@openpond/evals 0.11.16 → 0.11.18
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/dist/connected-evidence/contracts.js +72 -0
- package/dist/connected-evidence/imports.js +71 -0
- package/dist/connected-evidence/index.js +5 -0
- package/dist/connected-evidence/normalize.js +97 -0
- package/dist/connected-evidence/python-correctness.js +73 -0
- package/dist/connected-evidence/sources/claude-code.js +71 -0
- package/dist/connected-evidence/sources/codex.js +52 -0
- package/dist/connected-evidence/sources/history.js +86 -0
- package/dist/connected-evidence/sources/usage.js +15 -0
- package/dist/connected-evidence/usage-summary.js +52 -0
- package/dist/experiments.js +16 -4
- package/dist/graders.js +9 -6
- package/dist/human-review/comparison.js +161 -0
- package/dist/human-review/contracts.js +110 -0
- package/dist/human-review/execution.js +61 -0
- package/dist/human-review/handoff.js +18 -0
- package/dist/human-review/index.js +9 -0
- package/dist/human-review/local-publication.js +18 -0
- package/dist/human-review/repository.js +1 -0
- package/dist/human-review/results.js +18 -0
- package/dist/human-review/service.js +268 -0
- package/dist/human-review/task-proposals.js +39 -0
- package/dist/index.js +1 -0
- package/dist/javascript-verifier-contract.js +2 -2
- package/dist/learning/authoring.js +2 -1
- package/dist/learning/contracts.js +353 -81
- package/dist/learning/index.js +1 -0
- package/dist/learning/reward-authoring.js +5 -2
- package/dist/learning/training-handoff.js +32 -0
- package/dist/metrics-node.js +2 -0
- package/dist/profile-evaluation-target.js +7 -0
- package/dist/profile-evaluations.js +7 -10
- package/dist/profile-external-evaluation.js +35 -0
- package/dist/profile-source-candidate.js +5 -0
- package/dist/rewards.js +5 -3
- package/dist/taskset-run-contract.js +10 -0
- package/dist/tasksets.js +2 -0
- package/dist/types/builtin-benchmarks/harness-refiner.d.ts +75 -1
- package/dist/types/builtin-benchmarks/harness-refiner.d.ts.map +1 -1
- package/dist/types/conformance.d.ts +150 -2
- package/dist/types/conformance.d.ts.map +1 -1
- package/dist/types/connected-evidence/contracts.d.ts +317 -0
- package/dist/types/connected-evidence/contracts.d.ts.map +1 -0
- package/dist/types/connected-evidence/imports.d.ts +19 -0
- package/dist/types/connected-evidence/imports.d.ts.map +1 -0
- package/dist/types/connected-evidence/index.d.ts +6 -0
- package/dist/types/connected-evidence/index.d.ts.map +1 -0
- package/dist/types/connected-evidence/normalize.d.ts +25 -0
- package/dist/types/connected-evidence/normalize.d.ts.map +1 -0
- package/dist/types/connected-evidence/python-correctness.d.ts +196 -0
- package/dist/types/connected-evidence/python-correctness.d.ts.map +1 -0
- package/dist/types/connected-evidence/sources/claude-code.d.ts +74 -0
- package/dist/types/connected-evidence/sources/claude-code.d.ts.map +1 -0
- package/dist/types/connected-evidence/sources/codex.d.ts +74 -0
- package/dist/types/connected-evidence/sources/codex.d.ts.map +1 -0
- package/dist/types/connected-evidence/sources/history.d.ts +146 -0
- package/dist/types/connected-evidence/sources/history.d.ts.map +1 -0
- package/dist/types/connected-evidence/sources/usage.d.ts +3 -0
- package/dist/types/connected-evidence/sources/usage.d.ts.map +1 -0
- package/dist/types/connected-evidence/usage-summary.d.ts +162 -0
- package/dist/types/connected-evidence/usage-summary.d.ts.map +1 -0
- package/dist/types/evidence/conformance.d.ts +28 -28
- package/dist/types/evidence/contracts.d.ts +27 -27
- package/dist/types/execution-contracts.d.ts +159 -5
- package/dist/types/execution-contracts.d.ts.map +1 -1
- package/dist/types/experiments.d.ts +476 -7
- package/dist/types/experiments.d.ts.map +1 -1
- package/dist/types/graders.d.ts +21 -1
- package/dist/types/graders.d.ts.map +1 -1
- package/dist/types/human-review/comparison.d.ts +1959 -0
- package/dist/types/human-review/comparison.d.ts.map +1 -0
- package/dist/types/human-review/contracts.d.ts +2719 -0
- package/dist/types/human-review/contracts.d.ts.map +1 -0
- package/dist/types/human-review/execution.d.ts +298 -0
- package/dist/types/human-review/execution.d.ts.map +1 -0
- package/dist/types/human-review/handoff.d.ts +124 -0
- package/dist/types/human-review/handoff.d.ts.map +1 -0
- package/dist/types/human-review/index.d.ts +10 -0
- package/dist/types/human-review/index.d.ts.map +1 -0
- package/dist/types/human-review/local-publication.d.ts +165 -0
- package/dist/types/human-review/local-publication.d.ts.map +1 -0
- package/dist/types/human-review/repository.d.ts +43 -0
