/scorecard.js.map +1 -0
- package/dist/output/section.d.ts +15 -0
- package/dist/output/section.d.ts.map +1 -0
- package/dist/output/section.js +44 -0
- package/dist/output/section.js.map +1 -0
- package/dist/output/spinner.d.ts +24 -0
- package/dist/output/spinner.d.ts.map +1 -0
- package/dist/output/spinner.js +34 -0
- package/dist/output/spinner.js.map +1 -0
- package/dist/output/splash.d.ts +37 -0
- package/dist/output/splash.d.ts.map +1 -0
- package/dist/output/splash.js +169 -0
- package/dist/output/splash.js.map +1 -0
- package/dist/output/table.d.ts +33 -0
- package/dist/output/table.d.ts.map +1 -0
- package/dist/output/table.js +105 -0
- package/dist/output/table.js.map +1 -0
- package/dist/output/trend.d.ts +19 -0
- package/dist/output/trend.d.ts.map +1 -0
- package/dist/output/trend.js +38 -0
- package/dist/output/trend.js.map +1 -0
- package/dist/output/writer.d.ts +14 -0
- package/dist/output/writer.d.ts.map +1 -0
- package/dist/output/writer.js +33 -0
- package/dist/output/writer.js.map +1 -0
- package/dist/polling/clock.d.ts +9 -0
- package/dist/polling/clock.d.ts.map +1 -0
- package/dist/polling/clock.js +23 -0
- package/dist/polling/clock.js.map +1 -0
- package/dist/polling/wait-for-result.d.ts +16 -0
- package/dist/polling/wait-for-result.d.ts.map +1 -0
- package/dist/polling/wait-for-result.js +29 -0
- package/dist/polling/wait-for-result.js.map +1 -0
- package/dist/polling/wait-for-run.d.ts +14 -0
- package/dist/polling/wait-for-run.d.ts.map +1 -0
- package/dist/polling/wait-for-run.js +36 -0
- package/dist/polling/wait-for-run.js.map +1 -0
- package/dist/quickstart/prompt.d.ts +18 -0
- package/dist/quickstart/prompt.d.ts.map +1 -0
- package/dist/quickstart/prompt.js +37 -0
- package/dist/quickstart/prompt.js.map +1 -0
- package/dist/quickstart/sample.d.ts +28 -0
- package/dist/quickstart/sample.d.ts.map +1 -0
- package/dist/quickstart/sample.js +235 -0
- package/dist/quickstart/sample.js.map +1 -0
- package/dist/quickstart/selection.d.ts +53 -0
- package/dist/quickstart/selection.d.ts.map +1 -0
- package/dist/quickstart/selection.js +97 -0
- package/dist/quickstart/selection.js.map +1 -0
- package/dist/run.d.ts +28 -0
- package/dist/run.d.ts.map +1 -0
- package/dist/run.js +134 -0
- package/dist/run.js.map +1 -0
- package/dist/runtime/action-context.d.ts +19 -0
- package/dist/runtime/action-context.d.ts.map +1 -0
- package/dist/runtime/action-context.js +40 -0
- package/dist/runtime/action-context.js.map +1 -0
- package/dist/suite/execution.d.ts +46 -0
- package/dist/suite/execution.d.ts.map +1 -0
- package/dist/suite/execution.js +97 -0
- package/dist/suite/execution.js.map +1 -0
- package/dist/suite/manifest.d.ts +16 -0
- package/dist/suite/manifest.d.ts.map +1 -0
- package/dist/suite/manifest.js +92 -0
- package/dist/suite/manifest.js.map +1 -0
- package/dist/suite/reporting.d.ts +35 -0
- package/dist/suite/reporting.d.ts.map +1 -0
- package/dist/suite/reporting.js +131 -0
- package/dist/suite/reporting.js.map +1 -0
- package/dist/trace/deepseek-harness.d.ts +31 -0
- package/dist/trace/deepseek-harness.d.ts.map +1 -0
- package/dist/trace/deepseek-harness.js +412 -0
- package/dist/trace/deepseek-harness.js.map +1 -0
- package/dist/trace/import.d.ts +9 -0
- package/dist/trace/import.d.ts.map +1 -0
- package/dist/trace/import.js +45 -0
- package/dist/trace/import.js.map +1 -0
- package/dist/version.d.ts +3 -0
- package/dist/version.d.ts.map +1 -0
- package/dist/version.js +3 -0
- package/dist/version.js.map +1 -0
- package/package.json +82 -0
|
@@ -0,0 +1,97 @@
|
|
|
1
|
+
import type { JsonObject } from '../domain/common.js';
|
|
2
|
+
import type { EvaluationContextType, EvaluationResults } from '../domain/types.js';
|
|
3
|
+
import { type SuiteGatePolicy } from './policy.js';
|
|
4
|
+
/**
|
|
5
|
+
* Gate plane.
|
|
6
|
+
*
|
|
7
|
+
* Execution and result fetching happen elsewhere; this module is pure. It
|
|
8
|
+
* converts an already-fetched result payload plus the terminal run status into
|
|
9
|
+
* one release decision per eval, then rolls the eval decisions into a single
|
|
10
|
+
* suite decision. It never performs I/O and reads only fields the CLI already
|
|
11
|
+
* receives from existing result and run-status payloads.
|
|
12
|
+
*
|
|
13
|
+
* Deliberately unused for gating: conversation / agent-trace `outcome.*`,
|
|
14
|
+
* generated summaries and insights, and agent-trace model-performance CI.
|
|
15
|
+
*/
|
|
16
|
+
export type GateDecision = 'PASS' | 'FAIL' | 'INCONCLUSIVE' | 'ERROR';
|
|
17
|
+
export declare function worstDecision(decisions: readonly GateDecision[]): GateDecision;
|
|
18
|
+
export type GateCheckEvidence = {
|
|
19
|
+
mean?: number | null;
|
|
20
|
+
ci95Lower?: number | null;
|
|
21
|
+
ci95Upper?: number | null;
|
|
22
|
+
score?: number | null;
|
|
23
|
+
hasData?: boolean;
|
|
24
|
+
insufficientSamples?: boolean;
|
|
25
|
+
};
|
|
26
|
+
export type GateCheckResult = {
|
|
27
|
+
/** Stable machine identifier, e.g. `overall`, `scenario:Checkout`, `metric:factuality`. */
|
|
28
|
+
metric: string;
|
|
29
|
+
label: string;
|
|
30
|
+
threshold: string;
|
|
31
|
+
actual: string;
|
|
32
|
+
decision: GateDecision;
|
|
33
|
+
reason: string;
|
|
34
|
+
basis: 'mean' | 'confidence-interval';
|
|
35
|
+
evidence: GateCheckEvidence;
|
|
36
|
+
};
|
|
37
|
+
export type ConvergenceStatus = 'CONVERGED' | 'NOT_CONVERGED' | 'NOT_APPLICABLE' | 'ERROR' | 'UNKNOWN';
|
|
38
|
+
export type ConvergenceEvidence = {
|
|
39
|
+
status: ConvergenceStatus;
|
|
40
|
+
autoStopReason?: string;
|
|
41
|
+
achievedConsistency?: number;
|
|
42
|
+
targetConsistency?: number;
|
|
43
|
+
/** Comparison direction reported by the backend; `<` means lower is better. */
|
|
44
|
+
targetConsistencyOperator?: string;
|
|
45
|
+
failureReason?: string;
|
|
46
|
+
/** True when the run status carried none of the convergence fields. */
|
|
47
|
+
fieldsUnavailable: boolean;
|
|
48
|
+
};
|
|
49
|
+
export type EvalGateEvidence = {
|
|
50
|
+
runMode: 'single_run' | 'multi_run' | 'unknown';
|
|
51
|
+
runState?: string;
|
|
52
|
+
totalRuns?: number;
|
|
53
|
+
convergence?: ConvergenceEvidence;
|
|
54
|
+
};
|
|
55
|
+
export type EvalGateDecision = {
|
|
56
|
+
index: number;
|
|
57
|
+
inputFile: string;
|
|
58
|
+
evaluationName?: string;
|
|
59
|
+
evaluationId?: string;
|
|
60
|
+
runId?: string;
|
|
61
|
+
contextType?: EvaluationContextType;
|
|
62
|
+
decision: GateDecision;
|
|
63
|
+
checks: GateCheckResult[];
|
|
64
|
+
reason: string;
|
|
65
|
+
evidence: EvalGateEvidence;
|
|
66
|
+
};
|
|
67
|
+
export type SuiteGateReport = {
|
|
68
|
+
suiteDecision: GateDecision;
|
|
69
|
+
counts: Record<GateDecision, number>;
|
|
70
|
+
perEvalDecisions: EvalGateDecision[];
|
|
71
|
+
};
|
|
72
|
+
export type GateEvalInput = {
|
|
73
|
+
index: number;
|
|
74
|
+
inputFile: string;
|
|
75
|
+
evaluationName?: string;
|
|
76
|
+
evaluationId?: string;
|
|
77
|
+
runId?: string;
|
|
78
|
+
contextType?: EvaluationContextType;
|
|
79
|
+
/** Terminal state of the execution and results planes for this eval. */
|
|
80
|
+
status: 'completed' | 'execution_failed' | 'result_failed';
|
|
81
|
+
error?: string;
|
|
82
|
+
results?: EvaluationResults;
|
|
83
|
+
runState?: string;
|
|
84
|
+
runStatusRaw?: JsonObject;
|
|
85
|
+
policy: SuiteGatePolicy;
|
|
86
|
+
};
|
|
87
|
+
/**
|
|
88
|
+
* Classify convergence from the terminal run status. `MAX_RUNS_REACHED` alone
|
|
89
|
+
* does not prove a convergence failure, so the consistency target decides
|
|
90
|
+
* whether convergence was enabled at all.
