@mingyuxu/check 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/README.md +7 -0
- package/dist/critic/critique.d.ts +31 -0
- package/dist/critic/critique.d.ts.map +1 -0
- package/dist/critic/critique.js +68 -0
- package/dist/critic/critique.js.map +1 -0
- package/dist/critic/observe.d.ts +48 -0
- package/dist/critic/observe.d.ts.map +1 -0
- package/dist/critic/observe.js +62 -0
- package/dist/critic/observe.js.map +1 -0
- package/dist/domain/actions.d.ts +67 -0
- package/dist/domain/actions.d.ts.map +1 -0
- package/dist/domain/actions.js +123 -0
- package/dist/domain/actions.js.map +1 -0
- package/dist/domain/invalidation.d.ts +18 -0
- package/dist/domain/invalidation.d.ts.map +1 -0
- package/dist/domain/invalidation.js +44 -0
- package/dist/domain/invalidation.js.map +1 -0
- package/dist/domain/registers.d.ts +45 -0
- package/dist/domain/registers.d.ts.map +1 -0
- package/dist/domain/registers.js +121 -0
- package/dist/domain/registers.js.map +1 -0
- package/dist/domain/view.d.ts +48 -0
- package/dist/domain/view.d.ts.map +1 -0
- package/dist/domain/view.js +70 -0
- package/dist/domain/view.js.map +1 -0
- package/dist/forge/attribution.d.ts +18 -0
- package/dist/forge/attribution.d.ts.map +1 -0
- package/dist/forge/attribution.js +83 -0
- package/dist/forge/attribution.js.map +1 -0
- package/dist/forge/bakeoff.d.ts +35 -0
- package/dist/forge/bakeoff.d.ts.map +1 -0
- package/dist/forge/bakeoff.js +48 -0
- package/dist/forge/bakeoff.js.map +1 -0
- package/dist/forge/contracts.d.ts +67 -0
- package/dist/forge/contracts.d.ts.map +1 -0
- package/dist/forge/contracts.js +70 -0
- package/dist/forge/contracts.js.map +1 -0
- package/dist/forge/evaluate-scenario.d.ts +17 -0
- package/dist/forge/evaluate-scenario.d.ts.map +1 -0
- package/dist/forge/evaluate-scenario.js +93 -0
- package/dist/forge/evaluate-scenario.js.map +1 -0
- package/dist/forge/evaluate.d.ts +25 -0
- package/dist/forge/evaluate.d.ts.map +1 -0
- package/dist/forge/evaluate.js +81 -0
- package/dist/forge/evaluate.js.map +1 -0
- package/dist/forge/feedback.d.ts +21 -0
- package/dist/forge/feedback.d.ts.map +1 -0
- package/dist/forge/feedback.js +41 -0
- package/dist/forge/feedback.js.map +1 -0
- package/dist/forge/reflect.d.ts +35 -0
- package/dist/forge/reflect.d.ts.map +1 -0
- package/dist/forge/reflect.js +72 -0
- package/dist/forge/reflect.js.map +1 -0
- package/dist/forge/scorer.d.ts +67 -0
- package/dist/forge/scorer.d.ts.map +1 -0
- package/dist/forge/scorer.js +99 -0
- package/dist/forge/scorer.js.map +1 -0
- package/dist/forge/trace.d.ts +25 -0
- package/dist/forge/trace.d.ts.map +1 -0
- package/dist/forge/trace.js +51 -0
- package/dist/forge/trace.js.map +1 -0
- package/dist/index.d.ts +78 -0
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +47 -0
- package/dist/index.js.map +1 -0
- package/dist/knowledge/loader.d.ts +19 -0
- package/dist/knowledge/loader.d.ts.map +1 -0
- package/dist/knowledge/loader.js +76 -0
- package/dist/knowledge/loader.js.map +1 -0
- package/dist/knowledge/retrieval.d.ts +23 -0
- package/dist/knowledge/retrieval.d.ts.map +1 -0
- package/dist/knowledge/retrieval.js +100 -0
- package/dist/knowledge/retrieval.js.map +1 -0
- package/dist/knowledge/skill.d.ts +22 -0
- package/dist/knowledge/skill.d.ts.map +1 -0
- package/dist/knowledge/skill.js +112 -0
- package/dist/knowledge/skill.js.map +1 -0
- package/dist/knowledge/write.d.ts +35 -0
- package/dist/knowledge/write.d.ts.map +1 -0
- package/dist/knowledge/write.js +84 -0
- package/dist/knowledge/write.js.map +1 -0
- package/dist/manage.d.ts +7 -0
- package/dist/manage.d.ts.map +1 -0
- package/dist/manage.js +36 -0
- package/dist/manage.js.map +1 -0
- package/dist/methods/loader.d.ts +18 -0
- package/dist/methods/loader.d.ts.map +1 -0
- package/dist/methods/loader.js +65 -0
- package/dist/methods/loader.js.map +1 -0
- package/dist/nodes/aggregate.d.ts +2 -0
- package/dist/nodes/aggregate.d.ts.map +1 -0
- package/dist/nodes/aggregate.js +101 -0
- package/dist/nodes/aggregate.js.map +1 -0
- package/dist/nodes/base.d.ts +31 -0
- package/dist/nodes/base.d.ts.map +1 -0
- package/dist/nodes/base.js +44 -0
- package/dist/nodes/base.js.map +1 -0
- package/dist/nodes/cognition.d.ts +2 -0
- package/dist/nodes/cognition.d.ts.map +1 -0
- package/dist/nodes/cognition.js +263 -0
- package/dist/nodes/cognition.js.map +1 -0
- package/dist/nodes/confidence.d.ts +9 -0
- package/dist/nodes/confidence.d.ts.map +1 -0
- package/dist/nodes/confidence.js +19 -0
- package/dist/nodes/confidence.js.map +1 -0
- package/dist/nodes/deliver.d.ts +2 -0
- package/dist/nodes/deliver.d.ts.map +1 -0
- package/dist/nodes/deliver.js +37 -0
- package/dist/nodes/deliver.js.map +1 -0
- package/dist/nodes/evidence.d.ts +2 -0
- package/dist/nodes/evidence.d.ts.map +1 -0
- package/dist/nodes/evidence.js +118 -0
- package/dist/nodes/evidence.js.map +1 -0
- package/dist/nodes/ground.d.ts +2 -0
- package/dist/nodes/ground.d.ts.map +1 -0
- package/dist/nodes/ground.js +22 -0
- package/dist/nodes/ground.js.map +1 -0
- package/dist/nodes/parse.d.ts +2 -0
- package/dist/nodes/parse.d.ts.map +1 -0
- package/dist/nodes/parse.js +72 -0
- package/dist/nodes/parse.js.map +1 -0
- package/dist/nodes/scope.d.ts +4 -0
- package/dist/nodes/scope.d.ts.map +1 -0
- package/dist/nodes/scope.js +63 -0
- package/dist/nodes/scope.js.map +1 -0
- package/dist/nodes/shape.d.ts +2 -0
- package/dist/nodes/shape.d.ts.map +1 -0
- package/dist/nodes/shape.js +111 -0
- package/dist/nodes/shape.js.map +1 -0
- package/dist/ontology/checkskill.d.ts +26 -0
- package/dist/ontology/checkskill.d.ts.map +1 -0
- package/dist/ontology/checkskill.js +89 -0
- package/dist/ontology/checkskill.js.map +1 -0
- package/dist/ontology/functions.d.ts +29 -0
- package/dist/ontology/functions.d.ts.map +1 -0
- package/dist/ontology/functions.js +143 -0
- package/dist/ontology/functions.js.map +1 -0
- package/dist/ontology/loader.d.ts +44 -0
- package/dist/ontology/loader.d.ts.map +1 -0
- package/dist/ontology/loader.js +254 -0
- package/dist/ontology/loader.js.map +1 -0
- package/dist/ontology/runtime-graph.d.ts +22 -0
- package/dist/ontology/runtime-graph.d.ts.map +1 -0
- package/dist/ontology/runtime-graph.js +111 -0
- package/dist/ontology/runtime-graph.js.map +1 -0
- package/dist/ontology/store.d.ts +65 -0
- package/dist/ontology/store.d.ts.map +1 -0
- package/dist/ontology/store.js +179 -0
- package/dist/ontology/store.js.map +1 -0
- package/dist/parsers/exec.d.ts +9 -0
- package/dist/parsers/exec.d.ts.map +1 -0
- package/dist/parsers/exec.js +41 -0
- package/dist/parsers/exec.js.map +1 -0
- package/dist/parsers/runner.d.ts +2 -0
- package/dist/parsers/runner.d.ts.map +1 -0
- package/dist/parsers/runner.js +26 -0
- package/dist/parsers/runner.js.map +1 -0
- package/dist/parsers/store.d.ts +32 -0
- package/dist/parsers/store.d.ts.map +1 -0
- package/dist/parsers/store.js +64 -0
- package/dist/parsers/store.js.map +1 -0
- package/dist/parsers/synthesize.d.ts +26 -0
- package/dist/parsers/synthesize.d.ts.map +1 -0
- package/dist/parsers/synthesize.js +63 -0
- package/dist/parsers/synthesize.js.map +1 -0
- package/dist/planner.d.ts +89 -0
- package/dist/planner.d.ts.map +1 -0
- package/dist/planner.js +785 -0
- package/dist/planner.js.map +1 -0
- package/dist/project.d.ts +4 -0
- package/dist/project.d.ts.map +1 -0
- package/dist/project.js +42 -0
- package/dist/project.js.map +1 -0
- package/dist/scenario-builder.d.ts +56 -0
- package/dist/scenario-builder.d.ts.map +1 -0
- package/dist/scenario-builder.js +366 -0
- package/dist/scenario-builder.js.map +1 -0
- package/dist/schemas/index.d.ts +12 -0
- package/dist/schemas/index.d.ts.map +1 -0
- package/dist/schemas/index.js +8 -0
- package/dist/schemas/index.js.map +1 -0
- package/dist/schemas/judgment.d.ts +38 -0
- package/dist/schemas/judgment.d.ts.map +1 -0
- package/dist/schemas/judgment.js +45 -0
