@kindgi/api 0.1.3 → 0.1.4-rc.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/dist/agent-binding.d.ts +18 -4
- package/dist/agent-binding.d.ts.map +1 -1
- package/dist/agent-pins.d.ts +48 -0
- package/dist/agent-pins.d.ts.map +1 -0
- package/dist/agent-pins.js +102 -0
- package/dist/agent-pins.js.map +1 -0
- package/dist/app.d.ts +22 -0
- package/dist/app.d.ts.map +1 -1
- package/dist/app.js +20 -2
- package/dist/app.js.map +1 -1
- package/dist/block-binding.d.ts +132 -0
- package/dist/block-binding.d.ts.map +1 -0
- package/dist/block-binding.js +4 -0
- package/dist/block-binding.js.map +1 -0
- package/dist/block-pins.d.ts +22 -0
- package/dist/block-pins.d.ts.map +1 -0
- package/dist/block-pins.js +112 -0
- package/dist/block-pins.js.map +1 -0
- package/dist/cost-binding.d.ts +20 -1
- package/dist/cost-binding.d.ts.map +1 -1
- package/dist/cost-binding.js +4 -0
- package/dist/cost-binding.js.map +1 -1
- package/dist/deploy-versions.d.ts +58 -0
- package/dist/deploy-versions.d.ts.map +1 -0
- package/dist/deploy-versions.js +91 -0
- package/dist/deploy-versions.js.map +1 -0
- package/dist/deployment-binding.d.ts +24 -3
- package/dist/deployment-binding.d.ts.map +1 -1
- package/dist/derive-agent-version.d.ts +69 -0
- package/dist/derive-agent-version.d.ts.map +1 -0
- package/dist/derive-agent-version.js +139 -0
- package/dist/derive-agent-version.js.map +1 -0
- package/dist/errors.d.ts.map +1 -1
- package/dist/errors.js +17 -0
- package/dist/errors.js.map +1 -1
- package/dist/eval-case-binding.d.ts +65 -0
- package/dist/eval-case-binding.d.ts.map +1 -0
- package/dist/eval-case-binding.js +4 -0
- package/dist/eval-case-binding.js.map +1 -0
- package/dist/eval-run-binding.d.ts +30 -0
- package/dist/eval-run-binding.d.ts.map +1 -1
- package/dist/eval-run-dispatcher.d.ts +38 -3
- package/dist/eval-run-dispatcher.d.ts.map +1 -1
- package/dist/eval-run-dispatcher.js +21 -15
- package/dist/eval-run-dispatcher.js.map +1 -1
- package/dist/eval-suite-binding.d.ts +1 -1
- package/dist/eval-suite-binding.d.ts.map +1 -1
- package/dist/eval-suite-binding.js +2 -0
- package/dist/eval-suite-binding.js.map +1 -1
- package/dist/flow-binding.d.ts +10 -4
- package/dist/flow-binding.d.ts.map +1 -1
- package/dist/flow-pins.d.ts +36 -0
- package/dist/flow-pins.d.ts.map +1 -0
- package/dist/flow-pins.js +81 -0
- package/dist/flow-pins.js.map +1 -0
- package/dist/index.d.ts +17 -6
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +8 -2
- package/dist/index.js.map +1 -1
- package/dist/judged-dispatcher.d.ts +134 -0
- package/dist/judged-dispatcher.d.ts.map +1 -0
- package/dist/judged-dispatcher.js +297 -0
- package/dist/judged-dispatcher.js.map +1 -0
- package/dist/judged-items.d.ts +86 -0
- package/dist/judged-items.d.ts.map +1 -0
- package/dist/judged-items.js +184 -0
- package/dist/judged-items.js.map +1 -0
- package/dist/judgment-binding.d.ts +316 -0
- package/dist/judgment-binding.d.ts.map +1 -0
- package/dist/judgment-binding.js +19 -0
- package/dist/judgment-binding.js.map +1 -0
- package/dist/openapi/generate.d.ts.map +1 -1
- package/dist/openapi/generate.js +4 -1
- package/dist/openapi/generate.js.map +1 -1
- package/dist/openapi/operations.d.ts.map +1 -1
- package/dist/openapi/operations.js +461 -7
- package/dist/openapi/operations.js.map +1 -1
- package/dist/openapi/schemas.d.ts +40 -0
- package/dist/openapi/schemas.d.ts.map +1 -1
- package/dist/openapi/schemas.js +843 -7
- package/dist/openapi/schemas.js.map +1 -1
- package/dist/provider-binding.d.ts +12 -7
- package/dist/provider-binding.d.ts.map +1 -1
- package/dist/routes/agents.d.ts +9 -1
- package/dist/routes/agents.d.ts.map +1 -1
- package/dist/routes/agents.js +175 -11
- package/dist/routes/agents.js.map +1 -1
- package/dist/routes/blocks.d.ts +19 -0
- package/dist/routes/blocks.d.ts.map +1 -0
- package/dist/routes/blocks.js +281 -0
- package/dist/routes/blocks.js.map +1 -0
- package/dist/routes/cost.d.ts.map +1 -1
- package/dist/routes/cost.js +47 -2
- package/dist/routes/cost.js.map +1 -1
- package/dist/routes/deployments.d.ts +3 -0
- package/dist/routes/deployments.d.ts.map +1 -1
- package/dist/routes/deployments.js +208 -55
- package/dist/routes/deployments.js.map +1 -1
- package/dist/routes/eval-comparison.d.ts +14 -0
- package/dist/routes/eval-comparison.d.ts.map +1 -0
- package/dist/routes/eval-comparison.js +87 -0
- package/dist/routes/eval-comparison.js.map +1 -0
- package/dist/routes/eval-runs.d.ts.map +1 -1
- package/dist/routes/eval-runs.js +8 -0
- package/dist/routes/eval-runs.js.map +1 -1
- package/dist/routes/flows.d.ts +13 -1
- package/dist/routes/flows.d.ts.map +1 -1
- package/dist/routes/flows.js +44 -3
- package/dist/routes/flows.js.map +1 -1
- package/dist/routes/hierarchy-errors.d.ts +35 -0
- package/dist/routes/hierarchy-errors.d.ts.map +1 -0
- package/dist/routes/hierarchy-errors.js +39 -0
- package/dist/routes/hierarchy-errors.js.map +1 -0
- package/dist/routes/judged-suites.d.ts +20 -0
- package/dist/routes/judged-suites.d.ts.map +1 -0
- package/dist/routes/judged-suites.js +272 -0
- package/dist/routes/judged-suites.js.map +1 -0
- package/dist/routes/judgment-context.d.ts +22 -0
- package/dist/routes/judgment-context.d.ts.map +1 -0
- package/dist/routes/judgment-context.js +88 -0
- package/dist/routes/judgment-context.js.map +1 -0
- package/dist/routes/judgment-flow-context.d.ts +32 -0
- package/dist/routes/judgment-flow-context.d.ts.map +1 -0
- package/dist/routes/judgment-flow-context.js +195 -0
- package/dist/routes/judgment-flow-context.js.map +1 -0
- package/dist/routes/judgments.d.ts +41 -0
- package/dist/routes/judgments.d.ts.map +1 -0
- package/dist/routes/judgments.js +566 -0
- package/dist/routes/judgments.js.map +1 -0
- package/dist/routes/orgs.d.ts +5 -2
- package/dist/routes/orgs.d.ts.map +1 -1
- package/dist/routes/orgs.js +38 -22
- package/dist/routes/orgs.js.map +1 -1
- package/dist/routes/policies.d.ts.map +1 -1
- package/dist/routes/policies.js +12 -1
- package/dist/routes/policies.js.map +1 -1
- package/dist/routes/projects.d.ts +10 -2
- package/dist/routes/projects.d.ts.map +1 -1
- package/dist/routes/projects.js +87 -79
- package/dist/routes/projects.js.map +1 -1
- package/dist/routes/providers.d.ts.map +1 -1
- package/dist/routes/providers.js +6 -1
- package/dist/routes/providers.js.map +1 -1
- package/dist/routes/runs.js +24 -3
- package/dist/routes/runs.js.map +1 -1
- package/dist/routes/teams.d.ts +6 -2
- package/dist/routes/teams.d.ts.map +1 -1
- package/dist/routes/teams.js +77 -73
- package/dist/routes/teams.js.map +1 -1
- package/openapi.json +13545 -10103
- package/package.json +21 -21
- package/src/agent-binding.ts +19 -4
- package/src/agent-pins.ts +147 -0
- package/src/app.ts +71 -2
- package/src/block-binding.ts +137 -0
- package/src/block-pins.ts +148 -0
- package/src/cost-binding.ts +21 -1
- package/src/deploy-versions.ts +157 -0
- package/src/deployment-binding.ts +27 -3
- package/src/derive-agent-version.ts +206 -0
- package/src/errors.ts +17 -0
- package/src/eval-case-binding.ts +71 -0
- package/src/eval-run-binding.ts +33 -0
- package/src/eval-run-dispatcher.ts +57 -16
- package/src/eval-suite-binding.ts +2 -0
- package/src/flow-binding.ts +11 -4
- package/src/flow-pins.ts +113 -0
- package/src/index.ts +85 -2
- package/src/judged-dispatcher.ts +507 -0
- package/src/judged-items.ts +263 -0
- package/src/judgment-binding.ts +349 -0
- package/src/openapi/generate.ts +7 -1
- package/src/openapi/operations.ts +523 -7
- package/src/openapi/schemas.ts +1009 -88
- package/src/provider-binding.ts +12 -7
- package/src/routes/agents.ts +243 -19
- package/src/routes/blocks.ts +362 -0
- package/src/routes/cost.ts +55 -1
- package/src/routes/deployments.ts +266 -56
- package/src/routes/eval-comparison.ts +101 -0
- package/src/routes/eval-runs.ts +11 -0
- package/src/routes/flows.ts +63 -5
- package/src/routes/hierarchy-errors.ts +51 -0
- package/src/routes/judged-suites.ts +363 -0
- package/src/routes/judgment-context.ts +128 -0
- package/src/routes/judgment-flow-context.ts +245 -0
- package/src/routes/judgments.ts +743 -0
- package/src/routes/orgs.ts +44 -27
- package/src/routes/policies.ts +19 -0
- package/src/routes/projects.ts +106 -95
- package/src/routes/providers.ts +5 -0
- package/src/routes/runs.ts +28 -3
- package/src/routes/teams.ts +96 -90
|
@@ -0,0 +1,297 @@
|
|
|
1
|
+
// SPDX-License-Identifier: Apache-2.0
|
|
2
|
+
// Copyright (C) 2026 Kindgi Inc.
