@tangle-network/agent-eval 1.0.21 → 1.0.22
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 +28 -0
- package/dist/adapters/http.d.ts +1 -1
- package/dist/{agent-profile-Dngk3LZ0.js → agent-profile-4j6f2n-x.js} +4 -4
- package/dist/agent-profile-4j6f2n-x.js.map +1 -0
- package/dist/{agent-profile-BO8ruI6R.d.ts → agent-profile-DU5InUT9.d.ts} +4 -4
- package/dist/{agent-profile-BO8ruI6R.d.ts.map → agent-profile-DU5InUT9.d.ts.map} +1 -1
- package/dist/analyst/index.d.ts +2 -2
- package/dist/analyst/index.js +1 -1
- package/dist/app-grade/index.d.ts +1 -1
- package/dist/{benchmark-command-CoNndRCk.js → benchmark-command-DkUpd9Q6.js} +2 -2
- package/dist/{benchmark-command-CoNndRCk.js.map → benchmark-command-DkUpd9Q6.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +1 -1
- package/dist/benchmarks/index.js +1 -1
- package/dist/campaign/index.d.ts +2 -2
- package/dist/campaign/index.js +1 -1
- package/dist/{campaign-Bl07uIaf.js → campaign-C2mmydmx.js} +2 -2
- package/dist/{campaign-Bl07uIaf.js.map → campaign-C2mmydmx.js.map} +1 -1
- package/dist/cli.js +1 -1
- package/dist/contract/index.d.ts +3 -3
- package/dist/{define-agent-eval-Bx3b9B01.d.ts → define-agent-eval-DGWBgCid.d.ts} +2 -2
- package/dist/{define-agent-eval-Bx3b9B01.d.ts.map → define-agent-eval-DGWBgCid.d.ts.map} +1 -1
- package/dist/{index-DfqAy8ct.d.ts → index-Bp3w4fWc.d.ts} +2 -2
- package/dist/{index-DfqAy8ct.d.ts.map → index-Bp3w4fWc.d.ts.map} +1 -1
- package/dist/{index-DFso1b2W.d.ts → index-DX87wzX3.d.ts} +3 -3
- package/dist/{index-DFso1b2W.d.ts.map → index-DX87wzX3.d.ts.map} +1 -1
- package/dist/{index-CM12lr9E.d.ts → index-QLDWZTxI.d.ts} +2 -2
- package/dist/{index-CM12lr9E.d.ts.map → index-QLDWZTxI.d.ts.map} +1 -1
- package/dist/{index-7tWvC0aW.d.ts → index-YS58LHlx.d.ts} +5 -5
- package/dist/{index-7tWvC0aW.d.ts.map → index-YS58LHlx.d.ts.map} +1 -1
- package/dist/index.d.ts +5 -5
- package/dist/index.js +1 -1
- package/dist/openapi.json +1 -1
- package/dist/wire/index.d.ts +2 -2
- package/package.json +5 -5
- package/dist/agent-profile-Dngk3LZ0.js.map +0 -1
package/CHANGELOG.md
CHANGED
|
@@ -4,6 +4,34 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
|
|
|
4
4
|
|
|
5
5
|
---
|
|
6
6
|
|
|
7
|
+
## [1.0.22] — 2026-11-07
|
|
8
|
+
|
|
9
|
+
- chore(deps): refresh first-party library dependencies (#1221)
|
|
10
|
+
- docs(release): describe independent interval API and drop duplicate bumps
|
|
11
|
+
- chore(release): 1.0.21
|
|
12
|
+
- docs(test): cite printed Newcombe table pages
|
|
13
|
+
- feat(statistics): add independent risk differences from Wilson limits
|
|
14
|
+
- fix(test): mark disposable identities for host Git hooks
|
|
15
|
+
- fix(test): declare identity in disposable integration repositories
|
|
16
|
+
- fix(release): authenticate public evidence reads in verification
|
|
17
|
+
- fix(statistics): exclude steps through every selected cluster extent
|
|
18
|
+
- test(statistics): retain overlapping-cluster sealed-step regression
|
|
19
|
+
- docs(release): remove redundant release commit entries
|
|
20
|
+
- chore(release): 1.0.20
|
|
21
|
+
- docs(statistics): trim whitespace in interval guidance
|
|
22
|
+
- fix(verification): retain valid replay baselines and current release evidence
|
|
23
|
+
- chore(analyst): refresh manifest for shared statistics owner
|
|
24
|
+
- fix(statistics): preserve degenerate bootstrap results and mean seed
|
|
25
|
+
- feat(statistics): share stratified cluster resampling and calibrated placebo bands
|
|
26
|
+
- fix(adoption): retain unselected pnpm resolutions
|
|
27
|
+
- fix(exports): ship ./statistics subpath, gpt-5-mini pricing, sequentialCrossingHorizon
|
|
28
|
+
- fix(adopt-stack): qualify the committed caller tree
|
|
29
|
+
- fix(adopt-stack): read large tracked path inventories
|
|
30
|
+
- fix(adoption): read a Runtime release's evidence from its Publish run, and raise pins below a dependency floor (#1211)
|
|
31
|
+
- fix(adoption): hold a Runtime release whose tag evidence failed or never ran
|
|
32
|
+
- fix(adoption): gate stable-major candidates against installed versions
|
|
33
|
+
- feat(app-grade): probes take a method and JSON body (#1208)
|
|
34
|
+
|
|
7
35
|
## [1.0.21] — 2026-11-06
|
|
8
36
|
|
|
9
37
|
- Add `independentRiskDifference`: Newcombe/Wilson intervals for independent before/after counts, with explicit uncertainty for empty arms.