- package/dist/schemas/judgment.js.map +1 -0
- package/dist/schemas/payloads.d.ts +53 -0
- package/dist/schemas/payloads.d.ts.map +1 -0
- package/dist/schemas/payloads.js +54 -0
- package/dist/schemas/payloads.js.map +1 -0
- package/dist/schemas/result.d.ts +36 -0
- package/dist/schemas/result.d.ts.map +1 -0
- package/dist/schemas/result.js +26 -0
- package/dist/schemas/result.js.map +1 -0
- package/dist/schemas/spec.d.ts +40 -0
- package/dist/schemas/spec.d.ts.map +1 -0
- package/dist/schemas/spec.js +82 -0
- package/dist/schemas/spec.js.map +1 -0
- package/dist/schemas/target.d.ts +28 -0
- package/dist/schemas/target.d.ts.map +1 -0
- package/dist/schemas/target.js +36 -0
- package/dist/schemas/target.js.map +1 -0
- package/dist/schemas/workflow.d.ts +21 -0
- package/dist/schemas/workflow.d.ts.map +1 -0
- package/dist/schemas/workflow.js +100 -0
- package/dist/schemas/workflow.js.map +1 -0
- package/dist/skills/judge-runner.d.ts +2 -0
- package/dist/skills/judge-runner.d.ts.map +1 -0
- package/dist/skills/judge-runner.js +29 -0
- package/dist/skills/judge-runner.js.map +1 -0
- package/dist/skills/judge-synth.d.ts +31 -0
- package/dist/skills/judge-synth.d.ts.map +1 -0
- package/dist/skills/judge-synth.js +87 -0
- package/dist/skills/judge-synth.js.map +1 -0
- package/dist/subagents/worker.d.ts +5 -0
- package/dist/subagents/worker.d.ts.map +1 -0
- package/dist/subagents/worker.js +58 -0
- package/dist/subagents/worker.js.map +1 -0
- package/dist/workflows/context.d.ts +26 -0
- package/dist/workflows/context.d.ts.map +1 -0
- package/dist/workflows/context.js +23 -0
- package/dist/workflows/context.js.map +1 -0
- package/dist/workflows/executor.d.ts +44 -0
- package/dist/workflows/executor.d.ts.map +1 -0
- package/dist/workflows/executor.js +146 -0
- package/dist/workflows/executor.js.map +1 -0
- package/dist/workflows/merge.d.ts +4 -0
- package/dist/workflows/merge.d.ts.map +1 -0
- package/dist/workflows/merge.js +21 -0
- package/dist/workflows/merge.js.map +1 -0
- package/dist/workflows/run.d.ts +20 -0
- package/dist/workflows/run.d.ts.map +1 -0
- package/dist/workflows/run.js +66 -0
- package/dist/workflows/run.js.map +1 -0
- package/knowledge/methods/prose-rule.method.md +21 -0
- package/knowledge/methods/ratio-limit.method.md +31 -0
- package/knowledge/ontology/verification.yaml +268 -0
- package/package.json +36 -0
|
@@ -0,0 +1,121 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* v4/v5 领域运行时:把盘上的 Target 列表实例化为 Ontology 对象图,再由 Method
|
|
3
|
+
* 合成为 CheckSkill。这里没有流程编号,只有带类型的对象、关系和函数输入/输出。
|
|
4
|
+
*/
|
|
5
|
+
import { createHash } from "node:crypto";
|
|
6
|
+
import { basename } from "node:path";
|
|
7
|
+
import { methodRef, selectMethod } from "../methods/loader.js";
|
|
8
|
+
import { buildCheckSkill } from "../ontology/checkskill.js";
|
|
9
|
+
export function targetDigest(targets) {
|
|
10
|
+
const stable = targets.map((t) => ({
|
|
11
|
+
id: t.target_id, applicability: t.applicability, retrieval: t.retrieval,
|
|
12
|
+
judge: t.judge, outcome: t.outcome_domain, meta: t.meta,
|
|
13
|
+
}));
|
|
14
|
+
return createHash("sha256").update(JSON.stringify(stable)).digest("hex").slice(0, 16);
|
|
15
|
+
}
|
|
16
|
+
/** LLM 未标 rule_kind 时的保守分类。只识别高置信模式,其余交给通用 Method。 */
|
|
17
|
+
export function inferRuleKind(t) {
|
|
18
|
+
const stated = String(t.meta.rule_kind ?? "").trim();
|
|
19
|
+
if (stated)
|
|
20
|
+
return stated;
|
|
21
|
+
const text = [t.meta.name, t.judge.rubric_text, ...(t.retrieval.content_keywords ?? [])].join(" ");
|
|
22
|
+
if (/(%|%|百分之|比例).*(不得超过|不超过|不低于|至少|上限|下限)|(?:不得超过|不超过|不低于|至少).*(%|%|百分之|比例)/.test(text))
|
|
23
|
+
return "ratio_limit";
|
|
24
|
+
if (/(禁止|不得|严禁)/.test(text))
|
|
25
|
+
return "prohibition";
|
|
26
|
+
if (/(资质|资格|许可|牌照)/.test(text))
|
|
27
|
+
return "qualification";
|
|
28
|
+
if (/(披露|公示|告知)/.test(text))
|
|
29
|
+
return "disclosure";
|
|
30
|
+
if (/(阈值|不少于|不高于|至少|至多)/.test(text))
|
|
31
|
+
return "threshold";
|
|
32
|
+
return "procedure";
|
|
33
|
+
}
|
|
34
|
+
export function instantiateDomain(targets, methods, opts) {
|
|
35
|
+
const regulationId = basename(opts.regulationPath || "regulation");
|
|
36
|
+
const objects = [{
|
|
37
|
+
type: "Regulation", id: regulationId,
|
|
38
|
+
data: { regulation_id: regulationId, title: regulationId, source_file: opts.regulationPath },
|
|
39
|
+
}];
|
|
40
|
+
const links = [];
|
|
41
|
+
const selections = [];
|
|
42
|
+
const checkskills = {};
|
|
43
|
+
const unresolved = [];
|
|
44
|
+
for (const t of targets) {
|
|
45
|
+
const ruleKind = inferRuleKind(t);
|
|
46
|
+
const clauseId = String(t.meta.rule_clause_id ?? `${regulationId}:${t.target_id}`);
|
|
47
|
+
const prov = t.meta.provenance ?? {};
|
|
48
|
+
t.meta.rule_kind = ruleKind;
|
|
49
|
+
t.meta.rule_clause_id = clauseId;
|
|
50
|
+
objects.push({
|
|
51
|
+
type: "RuleClause", id: clauseId,
|
|
52
|
+
data: {
|
|
53
|
+
clause_id: clauseId,
|
|
54
|
+
source_ref: { regulation_id: regulationId, article: prov.section ?? prov.chunk_id ?? "unknown" },
|
|
55
|
+
text: prov.source_quote ?? t.judge.rubric_text ?? "",
|
|
56
|
+
rule_kind: ruleKind,
|
|
57
|
+
normative: "mandatory",
|
|
58
|
+
status: "active",
|
|
59
|
+
},
|
|
60
|
+
});
|
|
61
|
+
links.push({ type: "hasClause", from: regulationId, to: clauseId });
|
|
62
|
+
const targetId = t.target_id;
|
|
63
|
+
const rubricId = `${targetId}:rubric`;
|
|
64
|
+
const requirementId = `${targetId}:evidence`;
|
|
65
|
+
objects.push({ type: "VerificationTarget", id: targetId, data: { target_id: targetId, name: t.meta.name ?? targetId, task_type: t.task_type, applicability: t.applicability, outcome_domain: t.outcome_domain, status: "draft" } }, { type: "Rubric", id: rubricId, data: { rubric_id: rubricId, criteria: t.judge.rubric_text ?? "", constraint: t.judge.constraint ?? null } }, { type: "EvidenceRequirement", id: requirementId, data: { req_id: requirementId, description: "判定必须引用文档原文;无充分证据时 needs_review", evidence_kind: "quote" } });
|
|
66
|
+
links.push({ type: "derivesTo", from: clauseId, to: targetId }, { type: "judgedBy", from: targetId, to: rubricId }, { type: "requires", from: targetId, to: requirementId });
|
|
67
|
+
const method = selectMethod(methods, ruleKind);
|
|
68
|
+
if (!method) {
|
|
69
|
+
unresolved.push({ target_id: targetId, rule_kind: ruleKind, reason: "没有匹配的 Method" });
|
|
70
|
+
continue;
|
|
71
|
+
}
|
|
72
|
+
const mref = methodRef(method);
|
|
73
|
+
t.meta.built_by_method = mref;
|
|
74
|
+
t.meta.method_source = method.source;
|
|
75
|
+
if (method.emits_function)
|
|
76
|
+
t.judge.function_ref = method.emits_function;
|
|
77
|
+
applyTypedMethod(t, ruleKind);
|
|
78
|
+
const selection = {
|
|
79
|
+
target_id: targetId, clause_id: clauseId, rule_kind: ruleKind,
|
|
80
|
+
method_ref: mref, method_source: method.source,
|
|
81
|
+
emits_function: method.emits_function ?? "judge.prose_grounded",
|
|
82
|
+
};
|
|
83
|
+
selections.push(selection);
|
|
84
|
+
const skill = buildCheckSkill(t, {
|
|
85
|
+
ontologyVersion: opts.ontologyVersion,
|
|
86
|
+
builtByMethod: mref,
|
|
87
|
+
methodFunction: selection.emits_function,
|
|
88
|
+
lifecycle: opts.lifecycle ?? "draft",
|
|
89
|
+
winnerSpec: opts.winnerSpec,
|
|
90
|
+
frozenAt: opts.frozenAt,
|
|
91
|
+
eval: opts.eval,
|
|
92
|
+
});
|
|
93
|
+
checkskills[targetId] = skill;
|
|
94
|
+
objects.push({ type: "Method", id: mref, data: { method_id: method.method_id, version: method.version, applies_to_rule_kind: method.applies_to_rule_kind, produces: method.produces, emits_function: method.emits_function, source: method.source } }, { type: "CheckSkill", id: String(skill.id), data: skill });
|
|
95
|
+
links.push({ type: "realizedBy", from: targetId, to: String(skill.id) }, { type: "builtByMethod", from: String(skill.id), to: mref });
|
|
96
|
+
}
|
|
97
|
+
return {
|
