@tangle-network/agent-eval 0.125.0 → 0.126.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +62 -35
- package/README.md +270 -189
- package/dist/analyst/index.d.ts +15 -145
- package/dist/analyst/index.js +33 -47
- package/dist/analyst/index.js.map +1 -1
- package/dist/benchmarks/index.d.ts +45 -162
- package/dist/benchmarks/index.js +8 -9
- package/dist/campaign/index.d.ts +3674 -5393
- package/dist/campaign/index.js +21 -95
- package/dist/{chunk-R226UZOI.js → chunk-474LBSOX.js} +2 -2
- package/dist/{chunk-HM6V7F3M.js → chunk-FO7HEH76.js} +3 -3
- package/dist/chunk-IILEIWGW.js +635 -0
- package/dist/chunk-IILEIWGW.js.map +1 -0
- package/dist/{chunk-EQUK3RFS.js → chunk-J5SQWP6Y.js} +8 -5
- package/dist/chunk-J5SQWP6Y.js.map +1 -0
- package/dist/{chunk-W5B3ZGP3.js → chunk-KE2VWPZX.js} +8 -6
- package/dist/{chunk-W5B3ZGP3.js.map → chunk-KE2VWPZX.js.map} +1 -1
- package/dist/{chunk-DT7OXY3C.js → chunk-LUNF2SEL.js} +538 -851
- package/dist/chunk-LUNF2SEL.js.map +1 -0
- package/dist/chunk-NGUYT5CI.js +4637 -0
- package/dist/chunk-NGUYT5CI.js.map +1 -0
- package/dist/{chunk-QFQZ3U3X.js → chunk-OCFJACJU.js} +2 -2
- package/dist/{chunk-GID26AN4.js → chunk-P22LJ3Y2.js} +4 -6
- package/dist/{chunk-GID26AN4.js.map → chunk-P22LJ3Y2.js.map} +1 -1
- package/dist/{chunk-SJT4OBVL.js → chunk-SDPM6554.js} +3 -3
- package/dist/{chunk-D5JZ7UDZ.js → chunk-UCLVDLCH.js} +136 -50
- package/dist/chunk-UCLVDLCH.js.map +1 -0
- package/dist/chunk-VMUENW6F.js +7274 -0
- package/dist/chunk-VMUENW6F.js.map +1 -0
- package/dist/{chunk-JKDNAOF5.js → chunk-W4L6C2XT.js} +2 -2
- package/dist/chunk-WGXIEX7P.js +116 -0
- package/dist/chunk-WGXIEX7P.js.map +1 -0
- package/dist/{chunk-GRCDRKII.js → chunk-WS3NZZQQ.js} +58 -20
- package/dist/chunk-WS3NZZQQ.js.map +1 -0
- package/dist/cli.js +3 -3
- package/dist/contract/index.d.ts +3220 -3094
- package/dist/contract/index.js +173 -42
- package/dist/contract/index.js.map +1 -1
- package/dist/control.js +2 -3
- package/dist/fuzz.d.ts +14 -1
- package/dist/fuzz.js +1 -1
- package/dist/hosted/index.d.ts +8 -1
- package/dist/index.d.ts +71 -687
- package/dist/index.js +178 -497
- package/dist/index.js.map +1 -1
- package/dist/openapi.json +1 -1
- package/dist/rl.d.ts +5 -100
- package/dist/rl.js +4 -5
- package/dist/rl.js.map +1 -1
- package/dist/{run-campaign-I3JXKVAK.js → run-campaign-LVFKZCEU.js} +3 -3
- package/dist/traces.js +2 -3
- package/dist/wire/index.d.ts +14 -1
- package/dist/wire/index.js +3 -3
- package/docs/campaign-proposers.md +363 -168
- package/docs/design/loop-taxonomy.md +142 -190
- package/docs/design.md +1 -1
- package/docs/distributed-driver.md +8 -11
- package/docs/feature-guide.md +20 -19
- package/docs/knowledge-readiness.md +2 -5
- package/docs/multi-shot-optimization.md +35 -27
- package/docs/rollout.md +5 -5