|
package/dist/adapters/http.d.ts
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import { d as CostReceipt } from "../cost-ledger-Czro9wmo.js";
|
|
2
2
|
import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-ip2D3-Ib.js";
|
|
3
|
-
import "../index-
|
|
3
|
+
import "../index-DX87wzX3.js";
|
|
4
4
|
export interface HttpDispatchOptions<TScenario extends Scenario, _TArtifact> {
|
|
5
5
|
/** Static endpoint URL. Mutually exclusive with `resolveUrl`. */
|
|
6
6
|
url?: string;
|
|
@@ -18,9 +18,9 @@ const CODING_HARNESSES = [
|
|
|
18
18
|
];
|
|
19
19
|
/** Model sentinel for a vendor-locked harness that supports none of the swept models:
|
|
20
20
|
* it carries no provider prefix, so `harnessSupportsModel` accepts it and the harness
|
|
21
|
-
* resolves it to its own native default model at runtime
|
|
22
|
-
*
|
|
23
|
-
*
|
|
21
|
+
* resolves it to its own native default model at runtime. Lets `expandProfileAxes`
|
|
22
|
+
* snap-instead-of-drop without a per-harness flagship table that would rot as router
|
|
23
|
+
* catalogs change. */
|
|
24
24
|
const HARNESS_NATIVE_MODEL = "default";
|
|
25
25
|
/**
|
|
26
26
|
* Expand a base profile across the harness × model matrix into the `AgentProfile[]`
|
|
@@ -150,4 +150,4 @@ function agentProfileHash(profile) {
|
|
|
150
150
|
}
|
|
151
151
|
export { agentProfileModelId as a, agentProfileId as i, HARNESS_NATIVE_MODEL as n, expandProfileAxes as o, agentProfileHash as r, harnessAxisOf as s, CODING_HARNESSES as t };
|
|
152
152
|
|
|
153
|
-
//# sourceMappingURL=agent-profile-
|
|
153
|
+
//# sourceMappingURL=agent-profile-4j6f2n-x.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-profile-4j6f2n-x.js","names":[],"sources":["../src/agent-profile.ts"],"sourcesContent":["import { createHash } from 'node:crypto'\nimport type { AgentProfile } from '@tangle-network/agent-interface'\nimport { type HarnessType, harnessSupportsModel } from '@tangle-network/agent-interface'\nimport { ValidationError } from './errors'\nimport { canonicalString } from './ledger-core/canonical'\n\nexport type { AgentProfile, HarnessType } from '@tangle-network/agent-interface'\n\n/**\n * The agentic coding harnesses an eval sweeps by default — the ones we care about\n * ranking. This is the SINGLE source of that list; consumers import it instead of\n * re-declaring their own (a re-declared list is how the fleet drifts). Pass an\n * explicit `harnesses` (e.g. `harnessTypeSchema.options` for literally every known\n * harness) to widen beyond these.\n */\nexport const CODING_HARNESSES: readonly HarnessType[] = [\n 'opencode',\n 'claude-code',\n 'codex',\n 'kimi-code',\n]\n\nexport interface ProfileAxisSpec {\n /** The domain profile to sweep. Its prompt/tools/skills are held fixed; only the\n * harness and model vary. `model.default` is the fallback model. */\n base: AgentProfile\n /** Harnesses to cross. Default: {@link CODING_HARNESSES}. */\n harnesses?: readonly HarnessType[]\n /** Models to cross. Default: `[base.model.default]` — one model, i.e. today's\n * single-model behaviour, so omitting this never changes an existing run. */\n models?: readonly string[]\n /** Force every (harness, model) pair verbatim, even ones the harness can't run —\n * for deliberately testing failure modes. Default (false): SNAP instead — a\n * vendor-locked harness runs only the swept models in its family, or its native\n * default when it supports none, so no harness is dropped and none gets a\n * guaranteed-failing foreign-model cell. */\n keepIncompatible?: boolean\n}\n\n/** Model sentinel for a vendor-locked harness that supports none of the swept models:\n * it carries no provider prefix, so `harnessSupportsModel` accepts it and the harness\n * resolves it to its own native default model at runtime. Lets `expandProfileAxes`\n * snap-instead-of-drop without a per-harness flagship table that would rot as router\n * catalogs change. */\nexport const HARNESS_NATIVE_MODEL = 'default'\n\n/**\n * Expand a base profile across the harness × model matrix into the `AgentProfile[]`\n * that `runProfileMatrix` / `selfImprove` score — the ONE place \"which harnesses ×\n * which models do we evaluate\" lives, so no product hand-rolls its own harness list\n * or column→profile mapping (the pattern that let those copies drift and silently\n * break the harness pivot).