ralph-research 0.1.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/LICENSE +21 -0
- package/README.md +98 -0
- package/dist/adapters/extractor/command-extractor.d.ts +9 -0
- package/dist/adapters/extractor/command-extractor.js +93 -0
- package/dist/adapters/extractor/command-extractor.js.map +1 -0
- package/dist/adapters/extractor/llm-judge-extractor.d.ts +9 -0
- package/dist/adapters/extractor/llm-judge-extractor.js +12 -0
- package/dist/adapters/extractor/llm-judge-extractor.js.map +1 -0
- package/dist/adapters/fs/json-file-decision-store.d.ts +10 -0
- package/dist/adapters/fs/json-file-decision-store.js +53 -0
- package/dist/adapters/fs/json-file-decision-store.js.map +1 -0
- package/dist/adapters/fs/json-file-frontier-store.d.ts +8 -0
- package/dist/adapters/fs/json-file-frontier-store.js +29 -0
- package/dist/adapters/fs/json-file-frontier-store.js.map +1 -0
- package/dist/adapters/fs/json-file-run-store.d.ts +10 -0
- package/dist/adapters/fs/json-file-run-store.js +53 -0
- package/dist/adapters/fs/json-file-run-store.js.map +1 -0
- package/dist/adapters/fs/lockfile.d.ts +24 -0
- package/dist/adapters/fs/lockfile.js +110 -0
- package/dist/adapters/fs/lockfile.js.map +1 -0
- package/dist/adapters/fs/manifest-loader.d.ts +10 -0
- package/dist/adapters/fs/manifest-loader.js +43 -0
- package/dist/adapters/fs/manifest-loader.js.map +1 -0
- package/dist/adapters/git/git-client.d.ts +9 -0
- package/dist/adapters/git/git-client.js +23 -0
- package/dist/adapters/git/git-client.js.map +1 -0
- package/dist/adapters/index.d.ts +1 -0
- package/dist/adapters/index.js +3 -0
- package/dist/adapters/index.js.map +1 -0
- package/dist/adapters/judge/llm-judge-provider.d.ts +33 -0
- package/dist/adapters/judge/llm-judge-provider.js +90 -0
- package/dist/adapters/judge/llm-judge-provider.js.map +1 -0
- package/dist/adapters/proposer/command-proposer.d.ts +15 -0
- package/dist/adapters/proposer/command-proposer.js +29 -0
- package/dist/adapters/proposer/command-proposer.js.map +1 -0
- package/dist/app/context.d.ts +5 -0
- package/dist/app/context.js +7 -0
- package/dist/app/context.js.map +1 -0
- package/dist/app/services/manual-decision-service.d.ts +20 -0
- package/dist/app/services/manual-decision-service.js +143 -0
- package/dist/app/services/manual-decision-service.js.map +1 -0
- package/dist/app/services/project-state-service.d.ts +52 -0
- package/dist/app/services/project-state-service.js +92 -0
- package/dist/app/services/project-state-service.js.map +1 -0
- package/dist/app/services/run-cycle-service.d.ts +25 -0
- package/dist/app/services/run-cycle-service.js +69 -0
- package/dist/app/services/run-cycle-service.js.map +1 -0
- package/dist/cli/commands/accept.d.ts +10 -0
- package/dist/cli/commands/accept.js +54 -0
- package/dist/cli/commands/accept.js.map +1 -0
- package/dist/cli/commands/demo.d.ts +9 -0
- package/dist/cli/commands/demo.js +108 -0
- package/dist/cli/commands/demo.js.map +1 -0
- package/dist/cli/commands/frontier.d.ts +8 -0
- package/dist/cli/commands/frontier.js +48 -0
- package/dist/cli/commands/frontier.js.map +1 -0
- package/dist/cli/commands/init.d.ts +10 -0
- package/dist/cli/commands/init.js +123 -0
- package/dist/cli/commands/init.js.map +1 -0
- package/dist/cli/commands/inspect.d.ts +8 -0
- package/dist/cli/commands/inspect.js +55 -0
- package/dist/cli/commands/inspect.js.map +1 -0
- package/dist/cli/commands/reject.d.ts +10 -0
- package/dist/cli/commands/reject.js +54 -0
- package/dist/cli/commands/reject.js.map +1 -0
- package/dist/cli/commands/run.d.ts +13 -0
- package/dist/cli/commands/run.js +71 -0
- package/dist/cli/commands/run.js.map +1 -0
- package/dist/cli/commands/serve-mcp.d.ts +7 -0
- package/dist/cli/commands/serve-mcp.js +32 -0
- package/dist/cli/commands/serve-mcp.js.map +1 -0
- package/dist/cli/commands/status.d.ts +8 -0
- package/dist/cli/commands/status.js +53 -0
- package/dist/cli/commands/status.js.map +1 -0
- package/dist/cli/commands/validate.d.ts +11 -0
- package/dist/cli/commands/validate.js +56 -0
- package/dist/cli/commands/validate.js.map +1 -0
- package/dist/cli/main.d.ts +2 -0
- package/dist/cli/main.js +38 -0
- package/dist/cli/main.js.map +1 -0
- package/dist/core/engine/anchor-checker.d.ts +35 -0
- package/dist/core/engine/anchor-checker.js +84 -0
- package/dist/core/engine/anchor-checker.js.map +1 -0
- package/dist/core/engine/audit-sampler.d.ts +16 -0
- package/dist/core/engine/audit-sampler.js +25 -0
- package/dist/core/engine/audit-sampler.js.map +1 -0
- package/dist/core/engine/change-budget.d.ts +11 -0
- package/dist/core/engine/change-budget.js +10 -0
- package/dist/core/engine/change-budget.js.map +1 -0
- package/dist/core/engine/cycle-runner.d.ts +39 -0
- package/dist/core/engine/cycle-runner.js +652 -0
- package/dist/core/engine/cycle-runner.js.map +1 -0
- package/dist/core/engine/experiment-runner.d.ts +13 -0
- package/dist/core/engine/experiment-runner.js +24 -0
- package/dist/core/engine/experiment-runner.js.map +1 -0
- package/dist/core/engine/history-compactor.d.ts +15 -0
- package/dist/core/engine/history-compactor.js +76 -0
- package/dist/core/engine/history-compactor.js.map +1 -0
- package/dist/core/engine/judge-pack.d.ts +44 -0
- package/dist/core/engine/judge-pack.js +111 -0
- package/dist/core/engine/judge-pack.js.map +1 -0
- package/dist/core/engine/parallel-proposer.d.ts +21 -0
- package/dist/core/engine/parallel-proposer.js +58 -0
- package/dist/core/engine/parallel-proposer.js.map +1 -0
- package/dist/core/engine/scope-checker.d.ts +35 -0
- package/dist/core/engine/scope-checker.js +166 -0
- package/dist/core/engine/scope-checker.js.map +1 -0
- package/dist/core/engine/workspace-manager.d.ts +32 -0
- package/dist/core/engine/workspace-manager.js +145 -0
- package/dist/core/engine/workspace-manager.js.map +1 -0
- package/dist/core/index.d.ts +1 -0
- package/dist/core/index.js +3 -0
- package/dist/core/index.js.map +1 -0
- package/dist/core/manifest/defaults.d.ts +55 -0
- package/dist/core/manifest/defaults.js +56 -0
- package/dist/core/manifest/defaults.js.map +1 -0
- package/dist/core/manifest/schema.d.ts +647 -0
- package/dist/core/manifest/schema.js +254 -0
- package/dist/core/manifest/schema.js.map +1 -0
- package/dist/core/model/decision-record.d.ts +38 -0
- package/dist/core/model/decision-record.js +29 -0
- package/dist/core/model/decision-record.js.map +1 -0
- package/dist/core/model/frontier-entry.d.ts +24 -0
- package/dist/core/model/frontier-entry.js +15 -0
- package/dist/core/model/frontier-entry.js.map +1 -0
- package/dist/core/model/metric.d.ts +13 -0
- package/dist/core/model/metric.js +10 -0
- package/dist/core/model/metric.js.map +1 -0
- package/dist/core/model/run-record.d.ts +110 -0
- package/dist/core/model/run-record.js +104 -0
