@tangle-network/agent-eval 0.123.1 → 0.123.2
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 +5 -0
- package/README.md +151 -161
- package/dist/analyst/index.d.ts +9 -1
- package/dist/analyst/index.js +5 -5
- package/dist/authenticity/index.js +3 -2
- package/dist/authenticity/index.js.map +1 -1
- package/dist/benchmarks/index.d.ts +2 -1
- package/dist/benchmarks/index.js +6 -6
- package/dist/campaign/index.d.ts +29 -33
- package/dist/campaign/index.js +6 -6
- package/dist/{chunk-A5S77LSE.js → chunk-4SOQ4ND2.js} +2 -2
- package/dist/{chunk-VJ7T5WIO.js → chunk-5YMKIFYP.js} +3 -3
- package/dist/{chunk-U5CHZ5M3.js → chunk-DNVPOYUS.js} +4 -4
- package/dist/{chunk-6WX7CBAR.js → chunk-E3HAD4A3.js} +19 -8
- package/dist/chunk-E3HAD4A3.js.map +1 -0
- package/dist/{chunk-LBAHQOBI.js → chunk-EBDOTTZJ.js} +37 -11
- package/dist/chunk-EBDOTTZJ.js.map +1 -0
- package/dist/{chunk-XJYR7XFV.js → chunk-GC4ATIKK.js} +1 -1
- package/dist/chunk-GC4ATIKK.js.map +1 -0
- package/dist/{chunk-HZJF4IUO.js → chunk-HQY7LBV2.js} +3 -3
- package/dist/{chunk-NJC7U437.js → chunk-J7S4YM27.js} +6 -5
- package/dist/chunk-J7S4YM27.js.map +1 -0
- package/dist/{chunk-S3UZOQ5Y.js → chunk-LOW3U7JZ.js} +2 -2
- package/dist/{chunk-OYZAPX5G.js → chunk-R226UZOI.js} +2 -2
- package/dist/{chunk-GS3FJGUF.js → chunk-RQP5UTK5.js} +120 -14
- package/dist/chunk-RQP5UTK5.js.map +1 -0
- package/dist/{chunk-G2GPNLSX.js → chunk-WMJR67FX.js} +3 -3
- package/dist/{chunk-FC5NDO3E.js → chunk-WXQTVEKM.js} +3 -3
- package/dist/cli.js +100 -10
- package/dist/cli.js.map +1 -1
- package/dist/contract/index.d.ts +97 -5
- package/dist/contract/index.js +9 -7
- package/dist/contract/index.js.map +1 -1
- package/dist/control.js +3 -3
- package/dist/fuzz.js +3 -2
- package/dist/fuzz.js.map +1 -1
- package/dist/hosted/index.d.ts +8 -2
- package/dist/index.d.ts +10 -2
- package/dist/index.js +13 -13
- package/dist/index.js.map +1 -1
- package/dist/openapi.json +1 -1
- package/dist/rl.d.ts +9 -1
- package/dist/rl.js +4 -4
- package/dist/storyboard/index.js +1 -1
- package/dist/storyboard/index.js.map +1 -1
- package/dist/traces.js +3 -3
- package/dist/wire/index.d.ts +61 -4
- package/dist/wire/index.js +2 -2
- package/docs/adapters-observability.md +6 -6
- package/docs/building-doctrine.md +5 -5
- package/docs/concepts.md +29 -29
- package/docs/customer-journeys.md +80 -155
- package/docs/design/loop-taxonomy.md +26 -27
- package/docs/design.md +70 -0
- package/docs/distributed-driver.md +14 -14
- package/docs/eval-surface-map.md +11 -11
- package/docs/hosted-ingest-spec.md +4 -4
- package/docs/improvement-glossary.md +38 -38
- package/docs/insight-report.md +32 -27
- package/docs/multi-shot-optimization.md +8 -8
- package/docs/research-report-methodology.md +9 -9
- package/docs/self-improvement-map.md +13 -13
- package/docs/trace-analysis.md +2 -2
- package/docs/wire-protocol.md +16 -16
- package/package.json +2 -1
- package/dist/chunk-6WX7CBAR.js.map +0 -1
- package/dist/chunk-GS3FJGUF.js.map +0 -1
- package/dist/chunk-LBAHQOBI.js.map +0 -1
- package/dist/chunk-NJC7U437.js.map +0 -1
- package/dist/chunk-XJYR7XFV.js.map +0 -1
- package/docs/auto-research-loop-end-to-end.md +0 -186
- /package/dist/{chunk-A5S77LSE.js.map → chunk-4SOQ4ND2.js.map} +0 -0