|
98
|
+
ontology_version: opts.ontologyVersion,
|
|
99
|
+
target_digest: targetDigest(targets),
|
|
100
|
+
objects: dedupe(objects, (o) => `${o.type}:${o.id}`),
|
|
101
|
+
links: dedupe(links, (l) => `${l.type}:${l.from}:${l.to}`),
|
|
102
|
+
method_selection: selections,
|
|
103
|
+
checkskills,
|
|
104
|
+
unresolved,
|
|
105
|
+
};
|
|
106
|
+
}
|
|
107
|
+
function applyTypedMethod(t, ruleKind) {
|
|
108
|
+
if (ruleKind !== "ratio_limit")
|
|
109
|
+
return;
|
|
110
|
+
const text = String(t.judge.rubric_text ?? "");
|
|
111
|
+
const m = /(不得超过|不超过|不高于|至多|不低于|不少于|至少|等于|≤|>=|≥|<=|=)\s*(\d+(?:\.\d+)?)\s*(%|%|百分之)?/.exec(text);
|
|
112
|
+
if (!m)
|
|
113
|
+
return;
|
|
114
|
+
const op = /不低于|不少于|至少|≥|>=/.test(m[1]) ? ">=" : /等于|^=$/.test(m[1]) ? "==" : "<=";
|
|
115
|
+
t.judge.kind = "threshold_compare";
|
|
116
|
+
t.judge.constraint = { op, value: Number(m[2]), unit: m[3] ? "percent" : undefined };
|
|
117
|
+
}
|
|
118
|
+
function dedupe(rows, key) {
|
|
119
|
+
return [...new Map(rows.map((row) => [key(row), row])).values()];
|
|
120
|
+
}
|
|
121
|
+
//# sourceMappingURL=registers.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"registers.js","sourceRoot":"","sources":["../../src/domain/registers.ts"],"names":[],"mappings":"AAAA;;;GAGG;AACH,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,QAAQ,EAAE,MAAM,WAAW,CAAC;AAGrC,OAAO,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,sBAAsB,CAAC;AAC/D,OAAO,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAiC5D,MAAM,UAAU,YAAY,CAAC,OAAiB;IAC5C,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACjC,EAAE,EAAE,CAAC,CAAC,SAAS,EAAE,aAAa,EAAE,CAAC,CAAC,aAAa,EAAE,SAAS,EAAE,CAAC,CAAC,SAAS;QACvE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,OAAO,EAAE,CAAC,CAAC,cAAc,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI;KACxD,CAAC,CAAC,CAAC;IACJ,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AACxF,CAAC;AAED,sDAAsD;AACtD,MAAM,UAAU,aAAa,CAAC,CAAS;IACrC,MAAM,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACrD,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAC1B,MAAM,IAAI,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,gBAA4B,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC/G,IAAI,yEAAyE,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,aAAa,CAAC;IAC/G,IAAI,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,aAAa,CAAC;IAClD,IAAI,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,eAAe,CAAC;IACvD,IAAI,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,YAAY,CAAC;IACjD,IAAI,oBAAoB,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,WAAW,CAAC;IACxD,OAAO,WAAW,CAAC;AACrB,CAAC;AAED,MAAM,UAAU,iBAAiB,CAC/B,OAAiB,EACjB,OAAiB,EACjB,IAAiK;IAEjK,MAAM,YAAY,GAAG,QAAQ,CAAC,IAAI,CAAC,cAAc,IAAI,YAAY,CAAC,CAAC;IACnE,MAAM,OAAO,GAAsB,CAAC;YAClC,IAAI,EAAE,YAAY,EAAE,EAAE,EAAE,YAAY;YACpC,IAAI,EAAE,EAAE,aAAa,EAAE,YAAY,EAAE,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,IAAI,CAAC,cAAc,EAAE;SAC7F,CAAC,CAAC;IACH,MAAM,KAAK,GAAoB,EAAE,CAAC;IAClC,MAAM,UAAU,GAAsB,EAAE,CAAC;IACzC,MAAM,WAAW,GAA4C,EAAE,CAAC;IAChE,MAAM,UAAU,GAA8B,EAAE,CAAC;IAEjD,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,MAAM,QAAQ,GAAG,aAAa,CAAC,CAAC,CAAC,CAAC;QAClC,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,cAAc,IAAI,GAAG,YAAY,IAAI,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC;QACnF,MAAM,IAAI,GAAQ,CAAC,CAAC,IAAI,CAAC,UAAU,IAAI,EAAE,CAAC;QAC1C,CAAC,CAAC,IAAI,CAAC,SAAS,GAAG,QAAQ,CAAC;QAC5B,CAAC,CAAC,IAAI,CAAC,cAAc,GAAG,QAAQ,CAAC;QACjC,OAAO,CAAC,IAAI,CAAC;YACX,IAAI,EAAE,YAAY,EAAE,EAAE,EAAE,QAAQ;YAChC,IAAI,EAAE;gBACJ,SAAS,EAAE,QAAQ;gBACnB,UAAU,EAAE,EAAE,aAAa,EAAE,YAAY,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,QAAQ,IAAI,SAAS,EAAE;gBAChG,IAAI,EAAE,IAAI,CAAC,YAAY,IAAI,CAAC,CAAC,KAAK,CAAC,WAAW,IAAI,EAAE;gBACpD,SAAS,EAAE,QAAQ;gBACnB,SAAS,EAAE,WAAW;gBACtB,MAAM,EAAE,QAAQ;aACjB;SACF,CAAC,CAAC;QACH,KAAK,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,YAAY,EAAE,EAAE,EAAE,QAAQ,EAAE,CAAC,CAAC;QAEpE,MAAM,QAAQ,GAAG,CAAC,CAAC,SAAS,CAAC;QAC7B,MAAM,QAAQ,GAAG,GAAG,QAAQ,SAAS,CAAC;QACtC,MAAM,aAAa,GAAG,GAAG,QAAQ,WAAW,CAAC;QAC7C,OAAO,CAAC,IAAI,CACV,EAAE,IAAI,EAAE,oBAAoB,EAAE,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,EAAE,SAAS,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,IAAI,QAAQ,EAAE,SAAS,EAAE,CAAC,CAAC,SAAS,EAAE,aAAa,EAAE,CAAC,CAAC,aAAa,EAAE,cAAc,EAAE,CAAC,CAAC,cAAc,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,EACrN,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,EAAE,SAAS,EAAE,QAAQ,EAAE,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,WAAW,IAAI,EAAE,EAAE,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,UAAU,IAAI,IAAI,EAAE,EAAE,EAC5I,EAAE,IAAI,EAAE,qBAAqB,EAAE,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,EAAE,MAAM,EAAE,aAAa,EAAE,WAAW,EAAE,gCAAgC,EAAE,aAAa,EAAE,OAAO,EAAE,EAAE,CAC3J,CAAC;QACF,KAAK,CAAC,IAAI,CACR,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,QAAQ,EAAE,EACnD,EAAE,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,QAAQ,EAAE,EAClD,EAAE,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,aAAa,EAAE,CACxD,CAAC;QAEF,MAAM,MAAM,GAAG,YAAY,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;QAC/C,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,UAAU,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,QAAQ,EAAE,SAAS,EAAE,QAAQ,EAAE,MAAM,EAAE,cAAc,EAAE,CAAC,CAAC;YACtF,SAAS;QACX,CAAC;QACD,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;QAC/B,CAAC,CAAC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC;QAC9B,CAAC,CAAC,IAAI,CAAC,aAAa,GAAG,MAAM,CAAC,MAAM,CAAC;QACrC,IAAI,MAAM,CAAC,cAAc;YAAE,CAAC,CAAC,KAAK,CAAC,YAAY,GAAG,MAAM,CAAC,cAAc,CAAC;QACxE,gBAAgB,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC;QAC9B,MAAM,SAAS,GAAoB;YACjC,SAAS,EAAE,QAAQ,EAAE,SAAS,EAAE,QAAQ,EAAE,SAAS,EAAE,QAAQ;YAC7D,UAAU,EAAE,IAAI,EAAE,aAAa,EAAE,MAAM,CAAC,MAAM;YAC9C,cAAc,EAAE,MAAM,CAAC,cAAc,IAAI,sBAAsB;SAChE,CAAC;QACF,UAAU,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QAC3B,MAAM,KAAK,GAAG,eAAe,CAAC,CAAC,EAAE;YAC/B,eAAe,EAAE,IAAI,CAAC,eAAe;YACrC,aAAa,EAAE,IAAI;YACnB,cAAc,EAAE,SAAS,CAAC,cAAc;YACxC,SAAS,EAAE,IAAI,CAAC,SAAS,IAAI,OAAO;YACpC,UAAU,EAAE,IAAI,CAAC,UAAU;YAC3B,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,IAAI,EAAE,IAAI,CAAC,IAAI;SAChB,CAAC,CAAC;QACH,WAAW,CAAC,QAAQ,CAAC,GAAG,KAAK,CAAC;QAC9B,OAAO,CAAC,IAAI,CACV,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,EAAE,SAAS,EAAE,MAAM,CAAC,SAAS,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,oBAAoB,EAAE,MAAM,CAAC,oBAAoB,EAAE,QAAQ,EAAE,MAAM,CAAC,QAAQ,EAAE,cAAc,EAAE,MAAM,CAAC,cAAc,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,EAAE,EACxO,EAAE,IAAI,EAAE,YAAY,EAAE,EAAE,EAAE,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,CAC1D,CAAC;QACF,KAAK,CAAC,IAAI,CACR,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,EAAE,EAC5D,EAAE,IAAI,EAAE,eAAe,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,CAC5D,CAAC;IACJ,CAAC;IAED,OAAO;QACL,gBAAgB,EAAE,IAAI,CAAC,eAAe;QACtC,aAAa,EAAE,YAAY,CAAC,OAAO,CAAC;QACpC,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,EAAE,EAAE,CAAC;QACpD,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,EAAE,EAAE,CAAC;QAC1D,gBAAgB,EAAE,UAAU;QAC5B,WAAW;QACX,UAAU;KACX,CAAC;AACJ,CAAC;AAED,SAAS,gBAAgB,CAAC,CAAS,EAAE,QAAgB;IACnD,IAAI,QAAQ,KAAK,aAAa;QAAE,OAAO;IACvC,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC;IAC/C,MAAM,CAAC,GAAG,4EAA4E,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAClG,IAAI,CAAC,CAAC;QAAE,OAAO;IACf,MAAM,EAAE,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;IACnF,CAAC,CAAC,KAAK,CAAC,IAAI,GAAG,mBAAmB,CAAC;IACnC,CAAC,CAAC,KAAK,CAAC,UAAU,GAAG,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC;AACvF,CAAC;AAED,SAAS,MAAM,CAAI,IAAS,EAAE,GAAuB;IACnD,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC;AACnE,CAAC"}
|
|
@@ -0,0 +1,48 @@
|
|
|
1
|
+
import type { Session } from "@mingyuxu/core";
|
|
2
|
+
import type { VerificationAgent } from "../planner.js";