- package/dist/types/human-review/repository.d.ts.map +1 -0
- package/dist/types/human-review/results.d.ts +118 -0
- package/dist/types/human-review/results.d.ts.map +1 -0
- package/dist/types/human-review/service.d.ts +878 -0
- package/dist/types/human-review/service.d.ts.map +1 -0
- package/dist/types/human-review/task-proposals.d.ts +70 -0
- package/dist/types/human-review/task-proposals.d.ts.map +1 -0
- package/dist/types/index.d.ts +1 -0
- package/dist/types/index.d.ts.map +1 -1
- package/dist/types/javascript-environment-contract.d.ts +1 -1
- package/dist/types/javascript-verifier-contract.d.ts +1 -1
- package/dist/types/javascript-verifier-contract.d.ts.map +1 -1
- package/dist/types/learned-preference.d.ts +2 -2
- package/dist/types/learning/acceptance-group.d.ts +13 -13
- package/dist/types/learning/acceptance-plan.d.ts +1 -1
- package/dist/types/learning/admission.d.ts +78 -1
- package/dist/types/learning/admission.d.ts.map +1 -1
- package/dist/types/learning/authoring-service.d.ts +2 -0
- package/dist/types/learning/authoring-service.d.ts.map +1 -1
- package/dist/types/learning/authoring.d.ts +20 -0
- package/dist/types/learning/authoring.d.ts.map +1 -1
- package/dist/types/learning/batch-service.d.ts +76 -2
- package/dist/types/learning/batch-service.d.ts.map +1 -1
- package/dist/types/learning/contracts.d.ts +759 -24
- package/dist/types/learning/contracts.d.ts.map +1 -1
- package/dist/types/learning/index.d.ts +1 -0
- package/dist/types/learning/index.d.ts.map +1 -1
- package/dist/types/learning/iteration-candidate-decision.d.ts +2 -2
- package/dist/types/learning/iteration-dispatch-contracts.d.ts +2 -2
- package/dist/types/learning/operations.d.ts +1118 -8
- package/dist/types/learning/operations.d.ts.map +1 -1
- package/dist/types/learning/repository.d.ts +402 -15
- package/dist/types/learning/repository.d.ts.map +1 -1
- package/dist/types/learning/reward-authoring.d.ts.map +1 -1
- package/dist/types/learning/reward-calibration.d.ts +74 -0
- package/dist/types/learning/reward-calibration.d.ts.map +1 -1
- package/dist/types/learning/reward-checks.d.ts +14 -2
- package/dist/types/learning/reward-checks.d.ts.map +1 -1
- package/dist/types/learning/training-handoff.d.ts +201 -0
- package/dist/types/learning/training-handoff.d.ts.map +1 -0
- package/dist/types/learning/transport.d.ts +963 -21
- package/dist/types/learning/transport.d.ts.map +1 -1
- package/dist/types/metrics-node.d.ts +3 -1
- package/dist/types/metrics-node.d.ts.map +1 -1
- package/dist/types/model-improvement-qualification.d.ts +3 -3
- package/dist/types/native-harness-experiment-attempt.d.ts +2 -2
- package/dist/types/profile-evaluation-comparisons.d.ts +230 -0
- package/dist/types/profile-evaluation-comparisons.d.ts.map +1 -1
- package/dist/types/profile-evaluation-target.d.ts +14 -0
- package/dist/types/profile-evaluation-target.d.ts.map +1 -0
- package/dist/types/profile-evaluations.d.ts +117 -14
- package/dist/types/profile-evaluations.d.ts.map +1 -1
- package/dist/types/profile-external-evaluation.d.ts +404 -0
- package/dist/types/profile-external-evaluation.d.ts.map +1 -0
- package/dist/types/profile-source-candidate.d.ts +15 -0
- package/dist/types/profile-source-candidate.d.ts.map +1 -0
- package/dist/types/review-conformance.d.ts +7 -7
- package/dist/types/rewards.d.ts +253 -0
- package/dist/types/rewards.d.ts.map +1 -1
- package/dist/types/rollouts.d.ts +3 -3
- package/dist/types/taskset-run-contract.d.ts +230 -0
- package/dist/types/taskset-run-contract.d.ts.map +1 -1
- package/dist/types/tasksets.d.ts +311 -3
- package/dist/types/tasksets.d.ts.map +1 -1
- package/dist/types/telemetry-analysis.d.ts +2 -2
- package/dist/types/telemetry-bundle.d.ts +6 -6
- package/dist/types/telemetry.d.ts +3 -3
- package/package.json +13 -1
- package/schemas/learning/v1/command-request.schema.json +1745 -409
- package/schemas/learning/v1/decision.schema.json +11 -0
- package/schemas/learning/v1/draft.schema.json +11 -0
- package/schemas/learning/v1/grade.schema.json +11 -0
- package/schemas/learning/v1/package.schema.json +532 -12
- package/schemas/learning/v1/policy.schema.json +836 -28
- package/schemas/learning/v1/reward.schema.json +520 -0
- package/schemas/learning/v1/reward_check.schema.json +9 -0
|
@@ -0,0 +1,74 @@
|
|
|
1
|
+
import type { ConnectedFile } from "../contracts.js";
|
|
2
|
+
export declare function parseCodexSession(file: ConnectedFile, rows: Record<string, unknown>[]): {
|
|
3
|
+
schemaVersion: "openpond.connectedEvidence.v1";
|
|
4
|
+
normalizerVersion: "connected-evidence-1";
|
|
5
|
+
origin: "hermes" | "openclaw" | "native_chat" | "native_work" | "codex" | "claude_code";