\n *\n * Each cell clones `base`, sets the canonical top-level `harness` and `model.default`,\n * and stamps `metadata.harness` + `metadata.harnessModel` for matrix grouping. Both\n * metadata fields are hash-bearing, so every cell gets a distinct `agentProfileId` row\n * and results join back by harness/model via {@link harnessAxisOf} with no\n * hand-recomputed key. A vendor-locked harness snaps to its family's swept models — or\n * its native default ({@link HARNESS_NATIVE_MODEL}) when it supports none — so every\n * requested harness runs; `keepIncompatible` forces every pair verbatim.\n *\n * Omit `harnesses`/`models` to sweep the full default set — the \"turn it on for\n * everything we care about\" switch, identical in shape whether one harness or all.\n */\nexport function expandProfileAxes(spec: ProfileAxisSpec): AgentProfile[] {\n const harnesses = spec.harnesses ?? CODING_HARNESSES\n if (harnesses.length === 0) throw new ValidationError('expandProfileAxes: no harnesses to sweep')\n const baseModel = spec.base.model?.default\n const models = spec.models ?? (baseModel ? [baseModel] : [])\n if (models.length === 0) {\n throw new ValidationError(\n 'expandProfileAxes: no models to sweep — base profile has no model.default and none were supplied',\n )\n }\n const out: AgentProfile[] = []\n const seen = new Set<string>()\n for (const harness of harnesses) {\n // Keep each requested harness on supported models, or its native default when none fit.\n // `keepIncompatible` forces every (harness, model) pair verbatim (failure-mode runs).\n const supported = spec.keepIncompatible\n ? models\n : models.filter((model) => harnessSupportsModel(harness, model))\n const effective = supported.length > 0 ? supported : [HARNESS_NATIVE_MODEL]\n for (const model of effective) {\n const profile: AgentProfile = {\n ...spec.base,\n harness,\n name: `${spec.base.name ?? 'agent'}/${harness}/${model}`,\n model: { ...spec.base.model, default: model },\n metadata: { ...(spec.base.metadata ?? {}), harness, harnessModel: model },\n }\n const id = agentProfileId(profile)\n if (seen.has(id)) continue\n seen.add(id)\n out.push(profile)\n }\n }\n if (out.length === 0) {\n // Unreachable in normal use — snapping guarantees ≥1 cell per harness — but keep a\n // fail-closed guard so a future refactor can't silently produce an empty sweep.\n throw new ValidationError(\n `expandProfileAxes: produced no profiles (harnesses=[${harnesses.join(', ')}], models=[${models.join(', ')}]).`,\n )\n }\n return out\n}\n\n/**\n * Read the (harness, model) a matrix cell ran under, off a profile or a result row's\n * profile — the join-back for a `byHarness` pivot. Returns undefined when the profile\n * wasn't produced by {@link expandProfileAxes}. Callers group `result.byProfile` by\n * this instead of recomputing an id (recomputing the wrong key is what broke the pivot\n * in the hand-rolled copies).\n */\nexport function harnessAxisOf(\n profile: Pick<AgentProfile, 'metadata'>,\n): { harness: HarnessType; model: string } | undefined {\n const m = profile.metadata as Record<string, unknown> | undefined\n const harness = m?.harness\n const model = m?.harnessModel\n if (typeof harness === 'string' && typeof model === 'string') {\n return { harness: harness as HarnessType, model }\n }\n return undefined\n}\n\n/**\n * Collision-resistant, path-safe, human-readable profile id for eval artifacts.\n * Scorecard joins still use `agentProfileHash`; this id is for run ids, matrix\n * keys, and directory names where two profiles must not collapse onto one row.\n * The suffix is the first 64 bits of the behaviour hash, enough for ordinary\n * eval matrices while keeping filenames readable.\n */\nexport function agentProfileId(profile: AgentProfile): string {\n const label = pathSafeProfileLabel(agentProfileDisplayLabel(profile)) ?? 'profile'\n return `${label}-${agentProfileHash(profile).slice(0, 16)}`\n}\n\n/**\n * Model snapshot used for `RunRecord.model`. Eval surfaces require a concrete\n * model id because run records reject bare/missing model aliases.\n */\nexport function agentProfileModelId(profile: AgentProfile): string {\n const model = profile.model?.default?.trim()\n if (!model) {\n const label = agentProfileDisplayLabel(profile) ?? 'unnamed profile'\n throw new ValidationError(\n `AgentProfile \"${label}\" has no model.default — cannot record eval run`,\n )\n }\n return model\n}\n\nfunction agentProfileDisplayLabel(profile: AgentProfile): string | undefined {\n return profile.name?.trim() || profile.version?.trim() || undefined\n}\n\nfunction pathSafeProfileLabel(label: string | undefined): string | undefined {\n const safe = label\n ?.trim()\n .replace(/[^A-Za-z0-9._-]+/g, '-')\n .replace(/-+/g, '-')\n .replace(/^-|-$/g, '')\n return safe || undefined\n}\n\nfunction compact<T extends Record<string, unknown>>(input: T): Partial<T> {\n const out: Record<string, unknown> = {}\n for (const [key, value] of Object.entries(input)) {\n if (value !