|
|
3
|
+
import type { ActionDescription } from "./actions.js";
|
|
4
|
+
export interface DomainView {
|
|
5
|
+
scenario: string;
|
|
6
|
+
goal: string;
|
|
7
|
+
ontology: {
|
|
8
|
+
version: string;
|
|
9
|
+
source: string;
|
|
10
|
+
valid: boolean;
|
|
11
|
+
issues: string[];
|
|
12
|
+
layers: Array<{
|
|
13
|
+
id: string;
|
|
14
|
+
name: string;
|
|
15
|
+
description: string;
|
|
16
|
+
object_types: string[];
|
|
17
|
+
instance_count: number;
|
|
18
|
+
}>;
|
|
19
|
+
invariants: Array<{
|
|
20
|
+
id: string;
|
|
21
|
+
rule: string;
|
|
22
|
+
}>;
|
|
23
|
+
};
|
|
24
|
+
state: Record<string, unknown>;
|
|
25
|
+
available_actions: ActionDescription[];
|
|
26
|
+
available_functions: Array<{
|
|
27
|
+
id: string;
|
|
28
|
+
description: string;
|
|
29
|
+
input: unknown;
|
|
30
|
+
output: unknown;
|
|
31
|
+
}>;
|
|
32
|
+
promotion: {
|
|
33
|
+
ready: boolean;
|
|
34
|
+
reason: string;
|
|
35
|
+
summary?: string;
|
|
36
|
+
};
|
|
37
|
+
instances: {
|
|
38
|
+
graph_revision: number;
|
|
39
|
+
graph_valid: boolean;
|
|
40
|
+
objects: unknown[];
|
|
41
|
+
links: unknown[];
|
|
42
|
+
methods: unknown[];
|
|
43
|
+
};
|
|
44
|
+
stale: Record<string, unknown>;
|
|
45
|
+
knowledge_gaps: unknown[];
|
|
46
|
+
}
|
|
47
|
+
export declare function buildDomainView(agent: VerificationAgent, session: Session, focus?: string): DomainView;
|
|
48
|
+
//# sourceMappingURL=view.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"view.d.ts","sourceRoot":"","sources":["../../src/domain/view.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AAC9C,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAMvD,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,cAAc,CAAC;AAGtD,MAAM,WAAW,UAAU;IACzB,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,EAAE;QACR,OAAO,EAAE,MAAM,CAAC;QAChB,MAAM,EAAE,MAAM,CAAC;QACf,KAAK,EAAE,OAAO,CAAC;QACf,MAAM,EAAE,MAAM,EAAE,CAAC;QACjB,MAAM,EAAE,KAAK,CAAC;YAAE,EAAE,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAC;YAAC,WAAW,EAAE,MAAM,CAAC;YAAC,YAAY,EAAE,MAAM,EAAE,CAAC;YAAC,cAAc,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;QACjH,UAAU,EAAE,KAAK,CAAC;YAAE,EAAE,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;KACjD,CAAC;IACF,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC/B,iBAAiB,EAAE,iBAAiB,EAAE,CAAC;IACvC,mBAAmB,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,OAAO,CAAA;KAAE,CAAC,CAAC;IACjG,SAAS,EAAE;QAAE,KAAK,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAChE,SAAS,EAAE;QAAE,cAAc,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,OAAO,CAAC;QAAC,OAAO,EAAE,OAAO,EAAE,CAAC;QAAC,KAAK,EAAE,OAAO,EAAE,CAAC;QAAC,OAAO,EAAE,OAAO,EAAE,CAAA;KAAE,CAAC;IACtH,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC/B,cAAc,EAAE,OAAO,EAAE,CAAC;CAC3B;AAED,wBAAgB,eAAe,CAAC,KAAK,EAAE,iBAAiB,EAAE,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,EAAE,MAAM,GAAG,UAAU,CAkDtG"}
|
|
@@ -0,0 +1,70 @@
|
|
|
1
|
+
/** 为 Agent 投影当前场景最相关的 Ontology + 资产 + 可执行 Action,而不是灌入整库。 */
|
|
2
|
+
import { existsSync, readFileSync } from "node:fs";
|
|
3
|
+
import { join } from "node:path";
|
|
4
|
+
import { loadOntology, ontologyPath, validateOntology } from "../ontology/loader.js";
|
|
5
|
+
import { loadMethods, methodRef } from "../methods/loader.js";
|
|
6
|
+
import { loadStale } from "./invalidation.js";
|
|
7
|
+
import { ScenarioOntologyStore } from "../ontology/store.js";
|
|
8
|
+
import { createVerificationActionRegistry } from "./actions.js";
|
|
9
|
+
import { createFinanceFunctionRegistry } from "../ontology/functions.js";
|
|
10
|
+
export function buildDomainView(agent, session, focus) {
|
|
11
|
+
const ontology = loadOntology(agent.root);
|
|
12
|
+
const issues = validateOntology(ontology);
|
|
13
|
+
const home = session.home(agent.workspace);
|
|
14
|
+
const store = new ScenarioOntologyStore(home, ontology);
|
|
15
|
+
const graph = store.read();
|
|
16
|
+
const objects = graph.objects.length ? graph.objects : readRows(join(home, "build", "domain", "objects.json"), "objects");
|
|
17
|
+
const links = graph.links.length ? graph.links : readRows(join(home, "build", "domain", "links.json"), "links");
|
|
18
|
+
const methodSelections = readRows(join(home, "build", "method_selection.json"), "selections");
|
|
19
|
+
const unresolved = readRows(join(home, "build", "domain", "unresolved.json"), "unresolved");
|
|
20
|
+
const query = focus?.trim().toLowerCase();
|
|
21
|
+
const matches = (value) => !query || JSON.stringify(value).toLowerCase().includes(query);
|
|
22
|
+
const filteredObjects = objects.filter(matches);
|
|
23
|
+
const filteredLinks = links.filter(matches);
|
|
24
|
+
const counts = new Map();
|
|
25
|
+
for (const row of objects)
|
|
26
|
+
counts.set(String(row.type), (counts.get(String(row.type)) ?? 0) + 1);
|
|
27
|
+
return {
|
|
28
|
+
scenario: session.scenarioKey,
|
|
29
|
+
goal: session.goal,
|
|
30
|
+
ontology: {
|
|
31
|
+
version: ontology.version,
|
|
32
|
+
source: ontologyPath(agent.root) ?? "",
|
|
33
|
+
valid: issues.length === 0,
|
|
34
|
+
issues,
|
|
35
|
+
layers: Object.values(ontology.layers).map((layer) => ({
|
|
36
|
+
id: layer.id,
|
|
37
|
+
name: layer.name,
|
|
38
|
+
description: layer.description,
|
|
39
|
+
object_types: layer.objects,
|
|
40
|
+
instance_count: layer.objects.reduce((sum, type) => sum + (counts.get(type) ?? 0), 0),
|
|
41
|
+
})),
|
|
42
|
+
invariants: ontology.coreInvariants,
|
|
43
|
+
},
|
|
44
|
+
state: agent.stateSnapshot(session),
|
|
45
|
+
available_actions: createVerificationActionRegistry(agent, session).describe({ agent, session }, agent.capabilities(session).find((row) => row.recommended)?.name),
|
|
46
|
+
available_functions: createFinanceFunctionRegistry().list(),
|
|
47
|
+
promotion: agent.promotionState(session),
|
|
48
|
+
instances: {
|
|
49
|
+
graph_revision: graph.revision,
|
|
50
|
+
graph_valid: graph.validation.valid,
|
|
51
|
+
objects: filteredObjects,
|
|
52
|
+
links: filteredLinks,
|
|
53
|
+
methods: methodSelections.length ? methodSelections.filter(matches) : loadMethods(agent.root).map((m) => ({ ref: methodRef(m), rule_kind: m.applies_to_rule_kind, produces: m.produces })),
|
|
54
|
+
},
|
|
55
|
+
stale: loadStale(home).entries,
|
|
56
|
+
knowledge_gaps: unresolved,
|
|
57
|
+
};
|
|
58
|
+
}
|
|
59
|
+
function readRows(path, key) {
|
|
60
|
+
if (!existsSync(path))
|
|
61
|
+
return [];
|
|
62
|
+
try {
|
|
63
|
+
const value = JSON.parse(readFileSync(path, "utf-8"));
|
|
64
|
+
return Array.isArray(value?.[key]) ? value[key] : [];
|
|
65
|
+
}
|
|
66
|
+
catch {
|
|
67
|
+
return [];
|
|
68
|
+
}
|
|
69
|
+
}
|
|
70
|
+
//# sourceMappingURL=view.