|
|
91
|
+
*/
|
|
92
|
+
export declare function classifyConvergence(runStatusRaw: JsonObject | undefined, runState: string | undefined): ConvergenceEvidence;
|
|
93
|
+
/** Convert one completed (or failed) suite eval into a release decision. */
|
|
94
|
+
export declare function decideEvalGate(input: GateEvalInput): EvalGateDecision;
|
|
95
|
+
/** Roll per-eval decisions up with ERROR > FAIL > INCONCLUSIVE > PASS. */
|
|
96
|
+
export declare function rollUpSuiteDecision(perEvalDecisions: readonly EvalGateDecision[]): SuiteGateReport;
|
|
97
|
+
//# sourceMappingURL=decision.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"decision.d.ts","sourceRoot":"","sources":["../../src/gate/decision.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,qBAAqB,CAAC;AACtD,OAAO,KAAK,EAAE,qBAAqB,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AACnF,OAAO,EAAiB,KAAK,eAAe,EAAE,MAAM,aAAa,CAAC;AAElE;;;;;;;;;;;GAWG;AAEH,MAAM,MAAM,YAAY,GAAG,MAAM,GAAG,MAAM,GAAG,cAAc,GAAG,OAAO,CAAC;AAStE,wBAAgB,aAAa,CAAC,SAAS,EAAE,SAAS,YAAY,EAAE,GAAG,YAAY,CAK9E;AAED,MAAM,MAAM,iBAAiB,GAAG;IAC9B,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,mBAAmB,CAAC,EAAE,OAAO,CAAC;CAC/B,CAAC;AAEF,MAAM,MAAM,eAAe,GAAG;IAC5B,2FAA2F;IAC3F,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,YAAY,CAAC;IACvB,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,GAAG,qBAAqB,CAAC;IACtC,QAAQ,EAAE,iBAAiB,CAAC;CAC7B,CAAC;AAEF,MAAM,MAAM,iBAAiB,GAC3B,WAAW,GAAG,eAAe,GAAG,gBAAgB,GAAG,OAAO,GAAG,SAAS,CAAC;AAEzE,MAAM,MAAM,mBAAmB,GAAG;IAChC,MAAM,EAAE,iBAAiB,CAAC;IAC1B,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,+EAA+E;IAC/E,yBAAyB,CAAC,EAAE,MAAM,CAAC;IACnC,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,uEAAuE;IACvE,iBAAiB,EAAE,OAAO,CAAC;CAC5B,CAAC;AAEF,MAAM,MAAM,gBAAgB,GAAG;IAC7B,OAAO,EAAE,YAAY,GAAG,WAAW,GAAG,SAAS,CAAC;IAChD,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,mBAAmB,CAAC;CACnC,CAAC;AAEF,MAAM,MAAM,gBAAgB,GAAG;IAC7B,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,qBAAqB,CAAC;IACpC,QAAQ,EAAE,YAAY,CAAC;IACvB,MAAM,EAAE,eAAe,EAAE,CAAC;IAC1B,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,gBAAgB,CAAC;CAC5B,CAAC;AAEF,MAAM,MAAM,eAAe,GAAG;IAC5B,aAAa,EAAE,YAAY,CAAC;IAC5B,MAAM,EAAE,MAAM,CAAC,YAAY,EAAE,MAAM,CAAC,CAAC;IACrC,gBAAgB,EAAE,gBAAgB,EAAE,CAAC;CACtC,CAAC;AAEF,MAAM,MAAM,aAAa,GAAG;IAC1B,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,qBAAqB,CAAC;IACpC,wEAAwE;IACxE,MAAM,EAAE,WAAW,GAAG,kBAAkB,GAAG,eAAe,CAAC;IAC3D,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,iBAAiB,CAAC;IAC5B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,YAAY,CAAC,EAAE,UAAU,CAAC;IAC1B,MAAM,EAAE,eAAe,CAAC;CACzB,CAAC;AAoOF;;;;GAIG;AACH,wBAAgB,mBAAmB,CACjC,YAAY,EAAE,UAAU,GAAG,SAAS,EACpC,QAAQ,EAAE,MAAM,GAAG,SAAS,GAC3B,mBAAmB,CAoDrB;AA4GD,4EAA4E;AAC5E,wBAAgB,cAAc,CAAC,KAAK,EAAE,aAAa,GAAG,gBAAgB,CA6FrE;AAED,0EAA0E;AAC1E,wBAAgB,mBAAmB,CACjC,gBAAgB,EAAE,SAAS,gBAAgB,EAAE,GAC5C,eAAe,CAgBjB"}
|
|
@@ -0,0 +1,436 @@
|
|
|
1
|
+
import { z } from 'zod';
|
|
2
|
+
import { isEmptyPolicy } from './policy.js';
|
|
3
|
+
const PRECEDENCE = {
|
|
4
|
+
ERROR: 3,
|
|
5
|
+
FAIL: 2,
|
|
6
|
+
INCONCLUSIVE: 1,
|
|
7
|
+
PASS: 0,
|
|
8
|
+
};
|
|
9
|
+
export function worstDecision(decisions) {
|
|
10
|
+
return decisions.reduce((worst, decision) => (PRECEDENCE[decision] > PRECEDENCE[worst] ? decision : worst), 'PASS');
|
|
11
|
+
}
|
|
12
|
+
const NOT_AVAILABLE = 'Not available';
|
|
13
|
+
function formatNumber(value) {
|
|
14
|
+
return Number.isInteger(value) ? value.toFixed(2) : value.toFixed(2);
|
|
15
|
+
}
|
|
16
|
+
function formatActual(evidence, basis) {
|
|
17
|
+
if (basis === 'confidence-interval') {
|
|
18
|
+
const { ci95Lower: lower, ci95Upper: upper } = evidence;
|
|
19
|
+
if (typeof lower !== 'number' || typeof upper !== 'number')
|
|
20
|
+
return NOT_AVAILABLE;
|
|
21
|
+
const mean = typeof evidence.mean === 'number' ? `${formatNumber(evidence.mean)} ` : '';
|
|
22
|
+
return `${mean}95% CI ${formatNumber(lower)}-${formatNumber(upper)}`;
|
|
23
|
+
}
|
|
24
|
+
const value = evidence.score ?? evidence.mean;
|
|
25
|
+
return typeof value === 'number' ? formatNumber(value) : NOT_AVAILABLE;
|
|
26
|
+
}
|
|
27
|
+
/** Point-estimate comparison. A missing value is never treated as zero. */
|
|
28
|
+
function checkMean(metric, label, threshold, evidence) {
|
|
29
|
+
const value = evidence.score ?? evidence.mean;
|
|
30
|
+
const base = {
|
|
31
|
+
metric,
|
|
32
|
+
label,
|
|
33
|
+
threshold: `>= ${formatNumber(threshold)}`,
|
|
34
|
+
basis: 'mean',
|
|
35
|
+
evidence,
|
|
36
|
+
actual: formatActual(evidence, 'mean'),
|
|
37
|
+
};
|
|
38
|
+
if (evidence.hasData === false) {
|
|
39
|
+
return { ...base, decision: 'INCONCLUSIVE', reason: 'metric reported has_data: false' };
|
|
40
|
+
}
|
|
41
|
+
if (typeof value !== 'number') {
|
|
42
|
+
return { ...base, decision: 'INCONCLUSIVE', reason: 'required score is unavailable' };
|
|
43
|
+
}
|
|
44
|
+
return value >= threshold
|
|
45
|
+
? { ...base, decision: 'PASS', reason: 'score meets the required threshold' }
|
|
46
|
+
: { ...base, decision: 'FAIL', reason: 'score is below the required threshold' };
|
|
47
|
+
}
|
|
48
|
+
/** Interval comparison for multi-run evals. */
|
|
49
|
+
function checkConfidenceInterval(metric, label, threshold, evidence) {
|
|
50
|
+
const base = {
|
|
51
|
+
metric,
|
|
52
|
+
label,
|
|
53
|
+
threshold: `>= ${formatNumber(threshold)}`,
|
|
54
|
+
basis: 'confidence-interval',
|
|
55
|
+
evidence,
|
