engineering-behavior-observatory 0.2.1
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/README.md +151 -0
- package/contracts/codex-app-server-0.150.1/manifest.json +25 -0
- package/contracts/codex-app-server-0.150.1/schema/InitializeParams.json +84 -0
- package/contracts/codex-app-server-0.150.1/schema/JSONRPCMessage.json +137 -0
- package/contracts/codex-app-server-0.150.1/schema/v2/ThreadReadParams.json +17 -0
- package/contracts/codex-app-server-0.150.1/schema/v2/ThreadTokenUsageUpdatedNotification.json +82 -0
- package/contracts/codex-app-server-0.150.1/schema/v2/TurnInterruptParams.json +17 -0
- package/contracts/codex-app-server-0.150.1/types/AskForApproval.ts +5 -0
- package/contracts/codex-app-server-0.150.1/types/ClientNotification.ts +5 -0
- package/contracts/codex-app-server-0.150.1/types/SandboxMode.ts +5 -0
- package/contracts/codex-app-server-0.150.1/types/ThreadReadParams.ts +9 -0
- package/contracts/codex-app-server-0.150.1/types/TokenUsageBreakdown.ts +5 -0
- package/contracts/codex-app-server-0.150.1/types/TurnInterruptParams.ts +5 -0
- package/contracts/codex-app-server-0.153.4/manifest.json +27 -0
- package/contracts/codex-app-server-0.153.4/schema/JSONRPCMessage.json +137 -0
- package/contracts/codex-app-server-0.153.4/schema/v1/InitializeParams.json +84 -0
- package/contracts/codex-app-server-0.153.4/schema/v2/ThreadReadParams.json +17 -0
- package/contracts/codex-app-server-0.153.4/schema/v2/ThreadStartParams.json +515 -0
- package/contracts/codex-app-server-0.153.4/schema/v2/ThreadTokenUsageUpdatedNotification.json +82 -0
- package/contracts/codex-app-server-0.153.4/schema/v2/TurnInterruptParams.json +17 -0
- package/contracts/codex-app-server-0.153.4/schema/v2/TurnStartParams.json +921 -0
- package/contracts/codex-app-server-0.153.4/types/AskForApproval.ts +5 -0
- package/contracts/codex-app-server-0.153.4/types/ClientNotification.ts +5 -0
- package/contracts/codex-app-server-0.153.4/types/SandboxMode.ts +5 -0
- package/contracts/codex-app-server-0.153.4/types/ThreadReadParams.ts +12 -0
- package/contracts/codex-app-server-0.153.4/types/TokenUsageBreakdown.ts +5 -0
- package/contracts/codex-app-server-0.153.4/types/TurnInterruptParams.ts +5 -0
- package/contracts/openhands-agent-server-v1.44.1.json +65 -0
- package/contracts/openhands-agent-server-v1.46.0.json +46 -0
- package/dist/contracts/codex-app-server-0.153.4/types/AskForApproval.d.ts +9 -0
- package/dist/contracts/codex-app-server-0.153.4/types/AskForApproval.js +2 -0
- package/dist/contracts/codex-app-server-0.153.4/types/ClientNotification.d.ts +3 -0
- package/dist/contracts/codex-app-server-0.153.4/types/ClientNotification.js +2 -0
- package/dist/contracts/codex-app-server-0.153.4/types/SandboxMode.d.ts +1 -0
- package/dist/contracts/codex-app-server-0.153.4/types/SandboxMode.js +2 -0
- package/dist/contracts/codex-app-server-0.153.4/types/ThreadReadParams.d.ts +10 -0
- package/dist/contracts/codex-app-server-0.153.4/types/ThreadReadParams.js +2 -0
- package/dist/contracts/codex-app-server-0.153.4/types/TokenUsageBreakdown.d.ts +8 -0
- package/dist/contracts/codex-app-server-0.153.4/types/TokenUsageBreakdown.js +2 -0
- package/dist/contracts/codex-app-server-0.153.4/types/TurnInterruptParams.d.ts +4 -0
- package/dist/contracts/codex-app-server-0.153.4/types/TurnInterruptParams.js +2 -0
- package/dist/src/agent-sdk-normalizer.d.ts +22 -0
- package/dist/src/agent-sdk-normalizer.js +995 -0
- package/dist/src/agent-sdk-run.d.ts +31 -0
- package/dist/src/agent-sdk-run.js +242 -0
- package/dist/src/agent-sdk-runner.d.ts +96 -0
- package/dist/src/agent-sdk-runner.js +407 -0
- package/dist/src/agent-sdk.d.ts +294 -0
- package/dist/src/agent-sdk.js +644 -0
- package/dist/src/aggregation.d.ts +183 -0
- package/dist/src/aggregation.js +513 -0
- package/dist/src/artifacts.d.ts +36 -0
- package/dist/src/artifacts.js +1372 -0
- package/dist/src/atlas-grafana.d.ts +377 -0
- package/dist/src/atlas-grafana.js +47 -0
- package/dist/src/atlas-html.d.ts +2 -0
- package/dist/src/atlas-html.js +34 -0
- package/dist/src/atlas.d.ts +130 -0
- package/dist/src/atlas.js +235 -0
- package/dist/src/behavior-assertions.d.ts +95 -0
- package/dist/src/behavior-assertions.js +154 -0
- package/dist/src/cli.d.ts +5 -0
- package/dist/src/cli.js +957 -0
- package/dist/src/codex-judge.d.ts +44 -0
- package/dist/src/codex-judge.js +214 -0