|
|
6
|
+
sessionId: string;
|
|
7
|
+
branchId: string | null;
|
|
8
|
+
parentSessionId: string | null;
|
|
9
|
+
exporterVersion: string | null;
|
|
10
|
+
sourceFiles: {
|
|
11
|
+
path: string;
|
|
12
|
+
contentHash: string;
|
|
13
|
+
sizeBytes: number;
|
|
14
|
+
}[];
|
|
15
|
+
events: {
|
|
16
|
+
id: string;
|
|
17
|
+
sequence: number;
|
|
18
|
+
occurredAt: string | null;
|
|
19
|
+
kind: "message" | "unknown" | "artifact" | "terminal" | "usage" | "tool_call" | "tool_result" | "compaction";
|
|
20
|
+
role: "user" | "tool" | "assistant" | "system" | null;
|
|
21
|
+
callId: string | null;
|
|
22
|
+
parentId: string | null;
|
|
23
|
+
content: import("zod").JSONType;
|
|
24
|
+
usage: {
|
|
25
|
+
invocationId: string;
|
|
26
|
+
inputTokens: number | null;
|
|
27
|
+
outputTokens: number | null;
|
|
28
|
+
cachedInputTokens: number | null;
|
|
29
|
+
totalTokens: number | null;
|
|
30
|
+
provenance: "billed_invocation" | "reported_invocation" | "reported_cumulative";
|
|
31
|
+
} | null;
|
|
32
|
+
}[];
|
|
33
|
+
unmappedEvents: {
|
|
34
|
+
id: string;
|
|
35
|
+
sequence: number;
|
|
36
|
+
occurredAt: string | null;
|
|
37
|
+
kind: "message" | "unknown" | "artifact" | "terminal" | "usage" | "tool_call" | "tool_result" | "compaction";
|
|
38
|
+
role: "user" | "tool" | "assistant" | "system" | null;
|
|
39
|
+
callId: string | null;
|
|
40
|
+
parentId: string | null;
|
|
41
|
+
content: import("zod").JSONType;
|
|
42
|
+
usage: {
|
|
43
|
+
invocationId: string;
|
|
44
|
+
inputTokens: number | null;
|
|
45
|
+
outputTokens: number | null;
|
|
46
|
+
cachedInputTokens: number | null;
|
|
47
|
+
totalTokens: number | null;
|
|
48
|
+
provenance: "billed_invocation" | "reported_invocation" | "reported_cumulative";
|
|
49
|
+
} | null;
|
|
50
|
+
}[];
|
|
51
|
+
boundaries: {
|
|
52
|
+
id: string;
|
|
53
|
+
familyKey: string;
|
|
54
|
+
projection: "turn" | "conversation";
|
|
55
|
+
requestEventId: string;
|
|
56
|
+
start: number;
|
|
57
|
+
end: number;
|
|
58
|
+
inputHash: string;
|
|
59
|
+
outputHash: string | null;
|
|
60
|
+
revisionHash: string;
|
|
61
|
+
terminal: "unknown" | "cancelled" | "failed" | "completed";
|
|
62
|
+
coverage: {
|
|
63
|
+
answer: boolean;
|
|
64
|
+
context: "unknown" | "retained" | "compacted";
|
|
65
|
+
process: "absent" | "partial" | "retained";
|
|
66
|
+
artifacts: "absent" | "retained" | "references_only";
|
|
67
|
+
replay: "unavailable";
|
|
68
|
+
gaps: string[];
|
|
69
|
+
};
|
|
70
|
+
}[];
|
|
71
|
+
warnings: string[];
|
|
72
|
+
contentHash: string;
|
|
73
|
+
};
|
|
74
|
+
//# sourceMappingURL=codex.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"codex.d.ts","sourceRoot":"","sources":["../../../../../../src/connected-evidence/sources/codex.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAkB,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAIrE,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,aAAa,EAAE,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAgCrF"}
|
|
@@ -0,0 +1,146 @@
|
|
|
1
|
+
import type { ConnectedFile } from "../contracts.js";
|
|
2
|
+
export declare function parseHermesEvidence(file: ConnectedFile, row: Record<string, unknown>): {
|
|
3
|
+
schemaVersion: "openpond.connectedEvidence.v1";
|
|
4
|
+
normalizerVersion: "connected-evidence-1";
|
|
5
|
+
origin: "hermes" | "openclaw" | "native_chat" | "native_work" | "codex" | "claude_code";
|
|
6
|
+
sessionId: string;
|
|
7
|
+
branchId: string | null;
|
|
8
|
+
parentSessionId: string | null;
|
|
9
|
+
exporterVersion: string | null;
|
|
10
|
+
sourceFiles: {
|
|
11
|
+
path: string;
|
|
12
|
+
contentHash: string;
|
|
13
|
+
sizeBytes: number;
|
|
14
|
+
}[];
|
|
15
|
+
events: {
|
|
16
|
+
id: string;
|
|
17
|
+
sequence: number;
|
|
18
|
+
occurredAt: string | null;
|
|
19
|
+
kind: "message" | "unknown" | "artifact" | "terminal" | "usage" | "tool_call" | "tool_result" | "compaction";
|
|
20
|
+
role: "user" | "tool" | "assistant" | "system" | null;
|
|
21
|
+
callId: string | null;
|
|
22
|
+
parentId: string | null;
|
|
23
|
+
content: import("zod").JSONType;
|
|
24
|
+
usage: {
|
|
25
|
+
invocationId: string;
|
|
26
|
+
inputTokens: number | null;
|
|
27
|
+
outputTokens: number | null;
|
|
28
|
+
cachedInputTokens: number | null;
|
|
29
|
+
totalTokens: number | null;
|
|
30
|
+
provenance: "billed_invocation" | "reported_invocation" | "reported_cumulative";
|
|
31
|
+
} | null;
|
|
32
|
+
}[];
|
|
33
|
+
unmappedEvents: {
|
|
34
|
+
id: string;
|
|
35
|
+
sequence: number;
|
|
36
|
+
occurredAt: string | null;
|
|
37
|
+