|
56
|
+
actual: formatActual(evidence, 'confidence-interval'),
|
|
57
|
+
};
|
|
58
|
+
const lower = evidence.ci95Lower;
|
|
59
|
+
const upper = evidence.ci95Upper;
|
|
60
|
+
if (typeof lower !== 'number' || typeof upper !== 'number') {
|
|
61
|
+
return {
|
|
62
|
+
...base,
|
|
63
|
+
decision: 'INCONCLUSIVE',
|
|
64
|
+
reason: 'required 95% confidence interval is unavailable',
|
|
65
|
+
};
|
|
66
|
+
}
|
|
67
|
+
if (lower >= threshold) {
|
|
68
|
+
return {
|
|
69
|
+
...base,
|
|
70
|
+
decision: 'PASS',
|
|
71
|
+
reason: 'confidence interval is entirely at or above the threshold',
|
|
72
|
+
};
|
|
73
|
+
}
|
|
74
|
+
if (upper < threshold) {
|
|
75
|
+
return {
|
|
76
|
+
...base,
|
|
77
|
+
decision: 'FAIL',
|
|
78
|
+
reason: 'confidence interval is entirely below the threshold',
|
|
79
|
+
};
|
|
80
|
+
}
|
|
81
|
+
return {
|
|
82
|
+
...base,
|
|
83
|
+
decision: 'INCONCLUSIVE',
|
|
84
|
+
reason: 'confidence interval crosses release threshold',
|
|
85
|
+
};
|
|
86
|
+
}
|
|
87
|
+
const nullableNumber = z.number().nullable().optional();
|
|
88
|
+
const scoreCellSchema = z
|
|
89
|
+
.object({
|
|
90
|
+
mean: nullableNumber,
|
|
91
|
+
ci95_lower: nullableNumber,
|
|
92
|
+
ci95_upper: nullableNumber,
|
|
93
|
+
insufficient_samples: z.boolean().nullable().optional(),
|
|
94
|
+
has_data: z.boolean().nullable().optional(),
|
|
95
|
+
error: z.unknown().optional(),
|
|
96
|
+
})
|
|
97
|
+
.loose();
|
|
98
|
+
const modelPerformanceSchema = z
|
|
99
|
+
.object({
|
|
100
|
+
models: z
|
|
101
|
+
.array(z
|
|
102
|
+
.object({
|
|
103
|
+
display_name: z.string().optional(),
|
|
104
|
+
model_key: z.string().optional(),
|
|
105
|
+
is_primary: z.boolean().optional(),
|
|
106
|
+
scores: z.object({ overall: scoreCellSchema.optional() }).loose().optional(),
|
|
107
|
+
})
|
|
108
|
+
.loose())
|
|
109
|
+
.optional(),
|
|
110
|
+
})
|
|
111
|
+
.loose();
|
|
112
|
+
const scenarioComparisonSchema = z
|
|
113
|
+
.object({
|
|
114
|
+
scenarios: z
|
|
115
|
+
.array(z
|
|
116
|
+
.object({
|
|
117
|
+
scenario_name: z.string().optional(),
|
|
118
|
+
model_scores: z
|
|
119
|
+
.array(z
|
|
120
|
+
.object({
|
|
121
|
+
model_key: z.string().optional(),
|
|
122
|
+
is_primary: z.boolean().optional(),
|
|
123
|
+
score: scoreCellSchema.optional(),
|
|
124
|
+
error: z.unknown().optional(),
|
|
125
|
+
})
|
|
126
|
+
.loose())
|
|
127
|
+
.optional(),
|
|
128
|
+
})
|
|
129
|
+
.loose())
|
|
130
|
+
.optional(),
|
|
131
|
+
})
|
|
132
|
+
.loose();
|
|
133
|
+
const perMetricSchema = z
|
|
134
|
+
.object({
|
|
135
|
+
per_metric: z
|
|
136
|
+
.record(z.string(), z
|
|
137
|
+
.object({
|
|
138
|
+
score: nullableNumber,
|
|
139
|
+
has_data: z.boolean().nullable().optional(),
|
|
140
|
+
})
|
|
141
|
+
.loose())
|
|
142
|
+
.optional(),
|
|
143
|
+
})
|
|
144
|
+
.loose();
|
|
145
|
+
const runStatusGateSchema = z
|
|
146
|
+
.object({
|
|
147
|
+
evaluationState: z.string().optional(),
|
|
148
|
+
progress: z.object({ totalRuns: z.number().nullable().optional() }).loose().optional(),
|
|
149
|
+
evaluationMetrics: z
|
|
150
|
+
.object({
|
|
151
|
+
autoStopTriggered: z
|
|
152
|
+
.object({ reason: z.string().nullable().optional() })
|
|
153
|
+
.loose()
|
|
154
|
+
.nullable()
|
|
155
|
+
.optional(),
|
|
156
|
+
achievedConsistency: nullableNumber,
|
|
157
|
+
targetConsistency: nullableNumber,
|
|
158
|
+
targetConsistencyOperator: z.string().nullable().optional(),
|
|
159
|
+
})
|
|
160
|
+
.loose()
|
|
161
|
+
.nullable()
|
|
162
|
+
.optional(),
|
|
163
|
+
achievedConsistency: nullableNumber,
|
|
164
|
+
targetConsistency: nullableNumber,
|
|
165
|
+
targetConsistencyOperator: z.string().nullable().optional(),
|
|
166
|
+
failureDetails: z
|
|
167
|
+
.object({ failureReason: z.string().nullable().optional() })
|
|
168
|
+
.loose()
|
|
169
|
+
.nullable()
|
|
170
|
+
.optional(),
|
|
171
|
+
failure_reason: z.string().nullable().optional(),
|
|
172
|
+
failureReason: z.string().nullable().optional(),
|
|
173
|
+
})
|
|
174
|
+
.loose();
|
|
175
|
+
function toEvidence(cell) {
|
|
176
|
+
return {
|
|
177
|
+
mean: cell?.mean ?? null,
|
|
178
|
+
ci95Lower: cell?.ci95_lower ?? null,
|
|
179
|
+
ci95Upper: cell?.ci95_upper ?? null,
|
|
180
|
+
...(cell?.has_data === null || cell?.has_data === undefined ? {} : { hasData: cell.has_data }),
|
|
181
|
+
...(cell?.insufficient_samples === null || cell?.insufficient_samples === undefined
|
|
182
|
+
? {}
|
|
183
|
+
: { insufficientSamples: cell.insufficient_samples }),
|
|
184
|
+
};
|
|
185
|
+
}
|
|
186
|
+
/**
|
|
187
|
+
* Consistency can be reported either as a score to exceed or as a spread to
|
|
188
|
+
* stay under. The backend states the direction in `targetConsistencyOperator`;
|
|
189
|
+
* without it, a higher-is-better target is assumed.
|
|
190
|
+
*/
|
|
191
|
+
function meetsConsistencyTarget(achieved, target, operator) {
|
|
192
|
+
switch ((operator ?? '>=').trim()) {
|
|
193
|
+
case '<':
|
|
194
|
+
return achieved < target;
|
|
195
|
+
case '<=':
|
|
196
|
+
return achieved <= target;
|
|
197
|
+
case '>':
|
|
198
|
+
return achieved > target;
|
|
199
|
+
default:
|
|
200
|
+
return achieved >= target;
|
|
201
|
+
}
|
|
202
|
+
}
|
|
203
|
+
/**
|
|
204
|
+
* Classify convergence from the terminal run status. `MAX_RUNS_REACHED` alone
|
|
205
|
+
* does not prove a convergence failure, so the consistency target decides
|
|
206
|
+
* whether convergence was enabled at all.