|
|
3
|
+
import { itemChanges, matchJudged, outputItems, scoreItems, } from './judged-items.js';
|
|
4
|
+
export const DEFAULT_COMPARISON = {
|
|
5
|
+
baseline: 'recorded',
|
|
6
|
+
reads: 'recorded',
|
|
7
|
+
repetitions: 1,
|
|
8
|
+
k: 10,
|
|
9
|
+
};
|
|
10
|
+
/** Cases read per page. */
|
|
11
|
+
const CASE_PAGE = 100;
|
|
12
|
+
function validateComparison(suite, target, comparison) {
|
|
13
|
+
if (target.version === undefined) {
|
|
14
|
+
return {
|
|
15
|
+
kind: 'err',
|
|
16
|
+
message: 'agentId' in target
|
|
17
|
+
? '`agentRef.version` is required: the candidate is one version of the agent.'
|
|
18
|
+
: '`flowRef.version` is required: the candidate is one version of the flow.',
|
|
19
|
+
};
|
|
20
|
+
}
|
|
21
|
+
if (suite.spec.caseCount === 0) {
|
|
22
|
+
return { kind: 'err', message: 'The test set has no cases.' };
|
|
23
|
+
}
|
|
24
|
+
const c = comparison ?? DEFAULT_COMPARISON;
|
|
25
|
+
if (c.baseline !== 'recorded') {
|
|
26
|
+
return { kind: 'err', message: "Only `baseline: 'recorded'` runs today." };
|
|
27
|
+
}
|
|
28
|
+
return { kind: 'ok' };
|
|
29
|
+
}
|
|
30
|
+
export function createJudgedDispatcher(options) {
|
|
31
|
+
return {
|
|
32
|
+
kind: 'judged',
|
|
33
|
+
validate: validateComparison,
|
|
34
|
+
async dispatch(ctx) {
|
|
35
|
+
const comparison = ctx.comparison ?? DEFAULT_COMPARISON;
|
|
36
|
+
const all = await allCases(options.cases, ctx);
|
|
37
|
+
if (ctx.dryRun) {
|
|
38
|
+
return { result: { dryRun: true, cases: all.length, comparison } };
|
|
39
|
+
}
|
|
40
|
+
const results = [];
|
|
41
|
+
const models = new Map();
|
|
42
|
+
for (const judgedCase of all) {
|
|
43
|
+
if (ctx.abortSignal.aborted)
|
|
44
|
+
break;
|
|
45
|
+
const result = await runCase(ctx, comparison, judgedCase, models);
|
|
46
|
+
results.push(result);
|
|
47
|
+
ctx.onProgress(result);
|
|
48
|
+
}
|
|
49
|
+
const summary = summarize(ctx, comparison, all, results, [...models.values()]);
|
|
50
|
+
return {
|
|
51
|
+
result: { summary, perCase: results },
|
|
52
|
+
...(ctx.abortSignal.aborted && { error: 'cancelled' }),
|
|
53
|
+
};
|
|
54
|
+
},
|
|
55
|
+
};
|
|
56
|
+
}
|
|
57
|
+
async function allCases(store, ctx) {
|
|
58
|
+
const out = [];
|
|
59
|
+
let cursor;
|
|
60
|
+
for (;;) {
|
|
61
|
+
const page = await store.listCases({
|
|
62
|
+
tenantId: ctx.tenantId,
|
|
63
|
+
suiteId: ctx.suite.id,
|
|
64
|
+
version: ctx.suite.version,
|
|
65
|
+
limit: CASE_PAGE,
|
|
66
|
+
...(cursor !== undefined && { cursor }),
|
|
67
|
+
});
|
|
68
|
+
out.push(...page.data);
|
|
69
|
+
if (!page.hasMore || page.nextCursor === undefined)
|
|
70
|
+
return out;
|
|
71
|
+
cursor = page.nextCursor;
|
|
72
|
+
}
|
|
73
|
+
}
|
|
74
|
+
/** One case's repetitions as they come back. */
|
|
75
|
+
class CaseTally {
|
|
76
|
+
judgedCase;
|
|
77
|
+
comparison;
|
|
78
|
+
agentTurn;
|
|
79
|
+
runIds = [];
|
|
80
|
+
candidate = [];
|
|
81
|
+
first;
|
|
82
|
+
error;
|
|
83
|
+
stopped;
|
|
84
|
+
diverged = false;
|
|
85
|
+
refusedWrites = 0;
|
|
86
|
+
approvalSkipped = false;
|
|
87
|
+
constructor(judgedCase, comparison, agentTurn) {
|
|
88
|
+
this.judgedCase = judgedCase;
|
|
89
|
+
this.comparison = comparison;
|
|
90
|
+
this.agentTurn = agentTurn;
|
|
91
|
+
}
|
|
92
|
+
add(outcome) {
|
|
93
|
+
if (outcome.runId !== undefined)
|
|
94
|
+
this.runIds.push(outcome.runId);
|
|
95
|
+
if (outcome.stopped !== undefined) {
|
|
96
|
+
this.stopped ??= outcome.stopped;
|
|
97
|
+
this.refusedWrites = Math.max(this.refusedWrites, refusedCount(outcome));
|
|
98
|
+
this.first ??= { changes: { kept: [], dropped: [], new: [] }, ...replayOf(outcome) };
|
|
99
|
+
return 'stopped';
|
|
100
|
+
}
|
|
101
|
+
if (outcome.error !== undefined) {
|
|
102
|
+
this.error ??= outcome.error;
|
|
103
|
+
return 'error';
|
|
104
|
+
}
|
|
105
|
+
this.ran(outcome);
|
|
106
|
+
return 'ran';
|
|
107
|
+
}
|
|
108
|
+
ran(outcome) {
|
|
109
|
+
const { judgedCase, comparison } = this;
|
|
110
|
+
const matched = matchJudged(outputItems(outcome.output, this.agentTurn), judgedCase.items, judgedCase.output);
|
|
111
|
+
this.candidate.push(scoreItems(matched, comparison.k));
|
|
112
|
+
const tools = outcome.replay?.tools ?? [];
|
|
113
|
+
this.refusedWrites = Math.max(this.refusedWrites, refusedCount(outcome));
|
|
114
|
+
if (comparison.reads === 'recorded' && tools.some((t) => t.source === 'live')) {
|
|
115
|
+
this.diverged = true;
|
|
116
|
+
}
|
|
117
|
+
if (outcome.replay?.approval === 'skipped')
|
|
118
|
+
this.approvalSkipped = true;
|
|
119
|
+
this.first ??= { changes: itemChanges(matched, judgedCase.items), ...replayOf(outcome) };
|
|
120
|
+
}
|
|
121
|
+
result(baseline) {
|
|
122
|
+
const none = this.candidate.length === 0;
|
|
123
|
+
return {
|
|
124
|
+
caseId: this.judgedCase.caseId,
|
|
125
|
+
runIds: this.runIds,
|
|
126
|
+
baseline,
|
|
127
|
+
candidate: this.candidate,
|
|
128
|
+
...(this.first !== undefined && { changes: this.first.changes }),
|
|
129
|
+
...(this.first?.replay !== undefined && { tools: this.first.replay.tools }),
|
|
130
|
+
diverged: this.diverged,
|
|
131
|
+
refusedWrites: this.refusedWrites,
|
|
132
|
+
noContext: this.judgedCase.context === undefined,
|
|
133
|
+
approvalSkipped: this.approvalSkipped,
|
|
134
|
+
// A case none of whose repetitions ran stopped (at a refused write) or errored.
|
|
135
|
+
...(none && this.stopped !== undefined && { stopped: this.stopped }),
|
|
136
|
+
...(none && this.stopped === undefined && this.error !== undefined && { error: this.error }),
|
|
137
|
+
};
|
|
138
|
+
}
|
|
139
|
+
}
|
|
140
|
+
async function runCase(ctx, comparison, judgedCase, models) {
|
|
141
|
+
// An agent turn's items are its answer and typed result; a flow run's, its whole output.