== undefined) out[key] = value\n }\n return out as Partial<T>\n}\n\n/**\n * Deterministic behaviour identity for the canonical\n * `@tangle-network/agent-interface` AgentProfile.\n *\n * `name` and `description` are labels and do not affect the hash. Profile\n * `version`, prompt, model hints, tools, resources, hooks, modes, permissions,\n * and extensions do affect the hash. Resource array order is hash-bearing\n * because mount order can change agent behaviour. Undefined fields are treated\n * as absent; explicit `null` fields remain hash-bearing.\n */\nexport function agentProfileHash(profile: AgentProfile): string {\n const model = agentProfileModelId(profile)\n const behaviour = compact({\n ...profile,\n name: undefined,\n description: undefined,\n tags: profile.tags ? [...profile.tags].sort() : undefined,\n model: compact({ ...profile.model, default: model }),\n })\n return createHash('sha256').update(canonicalString(behaviour)).digest('hex')\n}\n"],"mappings":";;;;;;;;;;;;AAeA,MAAa,mBAA2C;CACtD;CACA;CACA;CACA;AACF;;;;;;AAwBA,MAAa,uBAAuB;;;;;;;;;;;;;;;;;;;AAoBpC,SAAgB,kBAAkB,MAAuC;CACvE,MAAM,YAAY,KAAK,aAAa;CACpC,IAAI,UAAU,WAAW,GAAG,MAAM,IAAI,gBAAgB,0CAA0C;CAChG,MAAM,YAAY,KAAK,KAAK,OAAO;CACnC,MAAM,SAAS,KAAK,WAAW,YAAY,CAAC,SAAS,IAAI,CAAC;CAC1D,IAAI,OAAO,WAAW,GACpB,MAAM,IAAI,gBACR,kGACF;CAEF,MAAM,MAAsB,CAAC;CAC7B,MAAM,uBAAO,IAAI,IAAY;CAC7B,KAAK,MAAM,WAAW,WAAW;EAG/B,MAAM,YAAY,KAAK,mBACnB,SACA,OAAO,QAAQ,UAAU,qBAAqB,SAAS,KAAK,CAAC;EACjE,MAAM,YAAY,UAAU,SAAS,IAAI,YAAY,CAAC,oBAAoB;EAC1E,KAAK,MAAM,SAAS,WAAW;GAC7B,MAAM,UAAwB;IAC5B,GAAG,KAAK;IACR;IACA,MAAM,GAAG,KAAK,KAAK,QAAQ,QAAQ,GAAG,QAAQ,GAAG;IACjD,OAAO;KAAE,GAAG,KAAK,KAAK;KAAO,SAAS;IAAM;IAC5C,UAAU;KAAE,GAAI,KAAK,KAAK,YAAY,CAAC;KAAI;KAAS,cAAc;IAAM;GAC1E;GACA,MAAM,KAAK,eAAe,OAAO;GACjC,IAAI,KAAK,IAAI,EAAE,GAAG;GAClB,KAAK,IAAI,EAAE;GACX,IAAI,KAAK,OAAO;EAClB;CACF;CACA,IAAI,IAAI,WAAW,GAGjB,MAAM,IAAI,gBACR,uDAAuD,UAAU,KAAK,IAAI,EAAE,aAAa,OAAO,KAAK,IAAI,EAAE,IAC7G;CAEF,OAAO;AACT;;;;;;;;AASA,SAAgB,cACd,SACqD;CACrD,MAAM,IAAI,QAAQ;CAClB,MAAM,UAAU,GAAG;CACnB,MAAM,QAAQ,GAAG;CACjB,IAAI,OAAO,YAAY,YAAY,OAAO,UAAU,UAClD,OAAO;EAAW;EAAwB;CAAM;AAGpD;;;;;;;;AASA,SAAgB,eAAe,SAA+B;CAE5D,OAAO,GADO,qBAAqB,yBAAyB,OAAO,CAAC,KAAK,UACzD,GAAG,iBAAiB,OAAO,CAAC,CAAC,MAAM,GAAG,EAAE;AAC1D;;;;;AAMA,SAAgB,oBAAoB,SAA+B;CACjE,MAAM,QAAQ,QAAQ,OAAO,SAAS,KAAK;CAC3C,IAAI,CAAC,OAAO;EACV,MAAM,QAAQ,yBAAyB,OAAO,KAAK;EACnD,MAAM,IAAI,gBACR,iBAAiB,MAAM,gDACzB;CACF;CACA,OAAO;AACT;AAEA,SAAS,yBAAyB,SAA2C;CAC3E,OAAO,QAAQ,MAAM,KAAK,KAAK,QAAQ,SAAS,KAAK,KAAK,KAAA;AAC5D;AAEA,SAAS,qBAAqB,OAA+C;CAM3E,OALa,OACT,KAAK,CAAC,CACP,QAAQ,qBAAqB,GAAG,CAAC,CACjC,QAAQ,OAAO,GAAG,CAAC,CACnB,QAAQ,UAAU,EAAE,KACR,KAAA;AACjB;AAEA,SAAS,QAA2C,OAAsB;CACxE,MAAM,MAA+B,CAAC;CACtC,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,KAAK,GAC7C,IAAI,UAAU,KAAA,GAAW,IAAI,OAAO;CAEtC,OAAO;AACT;;;;;;;;;;;AAYA,SAAgB,iBAAiB,SAA+B;CAC9D,MAAM,QAAQ,oBAAoB,OAAO;CACzC,MAAM,YAAY,QAAQ;EACxB,GAAG;EACH,MAAM,KAAA;EACN,aAAa,KAAA;EACb,MAAM,QAAQ,OAAO,CAAC,GAAG,QAAQ,IAAI,CAAC,CAAC,KAAK,IAAI,KAAA;EAChD,OAAO,QAAQ;GAAE,GAAG,QAAQ;GAAO,SAAS;EAAM,CAAC;CACrD,CAAC;CACD,OAAO,WAAW,QAAQ,CAAC,CAAC,OAAO,gBAAgB,SAAS,CAAC,CAAC,CAAC,OAAO,KAAK;AAC7E"}
|
|
@@ -412,9 +412,9 @@ interface ProfileAxisSpec {
|
|
|
412
412
|
}
|
|
413
413
|
/** Model sentinel for a vendor-locked harness that supports none of the swept models:
|
|
414
414
|
* it carries no provider prefix, so `harnessSupportsModel` accepts it and the harness
|
|
415
|
-
* resolves it to its own native default model at runtime
|
|
416
|
-
*
|
|
417
|
-
*
|
|
415
|
+
* resolves it to its own native default model at runtime. Lets `expandProfileAxes`
|
|
416
|
+
* snap-instead-of-drop without a per-harness flagship table that would rot as router
|
|
417
|
+
* catalogs change. */
|
|
418
418
|
declare const HARNESS_NATIVE_MODEL = "default";
|
|
419
419
|
/**
|
|
420
420
|
* Expand a base profile across the harness × model matrix into the `AgentProfile[]`
|
|
@@ -466,4 +466,4 @@ declare function agentProfileId(profile: AgentProfile): string;
|
|
|
466
466
|
*/
|
|
467
467
|
declare function agentProfileHash(profile: AgentProfile): string;
|
|
468
468
|
