openpond-sdk 0.9.2 → 0.9.4
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/README.md +24 -0
- package/dist/chunk-62LWLACP.js +155 -0
- package/dist/chunk-62LWLACP.js.map +7 -0
- package/dist/chunk-EMANJDJR.js +393 -0
- package/dist/chunk-EMANJDJR.js.map +7 -0
- package/dist/{chunk-QABCKPW6.js → chunk-TQ5ASK7D.js} +4 -4
- package/dist/{chunk-IKTZ2CD4.js → chunk-XR6CLSCS.js} +85 -43
- package/dist/chunk-XR6CLSCS.js.map +7 -0
- package/dist/dataset-workspaces.js +3 -3
- package/dist/experiments.js +57 -0
- package/dist/experiments.js.map +7 -0
- package/dist/index.js +12 -3
- package/dist/index.js.map +2 -2
- package/dist/learning.js +3 -3
- package/dist/model-starter-attempts.js +3 -1
- package/dist/model-taskset-runs.js +21 -113
- package/dist/model-taskset-runs.js.map +4 -4
- package/dist/taskset-drafts.js +2 -2
- package/dist/taskset-packages.js +2 -2
- package/dist/types/packages/sdk/src/experiment-client.d.ts +2441 -0
- package/dist/types/packages/sdk/src/experiment-client.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/experiment-contracts.d.ts +570 -0
- package/dist/types/packages/sdk/src/experiment-contracts.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/experiment-field-mappings.d.ts +43 -0
- package/dist/types/packages/sdk/src/experiment-field-mappings.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/experiment-grading-contracts.d.ts +79 -0
- package/dist/types/packages/sdk/src/experiment-grading-contracts.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/experiment-scoring-contracts.d.ts +192 -0
- package/dist/types/packages/sdk/src/experiment-scoring-contracts.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/experiments.d.ts +7 -0
- package/dist/types/packages/sdk/src/experiments.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/harness-experiment-policy.d.ts +44 -0
- package/dist/types/packages/sdk/src/harness-experiment-policy.d.ts.map +1 -0
- package/dist/types/packages/sdk/src/index.d.ts +2 -0
- package/dist/types/packages/sdk/src/index.d.ts.map +1 -1
- package/dist/types/packages/sdk/src/model-starter-attempts.d.ts +123 -0
- package/dist/types/packages/sdk/src/model-starter-attempts.d.ts.map +1 -1
- package/dist/types/packages/sdk/src/model-taskset-runs-contracts.d.ts +159 -0
- package/dist/types/packages/sdk/src/model-taskset-runs-contracts.d.ts.map +1 -1
- package/dist/types/packages/sdk/src/model-taskset-runs.d.ts +129 -5
- package/dist/types/packages/sdk/src/model-taskset-runs.d.ts.map +1 -1
- package/package.json +6 -2
- package/dist/chunk-IKTZ2CD4.js.map +0 -7
- /package/dist/{chunk-QABCKPW6.js.map → chunk-TQ5ASK7D.js.map} +0 -0
|
@@ -0,0 +1,79 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
export declare const ExperimentAttemptGradeSchema: z.ZodObject<{
|
|
3
|
+
schemaVersion: z.ZodLiteral<"openpond.experimentAttemptGrade.v1">;
|
|
4
|
+
selectionHash: z.ZodString;
|
|
5
|
+
grades: z.ZodArray<z.ZodObject<{
|
|
6
|
+
score: z.ZodNullable<z.ZodNumber>;
|
|
7
|
+
passed: z.ZodBoolean;
|
|
8
|
+
rewardEligible: z.ZodBoolean;
|
|
9
|
+
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
10
|
+
policy_failure: "policy_failure";
|
|
11
|
+
grader_failure: "grader_failure";
|
|
12
|
+
environment_failure: "environment_failure";
|
|
13
|
+
infrastructure_failure: "infrastructure_failure";
|
|
14
|
+
timeout: "timeout";
|
|
15
|
+
cancelled: "cancelled";
|
|
16
|
+
}>>;
|
|
17
|
+
gradingStatus: z.ZodEnum<{
|
|
18
|
+
scored: "scored";
|
|
19
|
+
unscorable: "unscorable";
|
|
20
|
+
not_configured: "not_configured";
|
|
21
|
+
}>;
|
|
22
|
+
schemaVersion: z.ZodLiteral<"openpond.taskGrade.v1">;
|
|
23
|
+
taskHash: z.ZodString;
|
|
24
|
+
evidenceHash: z.ZodString;
|
|
25
|
+
graderSetHash: z.ZodString;
|
|
26
|
+
components: z.ZodArray<z.ZodObject<{