|
|
207
|
+
*/
|
|
208
|
+
export function classifyConvergence(runStatusRaw, runState) {
|
|
209
|
+
const parsed = runStatusGateSchema.safeParse(runStatusRaw ?? {});
|
|
210
|
+
const data = parsed.success ? parsed.data : {};
|
|
211
|
+
const metrics = data.evaluationMetrics ?? undefined;
|
|
212
|
+
const autoStopReason = metrics?.autoStopTriggered?.reason ?? undefined;
|
|
213
|
+
const achieved = metrics?.achievedConsistency ?? data.achievedConsistency ?? undefined;
|
|
214
|
+
const target = metrics?.targetConsistency ?? data.targetConsistency ?? undefined;
|
|
215
|
+
const operator = metrics?.targetConsistencyOperator ?? data.targetConsistencyOperator ?? undefined;
|
|
216
|
+
const failureReason = data.failureDetails?.failureReason ?? data.failureReason ?? data.failure_reason ?? undefined;
|
|
217
|
+
const fieldsUnavailable = autoStopReason === undefined &&
|
|
218
|
+
achieved === null &&
|
|
219
|
+
target === null &&
|
|
220
|
+
failureReason === undefined
|
|
221
|
+
? true
|
|
222
|
+
: autoStopReason === undefined &&
|
|
223
|
+
achieved === undefined &&
|
|
224
|
+
target === undefined &&
|
|
225
|
+
failureReason === undefined;
|
|
226
|
+
const base = {
|
|
227
|
+
...(autoStopReason ? { autoStopReason } : {}),
|
|
228
|
+
...(typeof achieved === 'number' ? { achievedConsistency: achieved } : {}),
|
|
229
|
+
...(typeof target === 'number' ? { targetConsistency: target } : {}),
|
|
230
|
+
...(operator ? { targetConsistencyOperator: operator } : {}),
|
|
231
|
+
...(failureReason ? { failureReason } : {}),
|
|
232
|
+
fieldsUnavailable,
|
|
233
|
+
};
|
|
234
|
+
if (failureReason !== undefined && failureReason !== null) {
|
|
235
|
+
return { ...base, status: 'ERROR' };
|
|
236
|
+
}
|
|
237
|
+
const state = (runState ?? data.evaluationState ?? '').toUpperCase();
|
|
238
|
+
if (state !== '' && state !== 'COMPLETED' && state !== 'COMPLETE' && state !== 'SUCCEEDED') {
|
|
239
|
+
return { ...base, status: 'ERROR' };
|
|
240
|
+
}
|
|
241
|
+
if (fieldsUnavailable)
|
|
242
|
+
return { ...base, status: 'UNKNOWN' };
|
|
243
|
+
if (autoStopReason === 'OPTIMAL_CONFIDENCE_REACHED' ||
|
|
244
|
+
autoStopReason === 'CONSISTENCY_TARGET_MET') {
|
|
245
|
+
return { ...base, status: 'CONVERGED' };
|
|
246
|
+
}
|
|
247
|
+
if (typeof target !== 'number' || target <= 0)
|
|
248
|
+
return { ...base, status: 'NOT_APPLICABLE' };
|
|
249
|
+
if (typeof achieved !== 'number')
|
|
250
|
+
return { ...base, status: 'NOT_CONVERGED' };
|
|
251
|
+
return {
|
|
252
|
+
...base,
|
|
253
|
+
status: meetsConsistencyTarget(achieved, target, operator) ? 'CONVERGED' : 'NOT_CONVERGED',
|
|
254
|
+
};
|
|
255
|
+
}
|
|
256
|
+
function detectRunMode(runStatusRaw, cells) {
|
|
257
|
+
const parsed = runStatusGateSchema.safeParse(runStatusRaw ?? {});
|
|
258
|
+
const totalRuns = parsed.success ? (parsed.data.progress?.totalRuns ?? undefined) : undefined;
|
|
259
|
+
if (typeof totalRuns === 'number' && totalRuns > 0) {
|
|
260
|
+
return { runMode: totalRuns > 1 ? 'multi_run' : 'single_run', totalRuns };
|
|
261
|
+
}
|
|
262
|
+
const hasInterval = cells.some((cell) => typeof cell.ci95Lower === 'number' && typeof cell.ci95Upper === 'number');
|
|
263
|
+
return { runMode: hasInterval ? 'multi_run' : 'single_run' };
|
|
264
|
+
}
|
|
265
|
+
function scenarioChecks(results, policy) {
|
|
266
|
+
const cells = [];
|
|
267
|
+
const performance = modelPerformanceSchema.safeParse(results.modelPerformance);
|
|
268
|
+
const models = performance.success ? (performance.data.models ?? []) : [];
|
|
269
|
+
const primary = models.find((model) => model.is_primary === true) ?? models[0];
|
|
270
|
+
const primaryKey = primary?.model_key;
|
|
271
|
+
if (policy.minOverall !== undefined) {
|
|
272
|
+
cells.push({
|
|
273
|
+
metric: 'overall',
|
|
274
|
+
label: 'Primary model overall score',
|
|
275
|
+
threshold: policy.minOverall,
|
|
276
|
+
evidence: toEvidence(primary?.scores?.overall),
|
|
277
|
+
});
|
|
278
|
+
}
|
|
279
|
+
if (policy.minScenario !== undefined) {
|
|
280
|
+
const comparison = scenarioComparisonSchema.safeParse(results.scenarioComparison);
|
|
281
|
+
const scenarios = comparison.success ? (comparison.data.scenarios ?? []) : [];
|
|
282
|
+
for (const scenario of scenarios) {
|
|
283
|
+
const modelScores = scenario.model_scores ?? [];
|
|
284
|
+
const cell = modelScores.find((entry) => entry.is_primary === true) ??
|
|
285
|
+
(primaryKey === undefined
|
|
286
|
+
? modelScores[0]
|
|
287
|
+
: (modelScores.find((entry) => entry.model_key === primaryKey) ?? undefined));
|
|
288
|
+
if (cell === undefined)
|
|
289
|
+
continue;
|
|
290
|
+
const name = scenario.scenario_name ?? 'scenario';
|
|
291
|
+
cells.push({
|
|
292
|
+
metric: `scenario:${name}`,
|
|
293
|
+
label: `Scenario ${name}`,
|
|
294
|
+
threshold: policy.minScenario,
|
|
295
|
+
evidence: cell.error === undefined || cell.error === null
|
|
296
|
+
? toEvidence(cell.score)
|
|
297
|
+
: { ...toEvidence(cell.score), hasData: false },
|
|
298
|
+
});
|
|
299
|
+
}
|
|
300
|
+
}
|
|
301
|
+
return { cells };
|
|
302
|
+
}
|
|
303
|
+
function sessionChecks(payload, policy) {
|
|
304
|
+
const parsed = perMetricSchema.safeParse(payload);
|
|
305
|
+
const perMetric = parsed.success ? (parsed.data.per_metric ?? {}) : {};
|
|
306
|
+
return Object.entries(policy.metrics ?? {}).map(([name, threshold]) => {
|
|
307
|
+
const cell = perMetric[name];
|
|
308
|
+
const evidence = {
|
|
309
|
+
score: cell?.score ?? null,
|
|
310
|
+
...(cell?.has_data === null || cell?.has_data === undefined
|
|
311
|
+
? {}
|
|
312
|
+
: { hasData: cell.has_data }),
|
|
313
|
+
};
|
|
314
|
+
if (cell === undefined) {
|
|
315
|
+
return {
|
|
316
|
+
metric: `metric:${name}`,
|
|
317
|
+
label: name,
|
|
318
|
+
threshold: `>= ${formatNumber(threshold)}`,
|
|
319
|
+
actual: NOT_AVAILABLE,
|
|
320
|
+
decision: 'INCONCLUSIVE',
|
|
321
|
+
reason: 'required metric is missing from the result payload',
|
|
322
|
+
basis: 'mean',
|
|
323
|
+
evidence,
|
|
324
|
+
};
|
|
325
|
+
}
|
|
326
|
+
return checkMean(`metric:${name}`, name, threshold, evidence);
|
|
327
|
+
});
|
|
328
|
+
}
|
|
329
|
+
function summarize(decision, checks) {
|
|
330
|
+
if (decision === 'PASS') {
|
|
331
|
+
return checks.length === 0
|
|
332
|
+
? 'no blocking criteria configured'
|
|
333
|
+
: 'every configured criterion was met';
|
|
334
|
+
}
|
|
335
|
+
const offending = checks.find((check) => check.decision === decision);
|
|
336
|
+
return offending === undefined
|
|
337
|
+
? 'gate evidence is unusable'
|
|
338
|
+
: `${offending.label}: ${offending.reason}`;
|
|
339
|
+
}
|
|
340
|
+
/** Convert one completed (or failed) suite eval into a release decision. */
|
|
341
|
+
export function decideEvalGate(input) {
|
|
342
|
+
const identity = {
|
|
343
|
+
index: input.index,
|
|
344
|
+
inputFile: input.inputFile,
|
|
345
|
+
...(input.evaluationName ? { evaluationName: input.evaluationName } : {}),
|
|
346
|
+
...(input.evaluationId ? { evaluationId: input.evaluationId } : {}),
|
|
347
|
+
...(input.runId ? { runId: input.runId } : {}),
|
|
348
|
+
...(input.contextType ? { contextType: input.contextType } : {}),
|
|
349
|
+
};
|
|
350
|
+
if (input.status !== 'completed' || input.results === undefined) {
|
|
351
|
+
return {
|
|
352
|
+
...identity,
|
|
353
|
+
decision: 'ERROR',
|
|
354
|
+
checks: [],
|
|
355
|
+
reason: input.error ??