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"view.js","sourceRoot":"","sources":["../../src/domain/view.ts"],"names":[],"mappings":"AAAA,6DAA6D;AAC7D,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAGjC,OAAO,EAAE,YAAY,EAAE,YAAY,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AACrF,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,MAAM,sBAAsB,CAAC;AAC9D,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAC9C,OAAO,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAC7D,OAAO,EAAE,gCAAgC,EAAE,MAAM,cAAc,CAAC;AAEhE,OAAO,EAAE,6BAA6B,EAAE,MAAM,0BAA0B,CAAC;AAsBzE,MAAM,UAAU,eAAe,CAAC,KAAwB,EAAE,OAAgB,EAAE,KAAc;IACxF,MAAM,QAAQ,GAAG,YAAY,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC1C,MAAM,MAAM,GAAG,gBAAgB,CAAC,QAAQ,CAAC,CAAC;IAC1C,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IAC3C,MAAM,KAAK,GAAG,IAAI,qBAAqB,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;IACxD,MAAM,KAAK,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC;IAC3B,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,QAAQ,EAAE,cAAc,CAAC,EAAE,SAAS,CAAC,CAAC;IAC1H,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,QAAQ,EAAE,YAAY,CAAC,EAAE,OAAO,CAAC,CAAC;IAChH,MAAM,gBAAgB,GAAG,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,uBAAuB,CAAC,EAAE,YAAY,CAAC,CAAC;IAC9F,MAAM,UAAU,GAAG,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,QAAQ,EAAE,iBAAiB,CAAC,EAAE,YAAY,CAAC,CAAC;IAC5F,MAAM,KAAK,GAAG,KAAK,EAAE,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC1C,MAAM,OAAO,GAAG,CAAC,KAAc,EAAE,EAAE,CAAC,CAAC,KAAK,IAAI,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;IAClG,MAAM,eAAe,GAAG,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAChD,MAAM,aAAa,GAAG,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAC5C,MAAM,MAAM,GAAG,IAAI,GAAG,EAAkB,CAAC;IACzC,KAAK,MAAM,GAAG,IAAI,OAAyC;QAAE,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAEnI,OAAO;QACL,QAAQ,EAAE,OAAO,CAAC,WAAW;QAC7B,IAAI,EAAE,OAAO,CAAC,IAAI;QAClB,QAAQ,EAAE;YACR,OAAO,EAAE,QAAQ,CAAC,OAAO;YACzB,MAAM,EAAE,YAAY,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,EAAE;YACtC,KAAK,EAAE,MAAM,CAAC,MAAM,KAAK,CAAC;YAC1B,MAAM;YACN,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;gBACrD,EAAE,EAAE,KAAK,CAAC,EAAE;gBACZ,IAAI,EAAE,KAAK,CAAC,IAAI;gBAChB,WAAW,EAAE,KAAK,CAAC,WAAW;gBAC9B,YAAY,EAAE,KAAK,CAAC,OAAO;gBAC3B,cAAc,EAAE,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC;aACtF,CAAC,CAAC;YACH,UAAU,EAAE,QAAQ,CAAC,cAAc;SACpC;QACD,KAAK,EAAE,KAAK,CAAC,aAAa,CAAC,OAAO,CAAC;QACnC,iBAAiB,EAAE,gCAAgC,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC,QAAQ,CAC1E,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE,KAAK,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,WAAW,CAAC,EAAE,IAAI,CACrF;QACD,mBAAmB,EAAE,6BAA6B,EAAE,CAAC,IAAI,EAAE;QAC3D,SAAS,EAAE,KAAK,CAAC,cAAc,CAAC,OAAO,CAAC;QACxC,SAAS,EAAE;YACT,cAAc,EAAE,KAAK,CAAC,QAAQ;YAC9B,WAAW,EAAE,KAAK,CAAC,UAAU,CAAC,KAAK;YACnC,OAAO,EAAE,eAAe;YACxB,KAAK,EAAE,aAAa;YACpB,OAAO,EAAE,gBAAgB,CAAC,MAAM,CAAC,CAAC,CAAC,gBAAgB,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,EAAE,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,oBAAoB,EAAE,QAAQ,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC;SAC3L;QACD,KAAK,EAAE,SAAS,CAAC,IAAI,CAAC,CAAC,OAAO;QAC9B,cAAc,EAAE,UAAU;KAC3B,CAAC;AACJ,CAAC;AAED,SAAS,QAAQ,CAAC,IAAY,EAAE,GAAW;IACzC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO,EAAE,CAAC;IACjC,IAAI,CAAC;QACH,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC;QACtD,OAAO,KAAK,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IACvD,CAAC;IAAC,MAAM,CAAC;QAAC,OAAO,EAAE,CAAC;IAAC,CAAC;AACxB,CAAC"}
|
|
@@ -0,0 +1,18 @@
|
|
|
1
|
+
export interface FailureAttribution {
|
|
2
|
+
case_id: string;
|
|
3
|
+
target_id: string;
|
|
4
|
+
kind: "false_negative" | "false_positive" | "mismatch";
|
|
5
|
+
first_error_stage: "parse" | "scope" | "retrieval" | "extraction" | "judge" | "aggregate";
|
|
6
|
+
confidence: number;
|
|
7
|
+
rationale: string;
|
|
8
|
+
next_probe: string;
|
|
9
|
+
}
|
|
10
|
+
export interface AttemptDiff {
|
|
11
|
+
from?: string;
|
|
12
|
+
to?: string;
|
|
13
|
+
changed_paths: string[];
|
|
14
|
+
}
|
|
15
|
+
export declare function attributeEvaluation(result: Record<string, any>): FailureAttribution[];
|
|
16
|
+
export declare function latestAttemptDiff(home: string): AttemptDiff;
|
|
17
|
+
export declare function diffJson(before: unknown, after: unknown, prefix?: string): string[];
|
|
18
|
+
//# sourceMappingURL=attribution.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"attribution.d.ts","sourceRoot":"","sources":["../../src/forge/attribution.ts"],"names":[],"mappings":"AAIA,MAAM,WAAW,kBAAkB;IACjC,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,IAAI,EAAE,gBAAgB,GAAG,gBAAgB,GAAG,UAAU,CAAC;IACvD,iBAAiB,EAAE,OAAO,GAAG,OAAO,GAAG,WAAW,GAAG,YAAY,GAAG,OAAO,GAAG,WAAW,CAAC;IAC1F,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,WAAW;IAAG,IAAI,CAAC,EAAE,MAAM,CAAC;IAAC,EAAE,CAAC,EAAE,MAAM,CAAC;IAAC,aAAa,EAAE,MAAM,EAAE,CAAC;CAAE;AAErF,wBAAgB,mBAAmB,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,GAAG,kBAAkB,EAAE,CA+BrF;AAED,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,GAAG,WAAW,CAO3D;AAED,wBAAgB,QAAQ,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,SAAK,GAAG,MAAM,EAAE,CAc/E"}
|
|
@@ -0,0 +1,83 @@
|
|
|
1
|
+
/** 失败归因的确定性第一层:先定位最早可疑阶段,再交给 Agent/专家读取完整 trace。 */
|
|
2
|
+
import { existsSync, readFileSync, readdirSync } from "node:fs";
|
|
3
|
+
import { join } from "node:path";
|
|
4
|
+
export function attributeEvaluation(result) {
|
|
5
|
+
const fn = keys(result.false_negatives), fp = keys(result.false_positives);
|
|
6
|
+
return (result.mismatches ?? []).map((row) => {
|
|
7
|
+
const key = `${row.case}|${row.target}`;
|
|
8
|
+
const kind = fn.has(key) ? "false_negative" : fp.has(key) ? "false_positive" : "mismatch";
|
|
9
|
+
const actual = String(row.actual ?? "missing");
|
|
10
|
+
if (["missing", "needs_review", "review_required", "undefined", "null"].includes(actual))
|
|
11
|
+
return {
|
|
12
|
+
case_id: String(row.case), target_id: String(row.target), kind,
|
|
13
|
+
first_error_stage: "retrieval", confidence: 0.75,
|
|
14
|
+
rationale: "执行结果缺失或进入复核,最早应先检查适用性和证据召回。",
|
|
15
|
+
next_probe: "重放该 case,检查 scope、命中区块和 evidence 列表。",
|
|
16
|
+
};
|
|
17
|
+
if (kind === "false_negative")
|
|
18
|
+
return {
|
|
19
|
+
case_id: String(row.case), target_id: String(row.target), kind,
|
|
20
|
+
first_error_stage: "judge", confidence: 0.65,
|
|
21
|
+
rationale: "已产生明确结果但漏报,优先怀疑例外、阈值或判据过松。",
|
|
22
|
+
next_probe: "固定现有证据,单独重放 judge 并比较 rubric/constraint。",
|
|
23
|
+
};
|
|
24
|
+
if (kind === "false_positive")
|
|
25
|
+
return {
|
|
26
|
+
case_id: String(row.case), target_id: String(row.target), kind,
|
|
27
|
+
first_error_stage: "judge", confidence: 0.65,
|
|
28
|
+
rationale: "已产生明确误报,优先怀疑证据边界、例外或判据过严。",
|
|
29
|
+
next_probe: "检查 evidence 是否跨段污染,再单独重放 judge。",
|
|
30
|
+
};
|
|
31
|
+
return {
|
|
32
|
+
case_id: String(row.case), target_id: String(row.target), kind,
|
|
33
|
+
first_error_stage: "aggregate", confidence: 0.45,
|
|
34
|
+
rationale: "非 predicate 结果不一致,现有摘要不足以定位更早节点。",
|
|
35
|
+
next_probe: "读取节点级 trace,对比 extraction、单位和 aggregate 输入。",
|
|
36
|
+
};
|
|
37
|
+
});
|
|
38
|
+
}
|
|
39
|
+
export function latestAttemptDiff(home) {
|
|
40
|
+
const build = join(home, "build");
|
|
41
|
+
if (!existsSync(build))
|
|
42
|
+
return { changed_paths: [] };
|
|
43
|
+
const dirs = readdirSync(build).filter((name) => /^attempt-\d+$/.test(name)).sort();
|
|
44
|
+
if (dirs.length < 2)
|
|
45
|
+
return { to: dirs.at(-1), changed_paths: [] };
|
|
46
|
+
const [from, to] = dirs.slice(-2);
|
|
47
|
+
return { from, to, changed_paths: diffJson(read(join(build, from, "attempt.json")), read(join(build, to, "attempt.json"))) };
|
|
48
|
+
}
|
|
49
|
+
export function diffJson(before, after, prefix = "") {
|
|
50
|
+
if (Object.is(before, after))
|
|
51
|
+
return [];
|
|
52
|
+
if (Array.isArray(before) && Array.isArray(after)) {
|
|
53
|
+
if (before.length !== after.length)
|
|
54
|
+
return [prefix || "$"];
|
|
55
|
+
return before.flatMap((value, index) => diffJson(value, after[index], `${prefix}[${index}]`));
|
|
56
|
+
}
|
|
57
|
+
if (!isRecord(before) || !isRecord(after))
|
|
58
|
+
return [prefix || "$"];
|
|
59
|
+
const out = [];
|
|
60
|
+
for (const key of new Set([...Object.keys(before), ...Object.keys(after)])) {
|
|
61
|
+
const path = prefix ? `${prefix}.${key}` : key;
|
|
62
|
+
if (!(key in before) || !(key in after))
|
|
63
|
+
out.push(path);
|
|
64
|
+
else
|
|
65
|
+
out.push(...diffJson(before[key], after[key], path));
|
|
66
|
+
}
|
|
67
|
+
return out.slice(0, 500);
|
|
68
|
+
}
|
|
69
|
+
function read(path) {
|
|
70
|
+
try {
|
|
71
|
+
return JSON.parse(readFileSync(path, "utf-8"));
|
|
72
|
+
}
|
|
73
|
+
catch {
|
|
74
|
+
return null;
|
|
75
|
+
}
|
|
76
|
+
}
|
|
77
|
+
function keys(rows) {
|
|
78
|
+
return new Set((Array.isArray(rows) ? rows : []).map((row) => `${row.case}|${row.target}`));
|
|
79
|
+
}
|
|
80
|
+