- package/dist/src/codex-run.d.ts +114 -0
- package/dist/src/codex-run.js +519 -0
- package/dist/src/codex.d.ts +187 -0
- package/dist/src/codex.js +1028 -0
- package/dist/src/contracts.d.ts +155 -0
- package/dist/src/contracts.js +866 -0
- package/dist/src/corpus.d.ts +67 -0
- package/dist/src/corpus.js +540 -0
- package/dist/src/cursor-sdk-runner.d.ts +81 -0
- package/dist/src/cursor-sdk-runner.js +362 -0
- package/dist/src/cursor-sdk.d.ts +81 -0
- package/dist/src/cursor-sdk.js +1078 -0
- package/dist/src/deepseek-adapter.d.ts +152 -0
- package/dist/src/deepseek-adapter.js +777 -0
- package/dist/src/exports.d.ts +69 -0
- package/dist/src/exports.js +966 -0
- package/dist/src/human-calibration.d.ts +167 -0
- package/dist/src/human-calibration.js +618 -0
- package/dist/src/index.d.ts +71 -0
- package/dist/src/index.js +38 -0
- package/dist/src/lifecycle.d.ts +196 -0
- package/dist/src/lifecycle.js +1889 -0
- package/dist/src/normalization-integrity.d.ts +128 -0
- package/dist/src/normalization-integrity.js +429 -0
- package/dist/src/openhands-run.d.ts +36 -0
- package/dist/src/openhands-run.js +415 -0
- package/dist/src/openhands.d.ts +130 -0
- package/dist/src/openhands.js +859 -0
- package/dist/src/pi.d.ts +203 -0
- package/dist/src/pi.js +1345 -0
- package/dist/src/process-protocol.d.ts +245 -0
- package/dist/src/process-protocol.js +1261 -0
- package/dist/src/retained-evidence.d.ts +12 -0
- package/dist/src/retained-evidence.js +201 -0
- package/dist/src/run-bundles.d.ts +193 -0
- package/dist/src/run-bundles.js +993 -0
- package/dist/src/scheduler.d.ts +109 -0
- package/dist/src/scheduler.js +1080 -0
- package/dist/src/semantic-judge.d.ts +194 -0
- package/dist/src/semantic-judge.js +875 -0
- package/dist/src/structural-observations.d.ts +124 -0
- package/dist/src/structural-observations.js +671 -0
- package/dist/src/task-packets.d.ts +115 -0
- package/dist/src/task-packets.js +683 -0
- package/dist/src/uniform-events.d.ts +120 -0
- package/dist/src/uniform-events.js +158 -0
- package/dist/src/verifiers.d.ts +116 -0
- package/dist/src/verifiers.js +819 -0
- package/dist/src/workspaces.d.ts +56 -0
- package/dist/src/workspaces.js +1314 -0
- package/docs/README.md +60 -0
- package/docs/development/README.md +71 -0
- package/docs/development/documentation-sync.md +58 -0
- package/docs/development/extension-contracts.md +307 -0
- package/docs/evaluation/README.md +48 -0
- package/docs/evaluation/aggregation.md +112 -0
- package/docs/evaluation/behavior-assertions.md +45 -0
- package/docs/evaluation/human-calibration.md +170 -0
- package/docs/evaluation/normalization-integrity.md +80 -0
- package/docs/evaluation/semantic-judge.md +188 -0
- package/docs/evaluation/structural-observations.md +84 -0
- package/docs/evaluation/uniform-events.md +131 -0
- package/docs/guides/atlas.md +144 -0
- package/docs/guides/evidence-and-sharing.md +100 -0
- package/docs/guides/operator-guide.md +424 -0
- package/docs/guides/quickstart.md +152 -0
- package/docs/guides/telemetry.md +66 -0
- package/docs/harnesses/README.md +31 -0
- package/docs/harnesses/claude-agent-sdk.md +61 -0
- package/docs/harnesses/codex-harness.md +174 -0
- package/docs/harnesses/cursor-sdk.md +149 -0
- package/docs/harnesses/deepseek-harness.md +134 -0
- package/docs/harnesses/openhands-agent-server.md +153 -0
- package/docs/harnesses/pi-sdk.md +66 -0
- package/docs/reference/README.md +19 -0
- package/docs/reference/agent-sdk-operational-runner.md +118 -0
- package/docs/reference/cli.md +114 -0
- package/docs/reference/contracts.md +222 -0
- package/docs/reference/run-bundle-contract.md +354 -0
- package/docs/reference/run-lifecycle.md +54 -0
- package/examples/cursor-sdk/README.md +11 -0
- package/examples/cursor-sdk/capture-profile.json +1 -0
- package/examples/cursor-sdk/harness.json +1 -0
- package/examples/cursor-sdk/model.json +1 -0
- package/examples/cursor-sdk/native-limits.json +1 -0
- package/examples/cursor-sdk/native-tool-policy.json +1 -0
- package/ontology/behavior-categories.v1.json +46 -0
- package/package.json +66 -0
- package/release/0.1.0/KNOWN_LIMITATIONS.md +35 -0
- package/release/0.1.0/README.md +46 -0
- package/release/0.1.0/reproducibility.json +61 -0
- package/release/0.2.0/KNOWN_LIMITATIONS.md +40 -0
- package/release/0.2.0/README.md +42 -0
- package/release/0.2.0/reproducibility.json +73 -0
- package/release/0.2.1/KNOWN_LIMITATIONS.md +18 -0