|
|
356
|
+
(input.status === 'execution_failed'
|
|
357
|
+
? 'eval execution failed'
|
|
358
|
+
: 'eval results could not be fetched'),
|
|
359
|
+
evidence: {
|
|
360
|
+
runMode: 'unknown',
|
|
361
|
+
...(input.runState ? { runState: input.runState } : {}),
|
|
362
|
+
},
|
|
363
|
+
};
|
|
364
|
+
}
|
|
365
|
+
const results = input.results;
|
|
366
|
+
if (results.contextType !== 'scenario') {
|
|
367
|
+
const payload = results.contextType === 'conversation' ? results.conversation : results.agentTrace;
|
|
368
|
+
const checks = sessionChecks(payload, input.policy);
|
|
369
|
+
const decision = isEmptyPolicy(input.policy)
|
|
370
|
+
? 'INCONCLUSIVE'
|
|
371
|
+
: worstDecision(checks.map((check) => check.decision));
|
|
372
|
+
return {
|
|
373
|
+
...identity,
|
|
374
|
+
decision,
|
|
375
|
+
checks,
|
|
376
|
+
reason: isEmptyPolicy(input.policy)
|
|
377
|
+
? 'no metric thresholds are configured for this eval'
|
|
378
|
+
: summarize(decision, checks),
|
|
379
|
+
evidence: {
|
|
380
|
+
runMode: 'single_run',
|
|
381
|
+
...(input.runState ? { runState: input.runState } : {}),
|
|
382
|
+
},
|
|
383
|
+
};
|
|
384
|
+
}
|
|
385
|
+
const { cells } = scenarioChecks(results, input.policy);
|
|
386
|
+
const { runMode, totalRuns } = detectRunMode(input.runStatusRaw, cells.map((cell) => cell.evidence));
|
|
387
|
+
const convergence = runMode === 'multi_run' ? classifyConvergence(input.runStatusRaw, input.runState) : undefined;
|
|
388
|
+
const checks = cells.map((cell) => runMode === 'multi_run'
|
|
389
|
+
? checkConfidenceInterval(cell.metric, cell.label, cell.threshold, cell.evidence)
|
|
390
|
+
: checkMean(cell.metric, cell.label, cell.threshold, cell.evidence));
|
|
391
|
+
const evidence = {
|
|
392
|
+
runMode,
|
|
393
|
+
...(input.runState ? { runState: input.runState } : {}),
|
|
394
|
+
...(totalRuns === undefined ? {} : { totalRuns }),
|
|
395
|
+
...(convergence ? { convergence } : {}),
|
|
396
|
+
};
|
|
397
|
+
if (isEmptyPolicy(input.policy)) {
|
|
398
|
+
return {
|
|
399
|
+
...identity,
|
|
400
|
+
decision: 'INCONCLUSIVE',
|
|
401
|
+
checks,
|
|
402
|
+
reason: 'no score thresholds are configured for this eval',
|
|
403
|
+
evidence,
|
|
404
|
+
};
|
|
405
|
+
}
|
|
406
|
+
let decision = worstDecision(checks.map((check) => check.decision));
|
|
407
|
+
let reason = summarize(decision, checks);
|
|
408
|
+
if (convergence?.status === 'ERROR') {
|
|
409
|
+
decision = 'ERROR';
|
|
410
|
+
reason = convergence.failureReason ?? 'the multi-run evaluation did not complete successfully';
|
|
411
|
+
}
|
|
412
|
+
else if (convergence?.status === 'NOT_CONVERGED' && decision !== 'FAIL') {
|
|
413
|
+
decision = 'INCONCLUSIVE';
|
|
414
|
+
reason = 'convergence was enabled but the consistency target was not reached';
|
|
415
|
+
}
|
|
416
|
+
return { ...identity, decision, checks, reason, evidence };
|
|
417
|
+
}
|
|
418
|
+
/** Roll per-eval decisions up with ERROR > FAIL > INCONCLUSIVE > PASS. */
|
|
419
|
+
export function rollUpSuiteDecision(perEvalDecisions) {
|
|
420
|
+
const counts = {
|
|
421
|
+
PASS: 0,
|
|
422
|
+
FAIL: 0,
|
|
423
|
+
INCONCLUSIVE: 0,
|
|
424
|
+
ERROR: 0,
|
|
425
|
+
};
|
|
426
|
+
for (const entry of perEvalDecisions)
|
|
427
|
+
counts[entry.decision] += 1;
|
|
428
|
+
return {
|
|
429
|
+
suiteDecision: perEvalDecisions.length === 0
|
|
430
|
+
? 'INCONCLUSIVE'
|
|
431
|
+
: worstDecision(perEvalDecisions.map((entry) => entry.decision)),
|
|
432
|
+
counts,
|
|
433
|
+
perEvalDecisions: [...perEvalDecisions].sort((left, right) => left.index - right.index),
|
|
434
|
+
};
|
|
435
|
+
}
|
|
436
|
+
//# sourceMappingURL=decision.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"decision.js","sourceRoot":"","sources":["../../src/gate/decision.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,OAAO,EAAE,aAAa,EAAwB,MAAM,aAAa,CAAC;AAiBlE,MAAM,UAAU,GAA2C;IACzD,KAAK,EAAE,CAAC;IACR,IAAI,EAAE,CAAC;IACP,YAAY,EAAE,CAAC;IACf,IAAI,EAAE,CAAC;CACR,CAAC;AAEF,MAAM,UAAU,aAAa,CAAC,SAAkC;IAC9D,OAAO,SAAS,CAAC,MAAM,CACrB,CAAC,KAAK,EAAE,QAAQ,EAAE,EAAE,CAAC,CAAC,UAAU,CAAC,QAAQ,CAAC,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,EAClF,MAAM,CACP,CAAC;AACJ,CAAC;AAgFD,MAAM,aAAa,GAAG,eAAe,CAAC;AAEtC,SAAS,YAAY,CAAC,KAAa;IACjC,OAAO,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;AACvE,CAAC;AAED,SAAS,YAAY,CAAC,QAA2B,EAAE,KAA+B;IAChF,IAAI,KAAK,KAAK,qBAAqB,EAAE,CAAC;QACpC,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,GAAG,QAAQ,CAAC;QACxD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,OAAO,KAAK,KAAK,QAAQ;YAAE,OAAO,aAAa,CAAC;QACjF,MAAM,IAAI,GAAG,OAAO,QAAQ,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,YAAY,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QACxF,OAAO,GAAG,IAAI,UAAU,YAAY,CAAC,KAAK,CAAC,IAAI,YAAY,CAAC,KAAK,CAAC,EAAE,CAAC;IACvE,CAAC;IACD,MAAM,KAAK,GAAG,QAAQ,CAAC,KAAK,IAAI,QAAQ,CAAC,IAAI,CAAC;IAC9C,OAAO,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC;AACzE,CAAC;AAED,2EAA2E;AAC3E,SAAS,SAAS,CAChB,MAAc,EACd,KAAa,EACb,SAAiB,EACjB,QAA2B;IAE3B,MAAM,KAAK,GAAG,QAAQ,CAAC,KAAK,IAAI,QAAQ,CAAC,IAAI,CAAC;IAC9C,MAAM,IAAI,GAAG;QACX,MAAM;QACN,KAAK;QACL,SAAS,EAAE,MAAM,YAAY,CAAC,SAAS,CAAC,EAAE;QAC1C,KAAK,EAAE,MAAe;QACtB,QAAQ;QACR,MAAM,EAAE,YAAY,CAAC,QAAQ,EAAE,MAAM,CAAC;KACvC,CAAC;IACF,IAAI,QAAQ,CAAC,OAAO,KAAK,KAAK,EAAE,CAAC;QAC/B,OAAO,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,cAAc,EAAE,MAAM,EAAE,iCAAiC,EAAE,CAAC;IAC1F,CAAC;IACD,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QAC9B,OAAO,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,cAAc,EAAE,MAAM,EAAE,+BAA+B,EAAE,CAAC;IACxF,CAAC;IACD,OAAO,KAAK,IAAI,SAAS;QACvB,CAAC,CAAC,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,oCAAoC,EAAE;QAC7E,CAAC,CAAC,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,uCAAuC,EAAE,CAAC;AACrF,CAAC;AAED,+CAA+C;AAC/C,SAAS,uBAAuB,CAC9B,MAAc,EACd,KAAa,EACb,SAAiB,EACjB,QAA2B;IAE3B,MAAM,IAAI,GAAG;QACX,MAAM;QACN,KAAK;QACL,SAAS,EAAE,MAAM,YAAY,CAAC,SAAS,CAAC,EAAE;QAC1C,KAAK,EAAE,qBAA8B;QACrC,QAAQ;QACR,MAAM,EAAE,YAAY,CAAC,QAAQ,EAAE,qBAAqB,CAAC;KACtD,CAAC;IACF,MAAM,KAAK,GAAG,QAAQ,CAAC,SAAS,CAAC;IACjC,MAAM,KAAK,GAAG,QAAQ,CAAC,SAAS,CAAC;IACjC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QAC3D,OAAO;YACL,GAAG,IAAI;YACP,QAAQ,EAAE,cAAc;YACxB,MAAM,EAAE,iDAAiD;SAC1D,CAAC;IACJ,CAAC;IACD,IAAI,KAAK,IAAI,SAAS,EAAE,CAAC;QACvB,OAAO;YACL,GAAG,IAAI;YACP,QAAQ,EAAE,MAAM;YAChB,MAAM,EAAE,2DAA2D;SACpE,CAAC;IACJ,CAAC;IACD,IAAI,KAAK,GAAG,SAAS,EAAE,CAAC;QACtB,OAAO;YACL,GAAG,IAAI;YACP,QAAQ,EAAE,MAAM;YAChB,MAAM,EAAE,qDAAqD;SAC9D,CAAC;IACJ,CAAC;IACD,OAAO;QACL,GAAG,IAAI;QACP,QAAQ,EAAE,cAAc;QACxB,MAAM,EAAE,+CAA+C;KACxD,CAAC;AACJ,CAAC;AAED,MAAM,cAAc,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC;AAExD,MAAM,eAAe,GAAG,CAAC;KACtB,MAAM,CAAC;IACN,IAAI,EAAE,cAAc;IACpB,UAAU,EAAE,cAAc;IAC1B,UAAU,EAAE,cAAc;IAC1B,oBAAoB,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvD,QAAQ,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC3C,KAAK,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;CAC9B,CAAC;KACD,KAAK,EAAE,CAAC;AAEX,MAAM,sBAAsB,GAAG,CAAC;KAC7B,MAAM,CAAC;IACN,MAAM,EAAE,CAAC;SACN,KAAK,CACJ,CAAC;SACE,MAAM,CAAC;QACN,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;QACnC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;QAChC,UAAU,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;QAClC,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,eAAe,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE;KAC7E,CAAC;SACD,KAAK,EAAE,CACX;SACA,QAAQ,EAAE;CACd,CAAC;KACD,KAAK,EAAE,CAAC;AAEX,MAAM,wBAAwB,GAAG,CAAC;KAC/B,MAAM,CAAC;IACN,SAAS,EAAE,CAAC;SACT,KAAK,CACJ,CAAC;SACE,MAAM,CAAC;QACN,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;QACpC,YAAY,EAAE,CAAC;aACZ,KAAK,CACJ,CAAC;aACE,MAAM,CAAC;YACN,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;YAChC,UAAU,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;YAClC,KAAK,EAAE,eAAe,CAAC,QAAQ,EAAE;YACjC,KAAK,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;SAC9B,CAAC;aACD,KAAK,EAAE,CACX;aACA,QAAQ,EAAE;KACd,CAAC;SACD,KAAK,EAAE,CACX;SACA,QAAQ,EAAE;CACd,CAAC;KACD,KAAK,EAAE,CAAC;AAEX,MAAM,eAAe,GAAG,CAAC;KACtB,MAAM,CAAC;IACN,UAAU,EAAE,CAAC;SACV,MAAM,CACL,CAAC,CAAC,MAAM,EAAE,EACV,CAAC;SACE,MAAM,CAAC;QACN,KAAK,EAAE,cAAc;QACrB,QAAQ,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;KAC5C,CAAC;SACD,KAAK,EAAE,CACX;SACA,QAAQ,EAAE;CACd,CAAC;KACD,KAAK,EAAE,CAAC;AAEX,MAAM,mBAAmB,GAAG,CAAC;KAC1B,MAAM,CAAC;IACN,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACtC,QAAQ,EAAE,CAAC,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE;IACtF,iBAAiB,EAAE,CAAC;SACjB,MAAM,CAAC;QACN,iBAAiB,EAAE,CAAC;aACjB,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC;aACpD,KAAK,EAAE;aACP,QAAQ,EAAE;aACV,QAAQ,EAAE;QACb,mBAAmB,EAAE,cAAc;QACnC,iBAAiB,EAAE,cAAc;QACjC,yBAAyB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;KAC5D,CAAC;SACD,KAAK,EAAE;SACP,QAAQ,EAAE;SACV,QAAQ,EAAE;IACb,mBAAmB,EAAE,cAAc;IACnC,iBAAiB,EAAE,cAAc;IACjC,yBAAyB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC3D,cAAc,EAAE,CAAC;SACd,MAAM,CAAC,EAAE,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC;SAC3D,KAAK,EAAE;SACP,QAAQ,EAAE;SACV,QAAQ,EAAE;IACb,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAChD,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAChD,CAAC;KACD,KAAK,EAAE,CAAC;AAEX,SAAS,UAAU,CAAC,IAAiD;IACnE,OAAO;QACL,IAAI,EAAE,IAAI,EAAE,IAAI,IAAI,IAAI;QACxB,SAAS,EAAE,IAAI,EAAE,UAAU,IAAI,IAAI;QACnC,SAAS,EAAE,IAAI,EAAE,UAAU,IAAI,IAAI;QACnC,GAAG,CAAC,IAAI,EAAE,QAAQ,KAAK,IAAI,IAAI,IAAI,EAAE,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC;QAC9F,GAAG,CAAC,IAAI,EAAE,oBAAoB,KAAK,IAAI,IAAI,IAAI,EAAE,oBAAoB,KAAK,SAAS;YACjF,CAAC,CAAC,EAAE;YACJ,CAAC,CAAC,EAAE,mBAAmB,EAAE,IAAI,CAAC,oBAAoB,EAAE,CAAC;KACxD,CAAC;AACJ,CAAC;AAED;;;;GAIG;AACH,SAAS,sBAAsB,CAC7B,QAAgB,EAChB,MAAc,EACd,QAA4B;IAE5B,QAAQ,CAAC,QAAQ,IAAI,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC;QAClC,KAAK,GAAG;YACN,OAAO,QAAQ,GAAG,MAAM,CAAC;QAC3B,KAAK,IAAI;YACP,OAAO,QAAQ,IAAI,MAAM,CAAC;QAC5B,KAAK,GAAG;YACN,OAAO,QAAQ,GAAG,MAAM,CAAC;QAC3B;YACE,OAAO,QAAQ,IAAI,MAAM,CAAC;IAC9B,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,mBAAmB,CACjC,YAAoC,EACpC,QAA4B;IAE5B,MAAM,MAAM,GAAG,mBAAmB,CAAC,SAAS,CAAC,YAAY,IAAI,EAAE,CAAC,CAAC;IACjE,MAAM,IAAI,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;IAC/C,MAAM,OAAO,GAAG,IAAI,CAAC,iBAAiB,IAAI,SAAS,CAAC;IACpD,MAAM,cAAc,GAAG,OAAO,EAAE,iBAAiB,EAAE,MAAM,IAAI,SAAS,CAAC;IACvE,MAAM,QAAQ,GAAG,OAAO,EAAE,mBAAmB,IAAI,IAAI,CAAC,mBAAmB,IAAI,SAAS,CAAC;IACvF,MAAM,MAAM,GAAG,OAAO,EAAE,iBAAiB,IAAI,IAAI,CAAC,iBAAiB,IAAI,SAAS,CAAC;IACjF,MAAM,QAAQ,GACZ,OAAO,EAAE,yBAAyB,IAAI,IAAI,CAAC,yBAAyB,IAAI,SAAS,CAAC;IACpF,MAAM,aAAa,GACjB,IAAI,CAAC,cAAc,EAAE,aAAa,IAAI,IAAI,CAAC,aAAa,IAAI,IAAI,CAAC,cAAc,IAAI,SAAS,CAAC;IAE/F,MAAM,iBAAiB,GACrB,cAAc,KAAK,SAAS;QAC5B,QAAQ,KAAK,IAAI;QACjB,MAAM,KAAK,IAAI;QACf,aAAa,KAAK,SAAS;QACzB,CAAC,CAAC,IAAI;QACN,CAAC,CAAC,cAAc,KAAK,SAAS;YAC5B,QAAQ,KAAK,SAAS;YACtB,MAAM,KAAK,SAAS;YACpB,aAAa,KAAK,SAAS,CAAC;IAElC,MAAM,IAAI,GAAG;QACX,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7C,GAAG,CAAC,OAAO,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC1E,GAAG,CAAC,OAAO,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACpE,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,yBAAyB,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC5D,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC3C,iBAAiB;KAClB,CAAC;IAEF,IAAI,aAAa,KAAK,SAAS,IAAI,aAAa,KAAK,IAAI,EAAE,CAAC;QAC1D,OAAO,EAAE,GAAG,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;IACtC,CAAC;IACD,MAAM,KAAK,GAAG,CAAC,QAAQ,IAAI,IAAI,CAAC,eAAe,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;IACrE,IAAI,KAAK,KAAK,EAAE,IAAI,KAAK,KAAK,WAAW,IAAI,KAAK,KAAK,UAAU,IAAI,KAAK,KAAK,WAAW,EAAE,CAAC;QAC3F,OAAO,EAAE,GAAG,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;IACtC,CAAC;IACD,IAAI,iBAAiB;QAAE,OAAO,EAAE,GAAG,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC;IAC7D,IACE,cAAc,KAAK,4BAA4B;QAC/C,cAAc,KAAK,wBAAwB,EAC3C,CAAC;QACD,OAAO,EAAE,GAAG,IAAI,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC;IAC1C,CAAC;IACD,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,IAAI,CAAC;QAAE,OAAO,EAAE,GAAG,IAAI,EAAE,MAAM,EAAE,gBAAgB,EAAE,CAAC;IAC5F,IAAI,OAAO,QAAQ,KAAK,QAAQ;QAAE,OAAO,EAAE,GAAG,IAAI,EAAE,MAAM,EAAE,eAAe,EAAE,CAAC;IAC9E,OAAO;QACL,GAAG,IAAI;QACP,MAAM,EAAE,sBAAsB,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,eAAe;KAC3F,CAAC;AACJ,CAAC;AAED,SAAS,aAAa,CACpB,YAAoC,EACpC,KAAmC;IAEnC,MAAM,MAAM,GAAG,mBAAmB,CAAC,SAAS,CAAC,YAAY,IAAI,EAAE,CAAC,CAAC;IACjE,MAAM,SAAS,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,QAAQ,EAAE,SAAS,IAAI,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAC9F,IAAI,OAAO,SAAS,KAAK,QAAQ,IAAI,SAAS,GAAG,CAAC,EAAE,CAAC;QACnD,OAAO,EAAE,OAAO,EAAE,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,YAAY,EAAE,SAAS,EAAE,CAAC;IAC5E,CAAC;IACD,MAAM,WAAW,GAAG,KAAK,CAAC,IAAI,CAC5B,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,IAAI,CAAC,SAAS,KAAK,QAAQ,IAAI,OAAO,IAAI,CAAC,SAAS,KAAK,QAAQ,CACnF,CAAC;IACF,OAAO,EAAE,OAAO,EAAE,WAAW,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,YAAY,EAAE,CAAC;AAC/D,CAAC;AAED,SAAS,cAAc,CACrB,OAAgE,EAChE,MAAuB;IAEvB,MAAM,KAAK,GAKL,EAAE,CAAC;IAET,MAAM,WAAW,GAAG,sBAAsB,CAAC,SAAS,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;IAC/E,MAAM,MAAM,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,IAAI,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IAC1E,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