- package/dist/core/model/run-record.js.map +1 -0
- package/dist/core/ports/decision-store.d.ts +6 -0
- package/dist/core/ports/decision-store.js +2 -0
- package/dist/core/ports/decision-store.js.map +1 -0
- package/dist/core/ports/frontier-store.d.ts +5 -0
- package/dist/core/ports/frontier-store.js +2 -0
- package/dist/core/ports/frontier-store.js.map +1 -0
- package/dist/core/ports/run-store.d.ts +6 -0
- package/dist/core/ports/run-store.js +2 -0
- package/dist/core/ports/run-store.js.map +1 -0
- package/dist/core/state/constraint-engine.d.ts +18 -0
- package/dist/core/state/constraint-engine.js +42 -0
- package/dist/core/state/constraint-engine.js.map +1 -0
- package/dist/core/state/frontier-engine.d.ts +24 -0
- package/dist/core/state/frontier-engine.js +178 -0
- package/dist/core/state/frontier-engine.js.map +1 -0
- package/dist/core/state/ratchet-engine.d.ts +28 -0
- package/dist/core/state/ratchet-engine.js +177 -0
- package/dist/core/state/ratchet-engine.js.map +1 -0
- package/dist/core/state/run-state-machine.d.ts +17 -0
- package/dist/core/state/run-state-machine.js +94 -0
- package/dist/core/state/run-state-machine.js.map +1 -0
- package/dist/mcp/main.d.ts +1 -0
- package/dist/mcp/main.js +8 -0
- package/dist/mcp/main.js.map +1 -0
- package/dist/mcp/server.d.ts +6 -0
- package/dist/mcp/server.js +97 -0
- package/dist/mcp/server.js.map +1 -0
- package/dist/shared/fs-errors.d.ts +1 -0
- package/dist/shared/fs-errors.js +4 -0
- package/dist/shared/fs-errors.js.map +1 -0
- package/dist/shared/logger.d.ts +2 -0
- package/dist/shared/logger.js +5 -0
- package/dist/shared/logger.js.map +1 -0
- package/dist/shared/template-utils.d.ts +9 -0
- package/dist/shared/template-utils.js +50 -0
- package/dist/shared/template-utils.js.map +1 -0
- package/package.json +44 -0
- package/templates/writing/docs/draft.md +1 -0
- package/templates/writing/prompts/judge.md +15 -0
- package/templates/writing/ralph.yaml +63 -0
- package/templates/writing/scripts/experiment.mjs +6 -0
- package/templates/writing/scripts/metric.mjs +24 -0
- package/templates/writing/scripts/propose.mjs +13 -0
|
@@ -0,0 +1,254 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { DEFAULT_ALLOWED_GLOBS, DEFAULT_COMMAND_TIMEOUT_SEC, DEFAULT_LOW_CONFIDENCE_THRESHOLD, DEFAULT_MANIFEST_FILENAME, DEFAULT_MAX_FILES_CHANGED, DEFAULT_MAX_LINE_DELTA, DEFAULT_MAX_PATCH_COUNT, DEFAULT_PROPOSER_HISTORY_MAX_RUNS, DEFAULT_PROPOSER_EXPLORATION_RATIO, DEFAULT_PROJECT_BASELINE_REF, DEFAULT_PROJECT_WORKSPACE, DEFAULT_SCHEMA_VERSION, DEFAULT_STAGNATION_AFTER_REJECTIONS, DEFAULT_STORAGE_ROOT, manifestDefaults, } from "./defaults.js";
|
|
3
|
+
const recordOfStringsSchema = z.record(z.string(), z.string()).default({});
|
|
4
|
+
export const commandSpecSchema = z.object({
|
|
5
|
+
command: z.string().min(1),
|
|
6
|
+
cwd: z.string().optional(),
|
|
7
|
+
env: recordOfStringsSchema,
|
|
8
|
+
timeoutSec: z.number().int().positive().default(DEFAULT_COMMAND_TIMEOUT_SEC),
|
|
9
|
+
});
|
|
10
|
+
export const scopeSchema = z.object({
|
|
11
|
+
allowedGlobs: z.array(z.string().min(1)).min(1).default(DEFAULT_ALLOWED_GLOBS),
|
|
12
|
+
maxFilesChanged: z.number().int().positive().default(DEFAULT_MAX_FILES_CHANGED),
|
|
13
|
+
maxLineDelta: z.number().int().positive().default(DEFAULT_MAX_LINE_DELTA),
|
|
14
|
+
});
|
|
15
|
+
export const projectSchema = z.object({
|
|
16
|
+
name: z.string().min(1),
|
|
17
|
+
artifact: z.enum(["code", "manuscript", "literature_review", "prompt", "generic"]),
|
|
18
|
+
baselineRef: z.string().min(1).default(DEFAULT_PROJECT_BASELINE_REF),
|
|
19
|
+
workspace: z.enum(["git", "copy"]).default(DEFAULT_PROJECT_WORKSPACE),
|
|
20
|
+
});
|
|
21
|
+
export const proposerHistorySchema = z.object({
|
|
22
|
+
enabled: z.boolean().default(false),
|
|
23
|
+
maxRuns: z.number().int().min(1).default(DEFAULT_PROPOSER_HISTORY_MAX_RUNS),
|
|
24
|
+
});
|
|
25
|
+
export const commandProposerSchema = z.object({
|
|
26
|
+
type: z.literal("command"),
|
|
27
|
+
...commandSpecSchema.shape,
|
|
28
|
+
history: proposerHistorySchema.default(manifestDefaults.proposer.history),
|
|
29
|
+
});
|
|
30
|
+
export const operatorLlmProposerSchema = z.object({
|
|
31
|
+
type: z.literal("operator_llm"),
|
|
32
|
+
model: z.string().min(1),
|
|
33
|
+
prompt: z.string().min(1),
|
|
34
|
+
operators: z.array(z.string().min(1)).min(1),
|
|
35
|
+
explorationRatio: z.number().min(0).max(1).default(DEFAULT_PROPOSER_EXPLORATION_RATIO),
|
|
36
|
+
stagnationAfterRejections: z.number().int().min(1).default(DEFAULT_STAGNATION_AFTER_REJECTIONS),
|
|
37
|
+
maxPatchCount: z.number().int().min(1).default(DEFAULT_MAX_PATCH_COUNT),
|
|
38
|
+
history: proposerHistorySchema.default(manifestDefaults.proposer.history),
|
|
39
|
+
});
|
|
40
|
+
const leafProposerSchema = z.discriminatedUnion("type", [
|
|
41
|
+
commandProposerSchema,
|
|
42
|
+
operatorLlmProposerSchema,
|
|
43
|
+
]);
|
|
44
|
+
export const parallelProposerSchema = z.object({
|
|
45
|
+
type: z.literal("parallel"),
|
|
46
|
+
strategies: z.array(leafProposerSchema).min(2),
|
|
47
|
+
pickBest: z.enum(["highest_metric", "judge_pairwise"]),
|
|
48
|
+
history: proposerHistorySchema.default(manifestDefaults.proposer.history),
|
|
49
|
+
});
|
|
50
|
+
export const proposerSchema = z.discriminatedUnion("type", [
|
|
51
|
+
commandProposerSchema,
|
|
52
|
+
operatorLlmProposerSchema,
|
|
53
|
+
parallelProposerSchema,
|
|
54
|
+
]);
|
|
55
|
+
export const experimentSchema = z.object({
|
|
56
|
+
run: commandSpecSchema,
|
|
57
|
+
outputs: z
|
|
58
|
+
.array(z.object({
|
|
59
|
+
id: z.string().min(1),
|
|
60
|
+
path: z.string().min(1),
|
|
61
|
+
}))
|
|
62
|
+
.default([]),
|
|
63
|
+
});
|
|
64
|
+
export const judgePackSchema = z.object({
|
|
65
|
+
id: z.string().min(1),
|
|
66
|
+
mode: z.enum(["absolute", "pairwise"]),
|
|
67
|
+
blindPairwise: z.boolean().default(true),
|
|
68
|
+
orderRandomized: z.boolean().default(true),
|
|
69
|
+
repeats: z.number().int().min(1).default(manifestDefaults.judgePack.repeats),
|
|
70
|
+
aggregation: z.enum(["mean", "median", "majority_vote", "sign_test"]).default("majority_vote"),
|
|
71
|
+
judges: z
|
|
72
|
+
.array(z.object({
|
|
73
|
+
model: z.string().min(1),
|
|
74
|
+
weight: z.number().positive().default(1),
|
|
75
|
+
}))
|
|
76
|
+
.min(1),
|
|
77
|
+
lowConfidenceThreshold: z.number().min(0).max(1).default(DEFAULT_LOW_CONFIDENCE_THRESHOLD),
|
|
78
|
+
anchors: z
|
|
79
|
+
.object({
|
|
80
|
+
path: z.string().min(1),
|
|
81
|
+
minAgreementWithHuman: z.number().min(0).max(1).default(manifestDefaults.judgePack.anchors.minAgreementWithHuman),
|
|
82
|
+
})
|
|
83
|
+
.optional(),
|
|
84
|
+
audit: z
|
|
85
|
+
.object({
|
|
86
|
+
sampleRate: z.number().min(0).max(1).default(manifestDefaults.judgePack.audit.sampleRate),