export { traceStoreEvidenceResolver as A, ExternalOptimizerModelCallRequest as B, AnalystEvidenceResolutionError as C, AnalystLatencyDistribution as D, AnalystIssueExpectation as E, ExternalOptimizerEndpointFormat as F, ExternalOptimizerRunnerCommand as G, ExternalOptimizerModelExecutionObservation as H, ExternalOptimizerEvaluationObservation as I, ExternalTextEvaluationRequest as J, ExternalOptimizerWireCounts as K, ExternalOptimizerEvaluationRefusalReason as L, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS as M, ExternalOptimizerCallbackLimits as N, RunAnalystBenchmarkOptions as O, ExternalOptimizerChatRequest as P, ExternalOptimizerModelBudget as R, AnalystEvidenceResolution as S, AnalystFindingScore as T, ExternalOptimizerProcessLimits as U, ExternalOptimizerModelCallResult as V, ExternalOptimizerResumeMode as W, AnalystBenchmarkProvenance as _, ProfileAxisSpec as a, AnalystBenchmarkSummary as b, expandProfileAxes as c, AnalystBenchmarkDatasetRef as d, AnalystBenchmarkDescriptor as f, AnalystBenchmarkOutput as g, AnalystBenchmarkObservation as h, HarnessType$1 as i, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS as j, runAnalystBenchmark as k, harnessAxisOf as l, AnalystBenchmarkLabelState as m, CODING_HARNESSES as n, agentProfileHash as o, AnalystBenchmarkError as p, ExternalTextCandidate as q, HARNESS_NATIVE_MODEL as r, agentProfileId as s, AgentProfile$1 as t, AnalystBenchmarkCase as u, AnalystBenchmarkResult as v, AnalystEvidenceResolver as w, AnalystEvidenceExpectation as x, AnalystBenchmarkRunner as y, ExternalOptimizerModelCall as z };
|
|
469
|
-
//# sourceMappingURL=agent-profile-
|
|
469
|
+
//# sourceMappingURL=agent-profile-DU5InUT9.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"agent-profile-
|
|
1
|
+
{"version":3,"file":"agent-profile-DU5InUT9.d.ts","names":["_$___cost_ledger0.CostReceiptInput"],"sources":["../src/campaign/external-optimizer-contracts.ts","../src/analyst/benchmark.ts","../src/agent-profile.ts"],"mappings":";;;;UAciB;;EAEf;;EAEA;;EAEA;;cAGW,2CAA2C,SAAS;UAOhD;;EAEf;;EAEA;;cAGW,4CAA4C,SAAS;UAMjD;EACf;EACA;EACA,MAAM,OAAO;;EAEb,SAAS,QAAQ;;KAGP;KAEA,iCAAiC;UAE5B;EACf,WAAW;EACX;;KAqDG,aAAa,KAAK,eAAc,6BACjC,IACA,0BAA0B,gBACf,aAAa,OACtB,+BACc,WAAW,IAAI,aAAa,EAAE,SAC1C;;KAGI,+BAA+B,aACzC,KAAK;EAA0B;;KAGrB;;UAMK;;WAEN;;WAEA,SAAS;;WAET,iBAAiB;WACjB,QAAQ;;;KAIP;WAEG;;WAEA,UAAU;;;;;;;WAOV,SAPUA;;WASV;;WAGA;;WAEA;;WAEA,SATkCA;;WAWlC;;;;;;;;;;;;KAaH,8BACV,SAAS,sCACN,QAAQ;;KAGD;WAEG;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;WAGA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;KAGH;;;;;;KAUA;WAEG;WACA;WACA,WAAW;WACX;;WAGA;WACA;WACA,WAAW;WACX;WACA;;WAEA;WACA;;WAGA;WACA;WACA,QAAQ;WACR,YAAY;WACZ;WACA;;UAGE;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;EAUA;;;;;;EAMA,UAAU;;EAEV;;;UAIe;;EAEf;;EAEA;;UCpQe;EACf;EACA,OAAO;;UAGQ;EACf;EACA;EACA;EACA;EACA,oBAAoB;EACpB;;EAEA,4BAA4B;;KAGlB;UAEK,qBAAqB;EACpC;;EAEA;;EAEA,YAAY;EACZ,OAAO;EACP,yBAAyB;;EAEzB,2BAA2B;EAC3B;EACA,WAAW;;UAGI;EACf;EACA;EACA;EACA;EACA;EACA,mBAAmB;EACnB;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;EACA;;UAGe;EACf,UAAU;EACV;EACA;;UAGe;EACf;EACA;EACA,oBAAoB;EACpB,QAAQ;;EAER;;KAGU,wBAAwB,qBAAqB;EACvD;EACA,WAAW;EACX,UAAU;EACV,SAAS;gBACK;;;;;iBAMA,2BAA2B,QACzC,WAAW,OAAO,WAAW,qBAC5B,wBAAwB;UAoBV;EACf,mBAAmB;EACnB,QAAQ;EACR,WAAW;;;;;EAKX;;EAEA,QAAQ;;UAGO;EACf;EACA;EACA;EACA;;UAGe,uBAAuB;EACtC;EACA,QACE,OAAO,QACP;IAAW;IAAgB;IAAoB,SAAS;MACvD,yBAAyB,QAAQ;;UAGrB;EACf;EACA;EACA;EACA,YAAY;EACZ;EACA;EACA;EACA;EACA,mBAAmB;EACnB,OAAO;EACP,qBAAqB;EACrB;EACA,eAAe;EACf,QAAQ;EACR,iBAAiB;EACjB,QAAQ;;UAGO;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;EACA,WAAW;EACX;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe;EACf;EACA,UAAU;EACV;EACA,cAAc;EACd,WAAW;;UAGI,mCAAmC;EAClD;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,YAAY;EACZ,cAAc;EACd,WAAW;;UAGI,2BAA2B;EAC1C,gBAAgB,qBAAqB;EACrC,kBAAkB,uBAAuB;EACzC;EACA;EACA;EACA,kBAAkB,wBAAwB;EAC1C,YAAY;;EAEZ,+BAA+B;EAC/B,iBAAiB,aAAa,uCAAuC;EACrE,SAAS;;iBAGW,oBAAoB,QACxC,SAAS,2BAA2B,UACnC,QAAQ;;;;;;;;cCnQE,2BAA2B;UAOvB;;;EAGf,MAAM;;EAEN,qBAAqB;;;EAGrB;;;;;;EAMA;;;;;;;cAQW;;;;;;;;;;;;;;;;;;;iBAoBG,kBAAkB,MAAM,kBAAkB;;;;;;;;iBAkD1C,cACd,SAAS,KAAK;EACX,SAAS;EAAa;;;;;;;;;iBAiBX,eAAe,SAAS;;;;;;;;;;;iBAmDxB,iBAAiB,SAAS"}
|
package/dist/analyst/index.d.ts
CHANGED
|
@@ -2,7 +2,7 @@ import { m as CustomTokenPricing, o as CostLedgerHandle } from "../cost-ledger-C
|
|
|
2
2
|