- /package/dist/{chunk-VJ7T5WIO.js.map → chunk-5YMKIFYP.js.map} +0 -0
- /package/dist/{chunk-U5CHZ5M3.js.map → chunk-DNVPOYUS.js.map} +0 -0
- /package/dist/{chunk-HZJF4IUO.js.map → chunk-HQY7LBV2.js.map} +0 -0
- /package/dist/{chunk-S3UZOQ5Y.js.map → chunk-LOW3U7JZ.js.map} +0 -0
- /package/dist/{chunk-OYZAPX5G.js.map → chunk-R226UZOI.js.map} +0 -0
- /package/dist/{chunk-G2GPNLSX.js.map → chunk-WMJR67FX.js.map} +0 -0
- /package/dist/{chunk-FC5NDO3E.js.map → chunk-WXQTVEKM.js.map} +0 -0
package/dist/control.js
CHANGED
|
@@ -10,11 +10,11 @@ import {
|
|
|
10
10
|
stopOnNoProgress,
|
|
11
11
|
stopOnRepeatedAction,
|
|
12
12
|
subjectiveEval
|
|
13
|
-
} from "./chunk-
|
|
13
|
+
} from "./chunk-R226UZOI.js";
|
|
14
14
|
import "./chunk-VQMK5FMP.js";
|
|
15
|
-
import "./chunk-
|
|
15
|
+
import "./chunk-LOW3U7JZ.js";
|
|
16
16
|
import "./chunk-MA6HLL3S.js";
|
|
17
|
-
import "./chunk-
|
|
17
|
+
import "./chunk-GC4ATIKK.js";
|
|
18
18
|
import "./chunk-VSMTAMNK.js";
|
|
19
19
|
import "./chunk-ONWEPEDO.js";
|
|
20
20
|
import "./chunk-PZ5AY32C.js";
|
package/dist/fuzz.js
CHANGED
|
@@ -63,8 +63,9 @@ function buildCoverage(cells, log, threshold) {
|
|
|
63
63
|
const dimSamples = {};
|
|
64
64
|
for (const r of recs) {
|
|
65
65
|
for (const [k, v] of Object.entries(r.ev.scores ?? {})) {
|
|
66
|
-
;
|
|
67
|
-
|
|
66
|
+
const samples = dimSamples[k] ?? [];
|
|
67
|
+
samples.push(v);
|
|
68
|
+
dimSamples[k] = samples;
|
|
68
69
|
}
|
|
69
70
|
}
|
|
70
71
|
const dimensions = {};
|
package/dist/fuzz.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/fuzz/cube.ts","../src/fuzz/capsule.ts","../src/fuzz/policies.ts","../src/fuzz/explorer.ts","../src/fuzz/fuzz-agent.ts","../src/fuzz/gates.ts","../src/fuzz/tools.ts"],"sourcesContent":["/**\n * Input-space tiling + coverage projection.\n *\n * Cells are the cartesian product of the input axes — the stratification plan,\n * enumerable up front so the planned-vs-covered denominator is honest. Coverage\n * is projected from the evaluation log: per cell, the full DISTRIBUTION of the\n * headline score, of each scored dimension, and of evaluation latency — a bare\n * mean hides outliers, so every aggregate carries its spread. Per-cell cost is\n * split known-dollars vs unknown-runs, never folded into a fabricated $0.\n */\n\nimport type { BehaviorSpace, Cell, CoverageCell, Distribution, Evaluation } from './types'\n\n/** One recorded evaluation — the unit coverage and the capsule are built from. */\nexport interface EvalRecord {\n cell: Cell\n ev: Evaluation\n /** The objective's interest score for this evaluation. */\n interest: number\n /** Evaluation wall-clock — engine-measured unless `ev.latencyMs` overrode it. */\n latencyMs: number\n /** Known dollars for this run. `null` = cost tracking was wired but this\n * run's cost was unknowable (counted apart). Absent = not tracked at all. */\n costUsd?: number | null\n}\n\n/** Enumerate every input cell (cartesian product of the axes), in stable order. */\nexport function enumerateCells(space: BehaviorSpace): Cell[] {\n if (space.axes.length === 0) return []\n let partials: Array<Record<string, string>> = [{}]\n for (const axis of space.axes) {\n const next: Array<Record<string, string>> = []\n for (const partial of partials) {\n for (const value of axis.values) next.push({ ...partial, [axis.name]: value })\n }\n partials = next\n }\n return partials.map((coords) => ({ id: cellId(space, coords), coords }))\n}\n\n/** Deterministic id for a coordinate map, e.g. `matterType=nda|difficulty=hard`. */\nexport function cellId(space: BehaviorSpace, coords: Record<string, string>): string {\n return space.axes.map((a) => `${a.name}=${coords[a.name]}`).join('|')\n}\n\n/** Nearest-rank percentile on a pre-sorted ascending sample. */\nfunction percentile(sorted: number[], p: number): number {\n const idx = Math.min(sorted.length - 1, Math.max(0, Math.ceil(p * sorted.length) - 1))\n return sorted[idx] as number\n}\n\n/** Summarize a sample. Throws on an empty sample — callers represent \"no data\"\n * as `null`, never as a zeroed distribution. */\nexport function distribution(values: number[]): Distribution {\n if (values.length === 0)\n throw new Error('distribution: empty sample — represent missing data as null, not zeros')\n const sorted = [...values].sort((a, b) => a - b)\n const mean = sorted.reduce((a, b) => a + b, 0) / sorted.length\n return {\n mean,\n median: percentile(sorted, 0.5),\n p90: percentile(sorted, 0.9),\n min: sorted[0] as number,\n max: sorted[sorted.length - 1] as number,\n n: sorted.length,\n }\n}\n\n/**\n * Project the evaluation log into the per-input-cell coverage map. A cell with\n * no evaluations reports `score: null` (honestly uncovered), never zeros.\n */\nexport function buildCoverage(cells: Cell[], log: EvalRecord[], threshold: number): CoverageCell[] {\n const byCell = new Map<string, EvalRecord[]>()\n for (const r of log) {\n const arr = byCell.get(r.cell.id) ?? []\n arr.push(r)\n byCell.set(r.cell.id, arr)\n }\n return cells.map((cell) => {\n const recs = byCell.get(cell.id) ?? []\n const runs = recs.length\n if (runs === 0)\n return { cell, runs: 0, score: null, findingRate: 0, dimensions: {}, latencyMs: null }\n\n const score = distribution(recs.map((r) => r.ev.score))\n const latencyMs = distribution(recs.map((r) => r.latencyMs))\n const findingRate = recs.filter((r) => r.interest >= threshold).length / runs\n\n const dimSamples: Record<string, number[]> = {}\n for (const r of recs) {\n for (const [k, v] of Object.entries(r.ev.scores ?? {})) {\n ;(dimSamples[k] ??= []).push(v)\n }\n }\n const dimensions: Record<string, Distribution> = {}\n for (const [k, xs] of Object.entries(dimSamples)) dimensions[k] = distribution(xs)\n\n // Cost fields appear only when tracking was wired: known dollars sum, and\n // tracked-but-unknown runs counted apart — never folded in as $0.\n const tracked = recs.filter((r) => r.costUsd !== undefined)\n const known = tracked.filter((r) => r.costUsd !== null)\n const cost =\n tracked.length > 0\n ? {\n costUsd: known.reduce((a, r) => a + (r.costUsd as number), 0),\n ...(tracked.length > known.length\n ? { costUnknownRuns: tracked.length - known.length }\n : {}),\n }\n : {}\n\n return { cell, runs, score, findingRate, dimensions, latencyMs, ...cost }\n })\n}\n","/**\n * The capsule — the artifact every exploration produces.