|
|
27
|
+
schemaVersion: z.ZodLiteral<"openpond.graderEvidence.v1">;
|
|
28
|
+
graderId: z.ZodString;
|
|
29
|
+
graderVersion: z.ZodString;
|
|
30
|
+
score: z.ZodNullable<z.ZodNumber>;
|
|
31
|
+
passed: z.ZodBoolean;
|
|
32
|
+
rewardEligible: z.ZodBoolean;
|
|
33
|
+
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
34
|
+
policy_failure: "policy_failure";
|
|
35
|
+
grader_failure: "grader_failure";
|
|
36
|
+
environment_failure: "environment_failure";
|
|
37
|
+
infrastructure_failure: "infrastructure_failure";
|
|
38
|
+
timeout: "timeout";
|
|
39
|
+
cancelled: "cancelled";
|
|
40
|
+
}>>;
|
|
41
|
+
feedback: z.ZodArray<z.ZodString>;
|
|
42
|
+
visibleEvidenceRefs: z.ZodArray<z.ZodString>;
|
|
43
|
+
privilegedEvidenceRefs: z.ZodArray<z.ZodString>;
|
|
44
|
+
modelJudgeReceipt: z.ZodOptional<z.ZodObject<{
|
|
45
|
+
schemaVersion: z.ZodLiteral<"openpond.modelJudgeReceipt.v1">;
|
|
46
|
+
providerId: z.ZodString;
|
|
47
|
+
modelId: z.ZodString;
|
|
48
|
+
modelRevision: z.ZodNullable<z.ZodString>;
|
|
49
|
+
responseId: z.ZodNullable<z.ZodString>;
|
|
50
|
+
requestHash: z.ZodString;
|
|
51
|
+
responseHash: z.ZodString;
|
|
52
|
+
inputTokens: z.ZodNullable<z.ZodNumber>;
|
|
53
|
+
outputTokens: z.ZodNullable<z.ZodNumber>;
|
|
54
|
+
costUsd: z.ZodNullable<z.ZodNumber>;
|
|
55
|
+
}, z.core.$strict>>;
|
|
56
|
+
contentHash: z.ZodString;
|
|
57
|
+
}, z.core.$strict>>;
|
|
58
|
+
contentHash: z.ZodString;
|
|
59
|
+
}, z.core.$strict>>;
|
|
60
|
+
score: z.ZodNullable<z.ZodNumber>;
|
|
61
|
+
passed: z.ZodBoolean;
|
|
62
|
+
rewardEligible: z.ZodBoolean;
|
|
63
|
+
gradingStatus: z.ZodEnum<{
|
|
64
|
+
unscorable: "unscorable";
|
|
65
|
+
scored: "scored";
|
|
66
|
+
not_configured: "not_configured";
|
|
67
|
+
}>;
|
|
68
|
+
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
69
|
+
policy_failure: "policy_failure";
|
|
70
|
+
grader_failure: "grader_failure";
|
|
71
|
+
environment_failure: "environment_failure";
|
|
72
|
+
infrastructure_failure: "infrastructure_failure";
|
|
73
|
+
timeout: "timeout";
|
|
74
|
+
cancelled: "cancelled";
|
|
75
|
+
}>>;
|
|
76
|
+
contentHash: z.ZodString;
|
|
77
|
+
}, z.core.$strict>;
|
|
78
|
+
export type ExperimentAttemptGrade = z.infer<typeof ExperimentAttemptGradeSchema>;
|
|
79
|
+
//# sourceMappingURL=experiment-grading-contracts.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"experiment-grading-contracts.d.ts","sourceRoot":"","sources":["../../../../../src/experiment-grading-contracts.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,4BAA4B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAmBvC,CAAC;AACH,MAAM,MAAM,sBAAsB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,4BAA4B,CAAC,CAAC"}
|
|
@@ -0,0 +1,192 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
export declare const ExperimentScoringRequestSchema: z.ZodObject<{
|
|
3
|
+
operationId: z.ZodString;
|
|
4
|
+
execution: z.ZodObject<{
|
|
5
|
+
id: z.ZodString;
|
|
6
|
+
contentHash: z.ZodString;
|
|
7
|
+
}, z.core.$strict>;
|
|
8
|
+
graders: z.ZodArray<z.ZodObject<{
|
|
9
|
+
id: z.ZodString;
|
|
10
|
+
contentHash: z.ZodString;
|
|
11
|
+
revision: z.ZodNumber;
|
|
12
|
+
}, z.core.$strict>>;
|
|
13
|
+
maximumCostUsd: z.ZodNumber;
|
|
14
|
+
mappings: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
15
|
+
graderId: z.ZodString;
|
|
16
|
+
fields: z.ZodArray<z.ZodObject<{
|
|
17
|
+
destination: z.ZodEnum<{
|
|
18
|
+
input: "input";
|
|
19
|
+
output: "output";
|
|
20
|
+
expectedOutput: "expectedOutput";
|
|
21
|
+
evaluatorContext: "evaluatorContext";
|
|
22
|
+
}>;
|
|
23
|
+
path: z.ZodArray<z.ZodString>;
|
|
24
|
+
source: z.ZodEnum<{
|
|
25
|
+
input: "input";
|
|
26
|
+
output: "output";
|
|
27
|
+
expectedOutput: "expectedOutput";