function isRecord(value) {
|
|
81
|
+
return Boolean(value) && typeof value === "object" && !Array.isArray(value);
|
|
82
|
+
}
|
|
83
|
+
//# sourceMappingURL=attribution.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"attribution.js","sourceRoot":"","sources":["../../src/forge/attribution.ts"],"names":[],"mappings":"AAAA,oDAAoD;AACpD,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAChE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAcjC,MAAM,UAAU,mBAAmB,CAAC,MAA2B;IAC7D,MAAM,EAAE,GAAG,IAAI,CAAC,MAAM,CAAC,eAAe,CAAC,EAAE,EAAE,GAAG,IAAI,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC;IAC3E,OAAO,CAAC,MAAM,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,GAAQ,EAAE,EAAE;QAChD,MAAM,GAAG,GAAG,GAAG,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,MAAM,EAAE,CAAC;QACxC,MAAM,IAAI,GAAG,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,UAAU,CAAC;QAC1F,MAAM,MAAM,GAAG,MAAM,CAAC,GAAG,CAAC,MAAM,IAAI,SAAS,CAAC,CAAC;QAC/C,IAAI,CAAC,SAAS,EAAE,cAAc,EAAE,iBAAiB,EAAE,WAAW,EAAE,MAAM,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC;YAAE,OAAO;gBAC/F,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,IAAI;gBAC9D,iBAAiB,EAAE,WAAoB,EAAE,UAAU,EAAE,IAAI;gBACzD,SAAS,EAAE,6BAA6B;gBACxC,UAAU,EAAE,sCAAsC;aACnD,CAAC;QACF,IAAI,IAAI,KAAK,gBAAgB;YAAE,OAAO;gBACpC,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,IAAI;gBAC9D,iBAAiB,EAAE,OAAgB,EAAE,UAAU,EAAE,IAAI;gBACrD,SAAS,EAAE,4BAA4B;gBACvC,UAAU,EAAE,0CAA0C;aACvD,CAAC;QACF,IAAI,IAAI,KAAK,gBAAgB;YAAE,OAAO;gBACpC,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,IAAI;gBAC9D,iBAAiB,EAAE,OAAgB,EAAE,UAAU,EAAE,IAAI;gBACrD,SAAS,EAAE,2BAA2B;gBACtC,UAAU,EAAE,iCAAiC;aAC9C,CAAC;QACF,OAAO;YACL,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,IAAI;YAC9D,iBAAiB,EAAE,WAAoB,EAAE,UAAU,EAAE,IAAI;YACzD,SAAS,EAAE,kCAAkC;YAC7C,UAAU,EAAE,6CAA6C;SAC1D,CAAC;IACJ,CAAC,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,IAAY;IAC5C,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAClC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC;QAAE,OAAO,EAAE,aAAa,EAAE,EAAE,EAAE,CAAC;IACrD,MAAM,IAAI,GAAG,WAAW,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IACpF,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,EAAE,EAAE,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,EAAE,EAAE,CAAC;IACnE,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IAClC,OAAO,EAAE,IAAI,EAAE,EAAE,EAAE,aAAa,EAAE,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,EAAE,cAAc,CAAC,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,EAAE,EAAE,cAAc,CAAC,CAAC,CAAC,EAAE,CAAC;AAC/H,CAAC;AAED,MAAM,UAAU,QAAQ,CAAC,MAAe,EAAE,KAAc,EAAE,MAAM,GAAG,EAAE;IACnE,IAAI,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACxC,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QAClD,IAAI,MAAM,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM;YAAE,OAAO,CAAC,MAAM,IAAI,GAAG,CAAC,CAAC;QAC3D,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,EAAE,GAAG,MAAM,IAAI,KAAK,GAAG,CAAC,CAAC,CAAC;IAChG,CAAC;IACD,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC;QAAE,OAAO,CAAC,MAAM,IAAI,GAAG,CAAE,CAAC;IACnE,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,KAAK,MAAM,GAAG,IAAI,IAAI,GAAG,CAAC,CAAC,GAAG,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC;QAC3E,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,GAAG,MAAM,IAAI,GAAG,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC;QAC/C,IAAI,CAAC,CAAC,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,GAAG,IAAI,KAAK,CAAC;YAAE,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;;YACnD,GAAG,CAAC,IAAI,CAAC,GAAG,QAAQ,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;IAC5D,CAAC;IACD,OAAO,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;AAC3B,CAAC;AAED,SAAS,IAAI,CAAC,IAAY;IACxB,IAAI,CAAC;QAAC,OAAO,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC;IAAC,CAAC;IAAC,MAAM,CAAC;QAAC,OAAO,IAAI,CAAC;IAAC,CAAC;AAChF,CAAC;AAED,SAAS,IAAI,CAAC,IAAS;IACrB,OAAO,IAAI,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;AAC9F,CAAC;AAED,SAAS,QAAQ,CAAC,KAAc;IAC9B,OAAO,OAAO,CAAC,KAAK,CAAC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;AAC9E,CAAC"}
|
|
@@ -0,0 +1,35 @@
|
|
|
1
|
+
import { PipelineSpec } from "../schemas/spec.js";
|
|
2
|
+
import type { Target } from "../schemas/target.js";
|
|
3
|
+
import { type Oracle, type ScoreReport } from "./scorer.js";
|
|
4
|
+
import type { LLMLike } from "../workflows/context.js";
|
|
5
|
+
export interface PackLike {
|
|
6
|
+
profile: Record<string, unknown>;
|
|
7
|
+
refdata: Record<string, unknown>;
|
|
8
|
+
targets: Target[];
|
|
9
|
+
}
|
|
10
|
+
export interface BakeoffRow {
|
|
11
|
+
candidate: string;
|
|
12
|
+
accuracy: number;
|
|
13
|
+
n: number;
|
|
14
|
+
llm_calls: number;
|
|
15
|
+
mismatches: number;
|
|
16
|
+
recall?: number | null;
|
|
17
|
+
precision?: number | null;
|
|
18
|
+
fn?: number;
|
|
19
|
+
}
|
|
20
|
+
export interface BakeoffResult {
|
|
21
|
+
ranking: BakeoffRow[];
|
|
22
|
+
winner: string | null;
|
|
23
|
+
accuracy_bar: number;
|
|
24
|
+
stamp: string;
|
|
25
|
+
winner_mismatches: unknown[];
|
|
26
|
+
winner_report?: ScoreReport | null;
|
|
27
|
+
}
|
|
28
|
+
export declare function runBakeoff(pack: PackLike, candidates: Record<string, PipelineSpec>, oracle: Oracle, opts?: {
|
|
29
|
+
llm?: LLMLike | null;
|
|
30
|
+
accuracyBar?: number;
|
|
31
|
+
winnerMetric?: string;
|
|
32
|
+
historyPath?: string;
|
|
33
|
+
stamp?: string;
|
|
34
|
+
}): Promise<BakeoffResult>;
|
|
35
|
+
//# sourceMappingURL=bakeoff.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"bakeoff.d.ts","sourceRoot":"","sources":["../../src/forge/bakeoff.ts"],"names":[],"mappings":"AAMA,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,sBAAsB,CAAC;AAEnD,OAAO,EAAU,KAAK,MAAM,EAAE,KAAK,WAAW,EAAE,MAAM,aAAa,CAAC;AACpE,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,yBAAyB,CAAC;AAEvD,MAAM,WAAW,QAAQ;IACvB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACjC,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACjC,OAAO,EAAE,MAAM,EAAE,CAAC;CACnB;AAED,MAAM,WAAW,UAAU;IACzB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,CAAC,EAAE,MAAM,CAAC;IACV,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,EAAE,CAAC,EAAE,MAAM,CAAC;CACb;AAED,MAAM,WAAW,aAAa;IAC5B,OAAO,EAAE,UAAU,EAAE,CAAC;IACtB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,iBAAiB,EAAE,OAAO,EAAE,CAAC;IAC7B,aAAa,CAAC,EAAE,WAAW,GAAG,IAAI,CAAC;CACpC;AAED,wBAAsB,UAAU,CAC9B,IAAI,EAAE,QAAQ,EACd,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,EACxC,MAAM,EAAE,MAAM,EACd,IAAI,GAAE;IAAE,GAAG,CAAC,EAAE,OAAO,GAAG,IAAI,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAAC,YAAY,CAAC,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAAC,KAAK,CAAC,EAAE,MAAM,CAAA;CAAO,GACrH,OAAO,CAAC,aAAa,CAAC,CAiCxB"}
|
|
@@ -0,0 +1,48 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Bake-off —— N 个候选生产线在一个 oracle 上打分排名,选胜者。
|
|
3
|
+
* 胜负由沙箱分数定,不由说服力定。从 Python bakeoff.py 忠实移植(异步)。
|
|
4
|
+
*/
|
|
5
|
+
import { appendFileSync, mkdirSync } from "node:fs";
|
|
6
|
+
import { dirname } from "node:path";
|
|
7
|
+
import { Executor } from "../workflows/executor.js";
|
|
8
|
+
import { Scorer } from "./scorer.js";
|
|
9
|
+
export async function runBakeoff(pack, candidates, oracle, opts = {}) {
|
|
10
|
+
const accuracyBar = opts.accuracyBar ?? 0.9;
|
|
11
|
+
const winnerMetric = opts.winnerMetric ?? "accuracy";
|
|
12
|
+
const rows = [];
|
|
13
|
+
const detail = {};
|
|
14
|
+
for (const [name, spec] of Object.entries(candidates)) {
|
|
15
|
+
const ex = new Executor(spec, { profileVocab: pack.profile, refdata: pack.refdata, llm: opts.llm });
|
|
16
|
+
const rep = await new Scorer(ex).score(pack.targets, oracle);
|
|
17
|
+
const row = { candidate: name, accuracy: rep.accuracy, n: rep.n, llm_calls: rep.llm_calls, mismatches: rep.mismatches.length };
|
|
18
|
+
if (rep.recall != null) {
|
|
19
|
+