kind: "message" | "unknown" | "artifact" | "terminal" | "usage" | "tool_call" | "tool_result" | "compaction";
|
|
38
|
+
role: "user" | "tool" | "assistant" | "system" | null;
|
|
39
|
+
callId: string | null;
|
|
40
|
+
parentId: string | null;
|
|
41
|
+
content: import("zod").JSONType;
|
|
42
|
+
usage: {
|
|
43
|
+
invocationId: string;
|
|
44
|
+
inputTokens: number | null;
|
|
45
|
+
outputTokens: number | null;
|
|
46
|
+
cachedInputTokens: number | null;
|
|
47
|
+
totalTokens: number | null;
|
|
48
|
+
provenance: "billed_invocation" | "reported_invocation" | "reported_cumulative";
|
|
49
|
+
} | null;
|
|
50
|
+
}[];
|
|
51
|
+
boundaries: {
|
|
52
|
+
id: string;
|
|
53
|
+
familyKey: string;
|
|
54
|
+
projection: "turn" | "conversation";
|
|
55
|
+
requestEventId: string;
|
|
56
|
+
start: number;
|
|
57
|
+
end: number;
|
|
58
|
+
inputHash: string;
|
|
59
|
+
outputHash: string | null;
|
|
60
|
+
revisionHash: string;
|
|
61
|
+
terminal: "unknown" | "cancelled" | "failed" | "completed";
|
|
62
|
+
coverage: {
|
|
63
|
+
answer: boolean;
|
|
64
|
+
context: "unknown" | "retained" | "compacted";
|
|
65
|
+
process: "absent" | "partial" | "retained";
|
|
66
|
+
artifacts: "absent" | "retained" | "references_only";
|
|
67
|
+
replay: "unavailable";
|
|
68
|
+
gaps: string[];
|
|
69
|
+
};
|
|
70
|
+
}[];
|
|
71
|
+
warnings: string[];
|
|
72
|
+
contentHash: string;
|
|
73
|
+
};
|
|
74
|
+
export declare function parseOpenClawEvidence(files: ConnectedFile[]): {
|
|
75
|
+
schemaVersion: "openpond.connectedEvidence.v1";
|
|
76
|
+
normalizerVersion: "connected-evidence-1";
|
|
77
|
+
origin: "hermes" | "openclaw" | "native_chat" | "native_work" | "codex" | "claude_code";
|
|
78
|
+
sessionId: string;
|
|
79
|
+
branchId: string | null;
|
|
80
|
+
parentSessionId: string | null;
|
|
81
|
+
exporterVersion: string | null;
|
|
82
|
+
sourceFiles: {
|
|
83
|
+
path: string;
|
|
84
|
+
contentHash: string;
|
|
85
|
+
sizeBytes: number;
|
|
86
|
+
}[];
|
|
87
|
+
events: {
|
|
88
|
+
id: string;
|
|
89
|
+
sequence: number;
|
|
90
|
+
occurredAt: string | null;
|
|
91
|
+
kind: "message" | "unknown" | "artifact" | "terminal" | "usage" | "tool_call" | "tool_result" | "compaction";
|
|
92
|
+
role: "user" | "tool" | "assistant" | "system" | null;
|
|
93
|
+
callId: string | null;
|
|
94
|
+
parentId: string | null;
|
|
95
|
+
content: import("zod").JSONType;
|
|
96
|
+
usage: {
|
|
97
|
+
invocationId: string;
|
|
98
|
+
inputTokens: number | null;
|
|
99
|
+
outputTokens: number | null;
|
|
100
|
+
cachedInputTokens: number | null;
|
|
101
|
+
totalTokens: number | null;
|
|
102
|
+
provenance: "billed_invocation" | "reported_invocation" | "reported_cumulative";
|
|
103
|
+
} | null;
|
|
104
|
+
}[];
|
|
105
|
+
unmappedEvents: {
|
|
106
|
+
id: string;
|
|
107
|
+
sequence: number;
|
|
108
|
+
occurredAt: string | null;
|
|
109
|
+
kind: "message" | "unknown" | "artifact" | "terminal" | "usage" | "tool_call" | "tool_result" | "compaction";
|
|
110
|
+
role: "user" | "tool" | "assistant" | "system" | null;
|
|
111
|
+
callId: string | null;
|
|
112
|
+
parentId: string | null;
|
|
113
|
+
content: import("zod").JSONType;
|
|
114
|
+
usage: {
|
|
115
|
+
invocationId: string;
|
|
116
|
+
inputTokens: number | null;
|
|
117
|
+
outputTokens: number | null;
|
|
118
|
+
cachedInputTokens: number | null;
|
|
119
|
+
totalTokens: number | null;
|
|
120
|
+
provenance: "billed_invocation" | "reported_invocation" | "reported_cumulative";
|
|
121
|
+
} | null;
|
|
122
|
+
}[];
|
|
123
|
+
boundaries: {
|
|
124
|
+
id: string;
|
|
125
|
+
familyKey: string;
|
|
126
|
+
projection: "turn" | "conversation";
|
|
127
|
+
requestEventId: string;
|
|
128
|
+
start: number;
|
|
129
|
+
end: number;
|
|
130
|
+
inputHash: string;
|
|
131
|
+
outputHash: string | null;
|
|
132
|
+
revisionHash: string;
|
|
133
|
+
terminal: "unknown" | "cancelled" | "failed" | "completed";
|
|
134
|
+
coverage: {
|
|
135
|
+
answer: boolean;
|
|
136
|
+
context: "unknown" | "retained" | "compacted";
|
|
137
|
+
process: "absent" | "partial" | "retained";
|
|
138
|
+
artifacts: "absent" | "retained" | "references_only";
|
|
139
|
+
replay: "unavailable";
|
|
140
|
+
gaps: string[];
|
|
141
|
+
};
|
|
142
|
+
}[];
|
|
143
|
+
warnings: string[];
|
|
144
|
+
contentHash: string;
|
|
145
|
+
};
|
|
146
|
+