|
|
28
|
+
evaluatorContext: "evaluatorContext";
|
|
29
|
+
}>;
|
|
30
|
+
sourcePath: z.ZodArray<z.ZodString>;
|
|
31
|
+
}, z.core.$strict>>;
|
|
32
|
+
}, z.core.$strict>>>;
|
|
33
|
+
}, z.core.$strict>;
|
|
34
|
+
export type ExperimentScoringRequest = z.infer<typeof ExperimentScoringRequestSchema>;
|
|
35
|
+
export declare const ExperimentScoringPassSchema: z.ZodObject<{
|
|
36
|
+
schemaVersion: z.ZodLiteral<"openpond.experimentScoringPass.v1">;
|
|
37
|
+
id: z.ZodString;
|
|
38
|
+
teamId: z.ZodString;
|
|
39
|
+
revision: z.ZodNumber;
|
|
40
|
+
request: z.ZodObject<{
|
|
41
|
+
operationId: z.ZodString;
|
|
42
|
+
execution: z.ZodObject<{
|
|
43
|
+
id: z.ZodString;
|
|
44
|
+
contentHash: z.ZodString;
|
|
45
|
+
}, z.core.$strict>;
|
|
46
|
+
graders: z.ZodArray<z.ZodObject<{
|
|
47
|
+
id: z.ZodString;
|
|
48
|
+
contentHash: z.ZodString;
|
|
49
|
+
revision: z.ZodNumber;
|
|
50
|
+
}, z.core.$strict>>;
|
|
51
|
+
maximumCostUsd: z.ZodNumber;
|
|
52
|
+
mappings: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
53
|
+
graderId: z.ZodString;
|
|
54
|
+
fields: z.ZodArray<z.ZodObject<{
|
|
55
|
+
destination: z.ZodEnum<{
|
|
56
|
+
input: "input";
|
|
57
|
+
output: "output";
|
|
58
|
+
expectedOutput: "expectedOutput";
|
|
59
|
+
evaluatorContext: "evaluatorContext";
|
|
60
|
+
}>;
|
|
61
|
+
path: z.ZodArray<z.ZodString>;
|
|
62
|
+
source: z.ZodEnum<{
|
|
63
|
+
input: "input";
|
|
64
|
+
output: "output";
|
|
65
|
+
expectedOutput: "expectedOutput";
|
|
66
|
+
evaluatorContext: "evaluatorContext";
|
|
67
|
+
}>;
|
|
68
|
+
sourcePath: z.ZodArray<z.ZodString>;
|
|
69
|
+
}, z.core.$strict>>;
|
|
70
|
+
}, z.core.$strict>>>;
|
|
71
|
+
}, z.core.$strict>;
|
|
72
|
+
graders: z.ZodArray<z.ZodObject<{
|
|
73
|
+
id: z.ZodString;
|
|
74
|
+
version: z.ZodString;
|
|
75
|
+
contentHash: z.ZodString;
|
|
76
|
+
feedbackKey: z.ZodString;
|
|
77
|
+
release: z.ZodNullable<z.ZodObject<{
|
|
78
|
+
id: z.ZodString;
|
|
79
|
+
revision: z.ZodNumber;
|
|
80
|
+
contentHash: z.ZodString;
|
|
81
|
+
}, z.core.$strict>>;
|
|
82
|
+
mappings: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
83
|
+
destination: z.ZodEnum<{
|
|
84
|
+
input: "input";
|
|
85
|
+
output: "output";
|
|
86
|
+
expectedOutput: "expectedOutput";
|
|
87
|
+
evaluatorContext: "evaluatorContext";
|
|
88
|
+
}>;
|
|
89
|
+
path: z.ZodArray<z.ZodString>;
|
|
90
|
+
source: z.ZodEnum<{
|
|
91
|
+
input: "input";
|
|
92
|
+
output: "output";
|
|
93
|
+
expectedOutput: "expectedOutput";
|
|
94
|
+
evaluatorContext: "evaluatorContext";
|
|
95
|
+
}>;
|
|
96
|
+
sourcePath: z.ZodArray<z.ZodString>;
|
|
97
|
+
}, z.core.$strict>>>;
|
|
98
|
+
}, z.core.$strict>>;
|
|
99
|
+
status: z.ZodEnum<{
|
|
100
|
+
failed: "failed";
|
|
101
|
+
completed: "completed";
|
|
102
|
+
cancelled: "cancelled";
|
|
103
|
+
running: "running";
|
|
104
|
+
cancelling: "cancelling";
|
|
105
|
+
queued: "queued";
|
|
106
|
+
}>;
|
|
107
|
+
totalCount: z.ZodNumber;
|
|
108
|
+
counts: z.ZodObject<{
|
|
109
|
+
pending: z.ZodNumber;
|
|
110
|
+
running: z.ZodNumber;
|
|
111
|
+
scored: z.ZodNumber;
|
|
112
|
+
unavailable: z.ZodNumber;
|
|
113
|
+
failed: z.ZodNumber;
|
|
114
|
+
cancelled: z.ZodNumber;
|
|
115
|
+
}, z.core.$strict>;
|
|
116
|
+
resultAvailable: z.ZodBoolean;
|
|
117
|
+
createdAt: z.ZodISODateTime;
|
|
118
|
+
startedAt: z.ZodNullable<z.ZodISODateTime>;
|
|
119
|
+
completedAt: z.ZodNullable<z.ZodISODateTime>;
|
|
120
|
+
error: z.ZodNullable<z.ZodObject<{
|
|
121
|
+
code: z.ZodString;
|
|
122
|
+
message: z.ZodString;
|
|
123
|
+
}, z.core.$strict>>;
|
|
124
|
+
contentHash: z.ZodString;
|
|
125
|
+
}, z.core.$strict>;
|
|
126
|
+
export type ExperimentScoringPass = z.infer<typeof ExperimentScoringPassSchema>;
|
|
127
|
+