row.recall = rep.recall;
|
|
20
|
+
row.precision = rep.precision;
|
|
21
|
+
row.fn = rep.confusion.fn;
|
|
22
|
+
}
|
|
23
|
+
detail[name] = rep;
|
|
24
|
+
rows.push(row);
|
|
25
|
+
}
|
|
26
|
+
if (winnerMetric === "recall") {
|
|
27
|
+
rows.sort((a, b) => -((a.recall ?? -1) - (b.recall ?? -1)) || -(a.accuracy - b.accuracy) || a.llm_calls - b.llm_calls || a.candidate.localeCompare(b.candidate));
|
|
28
|
+
}
|
|
29
|
+
else {
|
|
30
|
+
rows.sort((a, b) => -(a.accuracy - b.accuracy) || a.llm_calls - b.llm_calls || a.candidate.localeCompare(b.candidate));
|
|
31
|
+
}
|
|
32
|
+
const passing = rows.filter((r) => r.accuracy >= accuracyBar);
|
|
33
|
+
let winner;
|
|
34
|
+
if (passing.length) {
|
|
35
|
+
winner = [...passing].sort((a, b) => a.llm_calls - b.llm_calls || a.candidate.localeCompare(b.candidate))[0].candidate;
|
|
36
|
+
}
|
|
37
|
+
else {
|
|
38
|
+
winner = rows.length ? rows[0].candidate : null;
|
|
39
|
+
}
|
|
40
|
+
const wrep = winner ? detail[winner] : null;
|
|
41
|
+
const result = { ranking: rows, winner, accuracy_bar: accuracyBar, stamp: opts.stamp ?? "", winner_mismatches: wrep?.mismatches ?? [], winner_report: wrep };
|
|
42
|
+
if (opts.historyPath) {
|
|
43
|
+
mkdirSync(dirname(opts.historyPath), { recursive: true });
|
|
44
|
+
appendFileSync(opts.historyPath, JSON.stringify(result) + "\n", "utf-8");
|
|
45
|
+
}
|
|
46
|
+
return result;
|
|
47
|
+
}
|
|
48
|
+
//# sourceMappingURL=bakeoff.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"bakeoff.js","sourceRoot":"","sources":["../../src/forge/bakeoff.ts"],"names":[],"mappings":"AAAA;;;GAGG;AACH,OAAO,EAAE,cAAc,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpD,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAGpC,OAAO,EAAE,QAAQ,EAAE,MAAM,0BAA0B,CAAC;AACpD,OAAO,EAAE,MAAM,EAAiC,MAAM,aAAa,CAAC;AA6BpE,MAAM,CAAC,KAAK,UAAU,UAAU,CAC9B,IAAc,EACd,UAAwC,EACxC,MAAc,EACd,OAAoH,EAAE;IAEtH,MAAM,WAAW,GAAG,IAAI,CAAC,WAAW,IAAI,GAAG,CAAC;IAC5C,MAAM,YAAY,GAAG,IAAI,CAAC,YAAY,IAAI,UAAU,CAAC;IACrD,MAAM,IAAI,GAAiB,EAAE,CAAC;IAC9B,MAAM,MAAM,GAAgC,EAAE,CAAC;IAC/C,KAAK,MAAM,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,EAAE,CAAC;QACtD,MAAM,EAAE,GAAG,IAAI,QAAQ,CAAC,IAAI,EAAE,EAAE,YAAY,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,GAAG,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;QACpG,MAAM,GAAG,GAAG,MAAM,IAAI,MAAM,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QAC7D,MAAM,GAAG,GAAe,EAAE,SAAS,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,CAAC,QAAQ,EAAE,CAAC,EAAE,GAAG,CAAC,CAAC,EAAE,SAAS,EAAE,GAAG,CAAC,SAAS,EAAE,UAAU,EAAE,GAAG,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC;QAC3I,IAAI,GAAG,CAAC,MAAM,IAAI,IAAI,EAAE,CAAC;YAAC,GAAG,CAAC,MAAM,GAAG,GAAG,CAAC,MAAM,CAAC;YAAC,GAAG,CAAC,SAAS,GAAG,GAAG,CAAC,SAAS,CAAC;YAAC,GAAG,CAAC,EAAE,GAAG,GAAG,CAAC,SAAU,CAAC,EAAE,CAAC;QAAC,CAAC;QAC/G,MAAM,CAAC,IAAI,CAAC,GAAG,GAAG,CAAC;QACnB,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACjB,CAAC;IACD,IAAI,YAAY,KAAK,QAAQ,EAAE,CAAC;QAC9B,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CACjB,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,SAAS,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC;IACjJ,CAAC;SAAM,CAAC;QACN,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,SAAS,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC;IACzH,CAAC;IACD,MAAM,OAAO,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,IAAI,WAAW,CAAC,CAAC;IAC9D,IAAI,MAAqB,CAAC;IAC1B,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;QACnB,MAAM,GAAG,CAAC,GAAG,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,SAAS,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACzH,CAAC;SAAM,CAAC;QACN,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC;IAClD,CAAC;IACD,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAC5C,MAAM,MAAM,GAAkB,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,YAAY,EAAE,WAAW,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,IAAI,EAAE,EAAE,iBAAiB,EAAE,IAAI,EAAE,UAAU,IAAI,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC;IAC5K,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;QACrB,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,WAAW,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC1D,cAAc,CAAC,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;IAC3E,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC"}
|
|
@@ -0,0 +1,67 @@
|
|
|
1
|
+
import type { FailureAttribution } from "./attribution.js";
|
|
2
|
+
export declare const EVALUATION_CONTRACT_VERSION = 1;
|
|
3
|
+
export interface EvaluationSuiteContract {
|
|
4
|
+
schema_version: 1;
|
|
5
|
+
suite_id: string;
|
|
6
|
+
name: string;
|
|
7
|
+
version: string;
|
|
8
|
+
oracle_ref: string;
|
|
9
|
+
case_count: number;
|
|
10
|
+
case_ids: string[];
|
|
11
|
+
source: string;
|
|
12
|
+
}
|
|
13
|
+
export interface EvaluationRunContract {
|
|
14
|
+
schema_version: 1;
|
|
15
|
+
eval_run_id: string;
|
|
16
|
+
scenario_ref: string;
|
|
17
|
+
suite_ref: string;
|
|
18
|
+
pack_ref: string;
|
|
19
|
+
pack_version: string;
|
|
20
|
+
started_at: number;
|
|
21
|
+
completed_at: number;
|
|
22
|
+
status: "completed";
|
|
23
|
+
}
|
|
24
|
+
export interface EvaluationReportContract {
|
|
25
|
+
schema_version: 1;
|
|
26
|
+
report_id: string;
|
|
27
|
+
eval_run_ref: string;
|
|
28
|
+
accuracy: number;
|
|
29
|
+
precision: number | null;
|
|
30
|
+
recall: number | null;
|
|
31
|
+
case_count: number;
|
|
32
|
+
gate_status: "passed" | "regression" | "not_evaluated";
|
|
33
|
+
metrics: Record<string, unknown>;
|
|
34
|
+
}
|
|
35
|
+
export interface FailurePatternContract {
|
|
36
|
+
schema_version: 1;
|
|
37
|
+
failure_id: string;
|
|
38
|
+
eval_run_ref: string;
|
|
39
|
+
kind: "false_negative" | "false_positive" | "mismatch";
|
|
40
|
+
case_id: string;
|
|
41
|
+
target_id: string;
|
|
42
|
+
expected: unknown;
|
|
43
|
+
actual: unknown;
|
|
44
|
+
diagnosis: string;
|
|
45
|
+
first_error_stage?: string;
|
|
46
|
+
attribution_confidence?: number;
|
|
47
|
+
}
|
|
48
|
+
export interface EvaluationContracts {
|
|
49
|
+
suite: EvaluationSuiteContract;
|
|
50
|
+
run: EvaluationRunContract;
|
|
51
|
+
report: EvaluationReportContract;
|
|
52
|
+
failures: FailurePatternContract[];
|
|
53
|
+
}
|
|
54
|
+
export declare function buildEvaluationContracts(input: {
|
|
55
|
+
scenarioKey: string;
|
|
56
|
+
suiteName: string;
|
|
57
|
+
oraclePath: string;
|
|
58
|
+
oracle: unknown;
|
|
59
|
+
caseIds: string[];
|
|
60
|
+
packRef: string;
|
|
61
|
+
packVersion: string;
|
|
62
|
+
attributions?: FailureAttribution[];
|
|
63
|
+
result: Record<string, any>;
|
|
64
|
+
startedAt: number;
|
|
65
|
+
completedAt: number;
|
|
66
|
+
}): EvaluationContracts;
|
|
67
|
+
//# sourceMappingURL=contracts.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"contracts.d.ts","sourceRoot":"","sources":["../../src/forge/contracts.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AAE3D,eAAO,MAAM,2BAA2B,IAAI,CAAC;AAE7C,MAAM,WAAW,uBAAuB;IACtC,cAAc,EAAE,CAAC,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,qBAAqB;IACpC,cAAc,EAAE,CAAC,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,WAAW,CAAC;CACrB;AAED,MAAM,WAAW,wBAAwB;IACvC,cAAc,EAAE,CAAC,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,QAAQ,GAAG,YAAY,GAAG,eAAe,CAAC;IACvD,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAClC;AAED,MAAM,WAAW,sBAAsB;IACrC,cAAc,EAAE,CAAC,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,IAAI,EAAE,gBAAgB,GAAG,gBAAgB,GAAG,UAAU,CAAC;IACvD,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,OAAO,CAAC;IAClB,MAAM,EAAE,OAAO,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,sBAAsB,CAAC,EAAE,MAAM,CAAC;CACjC;AAED,MAAM,WAAW,mBAAmB;IAClC,KAAK,EAAE,uBAAuB,CAAC;IAC/B,GAAG,EAAE,qBAAqB,CAAC;IAC3B,MAAM,EAAE,wBAAwB,CAAC;IACjC,QAAQ,EAAE,sBAAsB,EAAE,CAAC;CACpC;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE;IAC9C,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,OAAO,CAAC;IAChB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,CAAC,EAAE,kBAAkB,EAAE,CAAC;IACpC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC5B,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;CACrB,GAAG,mBAAmB,CAiDtB"}