|
|
87
|
+
freezeAutoAcceptIfAnchorFails: z.boolean().default(manifestDefaults.judgePack.audit.freezeAutoAcceptIfAnchorFails),
|
|
88
|
+
})
|
|
89
|
+
.default(manifestDefaults.judgePack.audit),
|
|
90
|
+
});
|
|
91
|
+
export const commandMetricExtractorSchema = z.object({
|
|
92
|
+
type: z.literal("command"),
|
|
93
|
+
...commandSpecSchema.shape,
|
|
94
|
+
parser: z.enum(["json_path", "regex", "plain_number"]).default("json_path"),
|
|
95
|
+
valuePath: z.string().optional(),
|
|
96
|
+
pattern: z.string().optional(),
|
|
97
|
+
});
|
|
98
|
+
export const llmJudgeMetricExtractorSchema = z.object({
|
|
99
|
+
type: z.literal("llm_judge"),
|
|
100
|
+
judgePack: z.string().min(1),
|
|
101
|
+
prompt: z.string().min(1),
|
|
102
|
+
mode: z.enum(["absolute", "pairwise"]).default("pairwise"),
|
|
103
|
+
compareAgainst: z.enum(["frontier.best", "none"]).default("frontier.best"),
|
|
104
|
+
inputs: z.record(z.string(), z.string()).default({}),
|
|
105
|
+
outputKey: z.string().min(1).default("score"),
|
|
106
|
+
});
|
|
107
|
+
export const metricExtractorSchema = z.discriminatedUnion("type", [
|
|
108
|
+
commandMetricExtractorSchema,
|
|
109
|
+
llmJudgeMetricExtractorSchema,
|
|
110
|
+
]);
|
|
111
|
+
const metricDefinitionBaseSchema = z.object({
|
|
112
|
+
id: z.string().min(1),
|
|
113
|
+
direction: z.enum(["maximize", "minimize"]),
|
|
114
|
+
});
|
|
115
|
+
export const numericMetricDefinitionSchema = metricDefinitionBaseSchema.extend({
|
|
116
|
+
kind: z.literal("numeric"),
|
|
117
|
+
extractor: commandMetricExtractorSchema,
|
|
118
|
+
});
|
|
119
|
+
export const llmScoreMetricDefinitionSchema = metricDefinitionBaseSchema.extend({
|
|
120
|
+
kind: z.literal("llm_score"),
|
|
121
|
+
extractor: llmJudgeMetricExtractorSchema,
|
|
122
|
+
});
|
|
123
|
+
export const metricDefinitionSchema = z.discriminatedUnion("kind", [
|
|
124
|
+
numericMetricDefinitionSchema,
|
|
125
|
+
llmScoreMetricDefinitionSchema,
|
|
126
|
+
]);
|
|
127
|
+
export const constraintSchema = z.object({
|
|
128
|
+
metric: z.string().min(1),
|
|
129
|
+
op: z.enum([">=", ">", "<=", "<", "=="]),
|
|
130
|
+
value: z.number(),
|
|
131
|
+
});
|
|
132
|
+
const singleBestFrontierSchema = z.object({
|
|
133
|
+
strategy: z.literal("single_best"),
|
|
134
|
+
primaryMetric: z.string().min(1),
|
|
135
|
+
});
|
|
136
|
+
export const paretoObjectiveSchema = z.object({
|
|
137
|
+
metric: z.string().min(1),
|
|
138
|
+
epsilon: z.number().nonnegative().default(0),
|
|
139
|
+
});
|
|
140
|
+
const paretoFrontierSchema = z.object({
|
|
141
|
+
strategy: z.literal("pareto"),
|
|
142
|
+
objectives: z.array(paretoObjectiveSchema).min(2),
|
|
143
|
+
tieBreaker: z.enum(["hypervolume", "none"]).default("hypervolume"),
|
|
144
|
+
referencePoint: z.record(z.string(), z.number()).optional(),
|
|
145
|
+
});
|
|
146
|
+
export const frontierSchema = z.discriminatedUnion("strategy", [
|
|
147
|
+
singleBestFrontierSchema,
|
|
148
|
+
paretoFrontierSchema,
|
|
149
|
+
]);
|
|
150
|
+
export const epsilonImproveRatchetSchema = z.object({
|
|
151
|
+
type: z.literal("epsilon_improve"),
|
|
152
|
+
metric: z.string().min(1).optional(),
|
|
153
|
+
epsilon: z.number().default(0),
|
|
154
|
+
});
|
|
155
|
+
export const approvalGateRatchetSchema = z.object({
|
|
156
|
+
type: z.literal("approval_gate"),
|
|
157
|
+
metric: z.string().min(1).optional(),
|
|
158
|
+
minConfidence: z.number().min(0).max(1).default(DEFAULT_LOW_CONFIDENCE_THRESHOLD),
|
|
159
|
+
graduation: z
|
|
160
|
+
.object({
|
|
161
|
+
consecutiveAccepts: z.number().int().min(1),
|
|
162
|
+
epsilon: z.number().default(0),
|
|
163
|
+
})
|
|
164
|
+
.optional(),
|
|
165
|
+
});
|
|
166
|
+
export const paretoDominanceRatchetSchema = z.object({
|
|
167
|
+
type: z.literal("pareto_dominance"),
|
|
168
|
+
});
|
|
169
|
+
export const ratchetSchema = z.discriminatedUnion("type", [
|
|
170
|
+
epsilonImproveRatchetSchema,
|
|
171
|
+
approvalGateRatchetSchema,
|
|
172
|
+
paretoDominanceRatchetSchema,
|
|
173
|
+
]);
|
|
174
|
+
export const storageSchema = z.object({
|
|
175
|
+
root: z.string().min(1).default(DEFAULT_STORAGE_ROOT),
|
|
176
|
+
});
|
|
177
|
+
const manifestShapeSchema = z.object({
|
|
178
|
+
schemaVersion: z.literal(DEFAULT_SCHEMA_VERSION),
|
|
179
|
+
project: projectSchema,
|
|
180
|
+
scope: scopeSchema.default(manifestDefaults.scope),
|
|
181
|
+
proposer: proposerSchema,
|
|
182
|
+
experiment: experimentSchema,
|
|
183
|
+
judgePacks: z.array(judgePackSchema).default([]),
|
|
184
|
+
metrics: z.object({
|
|
185
|
+
catalog: z.array(metricDefinitionSchema).min(1),
|
|
186
|
+
}),
|
|
187
|
+
constraints: z.array(constraintSchema).default([]),
|
|
188
|
+
frontier: frontierSchema,
|
|
189
|
+
ratchet: ratchetSchema,
|
|
190
|
+
storage: storageSchema.default(manifestDefaults.storage),
|
|
191
|
+
});
|
|
192
|
+
export const RalphManifestSchema = manifestShapeSchema.superRefine((manifest, ctx) => {
|
|
193
|
+
const metricIds = new Set(manifest.metrics.catalog.map((metric) => metric.id));
|
|
194
|
+
const judgePackIds = new Set(manifest.judgePacks.map((pack) => pack.id));
|
|
195
|
+
if (manifest.frontier.strategy === "single_best" && !metricIds.has(manifest.frontier.primaryMetric)) {
|
|
196
|
+
ctx.addIssue({
|
|
197
|
+
code: z.ZodIssueCode.custom,
|
|
198
|
+
message: `frontier.primaryMetric references unknown metric "${manifest.frontier.primaryMetric}"`,
|
|
199
|
+
path: ["frontier", "primaryMetric"],
|
|
200
|
+
});
|
|
201
|
+
}
|
|
202
|
+
if (manifest.frontier.strategy === "pareto") {
|
|
203
|
+
for (const [index, objective] of manifest.frontier.objectives.entries()) {
|
|
204
|
+
if (!metricIds.has(objective.metric)) {
|
|
205
|
+
ctx.addIssue({
|
|
206
|
+
code: z.ZodIssueCode.custom,
|
|
207
|
+
message: `frontier objective references unknown metric "${objective.metric}"`,
|
|
208
|
+
path: ["frontier", "objectives", index, "metric"],
|
|
209
|
+
});
|
|
210
|
+
}
|
|
211
|
+
}
|
|
212
|
+
if (manifest.ratchet.type !== "pareto_dominance") {
|
|
213
|
+
ctx.addIssue({
|
|
214
|
+
code: z.ZodIssueCode.custom,
|
|
215
|
+
message: "frontier.strategy=pareto requires ratchet.type=pareto_dominance",
|
|
216
|
+
path: ["ratchet", "type"],
|
|
217
|
+
});
|
|
218
|
+
}
|
|
219
|
+
}
|
|
220
|
+
if (manifest.frontier.strategy === "single_best" && manifest.ratchet.type === "pareto_dominance") {
|
|
221
|
+
ctx.addIssue({
|
|
222
|
+
code: z.ZodIssueCode.custom,
|
|
223
|
+
message: "ratchet.type=pareto_dominance requires frontier.strategy=pareto",
|
|
224
|
+
path: ["frontier", "strategy"],
|
|
225
|
+
});
|
|
226
|
+
}
|
|
227
|
+
if ("metric" in manifest.ratchet && manifest.ratchet.metric && !metricIds.has(manifest.ratchet.metric)) {
|