/** A pass grades owned output and carries no model-execution authority. */
|
|
128
|
+
export declare function verifyExperimentScoringPass(value: unknown): {
|
|
129
|
+
schemaVersion: "openpond.experimentScoringPass.v1";
|
|
130
|
+
id: string;
|
|
131
|
+
teamId: string;
|
|
132
|
+
revision: number;
|
|
133
|
+
request: {
|
|
134
|
+
operationId: string;
|
|
135
|
+
execution: {
|
|
136
|
+
id: string;
|
|
137
|
+
contentHash: string;
|
|
138
|
+
};
|
|
139
|
+
graders: {
|
|
140
|
+
id: string;
|
|
141
|
+
contentHash: string;
|
|
142
|
+
revision: number;
|
|
143
|
+
}[];
|
|
144
|
+
maximumCostUsd: number;
|
|
145
|
+
mappings?: {
|
|
146
|
+
graderId: string;
|
|
147
|
+
fields: {
|
|
148
|
+
destination: "input" | "output" | "expectedOutput" | "evaluatorContext";
|
|
149
|
+
path: string[];
|
|
150
|
+
source: "input" | "output" | "expectedOutput" | "evaluatorContext";
|
|
151
|
+
sourcePath: string[];
|
|
152
|
+
}[];
|
|
153
|
+
}[] | undefined;
|
|
154
|
+
};
|
|
155
|
+
graders: {
|
|
156
|
+
id: string;
|
|
157
|
+
version: string;
|
|
158
|
+
contentHash: string;
|
|
159
|
+
feedbackKey: string;
|
|
160
|
+
release: {
|
|
161
|
+
id: string;
|
|
162
|
+
revision: number;
|
|
163
|
+
contentHash: string;
|
|
164
|
+
} | null;
|
|
165
|
+
mappings?: {
|
|
166
|
+
destination: "input" | "output" | "expectedOutput" | "evaluatorContext";
|
|
167
|
+
path: string[];
|
|
168
|
+
source: "input" | "output" | "expectedOutput" | "evaluatorContext";
|
|
169
|
+
sourcePath: string[];
|
|
170
|
+
}[] | undefined;
|
|
171
|
+
}[];
|
|
172
|
+
status: "failed" | "completed" | "cancelled" | "running" | "cancelling" | "queued";
|
|
173
|
+
totalCount: number;
|
|
174
|
+
counts: {
|
|
175
|
+
pending: number;
|
|
176
|
+
running: number;
|
|
177
|
+
scored: number;
|
|
178
|
+
unavailable: number;
|
|
179
|
+
failed: number;
|
|
180
|
+
cancelled: number;
|
|
181
|
+
};
|
|
182
|
+
resultAvailable: boolean;
|
|
183
|
+
createdAt: string;
|
|
184
|
+
startedAt: string | null;
|
|
185
|
+
completedAt: string | null;
|
|
186
|
+
error: {
|
|
187
|
+
code: string;
|
|
188
|
+
message: string;
|
|
189
|
+
} | null;
|
|
190
|
+
contentHash: string;
|
|
191
|
+
};
|
|
192
|
+
//# sourceMappingURL=experiment-scoring-contracts.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"experiment-scoring-contracts.d.ts","sourceRoot":"","sources":["../../../../../src/experiment-scoring-contracts.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAOxB,eAAO,MAAM,8BAA8B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAYzC,CAAC;AACH,MAAM,MAAM,wBAAwB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,8BAA8B,CAAC,CAAC;AAgBtF,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAiF,CAAC;AAC1H,MAAM,MAAM,qBAAqB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,2BAA2B,CAAC,CAAC;AAEhF,2EAA2E;AAC3E,wBAAgB,2BAA2B,CAAC,KAAK,EAAE,OAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAezD"}
|
|
@@ -0,0 +1,7 @@
|
|
|
1
|
+
export * from "./experiment-contracts.js";
|
|
2
|
+
export * from "./experiment-client.js";
|
|
3
|
+
export * from "./experiment-scoring-contracts.js";
|
|
4
|
+
export * from "./experiment-field-mappings.js";
|
|
5
|
+
export * from "./experiment-grading-contracts.js";
|
|
6
|
+
export * from "./harness-experiment-policy.js";
|
|
7
|
+
//# sourceMappingURL=experiments.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"experiments.d.ts","sourceRoot":"","sources":["../../../../../src/experiments.ts"],"names":[],"mappings":"AAAA,cAAc,2BAA2B,CAAC;AAC1C,cAAc,wBAAwB,CAAC;AACvC,cAAc,mCAAmC,CAAC;AAClD,cAAc,gCAAgC,CAAC;AAC/C,cAAc,mCAAmC,CAAC;AAClD,cAAc,gCAAgC,CAAC"}
|
|
@@ -0,0 +1,44 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
/** The host admits this released component through its native Harness engine.