- package/release/0.2.1/README.md +63 -0
- package/release/0.2.1/reproducibility.json +73 -0
- package/release/README.md +13 -0
- package/schemas/aggregation.v1.json +293 -0
- package/schemas/behavior-assertions.v1.json +220 -0
- package/schemas/deepseek-runtime-composition.v1.schema.json +122 -0
- package/schemas/experiment.v1.schema.json +188 -0
- package/schemas/human-calibration.v1.json +336 -0
- package/schemas/normalization-integrity.v1.json +302 -0
- package/schemas/run-bundles/v1.json +1040 -0
- package/schemas/run-queue.v1.schema.json +206 -0
- package/schemas/semantic-judge.v1.json +380 -0
- package/schemas/structural-observations.v1.json +131 -0
- package/schemas/task-packet-freeze.v1.schema.json +106 -0
- package/schemas/task-packet.v1.schema.json +234 -0
- package/schemas/uniform-events/v1.json +289 -0
- package/scripts/atlas-grafana.sh +18 -0
|
@@ -0,0 +1,206 @@
|
|
|
1
|
+
{
|
|
2
|
+
"$schema": "https://json-schema.org/draft/2020-12/schema",
|
|
3
|
+
"$id": "https://ebo.dev/schemas/run-queue.v1.schema.json",
|
|
4
|
+
"title": "EBO run queue v1",
|
|
5
|
+
"type": "object",
|
|
6
|
+
"additionalProperties": false,
|
|
7
|
+
"required": ["schemaVersion", "experimentId", "experimentDigest", "schedulingDigest", "captureProfile", "coordinatorBudget", "matrix", "seed", "ordering", "entries"],
|
|
8
|
+
"properties": {
|
|
9
|
+
"schemaVersion": { "const": "ebo.run-queue/v1" },
|
|
10
|
+
"experimentId": { "type": "string", "pattern": "^[a-z0-9][a-z0-9-]*$" },
|
|
11
|
+
"experimentDigest": { "$ref": "#/$defs/digest" },
|
|
12
|
+
"schedulingDigest": { "$ref": "#/$defs/digest" },
|
|
13
|
+
"captureProfile": { "$ref": "#/$defs/reference" },
|
|
14
|
+
"coordinatorBudget": {
|
|
15
|
+
"type": "object",
|
|
16
|
+
"additionalProperties": false,
|
|
17
|
+
"required": ["maxWallClockMs"],
|
|
18
|
+
"properties": {
|
|
19
|
+
"maxWallClockMs": {
|
|
20
|
+
"type": "integer",
|
|
21
|
+
"minimum": 1,
|
|
22
|
+
"maximum": 2147483647
|
|
23
|
+
}
|
|
24
|
+
}
|
|
25
|
+
},
|
|
26
|
+
"matrix": { "$ref": "#/$defs/matrix" },
|
|
27
|
+
"seed": { "type": "string", "minLength": 1 },
|
|
28
|
+
"ordering": {
|
|
29
|
+
"type": "object",
|
|
30
|
+
"additionalProperties": false,
|
|
31
|
+
"required": ["strategy"],
|
|
32
|
+
"properties": {
|
|
33
|
+
"strategy": { "enum": ["sequential", "seeded-shuffle", "balanced"] },
|
|
34
|
+
"balanceBy": { "enum": ["task", "model", "harness"] },
|
|
35
|
+
"permutationAlgorithm": { "const": "fisher-yates-v1" },
|
|
36
|
+
"permutationAlgorithmRef": { "$ref": "#/$defs/reference" }
|
|
37
|
+
},
|
|
38
|
+
"allOf": [
|
|
39
|
+
{
|
|
40
|
+
"if": {
|
|
41
|
+
"properties": { "strategy": { "const": "sequential" } }
|
|
42
|
+
},
|
|
43
|
+
"then": {
|
|
44
|
+
"not": {
|
|
45
|
+
"anyOf": [
|
|
46
|
+
{ "required": ["balanceBy"] },
|
|
47
|
+
{ "required": ["permutationAlgorithmRef"] }
|
|
48
|
+
]
|
|
49
|
+
}
|
|
50
|
+
}
|
|
51
|
+
},
|
|
52
|
+
{
|
|
53
|
+
"if": {
|
|
54
|
+
"properties": { "strategy": { "const": "balanced" } }
|
|
55
|
+
},
|
|
56
|
+
"then": { "not": { "required": ["permutationAlgorithmRef"] } }
|
|
57
|
+
},
|
|
58
|
+
{
|
|
59
|
+
"if": {
|
|
60
|
+
"properties": { "strategy": { "enum": ["seeded-shuffle"] } }
|
|
61
|
+
},
|
|
62
|
+
"then": {
|
|
63
|
+
"allOf": [
|
|
64
|
+
{ "required": ["permutationAlgorithm"] },
|
|
65
|
+
{ "not": { "required": ["balanceBy"] } }
|
|
66
|
+
]
|
|
67
|
+
}
|
|
68
|
+
},
|
|
69
|
+
{
|
|
70
|
+
"if": {
|
|
71
|
+
"properties": { "strategy": { "enum": ["sequential", "balanced"] } }
|
|
72
|
+
},
|
|
73
|
+
"then": { "not": { "required": ["permutationAlgorithm"] } }
|
|
74
|
+
}
|
|
75
|
+
]
|
|
76
|
+
},
|
|
77
|
+
"entries": {
|
|
78
|
+
"type": "array",
|
|
79
|
+
"minItems": 1,
|
|
80
|
+
"maxItems": 100000,
|
|
81
|
+
"items": { "$ref": "#/$defs/entry" }
|
|
82
|
+
}
|
|
83
|
+
},
|
|
84
|
+
"$defs": {
|
|
85
|
+
"digest": {
|
|
86
|
+
"type": "object",
|
|
87
|
+
"additionalProperties": false,
|
|
88
|
+
"required": ["algorithm", "value"],
|
|
89
|
+
"properties": {
|
|
90
|
+
"algorithm": { "const": "sha256" },
|
|
91
|
+
"value": { "type": "string", "pattern": "^[a-f0-9]{64}$" }
|
|
92
|
+
}
|
|
93
|
+
},
|
|
94
|
+
"reference": {
|
|
95
|
+
"type": "object",
|
|
96
|
+
"additionalProperties": false,
|
|
97
|
+
"required": ["locator", "digest"],
|
|
98
|
+
"properties": {
|
|
99
|
+
"locator": {