|
|
@@ -0,0 +1,70 @@
|
|
|
1
|
+
/** 评测系统的版本化磁盘合同:评测集、运行、报告、失败归因彼此分离但可追溯。 */
|
|
2
|
+
import { createHash } from "node:crypto";
|
|
3
|
+
export const EVALUATION_CONTRACT_VERSION = 1;
|
|
4
|
+
export function buildEvaluationContracts(input) {
|
|
5
|
+
const suiteVersion = digest({ oracle: input.oracle, cases: input.caseIds });
|
|
6
|
+
const suiteId = `${input.suiteName}@${suiteVersion}`;
|
|
7
|
+
const evalRunId = `eval-${input.startedAt}-${suiteVersion.slice(0, 8)}`;
|
|
8
|
+
const reportId = `${evalRunId}:report`;
|
|
9
|
+
const falseNegatives = keysOf(input.result.false_negatives);
|
|
10
|
+
const falsePositives = keysOf(input.result.false_positives);
|
|
11
|
+
const failures = (input.result.mismatches ?? []).map((row, index) => {
|
|
12
|
+
const key = `${row.case}|${row.target}`;
|
|
13
|
+
const kind = falseNegatives.has(key) ? "false_negative" : falsePositives.has(key) ? "false_positive" : "mismatch";
|
|
14
|
+
const attribution = input.attributions?.find((item) => item.case_id === String(row.case) && item.target_id === String(row.target));
|
|
15
|
+
return {
|
|
16
|
+
schema_version: 1,
|
|
17
|
+
failure_id: `${evalRunId}:failure:${index + 1}`,
|
|
18
|
+
eval_run_ref: evalRunId,
|
|
19
|
+
kind,
|
|
20
|
+
case_id: String(row.case),
|
|
21
|
+
target_id: String(row.target),
|
|
22
|
+
expected: row.expected,
|
|
23
|
+
actual: row.actual,
|
|
24
|
+
diagnosis: attribution?.rationale ?? diagnosisFor(kind),
|
|
25
|
+
first_error_stage: attribution?.first_error_stage,
|
|
26
|
+
attribution_confidence: attribution?.confidence,
|
|
27
|
+
};
|
|
28
|
+
});
|
|
29
|
+
const regression = input.result.regression;
|
|
30
|
+
const gateStatus = regression == null ? "not_evaluated" : regression.is_regression ? "regression" : "passed";
|
|
31
|
+
return {
|
|
32
|
+
suite: {
|
|
33
|
+
schema_version: 1, suite_id: suiteId, name: input.suiteName, version: suiteVersion,
|
|
34
|
+
oracle_ref: input.oraclePath, case_count: input.caseIds.length, case_ids: input.caseIds, source: input.oraclePath,
|
|
35
|
+
},
|
|
36
|
+
run: {
|
|
37
|
+
schema_version: 1, eval_run_id: evalRunId, scenario_ref: input.scenarioKey,
|
|
38
|
+
suite_ref: suiteId, pack_ref: input.packRef, pack_version: input.packVersion, started_at: input.startedAt,
|
|
39
|
+
completed_at: input.completedAt, status: "completed",
|
|
40
|
+
},
|
|
41
|
+
report: {
|
|
42
|
+
schema_version: 1, report_id: reportId, eval_run_ref: evalRunId,
|
|
43
|
+
accuracy: Number(input.result.accuracy ?? 0), precision: nullableNumber(input.result.precision),
|
|
44
|
+
recall: nullableNumber(input.result.recall), case_count: Number(input.result.n ?? 0), gate_status: gateStatus,
|
|
45
|
+
metrics: {
|
|
46
|
+
accuracy: input.result.accuracy, precision: input.result.precision, recall: input.result.recall,
|
|
47
|
+
confusion: input.result.confusion, per_target: input.result.per_target, per_tier: input.result.per_tier,
|
|
48
|
+
llm_calls: input.result.llm_calls, regression: input.result.regression,
|
|
49
|
+
},
|
|
50
|
+
},
|
|
51
|
+
failures,
|
|
52
|
+
};
|
|
53
|
+
}
|
|
54
|
+
function digest(value) {
|
|
55
|
+
return createHash("sha256").update(JSON.stringify(value)).digest("hex").slice(0, 16);
|
|
56
|
+
}
|
|
57
|
+
function keysOf(rows) {
|
|
58
|
+
return new Set((Array.isArray(rows) ? rows : []).map((row) => `${row.case}|${row.target}`));
|
|
59
|
+
}
|
|
60
|
+
function diagnosisFor(kind) {
|
|
61
|
+
if (kind === "false_negative")
|
|
62
|
+
return "疑似漏报:优先检查召回、适用性与判定阈值";
|
|
63
|
+
if (kind === "false_positive")
|
|
64
|
+
return "疑似误报:优先检查证据边界、例外条件与判定阈值";
|
|
65
|
+
return "结果不一致:需结合 trace 归因到解析、召回、规则映射或判定";
|
|
66
|
+
}
|
|
67
|
+
function nullableNumber(value) {
|
|
68
|
+
return value == null ? null : Number(value);
|
|
69
|
+
}
|
|
70
|
+
//# sourceMappingURL=contracts.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"contracts.js","sourceRoot":"","sources":["../../src/forge/contracts.ts"],"names":[],"mappings":"AAAA,2CAA2C;AAC3C,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAGzC,MAAM,CAAC,MAAM,2BAA2B,GAAG,CAAC,CAAC;AA0D7C,MAAM,UAAU,wBAAwB,CAAC,KAYxC;IACC,MAAM,YAAY,GAAG,MAAM,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;IAC5E,MAAM,OAAO,GAAG,GAAG,KAAK,CAAC,SAAS,IAAI,YAAY,EAAE,CAAC;IACrD,MAAM,SAAS,GAAG,QAAQ,KAAK,CAAC,SAAS,IAAI,YAAY,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;IACxE,MAAM,QAAQ,GAAG,GAAG,SAAS,SAAS,CAAC;IACvC,MAAM,cAAc,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC;IAC5D,MAAM,cAAc,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC;IAC5D,MAAM,QAAQ,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,GAAQ,EAAE,KAAa,EAA0B,EAAE;QACvG,MAAM,GAAG,GAAG,GAAG,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,MAAM,EAAE,CAAC;QACxC,MAAM,IAAI,GAAG,cAAc,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,cAAc,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,UAAU,CAAC;QAClH,MAAM,WAAW,GAAG,KAAK,CAAC,YAAY,EAAE,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,KAAK,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,SAAS,KAAK,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC;QACnI,OAAO;YACL,cAAc,EAAE,CAAC;YACjB,UAAU,EAAE,GAAG,SAAS,YAAY,KAAK,GAAG,CAAC,EAAE;YAC/C,YAAY,EAAE,SAAS;YACvB,IAAI;YACJ,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC;YACzB,SAAS,EAAE,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC;YAC7B,QAAQ,EAAE,GAAG,CAAC,QAAQ;YACtB,MAAM,EAAE,GAAG,CAAC,MAAM;YAClB,SAAS,EAAE,WAAW,EAAE,SAAS,IAAI,YAAY,CAAC,IAAI,CAAC;YACvD,iBAAiB,EAAE,WAAW,EAAE,iBAAiB;YACjD,sBAAsB,EAAE,WAAW,EAAE,UAAU;SAChD,CAAC;IACJ,CAAC,CAAC,CAAC;IACH,MAAM,UAAU,GAAG,KAAK,CAAC,MAAM,CAAC,UAAiB,CAAC;IAClD,MAAM,UAAU,GAAG,UAAU,IAAI,IAAI,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,UAAU,CAAC,aAAa,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ,CAAC;IAC7G,OAAO;QACL,KAAK,EAAE;YACL,cAAc,EAAE,CAAC,EAAE,QAAQ,EAAE,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,SAAS,EAAE,OAAO,EAAE,YAAY;YAClF,UAAU,EAAE,KAAK,CAAC,UAAU,EAAE,UAAU,EAAE,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,CAAC,UAAU;SAClH;QACD,GAAG,EAAE;YACH,cAAc,EAAE,CAAC,EAAE,WAAW,EAAE,SAAS,EAAE,YAAY,EAAE,KAAK,CAAC,WAAW;YAC1E,SAAS,EAAE,OAAO,EAAE,QAAQ,EAAE,KAAK,CAAC,OAAO,EAAE,YAAY,EAAE,KAAK,CAAC,WAAW,EAAE,UAAU,EAAE,KAAK,CAAC,SAAS;YACzG,YAAY,EAAE,KAAK,CAAC,WAAW,EAAE,MAAM,EAAE,WAAW;SACrD;QACD,MAAM,EAAE;YACN,cAAc,EAAE,CAAC,EAAE,SAAS,EAAE,QAAQ,EAAE,YAAY,EAAE,SAAS;YAC/D,QAAQ,EAAE,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,QAAQ,IAAI,CAAC,CAAC,EAAE,SAAS,EAAE,cAAc,CAAC,KAAK,CAAC,MAAM,CAAC,SAAS,CAAC;YAC/F,MAAM,EAAE,cAAc,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,UAAU,EAAE,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,EAAE,WAAW,EAAE,UAAU;YAC7G,OAAO,EAAE;gBACP,QAAQ,EAAE,KAAK,CAAC,MAAM,CAAC,QAAQ,EAAE,SAAS,EAAE,KAAK,CAAC,MAAM,CAAC,SAAS,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,MAAM;gBAC/F,SAAS,EAAE,KAAK,CAAC,MAAM,CAAC,SAAS,EAAE,UAAU,EAAE,KAAK,CAAC,MAAM,CAAC,UAAU,EAAE,QAAQ,EAAE,KAAK,CAAC,MAAM,CAAC,QAAQ;gBACvG,SAAS,EAAE,KAAK,CAAC,MAAM,CAAC,SAAS,EAAE,UAAU,EAAE,KAAK,CAAC,MAAM,CAAC,UAAU;aACvE;SACF;QACD,QAAQ;KACT,CAAC;AACJ,CAAC;AAED,SAAS,MAAM,CAAC,KAAc;IAC5B,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AACvF,CAAC;AAED,SAAS,MAAM,CAAC,IAAS;IACvB,OAAO,IAAI,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;AAC9F,CAAC;AAED,SAAS,YAAY,CAAC,IAAoC;IACxD,IAAI,IAAI,KAAK,gBAAgB;QAAE,OAAO,sBAAsB,CAAC;IAC7D,IAAI,IAAI,KAAK,gBAAgB;QAAE,OAAO,yBAAyB,CAAC;IAChE,OAAO,kCAAkC,CAAC;AAC5C,CAAC;AAED,SAAS,cAAc,CAAC,KAAc;IACpC,OAAO,KAAK,IAAI,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAC9C,CAAC"}
|