,UAAU,KAAK,IAAI,CAAC,IAAI,MAAM,CAAC,CAAC,CAAC,CAAC;IAC/E,MAAM,UAAU,GAAG,OAAO,EAAE,SAAS,CAAC;IAEtC,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;QACpC,KAAK,CAAC,IAAI,CAAC;YACT,MAAM,EAAE,SAAS;YACjB,KAAK,EAAE,6BAA6B;YACpC,SAAS,EAAE,MAAM,CAAC,UAAU;YAC5B,QAAQ,EAAE,UAAU,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC;SAC/C,CAAC,CAAC;IACL,CAAC;IAED,IAAI,MAAM,CAAC,WAAW,KAAK,SAAS,EAAE,CAAC;QACrC,MAAM,UAAU,GAAG,wBAAwB,CAAC,SAAS,CAAC,OAAO,CAAC,kBAAkB,CAAC,CAAC;QAClF,MAAM,SAAS,GAAG,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QAC9E,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;YACjC,MAAM,WAAW,GAAG,QAAQ,CAAC,YAAY,IAAI,EAAE,CAAC;YAChD,MAAM,IAAI,GACR,WAAW,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,UAAU,KAAK,IAAI,CAAC;gBACtD,CAAC,UAAU,KAAK,SAAS;oBACvB,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC;oBAChB,CAAC,CAAC,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,SAAS,KAAK,UAAU,CAAC,IAAI,SAAS,CAAC,CAAC,CAAC;YAClF,IAAI,IAAI,KAAK,SAAS;gBAAE,SAAS;YACjC,MAAM,IAAI,GAAG,QAAQ,CAAC,aAAa,IAAI,UAAU,CAAC;YAClD,KAAK,CAAC,IAAI,CAAC;gBACT,MAAM,EAAE,YAAY,IAAI,EAAE;gBAC1B,KAAK,EAAE,YAAY,IAAI,EAAE;gBACzB,SAAS,EAAE,MAAM,CAAC,WAAW;gBAC7B,QAAQ,EACN,IAAI,CAAC,KAAK,KAAK,SAAS,IAAI,IAAI,CAAC,KAAK,KAAK,IAAI;oBAC7C,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC;oBACxB,CAAC,CAAC,EAAE,GAAG,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,OAAO,EAAE,KAAK,EAAE;aACpD,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,OAAO,EAAE,KAAK,EAAE,CAAC;AACnB,CAAC;AAED,SAAS,aAAa,CAAC,OAAgB,EAAE,MAAuB;IAC9D,MAAM,MAAM,GAAG,eAAe,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC;IAClD,MAAM,SAAS,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IACvE,OAAO,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,SAAS,CAAC,EAAE,EAAE;QACpE,MAAM,IAAI,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC;QAC7B,MAAM,QAAQ,GAAsB;YAClC,KAAK,EAAE,IAAI,EAAE,KAAK,IAAI,IAAI;YAC1B,GAAG,CAAC,IAAI,EAAE,QAAQ,KAAK,IAAI,IAAI,IAAI,EAAE,QAAQ,KAAK,SAAS;gBACzD,CAAC,CAAC,EAAE;gBACJ,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC;SAChC,CAAC;QACF,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;YACvB,OAAO;gBACL,MAAM,EAAE,UAAU,IAAI,EAAE;gBACxB,KAAK,EAAE,IAAI;gBACX,SAAS,EAAE,MAAM,YAAY,CAAC,SAAS,CAAC,EAAE;gBAC1C,MAAM,EAAE,aAAa;gBACrB,QAAQ,EAAE,cAAc;gBACxB,MAAM,EAAE,oDAAoD;gBAC5D,KAAK,EAAE,MAAM;gBACb,QAAQ;aACT,CAAC;QACJ,CAAC;QACD,OAAO,SAAS,CAAC,UAAU,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;IAChE,CAAC,CAAC,CAAC;AACL,CAAC;AAED,SAAS,SAAS,CAAC,QAAsB,EAAE,MAAkC;IAC3E,IAAI,QAAQ,KAAK,MAAM,EAAE,CAAC;QACxB,OAAO,MAAM,CAAC,MAAM,KAAK,CAAC;YACxB,CAAC,CAAC,iCAAiC;YACnC,CAAC,CAAC,oCAAoC,CAAC;IAC3C,CAAC;IACD,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAC;IACtE,OAAO,SAAS,KAAK,SAAS;QAC5B,CAAC,CAAC,2BAA2B;QAC7B,CAAC,CAAC,GAAG,SAAS,CAAC,KAAK,KAAK,SAAS,CAAC,MAAM,EAAE,CAAC;AAChD,CAAC;AAED,4EAA4E;AAC5E,MAAM,UAAU,cAAc,CAAC,KAAoB;IACjD,MAAM,QAAQ,GAAG;QACf,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,GAAG,CAAC,KAAK,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,KAAK,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACzE,GAAG,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,KAAK,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACnE,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC9C,GAAG,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACjE,CAAC;IAEF,IAAI,KAAK,CAAC,MAAM,KAAK,WAAW,IAAI,KAAK,CAAC,OAAO,KAAK,SAAS,EAAE,CAAC;QAChE,OAAO;YACL,GAAG,QAAQ;YACX,QAAQ,EAAE,OAAO;YACjB,MAAM,EAAE,EAAE;YACV,MAAM,EACJ,KAAK,CAAC,KAAK;gBACX,CAAC,KAAK,CAAC,MAAM,KAAK,kBAAkB;oBAClC,CAAC,CAAC,uBAAuB;oBACzB,CAAC,CAAC,mCAAmC,CAAC;YAC1C,QAAQ,EAAE;gBACR,OAAO,EAAE,SAAS;gBAClB,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aACxD;SACF,CAAC;IACJ,CAAC;IAED,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;IAE9B,IAAI,OAAO,CAAC,WAAW,KAAK,UAAU,EAAE,CAAC;QACvC,MAAM,OAAO,GACX,OAAO,CAAC,WAAW,KAAK,cAAc,CAAC,CAAC,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC;QACrF,MAAM,MAAM,GAAG,aAAa,CAAC,OAAO,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;QACpD,MAAM,QAAQ,GAAG,aAAa,CAAC,KAAK,CAAC,MAAM,CAAC;YAC1C,CAAC,CAAC,cAAc;YAChB,CAAC,CAAC,aAAa,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC;QACzD,OAAO;YACL,GAAG,QAAQ;YACX,QAAQ;YACR,MAAM;YACN,MAAM,EAAE,aAAa,CAAC,KAAK,CAAC,MAAM,CAAC;gBACjC,CAAC,CAAC,mDAAmD;gBACrD,CAAC,CAAC,SAAS,CAAC,QAAQ,EAAE,MAAM,CAAC;YAC/B,QAAQ,EAAE;gBACR,OAAO,EAAE,YAAY;gBACrB,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aACxD;SACF,CAAC;IACJ,CAAC;IAED,MAAM,EAAE,KAAK,EAAE,GAAG,cAAc,CAAC,OAAO,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IACxD,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,aAAa,CAC1C,KAAK,CAAC,YAAY,EAClB,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,CACnC,CAAC;IACF,MAAM,WAAW,GACf,OAAO,KAAK,WAAW,CAAC,CAAC,CAAC,mBAAmB,CAAC,KAAK,CAAC,YAAY,EAAE,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAEhG,MAAM,MAAM,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAChC,OAAO,KAAK,WAAW;QACrB,CAAC,CAAC,uBAAuB,CAAC,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,QAAQ,CAAC;QACjF,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,QAAQ,CAAC,CACtE,CAAC;IAEF,MAAM,QAAQ,GAAqB;QACjC,OAAO;QACP,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACvD,GAAG,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC;QACjD,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACxC,CAAC;IAEF,IAAI,aAAa,CAAC,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC;QAChC,OAAO;YACL,GAAG,QAAQ;YACX,QAAQ,EAAE,cAAc;YACxB,MAAM;YACN,MAAM,EAAE,kDAAkD;YAC1D,QAAQ;SACT,CAAC;IACJ,CAAC;IAED,IAAI,QAAQ,GAAG,aAAa,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC;IACpE,IAAI,MAAM,GAAG,SAAS,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IAEzC,IAAI,WAAW,EAAE,MAAM,KAAK,OAAO,EAAE,CAAC;QACpC,QAAQ,GAAG,OAAO,CAAC;QACnB,MAAM,GAAG,WAAW,CAAC,aAAa,IAAI,wDAAwD,CAAC;IACjG,CAAC;SAAM,IAAI,WAAW,EAAE,MAAM,KAAK,eAAe,IAAI,QAAQ,KAAK,MAAM,EAAE,CAAC;QAC1E,QAAQ,GAAG,cAAc,CAAC;QAC1B,MAAM,GAAG,oEAAoE,CAAC;IAChF,CAAC;IAED,OAAO,EAAE,GAAG,QAAQ,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;AAC7D,CAAC;AAED,0EAA0E;AAC1E,MAAM,UAAU,mBAAmB,CACjC,gBAA6C;IAE7C,MAAM,MAAM,GAAiC;QAC3C,IAAI,EAAE,CAAC;QACP,IAAI,EAAE,CAAC;QACP,YAAY,EAAE,CAAC;QACf,KAAK,EAAE,CAAC;KACT,CAAC;IACF,KAAK,MAAM,KAAK,IAAI,gBAAgB;QAAE,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAClE,OAAO;QACL,aAAa,EACX,gBAAgB,CAAC,MAAM,KAAK,CAAC;YAC3B,CAAC,CAAC,cAAc;YAChB,CAAC,CAAC,aAAa,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;QACpE,MAAM;QACN,gBAAgB,EAAE,CAAC,GAAG,gBAAgB,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC;KACxF,CAAC;AACJ,CAAC"}
|
|
@@ -0,0 +1,31 @@
|
|
|
1
|
+
import { z } from 'zod';
|
|
2
|
+
/**
|
|
3
|
+
* Release-gate policy for one eval in a suite.