|
|
142
|
+
const agentTurn = judgedCase.subject.kind === 'agent';
|
|
143
|
+
const baseline = scoreItems(matchJudged(outputItems(judgedCase.output, agentTurn), judgedCase.items, judgedCase.output), comparison.k);
|
|
144
|
+
const tally = new CaseTally(judgedCase, comparison, agentTurn);
|
|
145
|
+
for (let rep = 0; rep < comparison.repetitions; rep++) {
|
|
146
|
+
const outcome = await invokeCase(ctx, judgedCase);
|
|
147
|
+
if (tally.add(outcome) === 'ran')
|
|
148
|
+
countModel(models, outcome);
|
|
149
|
+
}
|
|
150
|
+
return tally.result(baseline);
|
|
151
|
+
}
|
|
152
|
+
function refusedCount(outcome) {
|
|
153
|
+
return (outcome.replay?.tools ?? []).filter((t) => t.source === 'refused').length;
|
|
154
|
+
}
|
|
155
|
+
function replayOf(outcome) {
|
|
156
|
+
return outcome.replay !== undefined ? { replay: outcome.replay } : {};
|
|
157
|
+
}
|
|
158
|
+
async function invokeCase(ctx, judgedCase) {
|
|
159
|
+
try {
|
|
160
|
+
return await ctx.subject.invoke({
|
|
161
|
+
tenantId: ctx.tenantId,
|
|
162
|
+
target: ctx.target,
|
|
163
|
+
// An agent turn's input is its user message.
|
|
164
|
+
input: judgedCase.input,
|
|
165
|
+
dryRun: false,
|
|
166
|
+
abortSignal: ctx.abortSignal,
|
|
167
|
+
...(ctx.projectId !== undefined && { projectId: ctx.projectId }),
|
|
168
|
+
replay: {
|
|
169
|
+
of: judgedCase.caseId,
|
|
170
|
+
evalRunId: ctx.runId,
|
|
171
|
+
},
|
|
172
|
+
...(judgedCase.subject.kind === 'agent' && { history: judgedCase.context?.history ?? [] }),
|
|
173
|
+
});
|
|
174
|
+
}
|
|
175
|
+
catch (cause) {
|
|
176
|
+
return { error: cause instanceof Error ? cause.message : String(cause) };
|
|
177
|
+
}
|
|
178
|
+
}
|
|
179
|
+
function countModel(models, outcome) {
|
|
180
|
+
if (outcome.provider === undefined)
|
|
181
|
+
return;
|
|
182
|
+
const key = `${outcome.provider.id}\u0000${outcome.provider.model}`;
|
|
183
|
+
const kept = models.get(key) ?? {
|
|
184
|
+
providerId: outcome.provider.id,
|
|
185
|
+
model: outcome.provider.model,
|
|
186
|
+
runs: 0,
|
|
187
|
+
};
|
|
188
|
+
kept.runs += 1;
|
|
189
|
+
models.set(key, kept);
|
|
190
|
+
}
|
|
191
|
+
const yesShare = (s) => ({ yes: s.yesWeight, total: s.totalWeight });
|
|
192
|
+
const precisionAtK = (s) => ({ yes: s.topK.yesWeight, total: s.topK.totalWeight });
|
|
193
|
+
/** Coverage's ratio is judged items over items; its evidence weight is the judged weight. */
|
|
194
|
+
const coverage = (s) => ({ yes: s.judgedItems, total: s.items });
|
|
195
|
+
/** A metric over cases: Σ numerator / Σ denominator, with its evidence. */
|
|
196
|
+
function pooled(scores, side, weightOf) {
|
|
197
|
+
let yes = 0;
|
|
198
|
+
let total = 0;
|
|
199
|
+
let n = 0;
|
|
200
|
+
let weight = 0;
|
|
201
|
+
for (const s of scores) {
|
|
202
|
+
const { yes: y, total: t } = side(s);
|
|
203
|
+
if (t <= 0)
|
|
204
|
+
continue;
|
|
205
|
+
yes += y;
|
|
206
|
+
total += t;
|
|
207
|
+
n += 1;
|
|
208
|
+
weight += weightOf(s);
|
|
209
|
+
}
|
|
210
|
+
return { value: total > 0 ? yes / total : null, n, weight };
|
|
211
|
+
}
|
|
212
|
+
function metric(results, repetitions, side, weightOf, extra = {}) {
|
|
213
|
+
// Only cases the candidate ran: errored and stopped cases leave both sides.
|
|
214
|
+
const scored = results.filter((r) => r.candidate.length > 0);
|
|
215
|
+
const baseline = pooled(scored.map((r) => r.baseline), side, weightOf);
|
|
216
|
+
// Each repetition is a full pass over the cases: its value, then the spread across them.
|
|
217
|
+
const perRep = Array.from({ length: repetitions }, (_, rep) => pooled(scored.flatMap((r) => (r.candidate[rep] !== undefined ? [r.candidate[rep]] : [])), side, weightOf));
|
|
218
|
+
const values = perRep.map((p) => p.value).filter((v) => v !== null);
|
|
219
|
+
const candidate = values.length > 0 ? values.reduce((a, b) => a + b, 0) / values.length : null;
|
|
220
|
+
const cases = new Set();
|
|
221
|
+
scored.forEach((r, i) => {
|
|
222
|
+
if (r.candidate.some((s) => side(s).total > 0))
|
|
223
|
+
cases.add(i);
|
|
224
|
+
});
|
|
225
|
+
const weight = perRep.reduce((a, p) => a + p.weight, 0) / Math.max(1, perRep.length);
|
|
226
|
+
return {
|
|
227
|
+
baseline: baseline.value,
|
|
228
|
+
candidate,
|
|
229
|
+
delta: baseline.value !== null && candidate !== null ? candidate - baseline.value : null,
|
|
230
|
+
n: cases.size,
|
|
231
|
+
weight,
|
|
232
|
+
baselineN: baseline.n,
|
|
233
|
+
baselineWeight: baseline.weight,
|
|
234
|
+
direction: 'higher',
|
|
235
|
+
...(extra.k !== undefined && { k: extra.k }),
|
|
236
|
+
...(repetitions > 1 &&
|
|
237
|
+
values.length > 1 && { spread: Math.max(...values) - Math.min(...values) }),
|
|
238
|
+
};
|
|
239
|
+
}
|
|
240
|
+
function summarize(ctx, comparison, cases, results, models) {
|
|
241
|
+
const errors = results.filter((r) => r.error !== undefined).length;
|
|
242
|
+
const stopped = results.filter((r) => r.stopped !== undefined).length;
|
|
243
|
+
const versions = new Map();
|
|
244
|
+
for (const c of cases) {
|
|
245
|
+
const key = `${c.subject.kind}:${c.subject.id}@${c.subject.version}`;
|
|
246
|
+
const kept = versions.get(key) ?? {
|
|
247
|
+
id: c.subject.id,
|
|
248
|
+
flow: c.subject.kind === 'flow',
|
|
249
|
+
version: c.subject.version,
|
|
250
|
+
cases: 0,
|
|
251
|
+
};
|
|
252
|
+
kept.cases += 1;
|
|
253
|
+
versions.set(key, kept);
|
|
254
|
+
}
|
|
255
|
+
const specProject = ctx.suite.spec.projectId;
|
|
256
|
+
const projectId = typeof specProject === 'string'
|
|
257
|
+
? specProject
|
|
258
|
+
: ctx.projectId;
|
|
259
|
+
const judgedWeight = (s) => s.totalWeight;
|
|
260
|
+
return {
|
|
261
|
+
evalRunId: ctx.runId,
|
|
262
|
+
status: results.length > 0 && errors === 0
|
|
263
|
+
? 'completed'
|
|
264
|
+
: errors < results.length
|
|
265
|
+
? 'partial'
|
|
266
|
+
: 'failed',
|
|
267
|
+
completedAt: new Date().toISOString(),
|
|
268
|
+
suite: { id: ctx.suite.id, version: ctx.suite.version },
|
|
269
|
+
candidate: candidateOf(ctx.target),
|
|
270
|
+
baseline: {
|
|
271
|
+
kind: 'recorded',
|
|
272
|
+
versions: [...versions.values()].map((v) => v.flow
|
|
273
|
+
? { flowId: v.id, version: v.version, cases: v.cases }
|
|
274
|
+
: { agentId: v.id, version: v.version, cases: v.cases }),
|
|
275
|
+
},
|
|
276
|
+
scope: { ...(projectId !== undefined && { projectId }) },
|
|
277
|
+
cases: results.length,
|
|
278
|
+
diverged: results.filter((r) => r.diverged).length,
|
|
279
|
+
refusedWrites: results.reduce((a, r) => a + r.refusedWrites, 0),
|
|
280
|
+
errors,
|
|
281
|
+
stopped,
|
|
282
|
+
reads: comparison.reads,
|
|
283
|
+
sampling: { models },
|
|
284
|
+
repetitions: comparison.repetitions,
|
|
285
|
+
metrics: {
|
|
286
|
+
weightedYesShare: metric(results, comparison.repetitions, yesShare, judgedWeight),
|
|
287
|
+
judgedCoverage: metric(results, comparison.repetitions, coverage, judgedWeight),
|
|
288
|
+
weightedPrecisionAtK: metric(results, comparison.repetitions, precisionAtK, (s) => s.topK.totalWeight, { k: comparison.k }),
|
|
289
|
+
},
|
|
290
|
+
};
|
|
291
|
+
}
|
|
292