//# sourceMappingURL=history.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"history.d.ts","sourceRoot":"","sources":["../../../../../../src/connected-evidence/sources/history.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAkB,aAAa,EAAE,MAAM,iBAAiB,CAAC;AA8BrE,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,aAAa,EAAE,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAUpF;AACD,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,aAAa,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAkC3D"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"usage.d.ts","sourceRoot":"","sources":["../../../../../../src/connected-evidence/sources/usage.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAEtD,wBAAgB,WAAW,CAAC,KAAK,EAAE,OAAO,EAAE,YAAY,EAAE,MAAM,EAAE,UAAU,GAAE,cAAc,CAAC,YAAY,CAAyB,GAAG,cAAc,GAAG,IAAI,CAYzJ"}
|
|
@@ -0,0 +1,162 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
export declare const ConnectedInvocationSchema: z.ZodObject<{
|
|
3
|
+
requestId: z.ZodString;
|
|
4
|
+
startedAt: z.ZodString;
|
|
5
|
+
provider: z.ZodString;
|
|
6
|
+
model: z.ZodString;
|
|
7
|
+
mode: z.ZodString;
|
|
8
|
+
status: z.ZodString;
|
|
9
|
+
source: z.ZodString;
|
|
10
|
+
projectId: z.ZodNullable<z.ZodString>;
|
|
11
|
+
inputTokens: z.ZodNullable<z.ZodNumber>;
|
|
12
|
+
cachedInputTokens: z.ZodNullable<z.ZodNumber>;
|
|
13
|
+
outputTokens: z.ZodNullable<z.ZodNumber>;
|
|
14
|
+
totalTokens: z.ZodNullable<z.ZodNumber>;
|
|
15
|
+
}, z.core.$strict>;
|
|
16
|
+
export type ConnectedInvocation = z.infer<typeof ConnectedInvocationSchema>;
|
|
17
|
+
declare const empty: () => {
|
|
18
|
+
invocations: number;
|
|
19
|
+
inputTokens: number;
|
|
20
|
+
cachedInputTokens: number;
|
|
21
|
+
outputTokens: number;
|
|
22
|
+
totalTokens: number;
|
|
23
|
+
missingInput: number;
|
|
24
|
+
missingCachedInput: number;
|
|
25
|
+
missingOutput: number;
|
|
26
|
+
missingTotal: number;
|
|
27
|
+
failures: number;
|
|
28
|
+
unattributed: number;
|
|
29
|
+
};
|
|
30
|
+
export type ConnectedUsageTotals = ReturnType<typeof empty>;
|
|
31
|
+
export declare const ConnectedUsageTotalsSchema: z.ZodObject<{
|
|
32
|
+
invocations: z.ZodNumber;
|
|
33
|
+
inputTokens: z.ZodNumber;
|
|
34
|
+
cachedInputTokens: z.ZodNumber;
|
|
35
|
+
outputTokens: z.ZodNumber;
|
|
36
|
+
totalTokens: z.ZodNumber;
|
|
37
|
+
missingInput: z.ZodNumber;
|
|
38
|
+
missingCachedInput: z.ZodNumber;
|
|
39
|
+
missingOutput: z.ZodNumber;
|
|
40
|
+
missingTotal: z.ZodNumber;
|
|
41
|
+
failures: z.ZodNumber;
|
|
42
|
+
unattributed: z.ZodNumber;
|
|
43
|
+
}, z.core.$strict>;
|
|
44
|
+
export declare const ConnectedUsageSummarySchema: z.ZodObject<{
|
|
45
|
+
total: z.ZodObject<{
|
|
46
|
+
invocations: z.ZodNumber;
|
|
47
|
+
inputTokens: z.ZodNumber;
|
|
48
|
+
cachedInputTokens: z.ZodNumber;
|
|
49
|
+
outputTokens: z.ZodNumber;
|
|
50
|
+
totalTokens: z.ZodNumber;
|
|
51
|
+
missingInput: z.ZodNumber;
|
|
52
|
+
missingCachedInput: z.ZodNumber;
|
|
53
|
+
missingOutput: z.ZodNumber;
|
|
54
|
+
missingTotal: z.ZodNumber;
|
|
55
|
+
failures: z.ZodNumber;
|
|
56
|
+
unattributed: z.ZodNumber;
|
|
57
|
+
}, z.core.$strict>;
|
|
58
|
+
days: z.ZodArray<z.ZodObject<{
|
|
59
|
+
invocations: z.ZodNumber;
|
|
60
|
+
inputTokens: z.ZodNumber;
|
|
61
|
+
cachedInputTokens: z.ZodNumber;
|
|
62
|
+
outputTokens: z.ZodNumber;
|
|
63
|
+
totalTokens: z.ZodNumber;
|
|
64
|
+
missingInput: z.ZodNumber;
|
|
65
|
+
missingCachedInput: z.ZodNumber;
|
|
66
|
+
missingOutput: z.ZodNumber;
|
|
67
|
+
missingTotal: z.ZodNumber;
|
|
68
|
+
failures: z.ZodNumber;
|
|
69
|
+
unattributed: z.ZodNumber;
|
|
70
|
+
key: z.ZodString;
|
|
71
|
+
}, z.core.$strict>>;
|
|
72
|
+
modes: z.ZodArray<z.ZodObject<{
|
|
73
|
+
invocations: z.ZodNumber;
|
|
74
|
+
inputTokens: z.ZodNumber;
|
|
75
|
+
cachedInputTokens: z.ZodNumber;
|
|
76
|
+
outputTokens: z.ZodNumber;
|
|
77
|
+
totalTokens: z.ZodNumber;
|
|
78
|
+
missingInput: z.ZodNumber;
|
|
79
|
+
missingCachedInput: z.ZodNumber;
|
|
80
|
+
missingOutput: z.ZodNumber;
|
|
81
|
+
missingTotal: z.ZodNumber;
|
|
82
|
+
failures: z.ZodNumber;
|
|
83
|
+
unattributed: z.ZodNumber;
|
|
84
|
+
key: z.ZodString;
|
|
85
|
+
}, z.core.$strict>>;
|
|
86
|
+
models: z.ZodArray<z.ZodObject<{
|
|
87
|
+
invocations: z.ZodNumber;
|
|
88
|
+
inputTokens: z.ZodNumber;
|
|
89
|
+
cachedInputTokens: z.ZodNumber;
|
|
90
|
+
outputTokens: z.ZodNumber;
|
|
91
|
+
totalTokens: z.ZodNumber;
|
|
92
|
+
missingInput: z.ZodNumber;
|
|
93
|
+
missingCachedInput: z.ZodNumber;
|
|
94
|
+
missingOutput: z.ZodNumber;
|
|
95
|
+
missingTotal: z.ZodNumber;
|
|
96
|
+
failures: z.ZodNumber;
|
|
97
|
+
unattributed: z.ZodNumber;
|
|
98
|
+
key: z.ZodString;
|
|
99
|
+
}, z.core.$strict>>;
|
|
100
|
+
}, z.core.$strict>;
|
|
101
|
+
/** A billed retry has a new invocation identity; repeated copies of one receipt do not add spend.