|
228
|
+
ctx.addIssue({
|
|
229
|
+
code: z.ZodIssueCode.custom,
|
|
230
|
+
message: `ratchet.metric references unknown metric "${manifest.ratchet.metric}"`,
|
|
231
|
+
path: ["ratchet", "metric"],
|
|
232
|
+
});
|
|
233
|
+
}
|
|
234
|
+
for (const [index, constraint] of manifest.constraints.entries()) {
|
|
235
|
+
if (!metricIds.has(constraint.metric)) {
|
|
236
|
+
ctx.addIssue({
|
|
237
|
+
code: z.ZodIssueCode.custom,
|
|
238
|
+
message: `constraint references unknown metric "${constraint.metric}"`,
|
|
239
|
+
path: ["constraints", index, "metric"],
|
|
240
|
+
});
|
|
241
|
+
}
|
|
242
|
+
}
|
|
243
|
+
for (const [index, metric] of manifest.metrics.catalog.entries()) {
|
|
244
|
+
if (metric.extractor.type === "llm_judge" && !judgePackIds.has(metric.extractor.judgePack)) {
|
|
245
|
+
ctx.addIssue({
|
|
246
|
+
code: z.ZodIssueCode.custom,
|
|
247
|
+
message: `metric extractor references unknown judgePack "${metric.extractor.judgePack}"`,
|
|
248
|
+
path: ["metrics", "catalog", index, "extractor", "judgePack"],
|
|
249
|
+
});
|
|
250
|
+
}
|
|
251
|
+
}
|
|
252
|
+
});
|
|
253
|
+
export { DEFAULT_MANIFEST_FILENAME };
|
|
254
|
+
//# sourceMappingURL=schema.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"schema.js","sourceRoot":"","sources":["../../../src/core/manifest/schema.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EACL,qBAAqB,EACrB,2BAA2B,EAC3B,gCAAgC,EAChC,yBAAyB,EACzB,yBAAyB,EACzB,sBAAsB,EACtB,uBAAuB,EACvB,iCAAiC,EACjC,kCAAkC,EAClC,4BAA4B,EAC5B,yBAAyB,EACzB,sBAAsB,EACtB,mCAAmC,EACnC,oBAAoB,EACpB,gBAAgB,GACjB,MAAM,eAAe,CAAC;AAEvB,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;AAE3E,MAAM,CAAC,MAAM,iBAAiB,GAAG,CAAC,CAAC,MAAM,CAAC;IACxC,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC1B,GAAG,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC1B,GAAG,EAAE,qBAAqB;IAC1B,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,2BAA2B,CAAC;CAC7E,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,WAAW,GAAG,CAAC,CAAC,MAAM,CAAC;IAClC,YAAY,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,qBAAqB,CAAC;IAC9E,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,yBAAyB,CAAC;IAC/E,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,sBAAsB,CAAC;CAC1E,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,aAAa,GAAG,CAAC,CAAC,MAAM,CAAC;IACpC,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACvB,QAAQ,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,YAAY,EAAE,mBAAmB,EAAE,QAAQ,EAAE,SAAS,CAAC,CAAC;IAClF,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,4BAA4B,CAAC;IACpE,SAAS,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,yBAAyB,CAAC;CACtE,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5C,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC;IACnC,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,iCAAiC,CAAC;CAC5E,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5C,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,SAAS,CAAC;IAC1B,GAAG,iBAAiB,CAAC,KAAK;IAC1B,OAAO,EAAE,qBAAqB,CAAC,OAAO,CAAC,gBAAgB,CAAC,QAAQ,CAAC,OAAO,CAAC;CAC1E,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,yBAAyB,GAAG,CAAC,CAAC,MAAM,CAAC;IAChD,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,cAAc,CAAC;IAC/B,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACxB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACzB,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC5C,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,kCAAkC,CAAC;IACtF,yBAAyB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,mCAAmC,CAAC;IAC/F,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,uBAAuB,CAAC;IACvE,OAAO,EAAE,qBAAqB,CAAC,OAAO,CAAC,gBAAgB,CAAC,QAAQ,CAAC,OAAO,CAAC;CAC1E,CAAC,CAAC;AAEH,MAAM,kBAAkB,GAAG,CAAC,CAAC,kBAAkB,CAAC,MAAM,EAAE;IACtD,qBAAqB;IACrB,yBAAyB;CAC1B,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC7C,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC;IAC3B,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC9C,QAAQ,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,gBAAgB,EAAE,gBAAgB,CAAC,CAAC;IACtD,OAAO,EAAE,qBAAqB,CAAC,OAAO,CAAC,gBAAgB,CAAC,QAAQ,CAAC,OAAO,CAAC;CAC1E,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,cAAc,GAAG,CAAC,CAAC,kBAAkB,CAAC,MAAM,EAAE;IACzD,qBAAqB;IACrB,yBAAyB;IACzB,sBAAsB;CACvB,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC,MAAM,CAAC;IACvC,GAAG,EAAE,iBAAiB;IACtB,OAAO,EAAE,CAAC;SACP,KAAK,CACJ,CAAC,CAAC,MAAM,CAAC;QACP,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACrB,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;KACxB,CAAC,CACH;SACA,OAAO,CAAC,EAAE,CAAC;CACf,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,CAAC,MAAM,CAAC;IACtC,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACrB,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IACtC,aAAa,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC;IACxC,eAAe,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC;IAC1C,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,gBAAgB,CAAC,SAAS,CAAC,OAAO,CAAC;IAC5E,WAAW,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,QAAQ,EAAE,eAAe,EAAE,WAAW,CAAC,CAAC,CAAC,OAAO,CAAC,eAAe,CAAC;IAC9F,MAAM,EAAE,CAAC;SACN,KAAK,CACJ,CAAC,CAAC,MAAM,CAAC;QACP,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACxB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;KACzC,CAAC,CACH;SACA,GAAG,CAAC,CAAC,CAAC;IACT,sBAAsB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,gCAAgC,CAAC;IAC1F,OAAO,EAAE,CAAC;SACP,MAAM,CAAC;QACN,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACvB,qBAAqB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,gBAAgB,CAAC,SAAS,CAAC,OAAO,CAAC,qBAAqB,CAAC;KAClH,CAAC;SACD,QAAQ,EAAE;IACb,KAAK,EAAE,CAAC;SACL,MAAM,CAAC;QACN,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,gBAAgB,CAAC,SAAS,CAAC,KAAK,CAAC,UAAU,CAAC;QACzF,6BAA6B,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,gBAAgB,CAAC,SAAS,CAAC,KAAK,CAAC,6BAA6B,CAAC;KACnH,CAAC;SACD,OAAO,CAAC,gBAAgB,CAAC,SAAS,CAAC,KAAK,CAAC;CAC7C,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,4BAA4B,GAAG,CAAC,CAAC,MAAM,CAAC;IACnD,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,SAAS,CAAC;IAC1B,GAAG,iBAAiB,CAAC,KAAK;IAC1B,MAAM,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,WAAW,EAAE,OAAO,EAAE,cAAc,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC;IAC3E,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAChC,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAC/B,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,6BAA6B,GAAG,CAAC,CAAC,MAAM,CAAC;IACpD,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC;IAC5B,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC5B,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACzB,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC;IAC1D,cAAc,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