|
+
function candidateOf(target) {
|
|
293
|
+
return 'agentId' in target
|
|
294
|
+
? { kind: 'agent', agentId: target.agentId, version: target.version ?? '' }
|
|
295
|
+
: { kind: 'flow', flowId: target.flowId, version: target.version ?? '' };
|
|
296
|
+
}
|
|
297
|
+
//# sourceMappingURL=judged-dispatcher.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"judged-dispatcher.js","sourceRoot":"","sources":["../src/judged-dispatcher.ts"],"names":[],"mappings":"AAAA,sCAAsC;AACtC,iCAAiC;AA2BjC,OAAO,EAGL,WAAW,EACX,WAAW,EACX,WAAW,EACX,UAAU,GACX,MAAM,mBAAmB,CAAC;AAE3B,MAAM,CAAC,MAAM,kBAAkB,GAAmB;IAChD,QAAQ,EAAE,UAAU;IACpB,KAAK,EAAE,UAAU;IACjB,WAAW,EAAE,CAAC;IACd,CAAC,EAAE,EAAE;CACN,CAAC;AA2GF,2BAA2B;AAC3B,MAAM,SAAS,GAAG,GAAG,CAAC;AAEtB,SAAS,kBAAkB,CACzB,KAA2D,EAC3D,MAA0B,EAC1B,UAAsC;IAEtC,IAAI,MAAM,CAAC,OAAO,KAAK,SAAS,EAAE,CAAC;QACjC,OAAO;YACL,IAAI,EAAE,KAAK;YACX,OAAO,EACL,SAAS,IAAI,MAAM;gBACjB,CAAC,CAAC,4EAA4E;gBAC9E,CAAC,CAAC,0EAA0E;SACjF,CAAC;IACJ,CAAC;IACD,IAAI,KAAK,CAAC,IAAI,CAAC,SAAS,KAAK,CAAC,EAAE,CAAC;QAC/B,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,OAAO,EAAE,4BAA4B,EAAE,CAAC;IAChE,CAAC;IACD,MAAM,CAAC,GAAG,UAAU,IAAI,kBAAkB,CAAC;IAC3C,IAAI,CAAC,CAAC,QAAQ,KAAK,UAAU,EAAE,CAAC;QAC9B,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,OAAO,EAAE,yCAAyC,EAAE,CAAC;IAC7E,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;AACxB,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,OAAgC;IACrE,OAAO;QACL,IAAI,EAAE,QAAQ;QACd,QAAQ,EAAE,kBAAkB;QAC5B,KAAK,CAAC,QAAQ,CAAC,GAAG;YAChB,MAAM,UAAU,GAAG,GAAG,CAAC,UAAU,IAAI,kBAAkB,CAAC;YACxD,MAAM,GAAG,GAAG,MAAM,QAAQ,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;YAC/C,IAAI,GAAG,CAAC,MAAM,EAAE,CAAC;gBACf,OAAO,EAAE,MAAM,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,CAAC,MAAM,EAAE,UAAU,EAAE,EAAE,CAAC;YACrE,CAAC;YACD,MAAM,OAAO,GAAuB,EAAE,CAAC;YACvC,MAAM,MAAM,GAAG,IAAI,GAAG,EAA+D,CAAC;YACtF,KAAK,MAAM,UAAU,IAAI,GAAG,EAAE,CAAC;gBAC7B,IAAI,GAAG,CAAC,WAAW,CAAC,OAAO;oBAAE,MAAM;gBACnC,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,GAAG,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,CAAC,CAAC;gBAClE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACrB,GAAG,CAAC,UAAU,CAAC,MAAsD,CAAC,CAAC;YACzE,CAAC;YACD,MAAM,OAAO,GAAG,SAAS,CAAC,GAAG,EAAE,UAAU,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;YAC/E,OAAO;gBACL,MAAM,EAAE,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE;gBACrC,GAAG,CAAC,GAAG,CAAC,WAAW,CAAC,OAAO,IAAI,EAAE,KAAK,EAAE,WAAW,EAAE,CAAC;aACvD,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,QAAQ,CACrB,KAA2B,EAC3B,GAAoB;IAEpB,MAAM,GAAG,GAAqB,EAAE,CAAC;IACjC,IAAI,MAAkE,CAAC;IACvE,SAAS,CAAC;QACR,MAAM,IAAI,GAAG,MAAM,KAAK,CAAC,SAAS,CAAC;YACjC,QAAQ,EAAE,GAAG,CAAC,QAAQ;YACtB,OAAO,EAAE,GAAG,CAAC,KAAK,CAAC,EAAE;YACrB,OAAO,EAAE,GAAG,CAAC,KAAK,CAAC,OAAO;YAC1B,KAAK,EAAE,SAAS;YAChB,GAAG,CAAC,MAAM,KAAK,SAAS,IAAI,EAAE,MAAM,EAAE,CAAC;SACxC,CAAC,CAAC;QACH,GAAG,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC;QACvB,IAAI,CAAC,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,UAAU,KAAK,SAAS;YAAE,OAAO,GAAG,CAAC;QAC/D,MAAM,GAAG,IAAI,CAAC,UAAU,CAAC;IAC3B,CAAC;AACH,CAAC;AAED,gDAAgD;AAChD,MAAM,SAAS;IAWM;IACA;IACA;IAZV,MAAM,GAAa,EAAE,CAAC;IACtB,SAAS,GAAkB,EAAE,CAAC;IACvC,KAAK,CAAkE;IACvE,KAAK,CAAqB;IAC1B,OAAO,CAA8B;IACrC,QAAQ,GAAG,KAAK,CAAC;IACjB,aAAa,GAAG,CAAC,CAAC;IAClB,eAAe,GAAG,KAAK,CAAC;IAExB,YACmB,UAA0B,EAC1B,UAA0B,EAC1B,SAAkB;QAFlB,eAAU,GAAV,UAAU,CAAgB;QAC1B,eAAU,GAAV,UAAU,CAAgB;QAC1B,cAAS,GAAT,SAAS,CAAS;IAClC,CAAC;IAEJ,GAAG,CAAC,OAAoC;QACtC,IAAI,OAAO,CAAC,KAAK,KAAK,SAAS;YAAE,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,KAA0B,CAAC,CAAC;QACtF,IAAI,OAAO,CAAC,OAAO,KAAK,SAAS,EAAE,CAAC;YAClC,IAAI,CAAC,OAAO,KAAK,OAAO,CAAC,OAAO,CAAC;YACjC,IAAI,CAAC,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,aAAa,EAAE,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC;YACzE,IAAI,CAAC,KAAK,KAAK,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE,GAAG,EAAE,EAAE,EAAE,EAAE,GAAG,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;YACrF,OAAO,SAAS,CAAC;QACnB,CAAC;QACD,IAAI,OAAO,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;YAChC,IAAI,CAAC,KAAK,KAAK,OAAO,CAAC,KAAK,CAAC;YAC7B,OAAO,OAAO,CAAC;QACjB,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAClB,OAAO,KAAK,CAAC;IACf,CAAC;IAEO,GAAG,CAAC,OAAoC;QAC9C,MAAM,EAAE,UAAU,EAAE,UAAU,EAAE,GAAG,IAAI,CAAC;QACxC,MAAM,OAAO,GAAG,WAAW,CACzB,WAAW,CAAC,OAAO,CAAC,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC,EAC3C,UAAU,CAAC,KAAK,EAChB,UAAU,CAAC,MAAM,CAClB,CAAC;QACF,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,UAAU,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;QACvD,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,EAAE,KAAK,IAAI,EAAE,CAAC;QAC1C,IAAI,CAAC,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,aAAa,EAAE,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC;QACzE,IAAI,UAAU,CAAC,KAAK,KAAK,UAAU,IAAI,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,EAAE,CAAC;YAC9E,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC;QACvB,CAAC;QACD,IAAI,OAAO,CAAC,MAAM,EAAE,QAAQ,KAAK,SAAS;YAAE,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC;QACxE,IAAI,CAAC,KAAK,KAAK,EAAE,OAAO,EAAE,WAAW,CAAC,OAAO,EAAE,UAAU,CAAC,KAAK,CAAC,EAAE,GAAG,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;IAC3F,CAAC;IAED,MAAM,CAAC,QAAqB;QAC1B,MAAM,IAAI,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,KAAK,CAAC,CAAC;QACzC,OAAO;YACL,MAAM,EAAE,IAAI,CAAC,UAAU,CAAC,MAAM;YAC9B,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,QAAQ;YACR,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,GAAG,CAAC,IAAI,CAAC,KAAK,KAAK,SAAS,IAAI,EAAE,OAAO,EAAE,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;YAChE,GAAG,CAAC,IAAI,CAAC,KAAK,EAAE,MAAM,KAAK,SAAS,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;YAC3E,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,aAAa,EAAE,IAAI,CAAC,aAAa;YACjC,SAAS,EAAE,IAAI,CAAC,UAAU,CAAC,OAAO,KAAK,SAAS;YAChD,eAAe,EAAE,IAAI,CAAC,eAAe;YACrC,gFAAgF;YAChF,GAAG,CAAC,IAAI,IAAI,IAAI,CAAC,OAAO,KAAK,SAAS,IAAI,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC;YACpE,GAAG,CAAC,IAAI,IAAI,IAAI,CAAC,OAAO,KAAK,SAAS,IAAI,IAAI,CAAC,KAAK,KAAK,SAAS,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC;SAC7F,CAAC;IACJ,CAAC;CACF;AAED,KAAK,UAAU,OAAO,CACpB,GAAoB,EACpB,UAA0B,EAC1B,UAA0B,EAC1B,MAAwE;IAExE,yFAAyF;IACzF,MAAM,SAAS,GAAG,UAAU,CAAC,OAAO,CAAC,IAAI,KAAK,OAAO,CAAC;IACtD,MAAM,QAAQ,GAAG,UAAU,CACzB,WAAW,CAAC,WAAW,CAAC,UAAU,CAAC,MAAM,EAAE,SAAS,CAAC,EAAE,UAAU,CAAC,KAAK,EAAE,UAAU,CAAC,MAAM,CAAC,EAC3F,UAAU,CAAC,CAAC,CACb,CAAC;IACF,MAAM,KAAK,GAAG,IAAI,SAAS,CAAC,UAAU,EAAE,UAAU,EAAE,SAAS,CAAC,CAAC;IAC/D,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,GAAG,GAAG,UAAU,CAAC,WAAW,EAAE,GAAG,EAAE,EAAE,CAAC;QACtD,MAAM,OAAO,GAAG,MAAM,UAAU,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;QAClD,IAAI,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,KAAK,KAAK;YAAE,UAAU,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;AAChC,CAAC;AAED,SAAS,YAAY,CAAC,OAAoC;IACxD,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,KAAK,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,MAAM,CAAC;AACpF,CAAC;AAED