|
|
100
|
+
"type": "string",
|
|
101
|
+
"maxLength": 960,
|
|
102
|
+
"pattern": "^(?!/)(?!(?:[^/]+/){64})(?!.*//)(?!.*(?:^|/)\\.{1,2}(?:/|$))(?!.*(?:^|/)\\.[Gg][Ii][Tt](?:/|$))(?!.*(?:^|/)[^/]{256})(?!.*(?:^|/)[^/]*\\.(?:/|$))(?!.*(?:^|/)(?:[Cc][Oo][Nn]|[Pp][Rr][Nn]|[Aa][Uu][Xx]|[Nn][Uu][Ll]|[Cc][Oo][Mm][1-9]|[Ll][Pp][Tt][1-9])(?:\\.[^/]*)?(?:/|$))(?!.*/$)[A-Za-z0-9][A-Za-z0-9._/-]*$"
|
|
103
|
+
},
|
|
104
|
+
"digest": { "$ref": "#/$defs/digest" }
|
|
105
|
+
}
|
|
106
|
+
},
|
|
107
|
+
"task": {
|
|
108
|
+
"type": "object",
|
|
109
|
+
"additionalProperties": false,
|
|
110
|
+
"required": ["id", "packetId", "packetRef", "freezeLocator", "aggregateDigest"],
|
|
111
|
+
"properties": {
|
|
112
|
+
"id": { "type": "string", "pattern": "^[a-z0-9][a-z0-9-]*$" },
|
|
113
|
+
"packetId": { "type": "string", "pattern": "^[a-z0-9][a-z0-9-]*$" },
|
|
114
|
+
"packetRef": { "$ref": "#/$defs/reference" },
|
|
115
|
+
"freezeLocator": {
|
|
116
|
+
"type": "string",
|
|
117
|
+
"maxLength": 960,
|
|
118
|
+
"pattern": "^(?!/)(?!(?:[^/]+/){64})(?!.*//)(?!.*(?:^|/)\\.{1,2}(?:/|$))(?!.*(?:^|/)\\.[Gg][Ii][Tt](?:/|$))(?!.*(?:^|/)[^/]{256})(?!.*(?:^|/)[^/]*\\.(?:/|$))(?!.*(?:^|/)(?:[Cc][Oo][Nn]|[Pp][Rr][Nn]|[Aa][Uu][Xx]|[Nn][Uu][Ll]|[Cc][Oo][Mm][1-9]|[Ll][Pp][Tt][1-9])(?:\\.[^/]*)?(?:/|$))(?!.*/$)[A-Za-z0-9][A-Za-z0-9._/-]*$"
|
|
119
|
+
},
|
|
120
|
+
"aggregateDigest": { "$ref": "#/$defs/digest" }
|
|
121
|
+
}
|
|
122
|
+
},
|
|
123
|
+
"model": {
|
|
124
|
+
"type": "object",
|
|
125
|
+
"additionalProperties": false,
|
|
126
|
+
"required": ["id", "configurationRef"],
|
|
127
|
+
"properties": {
|
|
128
|
+
"id": { "type": "string", "pattern": "^[a-z0-9][a-z0-9-]*$" },
|
|
129
|
+
"configurationRef": { "$ref": "#/$defs/reference" }
|
|
130
|
+
}
|
|
131
|
+
},
|
|
132
|
+
"harness": {
|
|
133
|
+
"type": "object",
|
|
134
|
+
"additionalProperties": false,
|
|
135
|
+
"required": ["id", "configurationRef", "nativeLimitsRef", "nativeToolPolicyRef"],
|
|
136
|
+
"properties": {
|
|
137
|
+
"id": { "type": "string", "pattern": "^[a-z0-9][a-z0-9-]*$" },
|
|
138
|
+
"configurationRef": { "$ref": "#/$defs/reference" },
|
|
139
|
+
"nativeLimitsRef": { "$ref": "#/$defs/reference" },
|
|
140
|
+
"nativeToolPolicyRef": { "$ref": "#/$defs/reference" }
|
|
141
|
+
}
|
|
142
|
+
},
|
|
143
|
+
"configuration": {
|
|
144
|
+
"type": "object",
|
|
145
|
+
"additionalProperties": false,
|
|
146
|
+
"required": ["model", "harness", "nativeLimits", "nativeToolPolicy"],
|
|
147
|
+
"properties": {
|
|
148
|
+
"model": { "$ref": "#/$defs/reference" },
|
|
149
|
+
"harness": { "$ref": "#/$defs/reference" },
|
|
150
|
+
"nativeLimits": { "$ref": "#/$defs/reference" },
|
|
151
|
+
"nativeToolPolicy": { "$ref": "#/$defs/reference" }
|
|
152
|
+
}
|
|
153
|
+
},
|
|
154
|
+
"trial": {
|
|
155
|
+
"type": "object",
|
|
156
|
+
"additionalProperties": false,
|
|
157
|
+
"required": ["index"],
|
|
158
|
+
"properties": {
|
|
159
|
+
"index": { "type": "integer", "minimum": 1, "maximum": 9007199254740991 }
|
|
160
|
+
}
|
|
161
|
+
},
|
|
162
|
+
"matrix": {
|
|
163
|
+
"type": "object",
|
|
164
|
+
"additionalProperties": false,
|
|
165
|
+
"required": ["taskIds", "modelIds", "harnessIds", "trialCount"],
|
|
166
|
+
"properties": {
|
|
167
|
+
"taskIds": {
|
|
168
|
+
"type": "array",
|
|
169
|
+
"minItems": 1,
|
|
170
|
+
"uniqueItems": true,
|
|
171
|
+
"items": { "type": "string", "pattern": "^[a-z0-9][a-z0-9-]*$" }
|
|
172
|
+
},
|
|
173
|
+
"modelIds": {
|
|
174
|
+
"type": "array",
|
|
175
|
+
"minItems": 1,
|
|
176
|
+
"uniqueItems": true,
|
|
177
|
+
"items": { "type": "string", "pattern": "^[a-z0-9][a-z0-9-]*$" }
|
|
178
|
+
},
|
|
179
|
+
"harnessIds": {
|
|
180
|
+
"type": "array",
|
|
181
|
+
"minItems": 1,
|
|
182
|
+
"uniqueItems": true,
|
|
183
|
+
"items": { "type": "string", "pattern": "^[a-z0-9][a-z0-9-]*$" }
|
|
184
|
+
},
|
|
185
|
+
"trialCount": { "type": "integer", "minimum": 1, "maximum": 9007199254740991 }
|
|
186
|
+
}
|
|
187
|
+
},
|
|
188
|
+
"entry": {
|
|
189
|
+
"type": "object",
|
|
190
|
+
"additionalProperties": false,
|
|
191
|
+
"required": ["runId", "taskId", "modelId", "harnessId", "trialIndex", "task", "model", "harness", "configuration", "trial"],
|
|
192
|
+
"properties": {
|
|
193
|
+
"runId": { "type": "string", "pattern": "^run-[a-f0-9]{64}$" },
|
|
194
|
+
"taskId": { "type": "string", "pattern": "^[a-z0-9][a-z0-9-]*$" },