- package/package.json +4 -4
- package/dist/chunk-A62YMFWA.js +0 -9269
- package/dist/chunk-A62YMFWA.js.map +0 -1
- package/dist/chunk-A6GT67HT.js +0 -550
- package/dist/chunk-A6GT67HT.js.map +0 -1
- package/dist/chunk-D5JZ7UDZ.js.map +0 -1
- package/dist/chunk-DT7OXY3C.js.map +0 -1
- package/dist/chunk-EQUK3RFS.js.map +0 -1
- package/dist/chunk-GC4ATIKK.js +0 -317
- package/dist/chunk-GC4ATIKK.js.map +0 -1
- package/dist/chunk-GRCDRKII.js.map +0 -1
- package/dist/chunk-LOW3U7JZ.js +0 -328
- package/dist/chunk-LOW3U7JZ.js.map +0 -1
- package/dist/chunk-PMITBABE.js +0 -3841
- package/dist/chunk-PMITBABE.js.map +0 -1
- /package/dist/{chunk-R226UZOI.js.map → chunk-474LBSOX.js.map} +0 -0
- /package/dist/{chunk-HM6V7F3M.js.map → chunk-FO7HEH76.js.map} +0 -0
- /package/dist/{chunk-QFQZ3U3X.js.map → chunk-OCFJACJU.js.map} +0 -0
- /package/dist/{chunk-SJT4OBVL.js.map → chunk-SDPM6554.js.map} +0 -0
- /package/dist/{chunk-JKDNAOF5.js.map → chunk-W4L6C2XT.js.map} +0 -0
- /package/dist/{run-campaign-I3JXKVAK.js.map → run-campaign-LVFKZCEU.js.map} +0 -0
|
@@ -1,9 +1,9 @@
|
|
|
1
1
|
import {
|
|
2
2
|
planCampaignRun,
|
|
3
3
|
runCampaign
|
|
4
|
-
} from "./chunk-
|
|
4
|
+
} from "./chunk-UCLVDLCH.js";
|
|
5
5
|
import "./chunk-PJQFMIOX.js";
|
|
6
|
-
import "./chunk-
|
|
6
|
+
import "./chunk-WS3NZZQQ.js";
|
|
7
7
|
import "./chunk-VI2UW6B6.js";
|
|
8
8
|
import "./chunk-ONWEPEDO.js";
|
|
9
9
|
import "./chunk-PZ5AY32C.js";
|
|
@@ -11,4 +11,4 @@ export {
|
|
|
11
11
|
planCampaignRun,
|
|
12
12
|
runCampaign
|
|
13
13
|
};
|
|
14
|
-
//# sourceMappingURL=run-campaign-
|
|
14
|
+
//# sourceMappingURL=run-campaign-LVFKZCEU.js.map
|
package/dist/traces.js
CHANGED
|
@@ -27,7 +27,7 @@ import {
|
|
|
27
27
|
scoreTraceInsightReadiness,
|
|
28
28
|
tokenizeDomainWords,
|
|
29
29
|
traceAnalystOnRunComplete
|
|
30
|
-
} from "./chunk-
|
|
30
|
+
} from "./chunk-OCFJACJU.js";
|
|
31
31
|
import "./chunk-H5UD2323.js";
|
|
32
32
|
import {
|
|
33
33
|
extractUsage,
|
|
@@ -91,7 +91,7 @@ import {
|
|
|
91
91
|
TraceEmitter,
|
|
92
92
|
llmSpanFromProvider
|
|
93
93
|
} from "./chunk-VQMK5FMP.js";
|
|
94
|
-
import "./chunk-
|
|
94
|
+
import "./chunk-IILEIWGW.js";
|
|
95
95
|
import {
|
|
96
96
|
FAILURE_CLASSES,
|
|
97
97
|
TRACE_SCHEMA_VERSION,
|
|
@@ -101,7 +101,6 @@ import {
|
|
|
101
101
|
isSandboxSpan,
|
|
102
102
|
isToolSpan
|
|
103
103
|
} from "./chunk-MA6HLL3S.js";
|
|
104
|
-
import "./chunk-GC4ATIKK.js";
|
|
105
104
|
import "./chunk-VSMTAMNK.js";
|
|
106
105
|
import "./chunk-ONWEPEDO.js";
|
|
107
106
|
import {
|
package/dist/wire/index.d.ts
CHANGED
|
@@ -37,29 +37,40 @@ interface CostReceipt extends CostCallBase, CostUsage {
|
|
|
37
37
|
costUsd: number;
|
|
38
38
|
costUnknown: boolean;
|
|
39
39
|
usageUnknown?: boolean;
|
|
40
|
+