,SAAS,QAAQ,CAAC,OAAoC;IACpD,OAAO,OAAO,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;AACxE,CAAC;AAED,KAAK,UAAU,UAAU,CACvB,GAAoB,EACpB,UAA0B;IAE1B,IAAI,CAAC;QACH,OAAO,MAAM,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC;YAC9B,QAAQ,EAAE,GAAG,CAAC,QAAQ;YACtB,MAAM,EAAE,GAAG,CAAC,MAAM;YAClB,6CAA6C;YAC7C,KAAK,EAAE,UAAU,CAAC,KAAK;YACvB,MAAM,EAAE,KAAK;YACb,WAAW,EAAE,GAAG,CAAC,WAAW;YAC5B,GAAG,CAAC,GAAG,CAAC,SAAS,KAAK,SAAS,IAAI,EAAE,SAAS,EAAE,GAAG,CAAC,SAAS,EAAE,CAAC;YAChE,MAAM,EAAE;gBACN,EAAE,EAAE,UAAU,CAAC,MAA0B;gBACzC,SAAS,EAAE,GAAG,CAAC,KAA0B;aAC1C;YACD,GAAG,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,KAAK,OAAO,IAAI,EAAE,OAAO,EAAE,UAAU,CAAC,OAAO,EAAE,OAAO,IAAI,EAAE,EAAE,CAAC;SAC3F,CAAC,CAAC;IACL,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO,EAAE,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC;IAC3E,CAAC;AACH,CAAC;AAED,SAAS,UAAU,CACjB,MAAwE,EACxE,OAAoC;IAEpC,IAAI,OAAO,CAAC,QAAQ,KAAK,SAAS;QAAE,OAAO;IAC3C,MAAM,GAAG,GAAG,GAAG,OAAO,CAAC,QAAQ,CAAC,EAAE,SAAS,OAAO,CAAC,QAAQ,CAAC,KAAK,EAAE,CAAC;IACpE,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI;QAC9B,UAAU,EAAE,OAAO,CAAC,QAAQ,CAAC,EAAE;QAC/B,KAAK,EAAE,OAAO,CAAC,QAAQ,CAAC,KAAK;QAC7B,IAAI,EAAE,CAAC;KACR,CAAC;IACF,IAAI,CAAC,IAAI,IAAI,CAAC,CAAC;IACf,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;AACxB,CAAC;AAID,MAAM,QAAQ,GAAS,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,SAAS,EAAE,KAAK,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;AAC3E,MAAM,YAAY,GAAS,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,EAAE,CAAC,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC;AACzF,6FAA6F;AAC7F,MAAM,QAAQ,GAAS,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,WAAW,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC;AAEvE,2EAA2E;AAC3E,SAAS,MAAM,CACb,MAA8B,EAC9B,IAAU,EACV,QAAoC;IAEpC,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,KAAK,MAAM,CAAC,IAAI,MAAM,EAAE,CAAC;QACvB,MAAM,EAAE,GAAG,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QACrC,IAAI,CAAC,IAAI,CAAC;YAAE,SAAS;QACrB,GAAG,IAAI,CAAC,CAAC;QACT,KAAK,IAAI,CAAC,CAAC;QACX,CAAC,IAAI,CAAC,CAAC;QACP,MAAM,IAAI,QAAQ,CAAC,CAAC,CAAC,CAAC;IACxB,CAAC;IACD,OAAO,EAAE,KAAK,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC;AAC9D,CAAC;AAED,SAAS,MAAM,CACb,OAAoC,EACpC,WAAmB,EACnB,IAAU,EACV,QAAoC,EACpC,QAAiC,EAAE;IAEnC,4EAA4E;IAC5E,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAC7D,MAAM,QAAQ,GAAG,MAAM,CACrB,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,EAC7B,IAAI,EACJ,QAAQ,CACT,CAAC;IACF,yFAAyF;IACzF,MAAM,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,EAAE,EAAE,CAC5D,MAAM,CACJ,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EACjF,IAAI,EACJ,QAAQ,CACT,CACF,CAAC;IACF,MAAM,MAAM,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC;IACjF,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC;IAC/F,MAAM,KAAK,GAAG,IAAI,GAAG,EAAU,CAAC;IAChC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACtB,IAAI,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC;YAAE,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;IAC/D,CAAC,CAAC,CAAC;IACH,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC;IACrF,OAAO;QACL,QAAQ,EAAE,QAAQ,CAAC,KAAK;QACxB,SAAS;QACT,KAAK,EAAE,QAAQ,CAAC,KAAK,KAAK,IAAI,IAAI,SAAS,KAAK,IAAI,CAAC,CAAC,CAAC,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI;QACxF,CAAC,EAAE,KAAK,CAAC,IAAI;QACb,MAAM;QACN,SAAS,EAAE,QAAQ,CAAC,CAAC;QACrB,cAAc,EAAE,QAAQ,CAAC,MAAM;QAC/B,SAAS,EAAE,QAAQ;QACnB,GAAG,CAAC,KAAK,CAAC,CAAC,KAAK,SAAS,IAAI,EAAE,CAAC,EAAE,KAAK,CAAC,CAAC,EAAE,CAAC;QAC5C,GAAG,CAAC,WAAW,GAAG,CAAC;YACjB,MAAM,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,EAAE,CAAC;KAC9E,CAAC;AACJ,CAAC;AAED,SAAS,SAAS,CAChB,GAAoB,EACpB,UAA0B,EAC1B,KAAgC,EAChC,OAAoC,EACpC,MAAsE;IAEtE,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,MAAM,CAAC;IACnE,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,MAAM,CAAC;IACtE,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAyE,CAAC;IAClG,KAAK,MAAM,CAAC,IAAI,KAAK,EAAE,CAAC;QACtB,MAAM,GAAG,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,CAAC,OAAO,CAAC,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC;QACrE,MAAM,IAAI,GAAG,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI;YAChC,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,EAAE;YAChB,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,IAAI,KAAK,MAAM;YAC/B,OAAO,EAAE,CAAC,CAAC,OAAO,CAAC,OAAO;YAC1B,KAAK,EAAE,CAAC;SACT,CAAC;QACF,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;QAChB,QAAQ,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;IAC1B,CAAC;IACD,MAAM,WAAW,GAAG,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC;IAC7C,MAAM,SAAS,GACb,OAAO,WAAW,KAAK,QAAQ;QAC7B,CAAC,CAAC,WAAW;QACb,CAAC,CAAE,GAAG,CAAC,SAA2C,CAAC;IACvD,MAAM,YAAY,GAAG,CAAC,CAAc,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC;IACvD,OAAO;QACL,SAAS,EAAE,GAAG,CAAC,KAA0B;QACzC,MAAM,EACJ,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,MAAM,KAAK,CAAC;YAChC,CAAC,CAAC,WAAW;YACb,CAAC,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM;gBACvB,CAAC,CAAC,SAAS;gBACX,CAAC,CAAC,QAAQ;QAChB,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACrC,KAAK,EAAE,EAAE,EAAE,EAAE,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,OAAO,EAAE,GAAG,CAAC,KAAK,CAAC,OAAO,EAAE;QACvD,SAAS,EAAE,WAAW,CAAC,GAAG,CAAC,MAAM,CAAC;QAClC,QAAQ,EAAE;YACR,IAAI,EAAE,UAAU;YAChB,QAAQ,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAClC,CAAC,CAAC,EAAmB,EAAE,CACrB,CAAC,CAAC,IAAI;gBACJ,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,EAAE,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE;gBACtD,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,EAAE,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,CAC5D;SACF;QACD,KAAK,EAAE,EAAE,GAAG,CAAC,SAAS,KAAK,SAAS,IAAI,EAAE,SAAS,EAAE,CAAC,EAAE;QACxD,KAAK,EAAE,OAAO,CAAC,MAAM;QACrB,QAAQ,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,MAAM;QAClD,aAAa,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,aAAa,EAAE,CAAC,CAAC;QAC/D,MAAM;QACN,OAAO;QACP,KAAK,EAAE,UAAU,CAAC,KAAK;QACvB,QAAQ,EAAE,EAAE,MAAM,EAAE;QACpB,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,OAAO,EAAE;YACP,gBAAgB,EAAE,MAAM,CAAC,OAAO,EAAE,UAAU,CAAC,WAAW,EAAE,QAAQ,EAAE,YAAY,CAAC;YACjF,cAAc,EAAE,MAAM,CAAC,OAAO,EAAE,UAAU,CAAC,WAAW,EAAE,QAAQ,EAAE,YAAY,CAAC;YAC/E,oBAAoB,EAAE,MAAM,CAC1B,OAAO,EACP,UAAU,CAAC,WAAW,EACtB,YAAY,EACZ,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,EACzB,EAAE,CAAC,EAAE,UAAU,CAAC,CAAC,EAAE,CACpB;SACF;KACF,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,MAA0B;IAC7C,OAAO,SAAS,IAAI,MAAM;QACxB,CAAC,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,OAA4B,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,IAAI,EAAE,EAAE;QAChG,CAAC,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,MAA2B,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,IAAI,EAAE,EAAE,CAAC;AAClG,CAAC"}
|
|
@@ -0,0 +1,86 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* The items of a run's output that people judge, and how a new output is
|
|
3
|
+
* scored against the judgments of an old one. Pure, with no dependencies,
|
|
4
|
+
* so a console can derive items the same way.