|
|
102
|
+
* Unknown token fields remain unknown and are never inferred from another field.
|
|
103
|
+
*/
|
|
104
|
+
export declare function summarizeConnectedInvocations(raw: unknown[]): {
|
|
105
|
+
total: {
|
|
106
|
+
invocations: number;
|
|
107
|
+
inputTokens: number;
|
|
108
|
+
cachedInputTokens: number;
|
|
109
|
+
outputTokens: number;
|
|
110
|
+
totalTokens: number;
|
|
111
|
+
missingInput: number;
|
|
112
|
+
missingCachedInput: number;
|
|
113
|
+
missingOutput: number;
|
|
114
|
+
missingTotal: number;
|
|
115
|
+
failures: number;
|
|
116
|
+
unattributed: number;
|
|
117
|
+
};
|
|
118
|
+
days: {
|
|
119
|
+
invocations: number;
|
|
120
|
+
inputTokens: number;
|
|
121
|
+
cachedInputTokens: number;
|
|
122
|
+
outputTokens: number;
|
|
123
|
+
totalTokens: number;
|
|
124
|
+
missingInput: number;
|
|
125
|
+
missingCachedInput: number;
|
|
126
|
+
missingOutput: number;
|
|
127
|
+
missingTotal: number;
|
|
128
|
+
failures: number;
|
|
129
|
+
unattributed: number;
|
|
130
|
+
key: string;
|
|
131
|
+
}[];
|
|
132
|
+
modes: {
|
|
133
|
+
invocations: number;
|
|
134
|
+
inputTokens: number;
|
|
135
|
+
cachedInputTokens: number;
|
|
136
|
+
outputTokens: number;
|
|
137
|
+
totalTokens: number;
|
|
138
|
+
missingInput: number;
|
|
139
|
+
missingCachedInput: number;
|
|
140
|
+
missingOutput: number;
|
|
141
|
+
missingTotal: number;
|
|
142
|
+
failures: number;
|
|
143
|
+
unattributed: number;
|
|
144
|
+
key: string;
|
|
145
|
+
}[];
|
|
146
|
+
models: {
|
|
147
|
+
invocations: number;
|
|
148
|
+
inputTokens: number;
|
|
149
|
+
cachedInputTokens: number;
|
|
150
|
+
outputTokens: number;
|
|
151
|
+
totalTokens: number;
|
|
152
|
+
missingInput: number;
|
|
153
|
+
missingCachedInput: number;
|
|
154
|
+
missingOutput: number;
|
|
155
|
+
missingTotal: number;
|
|
156
|
+
failures: number;
|
|
157
|
+
unattributed: number;
|
|
158
|
+
key: string;
|
|
159
|
+
}[];
|
|
160
|
+
};
|
|
161
|
+
export {};
|
|
162
|
+
//# sourceMappingURL=usage-summary.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"usage-summary.d.ts","sourceRoot":"","sources":["../../../../../src/connected-evidence/usage-summary.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;kBAE6D,CAAC;AACpG,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,QAAA,MAAM,KAAK;;;;;;;;;;;;CACiG,CAAC;AAC7G,MAAM,MAAM,oBAAoB,GAAG,UAAU,CAAC,OAAO,KAAK,CAAC,CAAC;AAE5D,eAAO,MAAM,0BAA0B;;;;;;;;;;;;kBAE8E,CAAC;AACtH,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAEqH,CAAC;AAQ9J;;GAEG;AACH,wBAAgB,6BAA6B,CAAC,GAAG,EAAE,OAAO,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAgB3D"}
|
|
@@ -9,9 +9,9 @@ export declare const completeWorkProcessTraceFixture: {
|
|
|
9
9
|
sequence: number;
|
|
10
10
|
timestamp: string;
|
|
11
11
|
layer: "agent" | "environment";
|
|
12
|
-
kind: "validation" | "
|
|
12
|
+
kind: "validation" | "approval" | "artifact" | "tool" | "state_transition" | "question" | "cleanup";
|
|
13
13
|
action: "environment_destroyed" | "turn_started" | "turn_completed" | "turn_failed" | "turn_cancelled" | "turn_timed_out" | "tool_invoked" | "tool_completed" | "tool_failed" | "validation_completed" | "validation_failed" | "approval_requested" | "approval_resolved" | "question_asked" | "question_answered" | "question_dismissed" | "artifact_created" | "workspace_changed" | "environment_created" | "environment_reset" | "cleanup_completed" | "cleanup_failed";
|
|
14
|
-
status: "
|
|
14
|
+
status: "cancelled" | "failed" | "completed" | "started";
|
|
15
15
|
inputHash: string | null;
|
|
16
16
|
outputHash: string | null;
|
|
17
17
|
receiptHash: string | null;
|
|
@@ -23,16 +23,16 @@ export declare const completeWorkProcessTraceFixture: {
|
|
|
23
23
|
id: string;
|
|
24
24
|
}[];
|
|
25
25
|
attributes: {
|
|
26
|
-
toolCategory: "
|
|
26
|
+
toolCategory: "command" | "agent" | "filesystem" | "source_control" | "browser" | "connected_app" | "sandbox" | "other" | null;
|
|
27
27
|
validationKind: "test" | "structural" | "visual" | "user_review" | "other" | null;
|
|
28
|
-
transitionState: "
|
|
28
|
+
transitionState: "cancelled" | "failed" | "completed" | "timeout" | "running" | null;
|
|
29
29
|
interventionOutcome: "requested" | "approved" | "denied" | "answered" | "dismissed" | null;
|
|
30
30
|
artifactCount: number;
|
|
31
31
|
exitCode: number | null;
|
|
32
32
|
durationMs: number | null;
|
|
33
33
|
cpuTimeMs: number | null;
|
|
34