,eAAe,EAAE,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,eAAe,CAAC;IAC1E,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;IACpD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC;CAC9C,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,kBAAkB,CAAC,MAAM,EAAE;IAChE,4BAA4B;IAC5B,6BAA6B;CAC9B,CAAC,CAAC;AAEH,MAAM,0BAA0B,GAAG,CAAC,CAAC,MAAM,CAAC;IAC1C,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACrB,SAAS,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;CAC5C,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,6BAA6B,GAAG,0BAA0B,CAAC,MAAM,CAAC;IAC7E,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,SAAS,CAAC;IAC1B,SAAS,EAAE,4BAA4B;CACxC,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,8BAA8B,GAAG,0BAA0B,CAAC,MAAM,CAAC;IAC9E,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC;IAC5B,SAAS,EAAE,6BAA6B;CACzC,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC,kBAAkB,CAAC,MAAM,EAAE;IACjE,6BAA6B;IAC7B,8BAA8B;CAC/B,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC,MAAM,CAAC;IACvC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACzB,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,CAAC,CAAC;IACxC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE;CAClB,CAAC,CAAC;AAEH,MAAM,wBAAwB,GAAG,CAAC,CAAC,MAAM,CAAC;IACxC,QAAQ,EAAE,CAAC,CAAC,OAAO,CAAC,aAAa,CAAC;IAClC,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;CACjC,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5C,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACzB,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,WAAW,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;CAC7C,CAAC,CAAC;AAEH,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;IACpC,QAAQ,EAAE,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC;IAC7B,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,qBAAqB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACjD,UAAU,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,aAAa,CAAC;IAClE,cAAc,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;CAC5D,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,cAAc,GAAG,CAAC,CAAC,kBAAkB,CAAC,UAAU,EAAE;IAC7D,wBAAwB;IACxB,oBAAoB;CACrB,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,2BAA2B,GAAG,CAAC,CAAC,MAAM,CAAC;IAClD,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,iBAAiB,CAAC;IAClC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IACpC,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;CAC/B,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,yBAAyB,GAAG,CAAC,CAAC,MAAM,CAAC;IAChD,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,eAAe,CAAC;IAChC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IACpC,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,gCAAgC,CAAC;IACjF,UAAU,EAAE,CAAC;SACV,MAAM,CAAC;QACN,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QAC3C,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;KAC/B,CAAC;SACD,QAAQ,EAAE;CACd,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,4BAA4B,GAAG,CAAC,CAAC,MAAM,CAAC;IACnD,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,kBAAkB,CAAC;CACpC,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,aAAa,GAAG,CAAC,CAAC,kBAAkB,CAAC,MAAM,EAAE;IACxD,2BAA2B;IAC3B,yBAAyB;IACzB,4BAA4B;CAC7B,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,aAAa,GAAG,CAAC,CAAC,MAAM,CAAC;IACpC,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,oBAAoB,CAAC;CACtD,CAAC,CAAC;AAEH,MAAM,mBAAmB,GAAG,CAAC,CAAC,MAAM,CAAC;IACnC,aAAa,EAAE,CAAC,CAAC,OAAO,CAAC,sBAAsB,CAAC;IAChD,OAAO,EAAE,aAAa;IACtB,KAAK,EAAE,WAAW,CAAC,OAAO,CAAC,gBAAgB,CAAC,KAAK,CAAC;IAClD,QAAQ,EAAE,cAAc;IACxB,UAAU,EAAE,gBAAgB;IAC5B,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;IAChD,OAAO,EAAE,CAAC,CAAC,MAAM,CAAC;QAChB,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,sBAAsB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;KAChD,CAAC;IACF,WAAW,EAAE,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;IAClD,QAAQ,EAAE,cAAc;IACxB,OAAO,EAAE,aAAa;IACtB,OAAO,EAAE,aAAa,CAAC,OAAO,CAAC,gBAAgB,CAAC,OAAO,CAAC;CACzD,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,mBAAmB,GAAG,mBAAmB,CAAC,WAAW,CAAC,CAAC,QAAQ,EAAE,GAAG,EAAE,EAAE;IACnF,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC;IAC/E,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;IAEzE,IAAI,QAAQ,CAAC,QAAQ,CAAC,QAAQ,KAAK,aAAa,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAC,QAAQ,CAAC,aAAa,CAAC,EAAE,CAAC;QACpG,GAAG,CAAC,QAAQ,CAAC;YACX,IAAI,EAAE,CAAC,CAAC,YAAY,CAAC,MAAM;YAC3B,OAAO,EAAE,qDAAqD,QAAQ,CAAC,QAAQ,CAAC,aAAa,GAAG;YAChG,IAAI,EAAE,CAAC,UAAU,EAAE,eAAe,CAAC;SACpC,CAAC,CAAC;IACL,CAAC;IAED,IAAI,QAAQ,CAAC,QAAQ,CAAC,QAAQ,KAAK,QAAQ,EAAE,CAAC;QAC5C,KAAK,MAAM,CAAC,KAAK,EAAE,SAAS,CAAC,IAAI,QAAQ,CAAC,QAAQ,CAAC,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC;YACxE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,SAAS,CAAC,MAAM,CAAC,EAAE,CAAC;gBACrC,GAAG,CAAC,QAAQ,CAAC;oBACX,IAAI,EAAE,CAAC,CAAC,YAAY,CAAC,MAAM;oBAC3B,OAAO,EAAE,iDAAiD,SAAS,CAAC,MAAM,GAAG;oBAC7E,IAAI,EAAE,CAAC,UAAU,EAAE,YAAY,EAAE,KAAK,EAAE,QAAQ,CAAC;iBAClD,CAAC,CAAC;YACL,CAAC;QACH,CAAC;QAED,IAAI,QAAQ,CAAC,OAAO,CAAC,IAAI,KAAK,kBAAkB,EAAE,CAAC;YACjD,GAAG,CAAC,QAAQ,CAAC;gBACX,IAAI,EAAE,CAAC,CAAC,YAAY,CAAC,MAAM;gBAC3B,OAAO,EAAE,iEAAiE;gBAC1E,IAAI,EAAE,CAAC,SAAS,EAAE,MAAM,CAAC;aAC1B,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,IAAI,QAAQ,CAAC,QAAQ,CAAC,QAAQ,KAAK,aAAa,IAAI,QAAQ,CAAC,OAAO,CAAC,IAAI,KAAK,kBAAkB,EAAE,CAAC;QACjG,GAAG,CAAC,QAAQ,CAAC;YACX,IAAI,EAAE,CAAC,CAAC,YAAY,CAAC,MAAM;YAC3B,OAAO,EAAE,iEAAiE;YAC1E,IAAI,EAAE,CAAC,UAAU,EAAE,UAAU,CAAC;SAC/B,CAAC,CAAC;IACL,CAAC;IAED,IAAI,QAAQ,IAAI,QAAQ,CAAC,OAAO,IAAI,QAAQ,CAAC,OAAO,CAAC,MAAM,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;QACvG,GAAG,CAAC,QAAQ,CAAC;YACX,IAAI,EAAE,CAAC,CAAC,YAAY,CAAC,MAAM;YAC3B,OAAO,EAAE,6CAA6C,QAAQ,CAAC,OAAO,CAAC,MAAM,GAAG;YAChF,IAAI,EAAE,CAAC,SAAS,EAAE,QAAQ,CAAC;SAC5B,CAAC,CAAC;IACL,CAAC;IAED,KAAK,MAAM,CAAC,KAAK,EAAE,UAAU,CAAC,IAAI,QAAQ,CAAC,WAAW,CAAC,OAAO,EAAE,EAAE,CAAC;QACjE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,UAAU,CAAC,MAAM,CAAC,EAAE,CAAC;YACtC,GAAG,CAAC,QAAQ,CAAC;gBACX,IAAI,EAAE,CAAC,CAAC,YAAY,CAAC,MAAM;gBAC3B,OAAO,EAAE,yCAAyC,UAAU,CAAC,MAAM,GAAG;gBACtE,IAAI,EAAE,CAAC,aAAa,EAAE,KAAK,EAAE,QAAQ,CAAC;aACvC,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,KAAK,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,IAAI,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC;QACjE,IAAI,MAAM,CAAC,SAAS,CAAC,IAAI,KAAK,WAAW,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,MAAM,CAAC,SAAS,CAAC,SAAS,CAAC,EAAE,CAAC;YAC3F,GAAG,CAAC,QAAQ,CAAC;gBACX,IAAI,EAAE,CAAC,CAAC,YAAY,CAAC,MAAM;gBAC3B,OAAO,EAAE,kDAAkD,MAAM,CAAC,SAAS,CAAC,SAAS,GAAG;gBACxF,IAAI,EAAE,CAAC,SAAS,EAAE,SAAS,EAAE,KAAK,EAAE,WAAW,EAAE,WAAW,CAAC;aAC9D,CAAC,CAAC;QACL,CAAC;IACH,CAAC;AACH,CAAC,CAAC,CAAC;AAkBH,OAAO,EAAE,yBAAyB,EAAE,CAAC"}