|
|
5
|
+
*
|
|
6
|
+
* Items:
|
|
7
|
+
* - an agent turn's answer (its last agent message) is `answer`;
|
|
8
|
+
* - a flow run's whole output is `output`;
|
|
9
|
+
* - each element of a list in the output (an agent's typed result sits
|
|
10
|
+
* under `output`) is an item, keyed by its own `id` or `key` when it
|
|
11
|
+
* has one, else by where it sits (a JSON Pointer), with its rank.
|
|
12
|
+
*
|
|
13
|
+
* A judgment carries over to a new output's item when it's the same item:
|
|
14
|
+
* an item keyed by its own id is the same item when the key matches; any
|
|
15
|
+
* other (the answer, an element keyed by its place) only when its content
|
|
16
|
+
* is the same too. So a changed answer is a new item, with no judgments.
|
|
17
|
+
*/
|
|
18
|
+
export interface OutputItem {
|
|
19
|
+
/** The key judgments use for this item. */
|
|
20
|
+
readonly key: string;
|
|
21
|
+
/** Where it sits in the output (JSON Pointer). */
|
|
22
|
+
readonly pointer: string;
|
|
23
|
+
/** Its place in its list (0 = first). */
|
|
24
|
+
readonly rank?: number;
|
|
25
|
+
/** `true` when the key is the item's own id, so it identifies the item by itself. */
|
|
26
|
+
readonly ownKey: boolean;
|
|
27
|
+
readonly value: unknown;
|
|
28
|
+
}
|
|
29
|
+
/** The items of a run's output (`agentTurn`: an agent turn's result). */
|
|
30
|
+
export declare function outputItems(output: unknown, agentTurn: boolean): readonly OutputItem[];
|
|
31
|
+
/** The value at a JSON Pointer in `doc`, or `undefined` when there's none. */
|
|
32
|
+
export declare function valueAt(doc: unknown, pointer: string): unknown;
|
|
33
|
+
/** What people said about one judged item (as a judged test-set case keeps it). */
|
|
34
|
+
export interface ItemJudgments {
|
|
35
|
+
readonly key: string;
|
|
36
|
+
readonly pointer?: string;
|
|
37
|
+
readonly rank?: number;
|
|
38
|
+
readonly yesWeight: number;
|
|
39
|
+
readonly totalWeight: number;
|
|
40
|
+
}
|
|
41
|
+
/** A new output's item and the judged item it is the same as, if any. */
|
|
42
|
+
export interface MatchedItem {
|
|
43
|
+
readonly item: OutputItem;
|
|
44
|
+
readonly judged?: ItemJudgments;
|
|
45
|
+
}
|
|
46
|
+
/**
|
|
47
|
+
* Match a new output's items to the items judged in a past output
|
|
48
|
+
* (`judgedOutput`, where the judged items' pointers point).
|
|
49
|
+
*/
|
|
50
|
+
export declare function matchJudged(items: readonly OutputItem[], judged: readonly ItemJudgments[], judgedOutput: unknown): readonly MatchedItem[];
|
|
51
|
+
/** An output scored against judgments: the sums the metrics are made of. */
|
|
52
|
+
export interface OutputScore {
|
|
53
|
+
/** Σ yesWeight and Σ totalWeight over the output's judged items. */
|
|
54
|
+
readonly yesWeight: number;
|
|
55
|
+
readonly totalWeight: number;
|
|
56
|
+
readonly items: number;
|
|
57
|
+
readonly judgedItems: number;
|
|
58
|
+
/** The same sums over the judged items among the first `k` ranked items. */
|
|
59
|
+
readonly topK: {
|
|
60
|
+
readonly yesWeight: number;
|
|
61
|
+
readonly totalWeight: number;
|
|
62
|
+
};
|
|
63
|
+
}
|
|
64
|
+
export declare function scoreItems(matched: readonly MatchedItem[], k: number): OutputScore;
|
|
65
|
+
/** How a new output's items moved against the judged ones. */
|
|
66
|
+
export interface ItemChanges {
|
|
67
|
+
/** Judged items the new output still has, with their rank before and now. */
|
|
68
|
+
readonly kept: readonly {
|
|
69
|
+
readonly key: string;
|
|
70
|
+
readonly rankBefore?: number;
|
|
71
|
+
readonly rank?: number;
|
|
72
|
+
}[];
|
|
73
|
+
/** Judged items the new output no longer has. */
|
|
74
|
+
readonly dropped: readonly {
|
|
75
|
+
readonly key: string;
|
|
76
|
+
readonly rankBefore?: number;
|
|
77
|
+
}[];
|
|
78
|
+
/** The new output's items no one judged (for experts to judge). */
|
|
79
|
+
readonly new: readonly {
|
|
80
|
+
readonly key: string;
|
|
81
|
+
readonly pointer: string;
|
|
82
|
+
readonly rank?: number;
|
|
83
|
+
}[];
|
|
84
|
+
}
|
|
85
|
+
export declare function itemChanges(matched: readonly MatchedItem[], judged: readonly ItemJudgments[]): ItemChanges;
|
|
86
|
+
//# sourceMappingURL=judged-items.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"judged-items.d.ts","sourceRoot":"","sources":["../src/judged-items.ts"],"names":[],"mappings":"AAGA;;;;;;;;;;;;;;;;GAgBG;AAEH,MAAM,WAAW,UAAU;IACzB,2CAA2C;IAC3C,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;IACrB,kDAAkD;IAClD,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,yCAAyC;IACzC,QAAQ,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC;IACvB,qFAAqF;IACrF,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC;IACzB,QAAQ,CAAC,KAAK,EAAE,OAAO,CAAC;CACzB;AAkED,yEAAyE;AACzE,wBAAgB,WAAW,CAAC,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,OAAO,GAAG,SAAS,UAAU,EAAE,CAetF;AAED,8EAA8E;AAC9E,wBAAgB,OAAO,CAAC,GAAG,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAe9D;AAgBD,mFAAmF;AACnF,MAAM,WAAW,aAAa;IAC5B,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;CAC9B;AAED,yEAAyE;AACzE,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,IAAI,EAAE,UAAU,CAAC;IAC1B,QAAQ,CAAC,MAAM,CAAC,EAAE,aAAa,CAAC;CACjC;AAED;;;GAGG;AACH,wBAAgB,WAAW,CACzB,KAAK,EAAE,SAAS,UAAU,EAAE,EAC5B,MAAM,EAAE,SAAS,aAAa,EAAE,EAChC,YAAY,EAAE,OAAO,GACpB,SAAS,WAAW,EAAE,CAWxB;AAED,4EAA4E;AAC5E,MAAM,WAAW,WAAW;IAC1B,oEAAoE;IACpE,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,4EAA4E;IAC5E,QAAQ,CAAC,IAAI,EAAE;QAAE,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC;CAC7E;AAED,wBAAgB,UAAU,CAAC,OAAO,EAAE,SAAS,WAAW,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,WAAW,CAqBlF;AAED,8DAA8D;AAC9D,MAAM,WAAW,WAAW;IAC1B,6EAA6E;IAC7E,QAAQ,CAAC,IAAI,EAAE,SAAS;QACtB,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;QACrB,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;QAC7B,QAAQ,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC;KACxB,EAAE,CAAC;IACJ,iDAAiD;IACjD,QAAQ,CAAC,OAAO,EAAE,SAAS;QAAE,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;IACpF,mEAAmE;IACnE,QAAQ,CAAC,GAAG,EAAE,SAAS;QACrB,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;QACrB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;QACzB,QAAQ,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC;KACxB,EAAE,CAAC;CACL;AAED,wBAAgB,WAAW,CACzB,OAAO,EAAE,SAAS,WAAW,EAAE,EAC/B,MAAM,EAAE,SAAS,aAAa,EAAE,GAC/B,WAAW,CAwBb"}
|
|
@@ -0,0 +1,184 @@
|
|
|
1
|
+
// SPDX-License-Identifier: Apache-2.0
|
|
2
|
+
// Copyright (C) 2026 Kindgi Inc.