|
|
3
|
+
* Chat prompt/settings are owned by the released source, not replaced here. */
|
|
4
|
+
export declare const HarnessExperimentPolicySchema: z.ZodObject<{
|
|
5
|
+
kind: z.ZodLiteral<"hosted_harness">;
|
|
6
|
+
modelId: z.ZodString;
|
|
7
|
+
reasoningEffort: z.ZodOptional<z.ZodEnum<{
|
|
8
|
+
high: "high";
|
|
9
|
+
low: "low";
|
|
10
|
+
off: "off";
|
|
11
|
+
medium: "medium";
|
|
12
|
+
xhigh: "xhigh";
|
|
13
|
+
max: "max";
|
|
14
|
+
}>>;
|
|
15
|
+
profileRepositoryId: z.ZodString;
|
|
16
|
+
source: z.ZodObject<{
|
|
17
|
+
profileId: z.ZodString;
|
|
18
|
+
sourceRevision: z.ZodString;
|
|
19
|
+
harnessRelease: z.ZodObject<{
|
|
20
|
+
id: z.ZodString;
|
|
21
|
+
contentHash: z.ZodString;
|
|
22
|
+
}, z.core.$strict>;
|
|
23
|
+
catalogHash: z.ZodString;
|
|
24
|
+
definitionId: z.ZodString;
|
|
25
|
+
definitionHash: z.ZodString;
|
|
26
|
+
target: z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
27
|
+
kind: z.ZodLiteral<"profile">;
|
|
28
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
29
|
+
kind: z.ZodLiteral<"workflow">;
|
|
30
|
+
workflowId: z.ZodString;
|
|
31
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
32
|
+
kind: z.ZodLiteral<"skill">;
|
|
33
|
+
skillPath: z.ZodString;
|
|
34
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
35
|
+
kind: z.ZodLiteral<"agent_action">;
|
|
36
|
+
actionId: z.ZodString;
|
|
37
|
+
}, z.core.$strict>], "kind">;
|
|
38
|
+
environmentHash: z.ZodString;
|
|
39
|
+
}, z.core.$strict>;
|
|
40
|
+
modelConfigurationHash: z.ZodString;
|
|
41
|
+
packageHash: z.ZodString;
|
|
42
|
+
}, z.core.$strict>;
|
|
43
|
+
export type HarnessExperimentPolicy = z.infer<typeof HarnessExperimentPolicySchema>;
|
|
44
|
+
//# sourceMappingURL=harness-experiment-policy.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"harness-experiment-policy.d.ts","sourceRoot":"","sources":["../../../../../src/harness-experiment-policy.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAMxB;+EAC+E;AAC/E,eAAO,MAAM,6BAA6B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAQ/B,CAAC;AACZ,MAAM,MAAM,uBAAuB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,6BAA6B,CAAC,CAAC"}
|
|
@@ -5,6 +5,7 @@ import { OpenPondProfileActionsClient } from "./profile-actions.js";
|
|
|
5
5
|
import { OpenPondWorkflowsClient } from "./workflows.js";
|
|
6
6
|
import type { OpenPondClientOptions } from "./types.js";
|
|
7
7
|
import { OpenPondLearningClient } from "./learning-client.js";
|
|
8
|
+
import { OpenPondExperimentsClient } from "./experiment-client.js";
|
|
8
9
|
export declare class OpenPondClient {
|
|
9
10
|
readonly sandboxes: OpenPondSandboxClient;
|
|
10
11
|
readonly work: OpenPondWorkClient;
|
|
@@ -12,6 +13,7 @@ export declare class OpenPondClient {
|
|
|
12
13
|
readonly actions: OpenPondProjectActionsClient;
|
|
13
14
|
readonly profileActions: OpenPondProfileActionsClient;
|
|
14
15
|
readonly learning: (scope: string) => OpenPondLearningClient;
|
|
16
|
+
readonly experiments: (teamId: string) => OpenPondExperimentsClient;
|
|
15
17
|
constructor(options: OpenPondClientOptions);
|
|
16
18
|
}
|
|
17
19
|