|
|
195
|
+
"modelId": { "type": "string", "pattern": "^[a-z0-9][a-z0-9-]*$" },
|
|
196
|
+
"harnessId": { "type": "string", "pattern": "^[a-z0-9][a-z0-9-]*$" },
|
|
197
|
+
"trialIndex": { "type": "integer", "minimum": 1, "maximum": 9007199254740991 },
|
|
198
|
+
"task": { "$ref": "#/$defs/task" },
|
|
199
|
+
"model": { "$ref": "#/$defs/model" },
|
|
200
|
+
"harness": { "$ref": "#/$defs/harness" },
|
|
201
|
+
"configuration": { "$ref": "#/$defs/configuration" },
|
|
202
|
+
"trial": { "$ref": "#/$defs/trial" }
|
|
203
|
+
}
|
|
204
|
+
}
|
|
205
|
+
}
|
|
206
|
+
}
|
|
@@ -0,0 +1,380 @@
|
|
|
1
|
+
{
|
|
2
|
+
"$schema": "https://json-schema.org/draft/2020-12/schema",
|
|
3
|
+
"$id": "urn:ebo:schema:semantic-judge:v1",
|
|
4
|
+
"$defs": {
|
|
5
|
+
"identifier": {
|
|
6
|
+
"type": "string",
|
|
7
|
+
"minLength": 1,
|
|
8
|
+
"maxLength": 1024,
|
|
9
|
+
"pattern": ".*\\S.*"
|
|
10
|
+
},
|
|
11
|
+
"text": {
|
|
12
|
+
"type": "string",
|
|
13
|
+
"minLength": 1,
|
|
14
|
+
"maxLength": 16384,
|
|
15
|
+
"pattern": ".*\\S.*"
|
|
16
|
+
},
|
|
17
|
+
"digest": {
|
|
18
|
+
"type": "string",
|
|
19
|
+
"pattern": "^sha256:[a-f0-9]{64}$"
|
|
20
|
+
},
|
|
21
|
+
"behavior": {
|
|
22
|
+
"type": "object",
|
|
23
|
+
"additionalProperties": false,
|
|
24
|
+
"required": ["vocabularyVersion", "categoryId", "dimensionId"],
|
|
25
|
+
"properties": {
|
|
26
|
+
"vocabularyVersion": { "$ref": "#/$defs/identifier" },
|
|
27
|
+
"categoryId": { "$ref": "#/$defs/identifier" },
|
|
28
|
+
"dimensionId": { "$ref": "#/$defs/identifier" }
|
|
29
|
+
}
|
|
30
|
+
},
|
|
31
|
+
"rubric": {
|
|
32
|
+
"type": "object",
|
|
33
|
+
"additionalProperties": false,
|
|
34
|
+
"required": ["id", "version", "instructions"],
|
|
35
|
+
"properties": {
|
|
36
|
+
"id": { "$ref": "#/$defs/identifier" },
|
|
37
|
+
"version": { "$ref": "#/$defs/identifier" },
|
|
38
|
+
"instructions": { "$ref": "#/$defs/text" }
|
|
39
|
+
}
|
|
40
|
+
},
|
|
41
|
+
"limits": {
|
|
42
|
+
"type": "object",
|
|
43
|
+
"additionalProperties": false,
|
|
44
|
+
"required": ["maxEvidenceItems", "maxRecordChars", "maxInputChars", "maxOutputChars", "maxCitations", "maxWallClockMs", "maxTurns"],
|
|
45
|
+
"properties": {
|
|
46
|
+
"maxEvidenceItems": { "type": "integer", "minimum": 1, "maximum": 128 },
|
|
47
|
+
"maxRecordChars": { "type": "integer", "minimum": 256, "maximum": 65536 },
|
|
48
|
+
"maxInputChars": { "type": "integer", "minimum": 1024, "maximum": 1000000 },
|
|
49
|
+
"maxOutputChars": { "type": "integer", "minimum": 256, "maximum": 65536 },
|
|
50
|
+
"maxCitations": { "type": "integer", "minimum": 1, "maximum": 64 },
|
|
51
|
+
"maxWallClockMs": { "type": "integer", "minimum": 100, "maximum": 300000 },
|
|
52
|
+
"maxTurns": { "type": "integer", "minimum": 1, "maximum": 10 },
|
|
53
|
+
"maxBudgetUsd": { "type": "number", "exclusiveMinimum": 0, "maximum": 100 }
|
|
54
|
+
}
|
|
55
|
+
},
|
|
56
|
+
"request": {
|
|
57
|
+
"type": "object",
|
|
58
|
+
"additionalProperties": false,
|
|
59
|
+
"required": ["schemaVersion", "id", "behavior", "rubric", "evaluator", "selection", "limits", "blinding"],
|
|
60
|
+
"allOf": [{
|
|
61
|
+
"if": { "properties": { "evaluator": { "required": ["backend"], "properties": { "backend": { "const": "codex-app-server" } } } } },
|
|
62
|
+
"then": { "properties": {
|
|
63
|
+
"evaluator": { "properties": { "provider": { "const": "openai" } } },
|
|
64
|
+
"limits": { "properties": { "maxTurns": { "const": 1 } }, "not": { "required": ["maxBudgetUsd"] } }
|
|
65
|
+
} },
|
|
66
|
+
"else": { "properties": { "evaluator": {
|
|
67
|
+
"properties": { "provider": { "const": "anthropic" }, "backend": { "const": "claude-agent-sdk" }, "effort": { "enum": ["low", "medium", "high", "xhigh", "max"] } },
|
|
68
|
+
"not": { "required": ["executable"] }
|
|
69
|
+
} } }
|
|
70
|
+
}],
|
|
71
|
+
"properties": {
|
|
72
|
+
"schemaVersion": { "const": "ebo.semantic-judge-request/v1" },
|
|
73
|
+
"id": { "type": "string", "minLength": 1, "maxLength": 220, "pattern": ".*\\S.*" },
|
|
74
|
+
"behavior": { "$ref": "#/$defs/behavior" },
|
|
75
|
+
"rubric": { "$ref": "#/$defs/rubric" },
|
|
76
|
+
"evaluator": {
|
|
77
|
+
"type": "object",
|
|
78
|
+
"additionalProperties": false,
|
|
79
|
+
"required": ["provider", "model", "effort"],
|
|
80
|
+
"properties": {
|
|
81
|
+