|
|
3
|
+
const MAX_DEPTH = 3;
|
|
4
|
+
const MAX_LIST = 100;
|
|
5
|
+
function obj(value) {
|
|
6
|
+
return typeof value === 'object' && value !== null && !Array.isArray(value)
|
|
7
|
+
? value
|
|
8
|
+
: undefined;
|
|
9
|
+
}
|
|
10
|
+
function escapeToken(token) {
|
|
11
|
+
return token.replace(/~/g, '~0').replace(/\//g, '~1');
|
|
12
|
+
}
|
|
13
|
+
function unescapeToken(token) {
|
|
14
|
+
return token.replace(/~1/g, '/').replace(/~0/g, '~');
|
|
15
|
+
}
|
|
16
|
+
/** An element's own id, if it has a usable one. */
|
|
17
|
+
function ownId(element) {
|
|
18
|
+
const o = obj(element);
|
|
19
|
+
for (const field of ['id', 'key']) {
|
|
20
|
+
const v = o?.[field];
|
|
21
|
+
if (typeof v === 'string' && v.length > 0)
|
|
22
|
+
return v;
|
|
23
|
+
if (typeof v === 'number')
|
|
24
|
+
return String(v);
|
|
25
|
+
}
|
|
26
|
+
return undefined;
|
|
27
|
+
}
|
|
28
|
+
function listItems(value, pointer, depth) {
|
|
29
|
+
if (depth > MAX_DEPTH)
|
|
30
|
+
return [];
|
|
31
|
+
if (Array.isArray(value)) {
|
|
32
|
+
if (value.length === 0 || value.length > MAX_LIST)
|
|
33
|
+
return [];
|
|
34
|
+
const seen = new Set();
|
|
35
|
+
return value.map((element, i) => {
|
|
36
|
+
const id = ownId(element);
|
|
37
|
+
// Duplicate ids within one list fall back to the element's place.
|
|
38
|
+
const own = id !== undefined && !seen.has(id);
|
|
39
|
+
const key = own ? id : `${pointer}/${i}`;
|
|
40
|
+
seen.add(key);
|
|
41
|
+
return { key, pointer: `${pointer}/${i}`, rank: i, ownKey: own, value: element };
|
|
42
|
+
});
|
|
43
|
+
}
|
|
44
|
+
const o = obj(value);
|
|
45
|
+
if (o === undefined)
|
|
46
|
+
return [];
|
|
47
|
+
return Object.entries(o).flatMap(([field, child]) => listItems(child, `${pointer}/${escapeToken(field)}`, depth + 1));
|
|
48
|
+
}
|
|
49
|
+
function turnAnswer(output) {
|
|
50
|
+
const appended = Array.isArray(output.appended) ? output.appended : [];
|
|
51
|
+
for (let i = appended.length - 1; i >= 0; i--) {
|
|
52
|
+
const m = obj(appended[i]);
|
|
53
|
+
if (m?.role === 'agent' && m.content !== undefined && !hasToolCalls(m.content)) {
|
|
54
|
+
return { key: 'answer', pointer: `/appended/${i}/content`, ownKey: false, value: m.content };
|
|
55
|
+
}
|
|
56
|
+
}
|
|
57
|
+
return undefined;
|
|
58
|
+
}
|
|
59
|
+
function hasToolCalls(content) {
|
|
60
|
+
return Array.isArray(obj(content)?.toolCalls);
|
|
61
|
+
}
|
|
62
|
+
/** The items of a run's output (`agentTurn`: an agent turn's result). */
|
|
63
|
+
export function outputItems(output, agentTurn) {
|
|
64
|
+
const o = obj(output);
|
|
65
|
+
if (agentTurn) {
|
|
66
|
+
if (o === undefined)
|
|
67
|
+
return [];
|
|
68
|
+
const answer = turnAnswer(o);
|
|
69
|
+
return [
|
|
70
|
+
...(answer !== undefined ? [answer] : []),
|
|
71
|
+
...(o.output !== undefined ? listItems(o.output, '/output', 0) : []),
|
|
72
|
+
];
|
|
73
|
+
}
|
|
74
|
+
if (output === undefined || output === null)
|
|
75
|
+
return [];
|
|
76
|
+
return [
|
|
77
|
+
{ key: 'output', pointer: '', ownKey: false, value: output },
|
|
78
|
+
...listItems(output, '', 0),
|
|
79
|
+
];
|
|
80
|
+
}
|
|
81
|
+
/** The value at a JSON Pointer in `doc`, or `undefined` when there's none. */
|
|
82
|
+
export function valueAt(doc, pointer) {
|
|
83
|
+
if (pointer === '')
|
|
84
|
+
return doc;
|
|
85
|
+
if (!pointer.startsWith('/'))
|
|
86
|
+
return undefined;
|
|
87
|
+
let current = doc;
|
|
88
|
+
for (const raw of pointer.slice(1).split('/')) {
|
|
89
|
+
const token = unescapeToken(raw);
|
|
90
|
+
if (Array.isArray(current)) {
|
|
91
|
+
const i = Number(token);
|
|
92
|
+
current = Number.isInteger(i) ? current[i] : undefined;
|
|
93
|
+
}
|
|
94
|
+
else {
|
|
95
|
+
current = obj(current)?.[token];
|
|
96
|
+
}
|
|
97
|
+
if (current === undefined)
|
|
98
|
+
return undefined;
|
|
99
|
+
}
|
|
100
|
+
return current;
|
|
101
|
+
}
|
|
102
|
+
/** JSON with sorted keys, so equal values compare equal. */
|
|
103
|
+
function canonical(value) {
|
|
104
|
+
if (Array.isArray(value))
|
|
105
|
+
return `[${value.map(canonical).join(',')}]`;
|
|
106
|
+
const o = obj(value);
|
|
107
|
+
if (o !== undefined) {
|
|
108
|
+
return `{${Object.keys(o)
|
|
109
|
+
.sort()
|
|
110
|
+
.filter((k) => o[k] !== undefined)
|
|
111
|
+
.map((k) => `${JSON.stringify(k)}:${canonical(o[k])}`)
|
|
112
|
+
.join(',')}}`;
|
|
113
|
+
}
|
|
114
|
+
return JSON.stringify(value) ?? 'null';
|
|
115
|
+
}
|
|
116
|
+
/**
|
|
117
|
+
* Match a new output's items to the items judged in a past output
|
|
118
|
+
* (`judgedOutput`, where the judged items' pointers point).
|
|
119
|
+
*/
|
|
120
|
+
export function matchJudged(items, judged, judgedOutput) {
|
|
121
|
+
const byKey = new Map(judged.map((j) => [j.key, j]));
|
|
122
|
+
return items.map((item) => {
|
|
123
|
+
const j = byKey.get(item.key);
|
|
124
|
+
if (j === undefined)
|
|
125
|
+
return { item };
|
|
126
|
+
if (item.ownKey)
|
|
127
|
+
return { item, judged: j };
|
|
128
|
+
const before = j.pointer === undefined ? undefined : valueAt(judgedOutput, j.pointer);
|
|
129
|
+
return before !== undefined && canonical(before) === canonical(item.value)
|
|
130
|
+
? { item, judged: j }
|
|
131
|
+
: { item };
|
|
132
|
+
});
|
|
133
|
+
}
|
|
134
|
+
export function scoreItems(matched, k) {
|
|
135
|
+
let yesWeight = 0;
|
|
136
|
+
let totalWeight = 0;
|
|
137
|
+
let judgedItems = 0;
|
|
138
|
+
for (const m of matched) {
|
|
139
|
+
if (m.judged === undefined)
|
|
140
|
+
continue;
|
|
141
|
+
judgedItems += 1;
|
|
142
|
+
yesWeight += m.judged.yesWeight;
|
|
143
|
+
totalWeight += m.judged.totalWeight;
|
|
144
|
+
}
|
|
145
|
+
const top = matched
|
|
146
|
+
.filter((m) => m.item.rank !== undefined)
|
|
147
|
+
.sort((a, b) => (a.item.rank ?? 0) - (b.item.rank ?? 0))
|
|
148
|
+
.slice(0, k);
|
|
149
|
+
const topK = { yesWeight: 0, totalWeight: 0 };
|
|
150
|
+
for (const m of top) {
|
|
151
|
+
if (m.judged === undefined)
|
|
152
|
+
continue;
|
|
153
|
+
topK.yesWeight += m.judged.yesWeight;
|
|
154
|
+
topK.totalWeight += m.judged.totalWeight;
|
|
155
|
+
}
|
|
156
|
+
return { yesWeight, totalWeight, items: matched.length, judgedItems, topK };
|
|
157
|
+
}
|
|
158
|
+
export function itemChanges(matched, judged) {
|
|
159
|
+
const keptKeys = new Set();
|
|
160
|
+
const kept = [];
|
|
161
|
+
const fresh = [];
|
|
162
|
+
for (const m of matched) {
|
|
163
|
+
if (m.judged !== undefined) {
|
|
164
|
+
keptKeys.add(m.judged.key);
|
|
165
|
+
kept.push({
|
|
166
|
+
key: m.judged.key,
|
|
167
|
+
...(m.judged.rank !== undefined && { rankBefore: m.judged.rank }),
|
|
168
|
+
...(m.item.rank !== undefined && { rank: m.item.rank }),
|
|
169
|
+
});
|
|
170
|
+
}
|
|
171
|
+
else {
|
|
172
|
+
fresh.push({
|
|
173
|
+
key: m.item.key,
|
|
174
|
+
pointer: m.item.pointer,
|
|
175
|
+
...(m.item.rank !== undefined && { rank: m.item.rank }),
|
|
176
|
+
});
|
|
177
|
+
}
|
|
178
|
+
}
|
|
179
|
+
const dropped = judged
|
|
180
|
+
.filter((j) => !keptKeys.has(j.key))
|
|
181
|
+
.map((j) => ({ key: j.key, ...(j.rank !== undefined && { rankBefore: j.rank }) }));
|
|
182
|
+
return { kept, dropped, new: fresh };
|
|
183
|
+
}
|
|
184
|
+