34
|
memoryPeakBytes: number | null;
|
|
35
|
-
errorClass: "unknown" | "policy" | "validation" | "
|
|
35
|
+
errorClass: "unknown" | "policy" | "validation" | "cancelled" | "environment" | "timeout" | "infrastructure" | null;
|
|
36
36
|
};
|
|
37
37
|
}[];
|
|
38
38
|
contentHash: string;
|
|
@@ -70,8 +70,8 @@ export declare const completeWorkEvidenceFixture: {
|
|
|
70
70
|
};
|
|
71
71
|
inputHash: string;
|
|
72
72
|
terminal: {
|
|
73
|
-
status: "
|
|
74
|
-
failureClass: "unknown" | "
|
|
73
|
+
status: "cancelled" | "failed" | "completed" | "timeout";
|
|
74
|
+
failureClass: "unknown" | "cancelled" | "policy_failure" | "environment_failure" | "infrastructure_failure" | "timeout" | "validation_failure" | "model_failure" | null;
|
|
75
75
|
};
|
|
76
76
|
trace: {
|
|
77
77
|
sanitizedRef: {
|
|
@@ -186,9 +186,9 @@ export declare const incompleteWorkProcessTraceFixture: {
|
|
|
186
186
|
sequence: number;
|
|
187
187
|
timestamp: string;
|
|
188
188
|
layer: "agent" | "environment";
|
|
189
|
-
kind: "validation" | "
|
|
189
|
+
kind: "validation" | "approval" | "artifact" | "tool" | "state_transition" | "question" | "cleanup";
|
|
190
190
|
action: "environment_destroyed" | "turn_started" | "turn_completed" | "turn_failed" | "turn_cancelled" | "turn_timed_out" | "tool_invoked" | "tool_completed" | "tool_failed" | "validation_completed" | "validation_failed" | "approval_requested" | "approval_resolved" | "question_asked" | "question_answered" | "question_dismissed" | "artifact_created" | "workspace_changed" | "environment_created" | "environment_reset" | "cleanup_completed" | "cleanup_failed";
|
|
191
|
-
status: "
|
|
191
|
+
status: "cancelled" | "failed" | "completed" | "started";
|
|
192
192
|
inputHash: string | null;
|
|
193
193
|
outputHash: string | null;
|
|
194
194
|
receiptHash: string | null;
|
|
@@ -200,16 +200,16 @@ export declare const incompleteWorkProcessTraceFixture: {
|
|
|
200
200
|
id: string;
|
|
201
201
|
}[];
|
|
202
202
|
attributes: {
|
|
203
|
-
toolCategory: "
|
|
203
|
+
toolCategory: "command" | "agent" | "filesystem" | "source_control" | "browser" | "connected_app" | "sandbox" | "other" | null;
|
|
204
204
|
validationKind: "test" | "structural" | "visual" | "user_review" | "other" | null;
|
|
205
|
-
transitionState: "
|
|
205
|
+
transitionState: "cancelled" | "failed" | "completed" | "timeout" | "running" | null;
|
|
206
206
|
interventionOutcome: "requested" | "approved" | "denied" | "answered" | "dismissed" | null;
|
|
207
207
|
artifactCount: number;
|
|
208
208
|
exitCode: number | null;
|
|
209
209
|
durationMs: number | null;
|
|
210
210
|
cpuTimeMs: number | null;
|
|
211
211
|
memoryPeakBytes: number | null;
|
|
212
|
-
errorClass: "unknown" | "policy" | "validation" | "
|
|
212
|
+
errorClass: "unknown" | "policy" | "validation" | "cancelled" | "environment" | "timeout" | "infrastructure" | null;
|
|
213
213
|
};
|
|
214
214
|
}[];
|
|
215
215
|
contentHash: string;
|
|
@@ -308,8 +308,8 @@ export declare const invalidRawEvidenceFixture: {
|
|
|
308
308
|
};
|
|
309
309
|
inputHash: string;
|
|
310
310
|
terminal: {
|
|
311
|
-
status: "
|
|
312
|
-
failureClass: "unknown" | "
|
|
311
|
+
status: "cancelled" | "failed" | "completed" | "timeout";
|
|
312
|
+
failureClass: "unknown" | "cancelled" | "policy_failure" | "environment_failure" | "infrastructure_failure" | "timeout" | "validation_failure" | "model_failure" | null;
|
|
313
313
|
};
|
|
314
314
|
trace: {
|
|
315
315
|
sanitizedRef: {
|
|
@@ -392,9 +392,9 @@ export declare const workEvidenceConformance: {
|
|
|
392
392
|
sequence: number;
|
|
393
393
|
timestamp: string;
|
|
394
394
|
layer: "agent" | "environment";
|
|
395
|
-
kind: "validation" | "
|
|
395
|
+
kind: "validation" | "approval" | "artifact" | "tool" | "state_transition" | "question" | "cleanup";
|
|
396
396
|
action: "environment_destroyed" | "turn_started" | "turn_completed" | "turn_failed" | "turn_cancelled" | "turn_timed_out" | "tool_invoked" | "tool_completed" | "tool_failed" | "validation_completed" | "validation_failed" | "approval_requested" | "approval_resolved" | "question_asked" | "question_answered" | "question_dismissed" | "artifact_created" | "workspace_changed" | "environment_created" | "environment_reset" | "cleanup_completed" | "cleanup_failed";
|
|
397
|
-
status: "
|
|
397
|
+
status: "cancelled" | "failed" | "completed" | "started";
|
|
398
398
|