"backend": { "enum": ["claude-agent-sdk", "codex-app-server"] },
|
|
82
|
+
"executable": { "type": "string", "minLength": 1, "maxLength": 4096, "pattern": ".*\\S.*" },
|
|
83
|
+
"provider": { "enum": ["anthropic", "openai"] },
|
|
84
|
+
"model": { "type": "string", "minLength": 1, "maxLength": 220, "pattern": ".*\\S.*" },
|
|
85
|
+
"effort": { "enum": ["none", "minimal", "low", "medium", "high", "xhigh", "max", "ultra"] }
|
|
86
|
+
}
|
|
87
|
+
},
|
|
88
|
+
"selection": {
|
|
89
|
+
"type": "object",
|
|
90
|
+
"additionalProperties": false,
|
|
91
|
+
"required": ["eventIds", "structuralObservationIds", "includeOutcomeObservations"],
|
|
92
|
+
"properties": {
|
|
93
|
+
"eventIds": { "type": "array", "maxItems": 128, "uniqueItems": true, "items": { "$ref": "#/$defs/identifier" } },
|
|
94
|
+
"structuralObservationIds": { "type": "array", "maxItems": 128, "uniqueItems": true, "items": { "$ref": "#/$defs/identifier" } },
|
|
95
|
+
"includeOutcomeObservations": { "type": "boolean" }
|
|
96
|
+
}
|
|
97
|
+
},
|
|
98
|
+
"limits": { "$ref": "#/$defs/limits" },
|
|
99
|
+
"blinding": {
|
|
100
|
+
"type": "object",
|
|
101
|
+
"additionalProperties": false,
|
|
102
|
+
"required": ["evaluatedModelIdentity"],
|
|
103
|
+
"properties": { "evaluatedModelIdentity": { "enum": ["redact", "retain"] } }
|
|
104
|
+
}
|
|
105
|
+
}
|
|
106
|
+
},
|
|
107
|
+
"inputLimits": {
|
|
108
|
+
"type": "object",
|
|
109
|
+
"additionalProperties": false,
|
|
110
|
+
"required": ["maxEvidenceItems", "maxRecordChars", "maxInputChars", "maxCitations"],
|
|
111
|
+
"properties": {
|
|
112
|
+
"maxEvidenceItems": { "type": "integer", "minimum": 1, "maximum": 128 },
|
|
113
|
+
"maxRecordChars": { "type": "integer", "minimum": 256, "maximum": 65536 },
|
|
114
|
+
"maxInputChars": { "type": "integer", "minimum": 1024, "maximum": 1000000 },
|
|
115
|
+
"maxCitations": { "type": "integer", "minimum": 1, "maximum": 64 }
|
|
116
|
+
}
|
|
117
|
+
},
|
|
118
|
+
"stringList": {
|
|
119
|
+
"type": "array",
|
|
120
|
+
"maxItems": 256,
|
|
121
|
+
"uniqueItems": true,
|
|
122
|
+
"items": { "$ref": "#/$defs/identifier" }
|
|
123
|
+
},
|
|
124
|
+
"input": {
|
|
125
|
+
"type": "object",
|
|
126
|
+
"additionalProperties": false,
|
|
127
|
+
"required": ["schemaVersion", "promptVersion", "runId", "attemptId", "dataset", "behavior", "rubric", "selection", "blinding", "limits", "evidence"],
|
|
128
|
+
"properties": {
|
|
129
|
+
"schemaVersion": { "const": "ebo.semantic-judge-input/v1" },
|
|
130
|
+
"promptVersion": { "$ref": "#/$defs/identifier" },
|
|
131
|
+
"runId": { "$ref": "#/$defs/identifier" },
|
|
132
|
+
"attemptId": { "$ref": "#/$defs/identifier" },
|
|
133
|
+
"dataset": {
|
|
134
|
+
"type": "object",
|
|
135
|
+
"additionalProperties": false,
|
|
136
|
+
"required": ["schemaVersion", "digest"],
|
|
137
|
+
"properties": {
|
|
138
|
+
"schemaVersion": { "const": "ebo.normalized-dataset/v1" },
|
|
139
|
+
"digest": { "$ref": "#/$defs/digest" }
|
|
140
|
+
}
|
|
141
|
+
},
|
|
142
|
+
"behavior": { "$ref": "#/$defs/behavior" },
|
|
143
|
+
"rubric": { "$ref": "#/$defs/rubric" },
|
|
144
|
+
"selection": {
|
|
145
|
+
"type": "object",
|
|
146
|
+
"additionalProperties": false,
|
|
147
|
+
"required": ["requestedEventIds", "requestedStructuralObservationIds", "includedEventIds", "includedStructuralObservationIds", "omitted", "truncated"],
|
|
148
|
+
"properties": {
|
|
149
|
+
"requestedEventIds": { "$ref": "#/$defs/stringList" },
|
|
150
|
+
"requestedStructuralObservationIds": { "$ref": "#/$defs/stringList" },
|
|
151
|
+
"includedEventIds": { "$ref": "#/$defs/stringList" },
|
|
152
|
+
"includedStructuralObservationIds": { "$ref": "#/$defs/stringList" },
|
|
153
|
+
"omitted": { "$ref": "#/$defs/stringList" },
|
|
154
|
+
"truncated": { "$ref": "#/$defs/stringList" }
|
|
155
|
+
}
|
|
156
|
+
},
|
|
157
|
+
"blinding": {
|
|
158
|
+
"type": "object",
|
|
159
|
+
"additionalProperties": false,
|
|
160
|
+
"required": ["evaluatedModelIdentity", "redactionCount", "sourceIdentityPreserved", "residualClues"],
|
|
161
|
+
"properties": {
|
|
162
|
+
"evaluatedModelIdentity": { "enum": ["redacted", "retained"] },
|
|
163
|
+
"redactionCount": { "type": "integer", "minimum": 0 },
|
|
164
|
+
"sourceIdentityPreserved": { "const": true },
|
|
165
|
+
"residualClues": { "$ref": "#/$defs/stringList" }
|
|
166
|
+
}
|
|
167
|
+
},
|
|
168
|
+