//# sourceMappingURL=judged-items.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"judged-items.js","sourceRoot":"","sources":["../src/judged-items.ts"],"names":[],"mappings":"AAAA,sCAAsC;AACtC,iCAAiC;AAgCjC,MAAM,SAAS,GAAG,CAAC,CAAC;AACpB,MAAM,QAAQ,GAAG,GAAG,CAAC;AAErB,SAAS,GAAG,CAAC,KAAc;IACzB,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QACzE,CAAC,CAAE,KAAiC;QACpC,CAAC,CAAC,SAAS,CAAC;AAChB,CAAC;AAED,SAAS,WAAW,CAAC,KAAa;IAChC,OAAO,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;AACxD,CAAC;AAED,SAAS,aAAa,CAAC,KAAa;IAClC,OAAO,KAAK,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;AACvD,CAAC;AAED,mDAAmD;AACnD,SAAS,KAAK,CAAC,OAAgB;IAC7B,MAAM,CAAC,GAAG,GAAG,CAAC,OAAO,CAAC,CAAC;IACvB,KAAK,MAAM,KAAK,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC;QAClC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC;QACrB,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,CAAC,CAAC;QACpD,IAAI,OAAO,CAAC,KAAK,QAAQ;YAAE,OAAO,MAAM,CAAC,CAAC,CAAC,CAAC;IAC9C,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,SAAS,SAAS,CAAC,KAAc,EAAE,OAAe,EAAE,KAAa;IAC/D,IAAI,KAAK,GAAG,SAAS;QAAE,OAAO,EAAE,CAAC;IACjC,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QACzB,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,IAAI,KAAK,CAAC,MAAM,GAAG,QAAQ;YAAE,OAAO,EAAE,CAAC;QAC7D,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;QAC/B,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,CAAC,EAAE,EAAE;YAC9B,MAAM,EAAE,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC;YAC1B,kEAAkE;YAClE,MAAM,GAAG,GAAG,EAAE,KAAK,SAAS,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YAC9C,MAAM,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,OAAO,IAAI,CAAC,EAAE,CAAC;YACzC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YACd,OAAO,EAAE,GAAG,EAAE,OAAO,EAAE,GAAG,OAAO,IAAI,CAAC,EAAE,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,EAAE,GAAG,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;QACnF,CAAC,CAAC,CAAC;IACL,CAAC;IACD,MAAM,CAAC,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC;IACrB,IAAI,CAAC,KAAK,SAAS;QAAE,OAAO,EAAE,CAAC;IAC/B,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,EAAE,KAAK,CAAC,EAAE,EAAE,CAClD,SAAS,CAAC,KAAK,EAAE,GAAG,OAAO,IAAI,WAAW,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,GAAG,CAAC,CAAC,CAChE,CAAC;AACJ,CAAC;AAED,SAAS,UAAU,CAAC,MAA+B;IACjD,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;IACvE,KAAK,IAAI,CAAC,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC9C,MAAM,CAAC,GAAG,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;QAC3B,IAAI,CAAC,EAAE,IAAI,KAAK,OAAO,IAAI,CAAC,CAAC,OAAO,KAAK,SAAS,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;YAC/E,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,aAAa,CAAC,UAAU,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC;QAC/F,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,SAAS,YAAY,CAAC,OAAgB;IACpC,OAAO,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,SAAS,CAAC,CAAC;AAChD,CAAC;AAED,yEAAyE;AACzE,MAAM,UAAU,WAAW,CAAC,MAAe,EAAE,SAAkB;IAC7D,MAAM,CAAC,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC;IACtB,IAAI,SAAS,EAAE,CAAC;QACd,IAAI,CAAC,KAAK,SAAS;YAAE,OAAO,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC;QAC7B,OAAO;YACL,GAAG,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YACzC,GAAG,CAAC,CAAC,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;SACrE,CAAC;IACJ,CAAC;IACD,IAAI,MAAM,KAAK,SAAS,IAAI,MAAM,KAAK,IAAI;QAAE,OAAO,EAAE,CAAC;IACvD,OAAO;QACL,EAAE,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE;QAC5D,GAAG,SAAS,CAAC,MAAM,EAAE,EAAE,EAAE,CAAC,CAAC;KAC5B,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,MAAM,UAAU,OAAO,CAAC,GAAY,EAAE,OAAe;IACnD,IAAI,OAAO,KAAK,EAAE;QAAE,OAAO,GAAG,CAAC;IAC/B,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC;QAAE,OAAO,SAAS,CAAC;IAC/C,IAAI,OAAO,GAAY,GAAG,CAAC;IAC3B,KAAK,MAAM,GAAG,IAAI,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC;QAC9C,MAAM,KAAK,GAAG,aAAa,CAAC,GAAG,CAAC,CAAC;QACjC,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;YAC3B,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;YACxB,OAAO,GAAG,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QACzD,CAAC;aAAM,CAAC;YACN,OAAO,GAAG,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC;QAClC,CAAC;QACD,IAAI,OAAO,KAAK,SAAS;YAAE,OAAO,SAAS,CAAC;IAC9C,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,4DAA4D;AAC5D,SAAS,SAAS,CAAC,KAAc;IAC/B,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,IAAI,KAAK,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC;IACvE,MAAM,CAAC,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC;IACrB,IAAI,CAAC,KAAK,SAAS,EAAE,CAAC;QACpB,OAAO,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;aACtB,IAAI,EAAE;aACN,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,SAAS,CAAC;aACjC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;aACrD,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC;IAClB,CAAC;IACD,OAAO,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,MAAM,CAAC;AACzC,CAAC;AAiBD;;;GAGG;AACH,MAAM,UAAU,WAAW,CACzB,KAA4B,EAC5B,MAAgC,EAChC,YAAqB;IAErB,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IACrD,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;QACxB,MAAM,CAAC,GAAG,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC9B,IAAI,CAAC,KAAK,SAAS;YAAE,OAAO,EAAE,IAAI,EAAE,CAAC;QACrC,IAAI,IAAI,CAAC,MAAM;YAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;QAC5C,MAAM,MAAM,GAAG,CAAC,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,YAAY,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC;QACtF,OAAO,MAAM,KAAK,SAAS,IAAI,SAAS,CAAC,MAAM,CAAC,KAAK,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC;YACxE,CAAC,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,EAAE;YACrB,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC;IACf,CAAC,CAAC,CAAC;AACL,CAAC;AAaD,MAAM,UAAU,UAAU,CAAC,OAA+B,EAAE,CAAS;IACnE,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,WAAW,GAAG,CAAC,CAAC;IACpB,IAAI,WAAW,GAAG,CAAC,CAAC;IACpB,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS;YAAE,SAAS;QACrC,WAAW,IAAI,CAAC,CAAC;QACjB,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC;QAChC,WAAW,IAAI,CAAC,CAAC,MAAM,CAAC,WAAW,CAAC;IACtC,CAAC;IACD,MAAM,GAAG,GAAG,OAAO;SAChB,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,KAAK,SAAS,CAAC;SACxC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC;SACvD,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACf,MAAM,IAAI,GAAG,EAAE,SAAS,EAAE,CAAC,EAAE,WAAW,EAAE,CAAC,EAAE,CAAC;IAC9C,KAAK,MAAM,CAAC,IAAI,GAAG,EAAE,CAAC;QACpB,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS;YAAE,SAAS;QACrC,IAAI,CAAC,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC;QACrC,IAAI,CAAC,WAAW,IAAI,CAAC,CAAC,MAAM,CAAC,WAAW,CAAC;IAC3C,CAAC;IACD,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,KAAK,EAAE,OAAO,CAAC,MAAM,EAAE,WAAW,EAAE,IAAI,EAAE,CAAC;AAC9E,CAAC;AAoBD,MAAM,UAAU,WAAW,CACzB,OAA+B,EAC/B,MAAgC;IAEhC,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAU,CAAC;IACnC,MAAM,IAAI,GAAkC,EAAE,CAAC;IAC/C,MAAM,KAAK,GAAiC,EAAE,CAAC;IAC/C,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;YAC3B,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC3B,IAAI,CAAC,IAAI,CAAC;gBACR,GAAG,EAAE,CAAC,CAAC,MAAM,CAAC,GAAG;gBACjB,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;gBACjE,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;aACxD,CAAC,CAAC;QACL,CAAC;aAAM,CAAC;YACN,KAAK,CAAC,IAAI,CAAC;gBACT,GAAG,EAAE,CAAC,CAAC,IAAI,CAAC,GAAG;gBACf,OAAO,EAAE,CAAC,CAAC,IAAI,CAAC,OAAO;gBACvB,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;aACxD,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IACD,MAAM,OAAO,GAAG,MAAM;SACnB,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;SACnC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,UAAU,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;IACrF,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,GAAG,EAAE,KAAK,EAAE,CAAC;AACvC,CAAC"}
|