|
|
4
|
+
*
|
|
5
|
+
* The policy is authored by the developer next to the suite manifest; it adds
|
|
6
|
+
* no backend fields. Scenario evals gate on score thresholds, conversation and
|
|
7
|
+
* agent-trace evals gate on configured `per_metric.<metric>.score` thresholds.
|
|
8
|
+
*/
|
|
9
|
+
export type SuiteGatePolicy = {
|
|
10
|
+
/** Minimum primary-model overall score. */
|
|
11
|
+
minOverall?: number;
|
|
12
|
+
/** Minimum score for every configured scenario cell of the primary model. */
|
|
13
|
+
minScenario?: number;
|
|
14
|
+
/** Per-metric minimums for conversation and agent-trace evals. */
|
|
15
|
+
metrics?: Readonly<Record<string, number>>;
|
|
16
|
+
};
|
|
17
|
+
export declare const suiteGatePolicySchema: z.ZodObject<{
|
|
18
|
+
minOverall: z.ZodOptional<z.ZodNumber>;
|
|
19
|
+
minScenario: z.ZodOptional<z.ZodNumber>;
|
|
20
|
+
metrics: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodNumber>>;
|
|
21
|
+
}, z.core.$strict>;
|
|
22
|
+
export declare function isEmptyPolicy(policy: SuiteGatePolicy | undefined): boolean;
|
|
23
|
+
/** Per-eval policy wins over the suite-level default, field by field. */
|
|
24
|
+
type LoosePolicy = {
|
|
25
|
+
minOverall?: number | undefined;
|
|
26
|
+
minScenario?: number | undefined;
|
|
27
|
+
metrics?: Record<string, number> | undefined;
|
|
28
|
+
};
|
|
29
|
+
export declare function mergePolicies(base: LoosePolicy | undefined, override: LoosePolicy | undefined): SuiteGatePolicy;
|
|
30
|
+
export {};
|
|
31
|
+
//# sourceMappingURL=policy.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"policy.d.ts","sourceRoot":"","sources":["../../src/gate/policy.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB;;;;;;GAMG;AACH,MAAM,MAAM,eAAe,GAAG;IAC5B,2CAA2C;IAC3C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,6EAA6E;IAC7E,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,kEAAkE;IAClE,OAAO,CAAC,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC;CAC5C,CAAC;AAEF,eAAO,MAAM,qBAAqB;;;;kBAMvB,CAAC;AAEZ,wBAAgB,aAAa,CAAC,MAAM,EAAE,eAAe,GAAG,SAAS,GAAG,OAAO,CAI1E;AAED,yEAAyE;AACzE,KAAK,WAAW,GAAG;IACjB,UAAU,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAChC,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACjC,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,SAAS,CAAC;CAC9C,CAAC;AAEF,wBAAgB,aAAa,CAC3B,IAAI,EAAE,WAAW,GAAG,SAAS,EAC7B,QAAQ,EAAE,WAAW,GAAG,SAAS,GAChC,eAAe,CASjB"}
|
|
@@ -0,0 +1,26 @@
|
|
|
1
|
+
import { z } from 'zod';
|
|
2
|
+
export const suiteGatePolicySchema = z
|
|
3
|
+
.object({
|
|
4
|
+
minOverall: z.number().optional(),
|
|
5
|
+
minScenario: z.number().optional(),
|
|
6
|
+
metrics: z.record(z.string().min(1), z.number()).optional(),
|
|
7
|
+
})
|
|
8
|
+
.strict();
|
|
9
|
+
export function isEmptyPolicy(policy) {
|
|
10
|
+
if (policy === undefined)
|
|
11
|
+
return true;
|
|
12
|
+
if (policy.minOverall !== undefined || policy.minScenario !== undefined)
|
|
13
|
+
return false;
|
|
14
|
+
return Object.keys(policy.metrics ?? {}).length === 0;
|
|
15
|
+
}
|
|
16
|
+
export function mergePolicies(base, override) {
|
|
17
|
+
const metrics = { ...(base?.metrics ?? {}), ...(override?.metrics ?? {}) };
|
|
18
|
+
const minOverall = override?.minOverall ?? base?.minOverall;
|
|
19
|
+
const minScenario = override?.minScenario ?? base?.minScenario;
|
|
20
|
+
return {
|
|
21
|
+
...(minOverall === undefined ? {} : { minOverall }),
|
|
22
|
+
...(minScenario === undefined ? {} : { minScenario }),
|
|
23
|
+
...(Object.keys(metrics).length === 0 ? {} : { metrics }),
|
|
24
|
+
};
|
|
25
|
+
}
|
|
26
|
+
//# sourceMappingURL=policy.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"policy.js","sourceRoot":"","sources":["../../src/gate/policy.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAkBxB,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC;KACnC,MAAM,CAAC;IACN,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACjC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAClC,OAAO,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;CAC5D,CAAC;KACD,MAAM,EAAE,CAAC;AAEZ,MAAM,UAAU,aAAa,CAAC,MAAmC;IAC/D,IAAI,MAAM,KAAK,SAAS;QAAE,OAAO,IAAI,CAAC;IACtC,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,WAAW,KAAK,SAAS;QAAE,OAAO,KAAK,CAAC;IACtF,OAAO,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC;AACxD,CAAC;AASD,MAAM,UAAU,aAAa,CAC3B,IAA6B,EAC7B,QAAiC;IAEjC,MAAM,OAAO,GAAG,EAAE,GAAG,CAAC,IAAI,EAAE,OAAO,IAAI,EAAE,CAAC,EAAE,GAAG,CAAC,QAAQ,EAAE,OAAO,IAAI,EAAE,CAAC,EAAE,CAAC;IAC3E,MAAM,UAAU,GAAG,QAAQ,EAAE,UAAU,IAAI,IAAI,EAAE,UAAU,CAAC;IAC5D,MAAM,WAAW,GAAG,QAAQ,EAAE,WAAW,IAAI,IAAI,EAAE,WAAW,CAAC;IAC/D,OAAO;QACL,GAAG,CAAC,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC;QACnD,GAAG,CAAC,WAAW,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;QACrD,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC;KAC1D,CAAC;AACJ,CAAC"}
|
|
@@ -0,0 +1,20 @@
|
|
|
1
|
+
import type { ActionContext } from '../actions/context.js';
|
|
2
|
+
import type { SupportedModel, Workspace } from '../domain/types.js';
|
|
3
|
+
/**
|
|
4
|
+
* Everything the file guide needs from the backend, expressed as a
|
|
5
|
+
* narrow port so the flow itself stays deterministic and testable. Every method
|
|
6
|
+
* is backed by an existing shared action; the flow adds no network boundary.
|
|
7
|
+
*/
|
|
8
|
+
export interface EvalCreationBackend {
|
|
9
|
+
listWorkspaces(signal?: AbortSignal): Promise<readonly Workspace[]>;
|
|
10
|
+
listModels(signal?: AbortSignal): Promise<readonly SupportedModel[]>;
|
|
11
|
+
}
|
|
12
|
+
export declare class ActionEvalCreationBackend implements EvalCreationBackend {
|
|
13
|
+
#private;
|
|
14
|
+
constructor(input: {
|
|
15
|
+
actions: ActionContext;
|
|
16
|
+
});
|
|
17
|
+
listWorkspaces(signal?: AbortSignal): Promise<readonly Workspace[]>;
|
|
18
|
+
listModels(signal?: AbortSignal): Promise<readonly SupportedModel[]>;
|
|
19
|
+
}
|
|
20
|
+
//# sourceMappingURL=eval-creation-backend.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"eval-creation-backend.d.ts","sourceRoot":"","sources":["../../src/guide/eval-creation-backend.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAG3D,OAAO,KAAK,EAAE,cAAc,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAC;AAEpE;;;;GAIG;AACH,MAAM,WAAW,mBAAmB;IAClC,cAAc,CAAC,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,SAAS,SAAS,EAAE,CAAC,CAAC;IACpE,UAAU,CAAC,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,SAAS,cAAc,EAAE,CAAC,CAAC;CACtE;AAED,qBAAa,yBAA0B,YAAW,mBAAmB;;gBAGvD,KAAK,EAAE;QAAE,OAAO,EAAE,aAAa,CAAA;KAAE;IAIvC,cAAc,CAAC,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,SAAS,SAAS,EAAE,CAAC;IAKzE,UAAU,CAAC,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,SAAS,cAAc,EAAE,CAAC;CAGrE"}
|