/** Rates used to estimate cost locally. Absent when cost is provider-reported or unknown. */
|
|
40
41
|
pricing?: {
|
|
41
42
|
inputUsdPerThousand: number;
|
|
43
|
+
cachedInputUsdPerThousand?: number;
|
|
44
|
+
cacheWriteUsdPerThousand?: number;
|
|
42
45
|
outputUsdPerThousand: number;
|
|
43
46
|
};
|
|
47
|
+
/** Cost reported by the provider, not a local token-price calculation. */
|
|
44
48
|
actualCostUsd?: number;
|
|
45
49
|
error?: string;
|
|
46
50
|
}
|
|
47
51
|
interface CostReceiptInput extends CostUsage {
|
|
48
52
|
model: string;
|
|
53
|
+
/** Caller-supplied rates for a local estimate when the provider does not report billed cost. */
|
|
54
|
+
customTokenPricing?: CustomTokenPricing;
|
|
49
55
|
actualCostUsd?: number;
|
|
50
56
|
costUnknown?: boolean;
|
|
51
57
|
usageUnknown?: boolean;
|
|
52
58
|
}
|
|
53
59
|
/** Per-million token rates for a model or endpoint not covered by package pricing. */
|
|
54
60
|
interface CustomTokenPricing {
|
|
61
|
+
/** Non-cached input tokens. */
|
|
55
62
|
inputUsdPerMillion: number;
|
|
63
|
+
/** Cache-read tokens. Falls back to the normal input rate when omitted. */
|
|
64
|
+
cachedInputUsdPerMillion?: number;
|
|
65
|
+
/** Cache-creation or cache-write tokens. Falls back to the normal input rate when omitted. */
|
|
66
|
+
cacheWriteUsdPerMillion?: number;
|
|
56
67
|
outputUsdPerMillion: number;
|
|
57
68
|
}
|
|
58
69
|
type MaximumCharge = {
|
|
59
70
|
externallyEnforcedMaximumUsd: number;
|
|
60
71
|
} | ({
|
|
61
72
|
customTokenPricing: CustomTokenPricing;
|
|
62
|
-
} & Pick<CostUsage, 'inputTokens' | 'outputTokens'>) | ({
|
|
73
|
+
} & Pick<CostUsage, 'inputTokens' | 'outputTokens' | 'cachedTokens' | 'cacheWriteTokens'>) | ({
|
|
63
74
|
model: string;
|
|
64
75
|
} & CostUsage);
|
|
65
76
|
interface RunPaidCallInput<T> {
|
|
@@ -127,6 +138,8 @@ interface CostLedgerFilter {
|
|
|
127
138
|
interface CostLedgerWaitOptions {
|
|
128
139
|
/** Maximum time to wait for active provider calls. Default 5 seconds. */
|
|
129
140
|
timeoutMs?: number;
|
|
141
|
+
/** Wait only for calls matching this attribution filter. */
|
|
142
|
+
filter?: CostLedgerFilter;
|
|
130
143
|
}
|
|
131
144
|
/** Append-only storage. `append` must atomically reject stale revisions. */
|
|
132
145
|
interface CostLedgerPersistence {
|
package/dist/wire/index.js
CHANGED
|
@@ -34,9 +34,9 @@ import {
|
|
|
34
34
|
runRpcOnce,
|
|
35
35
|
startServer,
|
|
36
36
|
startServerAsync
|
|
37
|
-
} from "../chunk-
|
|
38
|
-
import "../chunk-
|
|
39
|
-
import "../chunk-
|
|
37
|
+
} from "../chunk-FO7HEH76.js";
|
|
38
|
+
import "../chunk-J5SQWP6Y.js";
|
|
39
|
+
import "../chunk-WS3NZZQQ.js";
|
|
40
40
|
import "../chunk-VI2UW6B6.js";
|
|
41
41
|
import "../chunk-PC4UYEBM.js";
|
|
42
42
|
import "../chunk-ONWEPEDO.js";
|