inputHash: string | null;
|
|
399
399
|
outputHash: string | null;
|
|
400
400
|
receiptHash: string | null;
|
|
@@ -406,16 +406,16 @@ export declare const workEvidenceConformance: {
|
|
|
406
406
|
id: string;
|
|
407
407
|
}[];
|
|
408
408
|
attributes: {
|
|
409
|
-
toolCategory: "
|
|
409
|
+
toolCategory: "command" | "agent" | "filesystem" | "source_control" | "browser" | "connected_app" | "sandbox" | "other" | null;
|
|
410
410
|
validationKind: "test" | "structural" | "visual" | "user_review" | "other" | null;
|
|
411
|
-
transitionState: "
|
|
411
|
+
transitionState: "cancelled" | "failed" | "completed" | "timeout" | "running" | null;
|
|
412
412
|
interventionOutcome: "requested" | "approved" | "denied" | "answered" | "dismissed" | null;
|
|
413
413
|
artifactCount: number;
|
|
414
414
|
exitCode: number | null;
|
|
415
415
|
durationMs: number | null;
|
|
416
416
|
cpuTimeMs: number | null;
|
|
417
417
|
memoryPeakBytes: number | null;
|
|
418
|
-
errorClass: "unknown" | "policy" | "validation" | "
|
|
418
|
+
errorClass: "unknown" | "policy" | "validation" | "cancelled" | "environment" | "timeout" | "infrastructure" | null;
|
|
419
419
|
};
|
|
420
420
|
}[];
|
|
421
421
|
contentHash: string;
|
|
@@ -453,8 +453,8 @@ export declare const workEvidenceConformance: {
|
|
|
453
453
|
};
|
|
454
454
|
inputHash: string;
|
|
455
455
|
terminal: {
|
|
456
|
-
status: "
|
|
457
|
-
failureClass: "unknown" | "
|
|
456
|
+
status: "cancelled" | "failed" | "completed" | "timeout";
|
|
457
|
+
failureClass: "unknown" | "cancelled" | "policy_failure" | "environment_failure" | "infrastructure_failure" | "timeout" | "validation_failure" | "model_failure" | null;
|
|
458
458
|
};
|
|
459
459
|
trace: {
|
|
460
460
|
sanitizedRef: {
|
|
@@ -625,9 +625,9 @@ export declare const workEvidenceConformance: {
|
|
|
625
625
|
sequence: number;
|
|
626
626
|
timestamp: string;
|
|
627
627
|
layer: "agent" | "environment";
|
|
628
|
-
kind: "validation" | "
|
|
628
|
+
kind: "validation" | "approval" | "artifact" | "tool" | "state_transition" | "question" | "cleanup";
|
|
629
629
|
action: "environment_destroyed" | "turn_started" | "turn_completed" | "turn_failed" | "turn_cancelled" | "turn_timed_out" | "tool_invoked" | "tool_completed" | "tool_failed" | "validation_completed" | "validation_failed" | "approval_requested" | "approval_resolved" | "question_asked" | "question_answered" | "question_dismissed" | "artifact_created" | "workspace_changed" | "environment_created" | "environment_reset" | "cleanup_completed" | "cleanup_failed";
|
|
630
|
-
status: "
|
|
630
|
+
status: "cancelled" | "failed" | "completed" | "started";
|
|
631
631
|
inputHash: string | null;
|
|
632
632
|
outputHash: string | null;
|
|
633
633
|
receiptHash: string | null;
|
|
@@ -639,16 +639,16 @@ export declare const workEvidenceConformance: {
|
|
|
639
639
|
id: string;
|
|
640
640
|
}[];
|
|
641
641
|
attributes: {
|
|
642
|
-
toolCategory: "
|
|
642
|
+
toolCategory: "command" | "agent" | "filesystem" | "source_control" | "browser" | "connected_app" | "sandbox" | "other" | null;
|
|
643
643
|
validationKind: "test" | "structural" | "visual" | "user_review" | "other" | null;
|
|
644
|
-
transitionState: "
|
|
644
|
+
transitionState: "cancelled" | "failed" | "completed" | "timeout" | "running" | null;
|
|
645
645
|
interventionOutcome: "requested" | "approved" | "denied" | "answered" | "dismissed" | null;
|
|
646
646
|
artifactCount: number;
|
|
647
647
|
exitCode: number | null;
|
|
648
648
|
durationMs: number | null;
|
|
649
649
|
cpuTimeMs: number | null;
|
|
650
650
|
memoryPeakBytes: number | null;
|
|
651
|
-
errorClass: "unknown" | "policy" | "validation" | "
|
|
651
|
+
errorClass: "unknown" | "policy" | "validation" | "cancelled" | "environment" | "timeout" | "infrastructure" | null;
|
|
652
652
|
};
|
|
653
653
|
}[];
|
|
654
654
|
contentHash: string;
|
|
@@ -691,8 +691,8 @@ export declare const workEvidenceConformance: {
|
|
|
691
691
|
};
|
|
692
692
|
inputHash: string;
|
|
693
693
|
terminal: {
|
|
694
|
-
status: "
|
|
695
|
-
failureClass: "unknown" | "
|
|
694
|
+
status: "cancelled" | "failed" | "completed" | "timeout";
|
|
695
|
+
failureClass: "unknown" | "cancelled" | "policy_failure" | "environment_failure" | "infrastructure_failure" | "timeout" | "validation_failure" | "model_failure" | null;
|
|
696
696
|
};
|
|
697
697
|
trace: {
|
|
698
698
|
sanitizedRef: {
|