|
|
@@ -0,0 +1,38 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
export declare const decisionOutcomeSchema: z.ZodEnum<{
|
|
3
|
+
accepted: "accepted";
|
|
4
|
+
rejected: "rejected";
|
|
5
|
+
needs_human: "needs_human";
|
|
6
|
+
}>;
|
|
7
|
+
export declare const decisionRecordSchema: z.ZodObject<{
|
|
8
|
+
decisionId: z.ZodString;
|
|
9
|
+
runId: z.ZodString;
|
|
10
|
+
outcome: z.ZodEnum<{
|
|
11
|
+
accepted: "accepted";
|
|
12
|
+
rejected: "rejected";
|
|
13
|
+
needs_human: "needs_human";
|
|
14
|
+
}>;
|
|
15
|
+
actorType: z.ZodEnum<{
|
|
16
|
+
system: "system";
|
|
17
|
+
human: "human";
|
|
18
|
+
}>;
|
|
19
|
+
actorId: z.ZodOptional<z.ZodString>;
|
|
20
|
+
policyType: z.ZodString;
|
|
21
|
+
metricId: z.ZodOptional<z.ZodString>;
|
|
22
|
+
delta: z.ZodOptional<z.ZodNumber>;
|
|
23
|
+
reason: z.ZodString;
|
|
24
|
+
createdAt: z.ZodString;
|
|
25
|
+
frontierChanged: z.ZodBoolean;
|
|
26
|
+
beforeFrontierIds: z.ZodDefault<z.ZodArray<z.ZodString>>;
|
|
27
|
+
afterFrontierIds: z.ZodDefault<z.ZodArray<z.ZodString>>;
|
|
28
|
+
commitSha: z.ZodOptional<z.ZodString>;
|
|
29
|
+
auditRequired: z.ZodDefault<z.ZodBoolean>;
|
|
30
|
+
graduation: z.ZodOptional<z.ZodObject<{
|
|
31
|
+
activatedPolicy: z.ZodLiteral<"epsilon_improve">;
|
|
32
|
+
consecutiveAccepts: z.ZodNumber;
|
|
33
|
+
epsilon: z.ZodNumber;
|
|
34
|
+
effectiveNextCycle: z.ZodDefault<z.ZodBoolean>;
|
|
35
|
+
reason: z.ZodString;
|
|
36
|
+
}, z.core.$strip>>;
|
|
37
|
+
}, z.core.$strip>;
|
|
38
|
+
export type DecisionRecord = z.infer<typeof decisionRecordSchema>;
|
|
@@ -0,0 +1,29 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
export const decisionOutcomeSchema = z.enum(["accepted", "rejected", "needs_human"]);
|
|
3
|
+
export const decisionRecordSchema = z.object({
|
|
4
|
+
decisionId: z.string().min(1),
|
|
5
|
+
runId: z.string().min(1),
|
|
6
|
+
outcome: decisionOutcomeSchema,
|
|
7
|
+
actorType: z.enum(["system", "human"]),
|
|
8
|
+
actorId: z.string().min(1).optional(),
|
|
9
|
+
policyType: z.string().min(1),
|
|
10
|
+
metricId: z.string().min(1).optional(),
|
|
11
|
+
delta: z.number().optional(),
|
|
12
|
+
reason: z.string().min(1),
|
|
13
|
+
createdAt: z.string().datetime(),
|
|
14
|
+
frontierChanged: z.boolean(),
|
|
15
|
+
beforeFrontierIds: z.array(z.string().min(1)).default([]),
|
|
16
|
+
afterFrontierIds: z.array(z.string().min(1)).default([]),
|
|
17
|
+
commitSha: z.string().min(1).optional(),
|
|
18
|
+
auditRequired: z.boolean().default(false),
|
|
19
|
+
graduation: z
|
|
20
|
+
.object({
|
|
21
|
+
activatedPolicy: z.literal("epsilon_improve"),
|
|
22
|
+
consecutiveAccepts: z.number().int().min(1),
|
|
23
|
+
epsilon: z.number(),
|
|
24
|
+
effectiveNextCycle: z.boolean().default(true),
|
|
25
|
+
reason: z.string().min(1),
|
|
26
|
+
})
|
|
27
|
+
.optional(),
|
|
28
|
+
});
|
|
29
|
+
//# sourceMappingURL=decision-record.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"decision-record.js","sourceRoot":"","sources":["../../../src/core/model/decision-record.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,UAAU,EAAE,aAAa,CAAC,CAAC,CAAC;AAErF,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC3C,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7B,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACxB,OAAO,EAAE,qBAAqB;IAC9B,SAAS,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IACtC,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IACrC,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7B,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IACtC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC5B,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACzB,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAChC,eAAe,EAAE,CAAC,CAAC,OAAO,EAAE;IAC5B,iBAAiB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;IACzD,gBAAgB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;IACxD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IACvC,aAAa,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC;IACzC,UAAU,EAAE,CAAC;SACV,MAAM,CAAC;QACN,eAAe,EAAE,CAAC,CAAC,OAAO,CAAC,iBAAiB,CAAC;QAC7C,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QAC3C,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;QACnB,kBAAkB,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC;QAC7C,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;KAC1B,CAAC;SACD,QAAQ,EAAE;CACd,CAAC,CAAC"}
|
|
@@ -0,0 +1,24 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
export declare const frontierEntrySchema: z.ZodObject<{
|
|
3
|
+
frontierId: z.ZodString;
|
|
4
|
+
runId: z.ZodString;
|
|
5
|
+
candidateId: z.ZodString;
|
|
6
|
+
acceptedAt: z.ZodString;
|
|
7
|
+
commitSha: z.ZodOptional<z.ZodString>;
|
|
8
|
+
metrics: z.ZodRecord<z.ZodString, z.ZodObject<{
|
|
9
|
+
metricId: z.ZodString;
|
|
10
|
+
value: z.ZodNumber;
|
|
11
|
+
confidence: z.ZodOptional<z.ZodNumber>;
|
|
12
|
+
direction: z.ZodEnum<{
|
|
13
|
+
maximize: "maximize";
|
|
14
|
+
minimize: "minimize";
|
|
15
|
+
}>;
|
|
16
|
+
judgeTracePath: z.ZodOptional<z.ZodString>;
|
|
17
|
+
details: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
18
|
+
}, z.core.$strip>>;
|
|
19
|
+
artifacts: z.ZodArray<z.ZodObject<{
|
|
20
|
+
id: z.ZodString;
|
|
21
|
+
path: z.ZodString;
|
|
22
|
+
}, z.core.$strip>>;
|
|
23
|
+
}, z.core.$strip>;
|
|
24
|
+
export type FrontierEntry = z.infer<typeof frontierEntrySchema>;
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { metricResultSchema } from "./metric.js";
|
|
3
|
+
export const frontierEntrySchema = z.object({
|
|
4
|
+
frontierId: z.string().min(1),
|
|
5
|
+
runId: z.string().min(1),
|
|
6
|
+
candidateId: z.string().min(1),
|
|
7
|
+
acceptedAt: z.string().datetime(),
|
|
8
|
+
commitSha: z.string().min(1).optional(),
|
|
9
|
+
metrics: z.record(z.string(), metricResultSchema),
|
|
10
|
+
artifacts: z.array(z.object({
|
|
11
|
+
id: z.string().min(1),
|
|
12
|
+
path: z.string().min(1),
|
|
13
|
+
})),
|
|
14
|
+
});
|
|
15
|
+
//# sourceMappingURL=frontier-entry.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"frontier-entry.js","sourceRoot":"","sources":["../../../src/core/model/frontier-entry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,kBAAkB,EAAE,MAAM,aAAa,CAAC;AAEjD,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC1C,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7B,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACxB,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC9B,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACjC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IACvC,OAAO,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,kBAAkB,CAAC;IACjD,SAAS,EAAE,CAAC,CAAC,KAAK,CAChB,CAAC,CAAC,MAAM,CAAC;QACP,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACrB,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;KACxB,CAAC,CACH;CACF,CAAC,CAAC"}