export declare function createOpenPondClient(options: OpenPondClientOptions): OpenPondClient;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAEL,KAAK,qBAAqB,EAC3B,MAAM,gCAAgC,CAAC;AAGxC,OAAO,EAAE,kBAAkB,EAAE,MAAM,WAAW,CAAC;AAC/C,OAAO,EAAE,4BAA4B,EAAE,MAAM,sBAAsB,CAAC;AACpE,OAAO,EAAE,4BAA4B,EAAE,MAAM,sBAAsB,CAAC;AACpE,OAAO,EAAE,uBAAuB,EAAE,MAAM,gBAAgB,CAAC;AACzD,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,YAAY,CAAC;AACxD,OAAO,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAEL,KAAK,qBAAqB,EAC3B,MAAM,gCAAgC,CAAC;AAGxC,OAAO,EAAE,kBAAkB,EAAE,MAAM,WAAW,CAAC;AAC/C,OAAO,EAAE,4BAA4B,EAAE,MAAM,sBAAsB,CAAC;AACpE,OAAO,EAAE,4BAA4B,EAAE,MAAM,sBAAsB,CAAC;AACpE,OAAO,EAAE,uBAAuB,EAAE,MAAM,gBAAgB,CAAC;AACzD,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,YAAY,CAAC;AACxD,OAAO,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAC9D,OAAO,EAAE,yBAAyB,EAAE,MAAM,wBAAwB,CAAC;AAGnE,qBAAa,cAAc;IACzB,QAAQ,CAAC,SAAS,EAAE,qBAAqB,CAAC;IAC1C,QAAQ,CAAC,IAAI,EAAE,kBAAkB,CAAC;IAClC,QAAQ,CAAC,SAAS,EAAE,uBAAuB,CAAC;IAC5C,QAAQ,CAAC,OAAO,EAAE,4BAA4B,CAAC;IAC/C,QAAQ,CAAC,cAAc,EAAE,4BAA4B,CAAC;IACtD,QAAQ,CAAC,QAAQ,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,sBAAsB,CAAC;IAC7D,QAAQ,CAAC,WAAW,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,yBAAyB,CAAC;gBAExD,OAAO,EAAE,qBAAqB;CAuC3C;AAED,wBAAgB,oBAAoB,CAAC,OAAO,EAAE,qBAAqB,GAAG,cAAc,CAEnF;AAED,YAAY,EAAE,qBAAqB,EAAE,MAAM,YAAY,CAAC;AACxD,OAAO,EAAE,kBAAkB,EAAE,MAAM,WAAW,CAAC;AAC/C,OAAO,EAAE,uBAAuB,EAAE,MAAM,gBAAgB,CAAC;AACzD,OAAO,EAAE,4BAA4B,EAAE,MAAM,sBAAsB,CAAC;AACpE,OAAO,EAAE,4BAA4B,EAAE,MAAM,sBAAsB,CAAC;AACpE,YAAY,EACV,0BAA0B,EAC1B,uBAAuB,EACvB,oBAAoB,GACrB,MAAM,sBAAsB,CAAC;AAC9B,YAAY,EACV,4BAA4B,EAC5B,iCAAiC,EACjC,+BAA+B,EAC/B,qCAAqC,GACtC,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAAE,gBAAgB,EAAE,MAAM,0BAA0B,CAAC;AAC5D,YAAY,EACV,iBAAiB,EACjB,mBAAmB,EACnB,0BAA0B,EAC1B,qBAAqB,EACrB,qBAAqB,EACrB,kBAAkB,EAClB,oCAAoC,EACpC,oBAAoB,EACpB,qBAAqB,GACtB,MAAM,WAAW,CAAC;AACnB,YAAY,EACV,uBAAuB,EACvB,2BAA2B,EAC3B,4BAA4B,EAC5B,0BAA0B,EAC1B,4BAA4B,EAC5B,0BAA0B,EAC1B,8BAA8B,EAC9B,mBAAmB,EACnB,4BAA4B,EAC5B,wBAAwB,EACxB,2BAA2B,EAC3B,4BAA4B,EAC5B,uBAAuB,GACxB,MAAM,gBAAgB,CAAC;AACxB,cAAc,cAAc,CAAC;AAC7B,OAAO,EAAE,sBAAsB,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AACrF,YAAY,EAAE,6BAA6B,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAElG,cAAc,gCAAgC,CAAC;AAC/C,cAAc,+BAA+B,CAAC;AAC9C,OAAO,EACL,cAAc,EACd,iBAAiB,EACjB,gBAAgB,EAChB,mBAAmB,EACnB,uBAAuB,EACvB,kBAAkB,EAClB,oBAAoB,GACrB,MAAM,6BAA6B,CAAC;AACrC,YAAY,EACV,oBAAoB,EACpB,iBAAiB,EACjB,qBAAqB,EACrB,cAAc,EACd,kBAAkB,EAClB,eAAe,EACf,WAAW,EACX,oBAAoB,EACpB,cAAc,EACd,uBAAuB,GACxB,MAAM,6BAA6B,CAAC"}
|
|
@@ -470,6 +470,82 @@ export declare const ModelStarterAttemptResultSchema: z.ZodObject<{
|
|
|
470
470
|
}, z.core.$strict>>;
|
|
471
471
|
contentHash: z.ZodString;
|
|
472
472
|
}, z.core.$strict>>>;
|
|
473
|
+
experimentGrade: z.ZodOptional<z.ZodNullable<z.ZodObject<{
|
|
474
|
+
schemaVersion: z.ZodLiteral<"openpond.experimentAttemptGrade.v1">;
|
|
475
|
+
selectionHash: z.ZodString;
|
|
476
|
+
grades: z.ZodArray<z.ZodObject<{
|
|
477
|
+
score: z.ZodNullable<z.ZodNumber>;
|
|
478
|
+
passed: z.ZodBoolean;
|
|
479
|
+
rewardEligible: z.ZodBoolean;
|
|
480
|
+
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
481
|
+
policy_failure: "policy_failure";
|
|
482
|
+
grader_failure: "grader_failure";
|
|
483
|
+
environment_failure: "environment_failure";
|
|
484
|
+
infrastructure_failure: "infrastructure_failure";
|
|
485
|
+
timeout: "timeout";
|
|
486
|
+
cancelled: "cancelled";
|
|
487
|
+
}>>;
|
|
488
|
+
gradingStatus: z.ZodEnum<{
|
|
489
|
+
scored: "scored";
|
|
490
|
+
unscorable: "unscorable";
|
|
491
|
+
not_configured: "not_configured";
|
|
492
|
+
}>;
|
|
493
|
+
schemaVersion: z.ZodLiteral<"openpond.taskGrade.v1">;
|
|
494
|
+
taskHash: z.ZodString;
|
|
495
|
+
evidenceHash: z.ZodString;
|
|
496
|
+
graderSetHash: z.ZodString;
|
|
497
|
+
components: z.ZodArray<z.ZodObject<{
|
|
498
|
+
schemaVersion: z.ZodLiteral<"openpond.graderEvidence.v1">;
|
|
499
|
+
graderId: z.ZodString;
|
|
500
|
+
graderVersion: z.ZodString;