import { A as FindingParseOutcome, B as FindingTreatment, C as CalibrationInterval, D as FindingLayer, E as FindingEvidenceUnit, F as independentUnitCount, H as TreatmentLayer, I as lineageRoots, L as parseAnalystFinding, M as FindingScope, N as FindingScopeReading, O as FindingMeasure, P as ProposalFindingSchema, R as readFindingScope, S as makeProposalFinding, T as FindingCalibration, V as TREATMENT_LAYERS, Y as TraceAnalysisStore, _ as ExecutionProbeRequest, a as AnalystFindingSchemaVersion, b as computeFindingId, c as AnalystRunEvent, d as AnalystRunSummary, f as AnalystSeverity, g as ExecutionProbeOutcome, h as ExecutionProbe, i as AnalystFinding, j as FindingReaderLabel, k as FindingMethod, l as AnalystRunInputs, m as EvidenceRef, n as AnalystContext, o as AnalystInputKind, ot as TraceAnalystSpan, p as AnalystUsageReceipt, r as AnalystCost, s as AnalystRequirements, t as Analyst, u as AnalystRunResult, v as ProposalFinding, w as FINDING_LAYERS, x as makeFinding, y as ProposalFindingOrigin, z as FindingConditions } from "../types-kBMNdo1X.js";
|
|
3
3
|
import { A as ChatResponse, D as ChatCallOpts, F as OpenAiCompatibleTransportOpts, I as SandboxSdkTransportOpts, L as createChatClient, M as CreateChatClientOpts, N as CustomTransportOpts, O as ChatClient, P as MockTransportOpts, j as ChatTransport, k as ChatRequest, t as BootstrapInterval } from "../cluster-Cd0MRbr9.js";
|
|
4
4
|
import { Ft as NominalAgreementOptions } from "../index-DNXFzSpc.js";
|
|
5
|
-
import { A as traceStoreEvidenceResolver, C as AnalystEvidenceResolutionError, D as AnalystLatencyDistribution, E as AnalystIssueExpectation, G as ExternalOptimizerRunnerCommand, H as ExternalOptimizerModelExecutionObservation, O as RunAnalystBenchmarkOptions, S as AnalystEvidenceResolution, T as AnalystFindingScore, _ as AnalystBenchmarkProvenance, b as AnalystBenchmarkSummary, d as AnalystBenchmarkDatasetRef, f as AnalystBenchmarkDescriptor, g as AnalystBenchmarkOutput, h as AnalystBenchmarkObservation, k as runAnalystBenchmark, m as AnalystBenchmarkLabelState, p as AnalystBenchmarkError, t as AgentProfile, u as AnalystBenchmarkCase, v as AnalystBenchmarkResult, w as AnalystEvidenceResolver, x as AnalystEvidenceExpectation, y as AnalystBenchmarkRunner, z as ExternalOptimizerModelCall } from "../agent-profile-
|
|
5
|
+
import { A as traceStoreEvidenceResolver, C as AnalystEvidenceResolutionError, D as AnalystLatencyDistribution, E as AnalystIssueExpectation, G as ExternalOptimizerRunnerCommand, H as ExternalOptimizerModelExecutionObservation, O as RunAnalystBenchmarkOptions, S as AnalystEvidenceResolution, T as AnalystFindingScore, _ as AnalystBenchmarkProvenance, b as AnalystBenchmarkSummary, d as AnalystBenchmarkDatasetRef, f as AnalystBenchmarkDescriptor, g as AnalystBenchmarkOutput, h as AnalystBenchmarkObservation, k as runAnalystBenchmark, m as AnalystBenchmarkLabelState, p as AnalystBenchmarkError, t as AgentProfile, u as AnalystBenchmarkCase, v as AnalystBenchmarkResult, w as AnalystEvidenceResolver, x as AnalystEvidenceExpectation, y as AnalystBenchmarkRunner, z as ExternalOptimizerModelCall } from "../agent-profile-DU5InUT9.js";
|
|
6
6
|
import { C as BehavioralWindowPolicy, S as BehavioralTraceWindow, _ as BehavioralTraceEventMap, a as createTraceAnalyst, b as BehavioralTraceQuestionsOptions, c as BehavioralAnalystOptions, d as pinBehavioralFailureClass, f as BEHAVIORAL_TRACE_QUESTIONS, g as BehavioralTraceEvent, h as BehavioralTraceAnswer, i as TraceAnalystDefinition, l as assertFailureClasses, m as BehavioralQuestionCalibration, n as buildDefaultAnalystRegistry, o as runTraceAnalyst, p as BEHAVIORAL_TRACE_QUESTIONS_VERSION, r as CreateTraceAnalystOptions, s as BEHAVIORAL_FAILURE_CLASS, t as DefaultAnalystRegistryOptions, u as behavioralAnalyst, v as BehavioralTraceLayer, w as behavioralTraceQuestionDigest, x as BehavioralTraceSupport, y as BehavioralTraceQuestionId } from "../default-registry-aPvwtzEM.js";
|
|
7
7
|
import { a as ExactAnalystRunPolicySnapshot, c as ExactAnalystSnapshot, d as ExactExecutionComponentSnapshot, i as ExactAnalystRunEvent, l as ExactCapableAnalyst, n as ExactAnalystExecutionPlanSnapshot, o as ExactAnalystRunResult, r as ExactAnalystRunCompletion, s as ExactAnalystRunSummary, t as ExactAnalystBudgetSnapshot, u as ExactExecutionComponentIdentity } from "../exact-types-VAFjtOQN.js";
|
|
8
8
|
import { g as SystemOneState } from "../systemone-protocol-C4t7JH4Y.js";
|
|
@@ -10,7 +10,7 @@ import { B as DiffPolicy, F as FindingsTimeStep, H as FindingsStore, I as Ledger
|
|
|
10
10
|
import { a as RAW_FINDING_SCHEMA_PROMPT, c as RawAnalystFindingSchema, i as TraceAnalystLimits, l as evidenceRefsFromRawFinding, n as TraceAnalysisEngineRequest, o as RawAnalystEvidence, r as TraceAnalysisEngineResult, s as RawAnalystFinding, t as TraceAnalysisEngine, u as parseRawFinding } from "../engine-BF8naBn7.js";