"limits": { "$ref": "#/$defs/inputLimits" },
|
|
169
|
+
"evidence": {
|
|
170
|
+
"type": "array",
|
|
171
|
+
"maxItems": 128,
|
|
172
|
+
"items": {
|
|
173
|
+
"type": "object",
|
|
174
|
+
"additionalProperties": false,
|
|
175
|
+
"required": ["kind", "id", "content", "truncated"],
|
|
176
|
+
"properties": {
|
|
177
|
+
"kind": { "enum": ["event", "structural-observation"] },
|
|
178
|
+
"id": { "$ref": "#/$defs/identifier" },
|
|
179
|
+
"citation": {
|
|
180
|
+
"type": "object",
|
|
181
|
+
"additionalProperties": false,
|
|
182
|
+
"required": ["eventId", "nativeReference"],
|
|
183
|
+
"properties": {
|
|
184
|
+
"eventId": { "$ref": "#/$defs/identifier" },
|
|
185
|
+
"nativeReference": { "$ref": "urn:ebo:schema:uniform-events:v1#/$defs/nativeReference" }
|
|
186
|
+
}
|
|
187
|
+
},
|
|
188
|
+
"content": { "type": "string", "maxLength": 65568 },
|
|
189
|
+
"truncated": { "type": "boolean" }
|
|
190
|
+
}
|
|
191
|
+
}
|
|
192
|
+
}
|
|
193
|
+
}
|
|
194
|
+
},
|
|
195
|
+
"reference": {
|
|
196
|
+
"type": "object",
|
|
197
|
+
"additionalProperties": false,
|
|
198
|
+
"required": ["path", "digest", "sharingClass"],
|
|
199
|
+
"properties": {
|
|
200
|
+
"path": { "type": "string", "pattern": "^[^/]+$", "minLength": 1, "maxLength": 256 },
|
|
201
|
+
"digest": { "$ref": "#/$defs/digest" },
|
|
202
|
+
"sharingClass": { "const": "restricted" },
|
|
203
|
+
"truncated": { "type": "boolean" }
|
|
204
|
+
}
|
|
205
|
+
},
|
|
206
|
+
"timingAvailability": {
|
|
207
|
+
"oneOf": [
|
|
208
|
+
{
|
|
209
|
+
"type": "object",
|
|
210
|
+
"additionalProperties": false,
|
|
211
|
+
"required": ["status", "value"],
|
|
212
|
+
"properties": {
|
|
213
|
+
"status": { "const": "available" },
|
|
214
|
+
"value": {
|
|
215
|
+
"type": "object",
|
|
216
|
+
"additionalProperties": false,
|
|
217
|
+
"required": ["durationMs", "durationApiMs"],
|
|
218
|
+
"properties": {
|
|
219
|
+
"durationMs": { "type": "number", "minimum": 0 },
|
|
220
|
+
"durationApiMs": { "type": "number", "minimum": 0 }
|
|
221
|
+
}
|
|
222
|
+
}
|
|
223
|
+
}
|
|
224
|
+
},
|
|
225
|
+
{
|
|
226
|
+
"type": "object",
|
|
227
|
+
"additionalProperties": false,
|
|
228
|
+
"required": ["status", "reason"],
|
|
229
|
+
"properties": { "status": { "const": "unavailable" }, "reason": { "$ref": "#/$defs/text" } }
|
|
230
|
+
}
|
|
231
|
+
]
|
|
232
|
+
},
|
|
233
|
+
"usageAvailability": {
|
|
234
|
+
"oneOf": [
|
|
235
|
+
{
|
|
236
|
+
"type": "object",
|
|
237
|
+
"additionalProperties": false,
|
|
238
|
+
"required": ["status", "value"],
|
|
239
|
+
"properties": {
|
|
240
|
+
"status": { "const": "available" },
|
|
241
|
+
"value": {
|
|
242
|
+
"type": "object",
|
|
243
|
+
"additionalProperties": false,
|
|
244
|
+
"required": ["totalCostUsd", "numTurns", "mainLoop", "byModel"],
|
|
245
|
+
"properties": {
|
|
246
|
+
"totalCostUsd": { "type": "number", "minimum": 0 },
|
|
247
|
+
"numTurns": { "type": "integer", "minimum": 0 },
|
|
248
|
+
"mainLoop": {},
|
|
249
|
+
"byModel": {}
|
|
250
|
+
}
|
|
251
|
+
}
|
|
252
|
+
}
|
|
253
|
+
},
|
|
254
|
+
{
|
|
255
|
+
"type": "object",
|
|
256
|
+
"additionalProperties": false,
|
|
257
|
+
"required": ["status", "reason"],
|
|
258
|
+
"properties": { "status": { "const": "unavailable" }, "reason": { "$ref": "#/$defs/text" } }
|
|
259
|
+
}
|
|
260
|
+
]
|
|
261
|
+
},
|
|
262
|
+
"judgment": {
|
|
263
|
+
"type": "object",
|
|
264
|
+
"additionalProperties": false,
|
|
265
|
+
"required": ["schemaVersion", "id", "runId", "attemptId", "createdAt", "status", "input", "evaluator", "parse", "timing", "usage"],
|
|
266
|
+
"properties": {
|
|
267
|
+
"schemaVersion": { "const": "ebo.semantic-judgment/v1" },
|
|
268
|
+
"id": { "$ref": "#/$defs/identifier" },
|
|
269
|
+
"runId": { "$ref": "#/$defs/identifier" },
|
|
270
|
+
"attemptId": { "$ref": "#/$defs/identifier" },
|
|
271
|
+
"createdAt": { "type": "string", "format": "date-time" },
|
|
272
|
+
"status": { "enum": ["proposed", "failed"] },
|
|
273
|
+
"input": { "$ref": "#/$defs/reference" },
|
|
274
|
+
"evaluator": {
|
|
275
|
+
"type": "object",
|
|
276
|
+
"additionalProperties": false,
|
|
277
|
+
"required": ["provider", "model", "effort", "backend", "environment", "limits"],
|
|
278
|
+
"allOf": [{
|
|
279
|
+
"if": { "properties": { "backend": { "properties": { "id": { "const": "codex-app-server" } } } } },
|
|
280
|
+