|
|
501
|
+
score: z.ZodNullable<z.ZodNumber>;
|
|
502
|
+
passed: z.ZodBoolean;
|
|
503
|
+
rewardEligible: z.ZodBoolean;
|
|
504
|
+
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
505
|
+
policy_failure: "policy_failure";
|
|
506
|
+
grader_failure: "grader_failure";
|
|
507
|
+
environment_failure: "environment_failure";
|
|
508
|
+
infrastructure_failure: "infrastructure_failure";
|
|
509
|
+
timeout: "timeout";
|
|
510
|
+
cancelled: "cancelled";
|
|
511
|
+
}>>;
|
|
512
|
+
feedback: z.ZodArray<z.ZodString>;
|
|
513
|
+
visibleEvidenceRefs: z.ZodArray<z.ZodString>;
|
|
514
|
+
privilegedEvidenceRefs: z.ZodArray<z.ZodString>;
|
|
515
|
+
modelJudgeReceipt: z.ZodOptional<z.ZodObject<{
|
|
516
|
+
schemaVersion: z.ZodLiteral<"openpond.modelJudgeReceipt.v1">;
|
|
517
|
+
providerId: z.ZodString;
|
|
518
|
+
modelId: z.ZodString;
|
|
519
|
+
modelRevision: z.ZodNullable<z.ZodString>;
|
|
520
|
+
responseId: z.ZodNullable<z.ZodString>;
|
|
521
|
+
requestHash: z.ZodString;
|
|
522
|
+
responseHash: z.ZodString;
|
|
523
|
+
inputTokens: z.ZodNullable<z.ZodNumber>;
|
|
524
|
+
outputTokens: z.ZodNullable<z.ZodNumber>;
|
|
525
|
+
costUsd: z.ZodNullable<z.ZodNumber>;
|
|
526
|
+
}, z.core.$strict>>;
|
|
527
|
+
contentHash: z.ZodString;
|
|
528
|
+
}, z.core.$strict>>;
|
|
529
|
+
contentHash: z.ZodString;
|
|
530
|
+
}, z.core.$strict>>;
|
|
531
|
+
score: z.ZodNullable<z.ZodNumber>;
|
|
532
|
+
passed: z.ZodBoolean;
|
|
533
|
+
rewardEligible: z.ZodBoolean;
|
|
534
|
+
gradingStatus: z.ZodEnum<{
|
|
535
|
+
unscorable: "unscorable";
|
|
536
|
+
scored: "scored";
|
|
537
|
+
not_configured: "not_configured";
|
|
538
|
+
}>;
|
|
539
|
+
failureClass: z.ZodNullable<z.ZodEnum<{
|
|
540
|
+
policy_failure: "policy_failure";
|
|
541
|
+
grader_failure: "grader_failure";
|
|
542
|
+
environment_failure: "environment_failure";
|
|
543
|
+
infrastructure_failure: "infrastructure_failure";
|
|
544
|
+
timeout: "timeout";
|
|
545
|
+
cancelled: "cancelled";
|
|
546
|
+
}>>;
|
|
547
|
+
contentHash: z.ZodString;
|
|
548
|
+
}, z.core.$strict>>>;
|
|
473
549
|
environment: z.ZodObject<{
|
|
474
550
|
status: z.ZodEnum<{
|
|
475
551
|
completed: "completed";
|
|
@@ -888,6 +964,53 @@ export declare class OpenPondModelStarterAttemptsClient {
|
|
|
888
964
|
}[];
|
|
889
965
|
contentHash: string;
|
|
890
966
|
} | null | undefined;
|
|
967
|
+
experimentGrade?: {
|
|
968
|
+
schemaVersion: "openpond.experimentAttemptGrade.v1";
|
|
969
|
+
selectionHash: string;
|
|
970
|
+
grades: {
|
|
971
|
+
score: number | null;
|
|
972
|
+
passed: boolean;
|
|
973
|
+
rewardEligible: boolean;
|
|
974
|
+
failureClass: "infrastructure_failure" | "cancelled" | "policy_failure" | "environment_failure" | "grader_failure" | "timeout" | null;
|
|
975
|
+
gradingStatus: "unscorable" | "scored" | "not_configured";
|
|
976
|
+
schemaVersion: "openpond.taskGrade.v1";
|
|
977
|
+
taskHash: string;
|
|
978
|
+
evidenceHash: string;
|
|
979
|
+
graderSetHash: string;
|
|
980
|
+
components: {
|
|
981
|
+
schemaVersion: "openpond.graderEvidence.v1";
|
|
982
|
+
graderId: string;
|
|
983
|
+
graderVersion: string;
|
|
984
|
+
score: number | null;
|
|
985
|
+
passed: boolean;
|
|
986
|
+
rewardEligible: boolean;
|
|
987
|
+
failureClass: "infrastructure_failure" | "cancelled" | "policy_failure" | "environment_failure" | "grader_failure" | "timeout" | null;
|
|
988
|
+
feedback: string[];
|
|
989
|
+
visibleEvidenceRefs: string[];
|
|
990
|
+
privilegedEvidenceRefs: string[];
|
|
991
|
+
contentHash: string;
|
|
992
|
+
modelJudgeReceipt?: {
|
|
993
|
+
schemaVersion: "openpond.modelJudgeReceipt.v1";