|
|
11
11
|
import { n as buildTraceToolsForGroup, t as TraceToolGroupName } from "../tool-groups-R5RO9OHn.js";
|
|
12
12
|
import { a as BudgetPolicy, c as RegistryRunOpts, i as AnalystRegistryOptions, n as ExactAnalystRunExecutionError, o as ExactAnalystBudgetPolicy, r as AnalystHooks, s as ExactRegistryRunOpts, t as AnalystRegistry } from "../registry-BAcEbzKm.js";
|
|
13
|
-
import { T as SemanticConceptJudgeOptions, _ as DspyRlmTraceEngineOptions, a as FAILURE_MODE_KIND_SPEC, b as createChatTraceEngine, c as emitControlIntegrityFindings, d as FindingSubject, f as FindingSubjectKind, g as renderFindingSubject, h as parseFindingSubject, i as IMPROVEMENT_KIND_SPEC, l as FINDING_SUBJECT_KINDS, m as findingSubjectGrammarPromptFor, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as KIND_EXPECTED_SUBJECTS, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FINDING_SUBJECT_SYNTAX, v as createDspyRlmTraceEngine, y as ChatTraceEngineOptions } from "../index-
|
|
13
|
+
import { T as SemanticConceptJudgeOptions, _ as DspyRlmTraceEngineOptions, a as FAILURE_MODE_KIND_SPEC, b as createChatTraceEngine, c as emitControlIntegrityFindings, d as FindingSubject, f as FindingSubjectKind, g as renderFindingSubject, h as parseFindingSubject, i as IMPROVEMENT_KIND_SPEC, l as FINDING_SUBJECT_KINDS, m as findingSubjectGrammarPromptFor, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as KIND_EXPECTED_SUBJECTS, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FINDING_SUBJECT_SYNTAX, v as createDspyRlmTraceEngine, y as ChatTraceEngineOptions } from "../index-QLDWZTxI.js";
|
|
14
14
|
import { C as runPrimeExchange, D as ReplyRowDecoded, E as ReplyEnvelope, O as ReplyRowRejection, S as projectPrimeTrajectory, T as ReplyContract, _ as buildPrimePrompt, a as PrimeProjectionOutcome, b as normalizePrimeUsage, c as PrimeProtocolIdentity, d as PrimeRepairPromptSpec, f as PrimeRepairState, g as analystUsageReceiptFromPrimeUsage, h as PrimeTurnRecord, i as PrimeProjectionDelivery, l as PrimeRawUsage, m as PrimeRowDecoded, n as PrimeExchangeOutcome, o as PrimeProjectionSource, p as PrimeReplyContract, r as PrimeFailure, s as PrimePromptSpec, t as PrimeExchangeOptions, u as PrimeRejectedRow, v as buildPrimeRepairPrompt, w as DecodedReply, x as primeProtocolSha256, y as mergePrimeRawUsage } from "../prime-protocol-D1NkJRuD.js";
|
|
15
15
|
import { i as nodeHttpPrimeBridgeTransport, n as PrimeBridgeTransportRequest, r as PrimeBridgeTransportResult, t as PrimeBridgeTransport } from "../prime-bridge-transport-Cfqir_Im.js";
|
|
16
16
|
import { b as SupervisorRunSources } from "../types-DUVhW5V1.js";
|
package/dist/analyst/index.js
CHANGED
|
@@ -11,7 +11,7 @@ import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-cWyn8G1a.js";
|
|
|
11
11
|
import { n as diffFindings, r as createChatTraceEngine, t as FindingsStore } from "../findings-store-BSuaV3dN.js";
|
|
12
12
|
import { m as validDecisionIdentity, o as systemOneAnalyst, t as DECISION_POINT_AUC_BAR } from "../decision-point-Di2LATNH.js";
|
|
13
13
|
import { t as createChatClient } from "../chat-client-Zz97O3NB.js";
|
|
14
|
-
import { A as agentRxBenchmarkCase, C as ANALYST_BENCHMARK_COST_LEDGER_FILE, D as AGENT_RX_UPSTREAM_REVISION, E as ANALYST_BENCHMARK_OBSERVATIONS_FILE, F as runAnalystBenchmark, I as traceStoreEvidenceResolver, M as normalizeAgentRxCategory, N as externalId, O as codeTraceBenchCase, P as positiveStep, S as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, T as ANALYST_BENCHMARK_MANIFEST_FILE, _ as readAnalystBenchmarkArtifact, a as runInlineAnalystDefinition, b as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, c as publicRlmAnalystDefinition, d as publicDirectAnalystDefinition, f as runChunkedAnalystDefinition, g as expandCodeTraceFailureBlocks, h as analystDefinitionProtocolSha256, i as primeCodeTraceAnalystDefinition, j as agentRxPredictionsToFindings, k as codeTracerPredictionsToFindings, l as runReplVariableAnalystDefinition, m as analystDefinitionAsymmetries, n as createPrimeBenchmarkRunner, o as loadPublicBenchmarkRows, p as AnalystExpressivenessError, r as primeAnalystProtocolSha256, s as createPublicBenchmarkRlmRunner, t as runAnalystBenchmarkCommand, u as createPublicBenchmarkDirectRunner, v as CODE_TRACE_BENCH_ANALYST_PROMPT, w as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, x as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, y as publicBenchmarkProtocolSha256 } from "../benchmark-command-
|
|
14
|
+