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
export declare const metricResultSchema: z.ZodObject<{
|
|
3
|
+
metricId: z.ZodString;
|
|
4
|
+
value: z.ZodNumber;
|
|
5
|
+
confidence: z.ZodOptional<z.ZodNumber>;
|
|
6
|
+
direction: z.ZodEnum<{
|
|
7
|
+
maximize: "maximize";
|
|
8
|
+
minimize: "minimize";
|
|
9
|
+
}>;
|
|
10
|
+
judgeTracePath: z.ZodOptional<z.ZodString>;
|
|
11
|
+
details: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
12
|
+
}, z.core.$strip>;
|
|
13
|
+
export type MetricResult = z.infer<typeof metricResultSchema>;
|
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
export const metricResultSchema = z.object({
|
|
3
|
+
metricId: z.string().min(1),
|
|
4
|
+
value: z.number(),
|
|
5
|
+
confidence: z.number().min(0).max(1).optional(),
|
|
6
|
+
direction: z.enum(["maximize", "minimize"]),
|
|
7
|
+
judgeTracePath: z.string().min(1).optional(),
|
|
8
|
+
details: z.record(z.string(), z.unknown()).default({}),
|
|
9
|
+
});
|
|
10
|
+
//# sourceMappingURL=metric.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"metric.js","sourceRoot":"","sources":["../../../src/core/model/metric.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC3B,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE;IACjB,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IAC/C,SAAS,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAC3C,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IAC5C,OAAO,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;CACvD,CAAC,CAAC"}
|
|
@@ -0,0 +1,110 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
export declare const runPhaseSchema: z.ZodEnum<{
|
|
3
|
+
proposed: "proposed";
|
|
4
|
+
executed: "executed";
|
|
5
|
+
evaluated: "evaluated";
|
|
6
|
+
decision_written: "decision_written";
|
|
7
|
+
committed: "committed";
|
|
8
|
+
frontier_updated: "frontier_updated";
|
|
9
|
+
completed: "completed";
|
|
10
|
+
failed: "failed";
|
|
11
|
+
}>;
|
|
12
|
+
export declare const pendingActionSchema: z.ZodEnum<{
|
|
13
|
+
none: "none";
|
|
14
|
+
execute_experiment: "execute_experiment";
|
|
15
|
+
evaluate_metrics: "evaluate_metrics";
|
|
16
|
+
write_decision: "write_decision";
|
|
17
|
+
commit_candidate: "commit_candidate";
|
|
18
|
+
update_frontier: "update_frontier";
|
|
19
|
+
cleanup_workspace: "cleanup_workspace";
|
|
20
|
+
}>;
|
|
21
|
+
export declare const runStatusSchema: z.ZodEnum<{
|
|
22
|
+
accepted: "accepted";
|
|
23
|
+
rejected: "rejected";
|
|
24
|
+
needs_human: "needs_human";
|
|
25
|
+
evaluated: "evaluated";
|
|
26
|
+
failed: "failed";
|
|
27
|
+
running: "running";
|
|
28
|
+
}>;
|
|
29
|
+
export declare const runRecordSchema: z.ZodObject<{
|
|
30
|
+
runId: z.ZodString;
|
|
31
|
+
cycle: z.ZodNumber;
|
|
32
|
+
candidateId: z.ZodString;
|
|
33
|
+
status: z.ZodEnum<{
|
|
34
|
+
accepted: "accepted";
|
|
35
|
+
rejected: "rejected";
|
|
36
|
+
needs_human: "needs_human";
|
|
37
|
+
evaluated: "evaluated";
|
|
38
|
+
failed: "failed";
|
|
39
|
+
running: "running";
|
|
40
|
+
}>;
|
|
41
|
+
phase: z.ZodEnum<{
|
|
42
|
+
proposed: "proposed";
|
|
43
|
+
executed: "executed";
|
|
44
|
+
evaluated: "evaluated";
|
|
45
|
+
decision_written: "decision_written";
|
|
46
|
+
committed: "committed";
|
|
47
|
+
frontier_updated: "frontier_updated";
|
|
48
|
+
completed: "completed";
|
|
49
|
+
failed: "failed";
|
|
50
|
+
}>;
|
|
51
|
+
pendingAction: z.ZodDefault<z.ZodEnum<{
|
|
52
|
+
none: "none";
|
|
53
|
+
execute_experiment: "execute_experiment";
|
|
54
|
+
evaluate_metrics: "evaluate_metrics";
|
|
55
|
+
write_decision: "write_decision";
|
|
56
|
+
commit_candidate: "commit_candidate";
|
|
57
|
+
update_frontier: "update_frontier";
|
|
58
|
+
cleanup_workspace: "cleanup_workspace";
|
|
59
|
+
}>>;
|
|
60
|
+
startedAt: z.ZodString;
|
|
61
|
+
endedAt: z.ZodOptional<z.ZodString>;
|
|
62
|
+
manifestHash: z.ZodString;
|
|
63
|
+
workspaceRef: z.ZodString;
|
|
64
|
+
workspacePath: z.ZodOptional<z.ZodString>;
|
|
65
|
+
proposal: z.ZodObject<{
|
|
66
|
+
proposerType: z.ZodString;
|
|
67
|
+
summary: z.ZodString;
|
|
68
|
+
operators: z.ZodDefault<z.ZodArray<z.ZodString>>;
|
|
69
|
+
patchPath: z.ZodOptional<z.ZodString>;
|
|
70
|
+
diffLines: z.ZodOptional<z.ZodNumber>;
|
|
71
|
+
filesChanged: z.ZodOptional<z.ZodNumber>;
|
|
72
|
+
changedPaths: z.ZodOptional<z.ZodArray<z.ZodString>>;
|
|
73
|
+
withinBudget: z.ZodOptional<z.ZodBoolean>;
|
|
74
|
+
}, z.core.$strip>;
|
|
75
|
+
artifacts: z.ZodArray<z.ZodObject<{
|
|
76
|
+
id: z.ZodString;
|
|
77
|
+
path: z.ZodString;
|
|
78
|
+
}, z.core.$strip>>;
|
|
79
|
+
metrics: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodObject<{
|
|
80
|
+
metricId: z.ZodString;
|
|
81
|
+
value: z.ZodNumber;
|
|
82
|
+
confidence: z.ZodOptional<z.ZodNumber>;
|
|
83
|
+
direction: z.ZodEnum<{
|
|
84
|
+
maximize: "maximize";
|
|
85
|
+
minimize: "minimize";
|
|
86
|
+
}>;
|
|
87
|
+
judgeTracePath: z.ZodOptional<z.ZodString>;
|
|
88
|
+
details: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
89
|
+
}, z.core.$strip>>>;
|
|
90
|
+
constraints: z.ZodDefault<z.ZodArray<z.ZodObject<{
|
|
91
|
+
metric: z.ZodString;
|
|
92
|
+
passed: z.ZodBoolean;
|
|
93
|
+
actual: z.ZodNumber;
|
|
94
|
+
expected: z.ZodNumber;
|
|
95
|
+
op: z.ZodString;
|
|
96
|
+
}, z.core.$strip>>>;
|
|
97
|
+
decisionId: z.ZodOptional<z.ZodString>;
|
|
98
|
+
logs: z.ZodDefault<z.ZodObject<{
|
|
99
|
+
proposeStdoutPath: z.ZodOptional<z.ZodString>;
|
|
100
|
+
runStdoutPath: z.ZodOptional<z.ZodString>;
|
|
101
|
+
}, z.core.$strip>>;
|
|
102
|
+
error: z.ZodOptional<z.ZodObject<{
|
|
103
|
+
message: z.ZodString;
|
|
104
|
+
stack: z.ZodOptional<z.ZodString>;
|
|
105
|
+
}, z.core.$strip>>;
|
|
106
|
+
}, z.core.$strip>;
|
|
107
|
+
export type RunRecord = z.infer<typeof runRecordSchema>;
|
|
108
|
+
export type RunPhase = z.infer<typeof runPhaseSchema>;
|
|
109
|
+
export type RunStatus = z.infer<typeof runStatusSchema>;
|
|
110
|
+
export type PendingAction = z.infer<typeof pendingActionSchema>;
|
|
@@ -0,0 +1,104 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { metricResultSchema } from "./metric.js";
|
|
3
|
+
export const runPhaseSchema = z.enum([
|
|
4
|
+
"proposed",
|
|
5
|
+
"executed",
|
|
6
|
+
"evaluated",
|
|
7
|
+
"decision_written",
|
|
8
|
+
"committed",
|
|
9
|
+
"frontier_updated",
|
|
10
|
+
"completed",
|
|
11
|
+
"failed",
|
|
12
|
+
]);
|
|
13
|
+
export const pendingActionSchema = z.enum([
|
|
14
|
+
"none",
|
|
15
|
+
"execute_experiment",
|
|
16
|
+
"evaluate_metrics",
|
|
17
|
+
"write_decision",
|
|
18
|
+
"commit_candidate",
|
|
19
|
+
"update_frontier",
|
|
20
|
+
"cleanup_workspace",
|