|
|
994
|
+
providerId: string;
|
|
995
|
+
modelId: string;
|
|
996
|
+
modelRevision: string | null;
|
|
997
|
+
responseId: string | null;
|
|
998
|
+
requestHash: string;
|
|
999
|
+
responseHash: string;
|
|
1000
|
+
inputTokens: number | null;
|
|
1001
|
+
outputTokens: number | null;
|
|
1002
|
+
costUsd: number | null;
|
|
1003
|
+
} | undefined;
|
|
1004
|
+
}[];
|
|
1005
|
+
contentHash: string;
|
|
1006
|
+
}[];
|
|
1007
|
+
score: number | null;
|
|
1008
|
+
passed: boolean;
|
|
1009
|
+
rewardEligible: boolean;
|
|
1010
|
+
gradingStatus: "unscorable" | "scored" | "not_configured";
|
|
1011
|
+
failureClass: "infrastructure_failure" | "cancelled" | "policy_failure" | "environment_failure" | "grader_failure" | "timeout" | null;
|
|
1012
|
+
contentHash: string;
|
|
1013
|
+
} | null | undefined;
|
|
891
1014
|
}>;
|
|
892
1015
|
list(value: z.input<typeof ModelStarterAttemptListQuerySchema>, options?: {
|
|
893
1016
|
signal?: AbortSignal;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"model-starter-attempts.d.ts","sourceRoot":"","sources":["../../../../../src/model-starter-attempts.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,8BAA8B,EAAE,wBAAwB,EAAE,KAAK,wBAAwB,EAAE,MAAM,iCAAiC,CAAC;
|
|
1
|
+
{"version":3,"file":"model-starter-attempts.d.ts","sourceRoot":"","sources":["../../../../../src/model-starter-attempts.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,8BAA8B,EAAE,wBAAwB,EAAE,KAAK,wBAAwB,EAAE,MAAM,iCAAiC,CAAC;AAS1I,eAAO,MAAM,+BAA+B;;;;;;;;;;;;;;;;;4BAG1C,CAAC;AACH;kFACkF;AAClF,eAAO,MAAM,gCAAgC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAOlC,CAAC;AACZ,MAAM,MAAM,0BAA0B,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,gCAAgC,CAAC,CAAC;AAC1F,eAAO,MAAM,qCAAqC;;;;;;;;;kBAA6L,CAAC;AAChP,eAAO,MAAM,gCAAgC;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAKlC,CAAC;AACZ,eAAO,MAAM,gCAAgC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAa3C,CAAC;AACH,MAAM,MAAM,0BAA0B,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,gCAAgC,CAAC,CAAC;AAC1F,eAAO,MAAM,kCAAkC;;;;kBAAqI,CAAC;AACrL,eAAO,MAAM,6BAA6B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAoH,CAAC;AAC/J,eAAO,MAAM,+BAA+B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAwB1C,CAAC;AACH,MAAM,MAAM,yBAAyB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,+BAA+B,CAAC,CAAC;AAExF,qBAAa,gCAAiC,SAAQ,KAAK;IAC7C,QAAQ,CAAC,MAAM,EAAE,MAAM;IAAE,QAAQ,CAAC,IAAI,EAAE,MAAM;gBAArC,MAAM,EAAE,MAAM,EAAW,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM;CAC5E;AAED,qBAAa,kCAAkC;;gBAEjC,OAAO,EAAE;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,KAAK,CAAC,EAAE,OAAO,UAAU,CAAC,KAAK,CAAA;KAAE;IAKnG,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,gCAAgC,CAAC,EAAE,OAAO,GAAE;QAAE,MAAM,CAAC,EAAE,WAAW,CAAA;KAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAOtG,GAAG,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,GAAE;QAAE,MAAM,CAAC,EAAE,WAAW,CAAA;KAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IACtD,OAAO,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,qCAAqC,CAAC,EAAE,OAAO,GAAE;QAAE,MAAM,CAAC,EAAE,WAAW,CAAA;KAAO;;;;;;;;;;;;;;;;;;;;;;;;IAQ5G,MAAM,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,GAAE;QAAE,MAAM,CAAC,EAAE,WAAW,CAAA;KAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IACzD,MAAM,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,GAAE;QAAE,MAAM,CAAC,EAAE,WAAW,CAAA;KAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAOzD,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,kCAAkC,CAAC,EAAE,OAAO,GAAE;QAAE,MAAM,CAAC,EAAE,WAAW,CAAA;KAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAkC7G"}
|