"then": { "properties": {
|
|
281
|
+
"provider": { "const": "openai" },
|
|
282
|
+
"limits": { "properties": { "maxTurns": { "const": 1 } }, "not": { "required": ["maxBudgetUsd"] } },
|
|
283
|
+
"environment": {
|
|
284
|
+
"required": ["mode", "allowedKeys", "authentication"],
|
|
285
|
+
"properties": { "parentPreserved": { "const": false }, "removedKeys": { "const": ["*"] },
|
|
286
|
+
"allowedKeys": { "const": ["HOME", "CODEX_HOME", "PATH", "LANG", "LC_ALL", "TMPDIR"] } }
|
|
287
|
+
}
|
|
288
|
+
} },
|
|
289
|
+
"else": { "properties": {
|
|
290
|
+
"provider": { "const": "anthropic" },
|
|
291
|
+
"effort": { "enum": ["low", "medium", "high", "xhigh", "max"] },
|
|
292
|
+
"environment": { "properties": { "parentPreserved": { "const": true } },
|
|
293
|
+
"not": { "anyOf": [{ "required": ["mode"] }, { "required": ["allowedKeys"] }, { "required": ["authentication"] }] } }
|
|
294
|
+
} }
|
|
295
|
+
}],
|
|
296
|
+
"properties": {
|
|
297
|
+
"provider": { "enum": ["anthropic", "openai"] },
|
|
298
|
+
"model": { "$ref": "#/$defs/identifier" },
|
|
299
|
+
"effort": { "enum": ["none", "minimal", "low", "medium", "high", "xhigh", "max", "ultra"] },
|
|
300
|
+
"backend": {
|
|
301
|
+
"type": "object",
|
|
302
|
+
"additionalProperties": false,
|
|
303
|
+
"required": ["id", "version"],
|
|
304
|
+
"properties": { "id": { "enum": ["claude-agent-sdk", "codex-app-server"] }, "version": { "$ref": "#/$defs/identifier" } }
|
|
305
|
+
},
|
|
306
|
+
"environment": {
|
|
307
|
+
"type": "object",
|
|
308
|
+
"additionalProperties": false,
|
|
309
|
+
"required": ["parentPreserved", "modelEffortOverrides", "ambientTelemetry", "removedKeys"],
|
|
310
|
+
"properties": {
|
|
311
|
+
"parentPreserved": { "type": "boolean" },
|
|
312
|
+
"mode": { "const": "replace" },
|
|
313
|
+
"allowedKeys": { "$ref": "#/$defs/stringList" },
|
|
314
|
+
"authentication": { "const": "existing-auth-json-only" },
|
|
315
|
+
"modelEffortOverrides": { "const": "removed" },
|
|
316
|
+
"ambientTelemetry": { "const": "removed" },
|
|
317
|
+
"removedKeys": { "$ref": "#/$defs/stringList" }
|
|
318
|
+
}
|
|
319
|
+
},
|
|
320
|
+
"limits": { "$ref": "#/$defs/limits" }
|
|
321
|
+
}
|
|
322
|
+
},
|
|
323
|
+
"rawResponse": { "$ref": "#/$defs/reference" },
|
|
324
|
+
"rawModelResponse": { "$ref": "#/$defs/reference" },
|
|
325
|
+
"parse": {
|
|
326
|
+
"oneOf": [
|
|
327
|
+
{
|
|
328
|
+
"type": "object",
|
|
329
|
+
"additionalProperties": false,
|
|
330
|
+
"required": ["status"],
|
|
331
|
+
"properties": { "status": { "const": "valid" } }
|
|
332
|
+
},
|
|
333
|
+
{
|
|
334
|
+
"type": "object",
|
|
335
|
+
"additionalProperties": false,
|
|
336
|
+
"required": ["status", "kind", "message"],
|
|
337
|
+
"properties": {
|
|
338
|
+
"status": { "const": "failed" },
|
|
339
|
+
"kind": { "$ref": "#/$defs/identifier" },
|
|
340
|
+
"message": { "$ref": "#/$defs/text" }
|
|
341
|
+
}
|
|
342
|
+
}
|
|
343
|
+
]
|
|
344
|
+
},
|
|
345
|
+
"timing": { "$ref": "#/$defs/timingAvailability" },
|
|
346
|
+
"usage": { "$ref": "#/$defs/usageAvailability" },
|
|
347
|
+
"assertion": {
|
|
348
|
+
"type": "object",
|
|
349
|
+
"additionalProperties": false,
|
|
350
|
+
"required": ["path", "digest", "sharingClass", "id"],
|
|
351
|
+
"properties": {
|
|
352
|
+
"path": { "type": "string", "pattern": "^[^/]+$", "minLength": 1, "maxLength": 256 },
|
|
353
|
+
"digest": { "$ref": "#/$defs/digest" },
|
|
354
|
+
"sharingClass": { "const": "restricted" },
|
|
355
|
+
"truncated": { "type": "boolean" },
|
|
356
|
+
"id": { "$ref": "#/$defs/identifier" }
|
|
357
|
+
}
|
|
358
|
+
}
|
|
359
|
+
},
|
|
360
|
+
"allOf": [
|
|
361
|
+
{
|
|
362
|
+
"if": { "properties": { "status": { "const": "proposed" } }, "required": ["status"] },
|
|
363
|
+
"then": {
|
|
364
|
+
"required": ["assertion"],
|
|
365
|
+
"properties": { "parse": { "properties": { "status": { "const": "valid" } }, "required": ["status"] } }
|
|
366
|
+
},
|
|
367
|
+
"else": {
|
|
368
|
+
"not": { "required": ["assertion"] },
|
|
369
|
+
"properties": { "parse": { "properties": { "status": { "const": "failed" } }, "required": ["status"] } }
|
|
370
|
+
}
|
|
371
|
+
}
|
|
372
|
+
]
|
|
373
|
+
}
|
|
374
|
+
},
|
|
375
|
+
"oneOf": [
|
|
376
|
+
{ "$ref": "#/$defs/request" },
|
|
377
|
+
{ "$ref": "#/$defs/input" },
|
|
378
|
+
{ "$ref": "#/$defs/judgment" }
|
|
379
|
+
]
|
|
380
|
+
}
|