import { A as agentRxBenchmarkCase, C as ANALYST_BENCHMARK_COST_LEDGER_FILE, D as AGENT_RX_UPSTREAM_REVISION, E as ANALYST_BENCHMARK_OBSERVATIONS_FILE, F as runAnalystBenchmark, I as traceStoreEvidenceResolver, M as normalizeAgentRxCategory, N as externalId, O as codeTraceBenchCase, P as positiveStep, S as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, T as ANALYST_BENCHMARK_MANIFEST_FILE, _ as readAnalystBenchmarkArtifact, a as runInlineAnalystDefinition, b as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, c as publicRlmAnalystDefinition, d as publicDirectAnalystDefinition, f as runChunkedAnalystDefinition, g as expandCodeTraceFailureBlocks, h as analystDefinitionProtocolSha256, i as primeCodeTraceAnalystDefinition, j as agentRxPredictionsToFindings, k as codeTracerPredictionsToFindings, l as runReplVariableAnalystDefinition, m as analystDefinitionAsymmetries, n as createPrimeBenchmarkRunner, o as loadPublicBenchmarkRows, p as AnalystExpressivenessError, r as primeAnalystProtocolSha256, s as createPublicBenchmarkRlmRunner, t as runAnalystBenchmarkCommand, u as createPublicBenchmarkDirectRunner, v as CODE_TRACE_BENCH_ANALYST_PROMPT, w as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, x as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, y as publicBenchmarkProtocolSha256 } from "../benchmark-command-DkUpd9Q6.js";
|
|
15
15
|
import { a as normalizePrimeUsage, c as runPrimeExchange, i as mergePrimeRawUsage, n as buildPrimePrompt, o as primeProtocolSha256, r as buildPrimeRepairPrompt, s as projectPrimeTrajectory, t as analystUsageReceiptFromPrimeUsage } from "../prime-protocol-CrhpXRz9.js";
|
|
16
16
|
import { t as nodeHttpPrimeBridgeTransport } from "../prime-bridge-transport-Csd5SQg7.js";
|
|
17
17
|
import { t as clusterBootstrapMean } from "../cluster-bootstrap-B0MTOBRU.js";
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { m as CustomTokenPricing, t as CostLedger } from "../cost-ledger-Czro9wmo.js";
|
|
2
|
-
import "../index-
|
|
2
|
+
import "../index-YS58LHlx.js";
|
|
3
3
|
import { O as ChatClient, j as ChatTransport } from "../cluster-Cd0MRbr9.js";
|
|
4
4
|
import { d as SystemOneRequest, g as SystemOneState, s as SystemOneQuestions } from "../systemone-protocol-C4t7JH4Y.js";
|
|
5
5
|
import "../systemone-DVk-Crtu.js";
|
|
@@ -1861,7 +1861,7 @@ Object.freeze([
|
|
|
1861
1861
|
"package.json",
|
|
1862
1862
|
"pnpm-lock.yaml"
|
|
1863
1863
|
]);
|
|
1864
|
-
const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "
|
|
1864
|
+
const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "e2d8beb075fcb4c4ec41d103b4b1a8c677ee0f5253097197868208cf94387cf0";
|
|
1865
1865
|
const ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256 = "1e03f2daed356d60316aabefb407ec1e437ac94d408d61eea4ae096e9c6fbb5b";
|
|
1866
1866
|
const ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256 = "4dba263b6256a30d56c7fdb2d992d3a953c0035d731f359b704db806f68f75ac";
|
|
1867
1867
|
const ANALYST_BENCHMARK_IMPLEMENTATION_FILES = Object.freeze([
|
|
@@ -6855,4 +6855,4 @@ function shellQuote(value) {
|
|
|
6855
6855
|
}
|
|
6856
6856
|
export { agentRxBenchmarkCase as A, ANALYST_BENCHMARK_COST_LEDGER_FILE as C, AGENT_RX_UPSTREAM_REVISION as D, ANALYST_BENCHMARK_OBSERVATIONS_FILE as E, runAnalystBenchmark as F, traceStoreEvidenceResolver as I, normalizeAgentRxCategory as M, externalId as N, codeTraceBenchCase as O, positiveStep as P, ANALYST_BENCHMARK_IMPLEMENTATION_FILES as S, ANALYST_BENCHMARK_MANIFEST_FILE as T, readAnalystBenchmarkArtifact as _, runInlineAnalystDefinition as a, ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256 as b, publicRlmAnalystDefinition as c, publicDirectAnalystDefinition as d, runChunkedAnalystDefinition as f, expandCodeTraceFailureBlocks as g, analystDefinitionProtocolSha256 as h, primeCodeTraceAnalystDefinition as i, agentRxPredictionsToFindings as j, codeTracerPredictionsToFindings as k, runReplVariableAnalystDefinition as l, analystDefinitionAsymmetries as m, createPrimeBenchmarkRunner as n, loadPublicBenchmarkRows as o, AnalystExpressivenessError as p, primeAnalystProtocolSha256 as r, createPublicBenchmarkRlmRunner as s, runAnalystBenchmarkCommand as t, createPublicBenchmarkDirectRunner as u, CODE_TRACE_BENCH_ANALYST_PROMPT as v, ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE as w, ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256 as x, publicBenchmarkProtocolSha256 as y };
|
|
6857
6857
|
|
|
6858
|
-
//# sourceMappingURL=benchmark-command-
|
|
6858
|
+
//# sourceMappingURL=benchmark-command-DkUpd9Q6.js.map
|