|
21
|
+
]);
|
|
22
|
+
export const runStatusSchema = z.enum([
|
|
23
|
+
"running",
|
|
24
|
+
"evaluated",
|
|
25
|
+
"accepted",
|
|
26
|
+
"rejected",
|
|
27
|
+
"needs_human",
|
|
28
|
+
"failed",
|
|
29
|
+
]);
|
|
30
|
+
export const runRecordSchema = z
|
|
31
|
+
.object({
|
|
32
|
+
runId: z.string().min(1),
|
|
33
|
+
cycle: z.number().int().nonnegative(),
|
|
34
|
+
candidateId: z.string().min(1),
|
|
35
|
+
status: runStatusSchema,
|
|
36
|
+
phase: runPhaseSchema,
|
|
37
|
+
pendingAction: pendingActionSchema.default("none"),
|
|
38
|
+
startedAt: z.string().datetime(),
|
|
39
|
+
endedAt: z.string().datetime().optional(),
|
|
40
|
+
manifestHash: z.string().min(1),
|
|
41
|
+
workspaceRef: z.string().min(1),
|
|
42
|
+
workspacePath: z.string().min(1).optional(),
|
|
43
|
+
proposal: z.object({
|
|
44
|
+
proposerType: z.string().min(1),
|
|
45
|
+
summary: z.string().min(1),
|
|
46
|
+
operators: z.array(z.string().min(1)).default([]),
|
|
47
|
+
patchPath: z.string().min(1).optional(),
|
|
48
|
+
diffLines: z.number().int().nonnegative().optional(),
|
|
49
|
+
filesChanged: z.number().int().nonnegative().optional(),
|
|
50
|
+
changedPaths: z.array(z.string().min(1)).optional(),
|
|
51
|
+
withinBudget: z.boolean().optional(),
|
|
52
|
+
}),
|
|
53
|
+
artifacts: z.array(z.object({
|
|
54
|
+
id: z.string().min(1),
|
|
55
|
+
path: z.string().min(1),
|
|
56
|
+
})),
|
|
57
|
+
metrics: z.record(z.string(), metricResultSchema).default({}),
|
|
58
|
+
constraints: z
|
|
59
|
+
.array(z.object({
|
|
60
|
+
metric: z.string().min(1),
|
|
61
|
+
passed: z.boolean(),
|
|
62
|
+
actual: z.number(),
|
|
63
|
+
expected: z.number(),
|
|
64
|
+
op: z.string().min(1),
|
|
65
|
+
}))
|
|
66
|
+
.default([]),
|
|
67
|
+
decisionId: z.string().min(1).optional(),
|
|
68
|
+
logs: z
|
|
69
|
+
.object({
|
|
70
|
+
proposeStdoutPath: z.string().min(1).optional(),
|
|
71
|
+
runStdoutPath: z.string().min(1).optional(),
|
|
72
|
+
})
|
|
73
|
+
.default({}),
|
|
74
|
+
error: z
|
|
75
|
+
.object({
|
|
76
|
+
message: z.string().min(1),
|
|
77
|
+
stack: z.string().min(1).optional(),
|
|
78
|
+
})
|
|
79
|
+
.optional(),
|
|
80
|
+
})
|
|
81
|
+
.superRefine((record, ctx) => {
|
|
82
|
+
if (record.phase === "completed" && !record.endedAt) {
|
|
83
|
+
ctx.addIssue({
|
|
84
|
+
code: z.ZodIssueCode.custom,
|
|
85
|
+
message: "completed runs must include endedAt",
|
|
86
|
+
path: ["endedAt"],
|
|
87
|
+
});
|
|
88
|
+
}
|
|
89
|
+
if (record.phase === "failed" && !record.error) {
|
|
90
|
+
ctx.addIssue({
|
|
91
|
+
code: z.ZodIssueCode.custom,
|
|
92
|
+
message: "failed runs must include error information",
|
|
93
|
+
path: ["error"],
|
|
94
|
+
});
|
|
95
|
+
}
|
|
96
|
+
if (record.phase === "decision_written" && !record.decisionId) {
|
|
97
|
+
ctx.addIssue({
|
|
98
|
+
code: z.ZodIssueCode.custom,
|
|
99
|
+
message: "decision_written runs must include a decisionId",
|
|
100
|
+
path: ["decisionId"],
|
|
101
|
+
});
|
|
102
|
+
}
|
|
103
|
+
});
|
|
104
|
+
//# sourceMappingURL=run-record.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"run-record.js","sourceRoot":"","sources":["../../../src/core/model/run-record.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,kBAAkB,EAAE,MAAM,aAAa,CAAC;AAEjD,MAAM,CAAC,MAAM,cAAc,GAAG,CAAC,CAAC,IAAI,CAAC;IACnC,UAAU;IACV,UAAU;IACV,WAAW;IACX,kBAAkB;IAClB,WAAW;IACX,kBAAkB;IAClB,WAAW;IACX,QAAQ;CACT,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC,CAAC,IAAI,CAAC;IACxC,MAAM;IACN,oBAAoB;IACpB,kBAAkB;IAClB,gBAAgB;IAChB,kBAAkB;IAClB,iBAAiB;IACjB,mBAAmB;CACpB,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,CAAC,IAAI,CAAC;IACpC,SAAS;IACT,WAAW;IACX,UAAU;IACV,UAAU;IACV,aAAa;IACb,QAAQ;CACT,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC;KAC7B,MAAM,CAAC;IACN,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACxB,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;IACrC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC9B,MAAM,EAAE,eAAe;IACvB,KAAK,EAAE,cAAc;IACrB,aAAa,EAAE,mBAAmB,CAAC,OAAO,CAAC,MAAM,CAAC;IAClD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAChC,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACzC,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC/B,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC/B,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IAC3C,QAAQ,EAAE,CAAC,CAAC,MAAM,CAAC;QACjB,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QAC/B,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QAC1B,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;QACjD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;QACvC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE,CAAC,QAAQ,EAAE;QACpD,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE,CAAC,QAAQ,EAAE;QACvD,YAAY,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;QACnD,YAAY,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;KACrC,CAAC;IACF,SAAS,EAAE,CAAC,CAAC,KAAK,CAChB,CAAC,CAAC,MAAM,CAAC;QACP,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACrB,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;KACxB,CAAC,CACH;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,kBAAkB,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;IAC7D,WAAW,EAAE,CAAC;SACX,KAAK,CACJ,CAAC,CAAC,MAAM,CAAC;QACP,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACzB,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE;QACnB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;QAClB,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE;QACpB,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;KACtB,CAAC,CACH;SACA,OAAO,CAAC,EAAE,CAAC;IACd,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IACxC,IAAI,EAAE,CAAC;SACJ,MAAM,CAAC;QACN,iBAAiB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;QAC/C,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;KAC5C,CAAC;SACD,OAAO,CAAC,EAAE,CAAC;IACd,KAAK,EAAE,CAAC;SACL,MAAM,CAAC;QACN,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QAC1B,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;KACpC,CAAC;SACD,QAAQ,EAAE;CACd,CAAC;KACD,WAAW,CAAC,CAAC,MAAM,EAAE,GAAG,EAAE,EAAE;IAC3B,IAAI,MAAM,CAAC,KAAK,KAAK,WAAW,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpD,GAAG,CAAC,QAAQ,CAAC;YACX,IAAI,EAAE,CAAC,CAAC,YAAY,CAAC,MAAM;YAC3B,OAAO,EAAE,qCAAqC;YAC9C,IAAI,EAAE,CAAC,SAAS,CAAC;SAClB,CAAC,CAAC;IACL,CAAC;IAED,IAAI,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;QAC/C,GAAG,CAAC,QAAQ,CAAC;YACX,IAAI,EAAE,CAAC,CAAC,YAAY,CAAC,MAAM;YAC3B,OAAO,EAAE,4CAA4C;YACrD,IAAI,EAAE,CAAC,OAAO,CAAC;SAChB,CAAC,CAAC;IACL,CAAC;IAED,IAAI,MAAM,CAAC,KAAK,KAAK,kBAAkB,IAAI,CAAC,MAAM,CAAC,UAAU,EAAE,CAAC;QAC9D,GAAG,CAAC,QAAQ,CAAC;YACX,IAAI,EAAE,CAAC,CAAC,YAAY,CAAC,MAAM;YAC3B,OAAO,EAAE,iDAAiD;YAC1D,IAAI,EAAE,CAAC,YAAY,CAAC;SACrB,CAAC,CAAC;IACL,CAAC;AACH,CAAC,CAAC,CAAC"}
|