\n *\n * `buildCapsule` assembles coverage + verified findings + the QD archive into a\n * pure `CapsuleData` (no clock, no I/O — deterministic and snapshot-testable).\n * `renderCapsuleHtml` turns it into a standalone page: the input-cell heat-map\n * (planned vs covered), per-dimension weakness chips, and the minimized finding\n * exemplars. One artifact — the hardening map and the shareable proof object.\n */\n\nimport type { EvalRecord } from './cube'\nimport { buildCoverage, distribution } from './cube'\nimport type { ArchiveEntry, CapsuleData, Cell, CoverageCell, Distribution, Finding } from './types'\n\nexport interface BuildCapsuleInput<S> {\n target: string\n objective: string\n cells: Cell[]\n log: EvalRecord[]\n /** The objective's notable threshold — drives findingRate. */\n threshold: number\n archive: ArchiveEntry<S>[]\n findings: Finding<S>[]\n candidateFindings: number\n runsUsed: number\n /** Known-dollar / unknown-run split — present only when cost tracking was\n * wired; the capsule never fabricates a $0 total. */\n cost?: { costUsd: number; costUnknownRuns: number }\n /** Evaluations that threw — infra outcomes, never folded into robustness. */\n evalErrors: number\n /** Set when the consecutive-error circuit breaker stopped the run early. */\n stoppedEarly?: { reason: 'eval-errors'; detail: string }\n}\n\nexport function buildCapsule<S>(input: BuildCapsuleInput<S>): CapsuleData<S> {\n const coverage = buildCoverage(input.cells, input.log, input.threshold)\n const covered = coverage.filter((c) => c.runs > 0)\n // Cells weigh equally: variance steering sends more runs to weak cells, so a\n // run-weighted average would bias the headline low.\n const robustness =\n covered.length === 0 ? null : distribution(covered.map((c) => (c.score as Distribution).mean))\n const latencyMs = input.log.length === 0 ? null : distribution(input.log.map((r) => r.latencyMs))\n // Measured-descriptor bins beyond the bare input cell — observed, never planned.\n const behaviorBinsObserved = input.archive.filter((e) => e.binId !== e.cell.id).length\n\n return {\n target: input.target,\n objective: input.objective,\n coverage,\n findings: [...input.findings].sort((a, b) => b.interest - a.interest),\n archive: [...input.archive].sort((a, b) => b.interest - a.interest),\n stats: {\n totalRuns: input.runsUsed,\n cellsTotal: input.cells.length,\n cellsCovered: covered.length,\n behaviorBinsObserved,\n candidateFindings: input.candidateFindings,\n verifiedFindings: input.findings.length,\n robustness,\n latencyMs,\n ...(input.cost\n ? { costUsd: input.cost.costUsd, costUnknownRuns: input.cost.costUnknownRuns }\n : {}),\n evalErrors: input.evalErrors,\n ...(input.stoppedEarly ? { stoppedEarly: input.stoppedEarly } : {}),\n },\n }\n}\n\n// ── HTML capsule ──────────────────────────────────────────────────────────────\n\nfunction esc(s: string): string {\n return s.replace(\n /[&<>\"]/g,\n (c) => ({ '&': '&', '<': '<', '>': '>', '\"': '"' })[c] as string,\n )\n}\n\n/** red (0) → amber (.5) → green (1); uncovered cells render gray. */\nfunction robustnessColor(r: number | null): string {\n if (r == null) return '#2a2a2e'\n return `hsl(${Math.round(r * 120)} 70% 42%)`\n}\n\nfunction pct(x: number): string {\n return `${Math.round(x * 100)}%`\n}\n\nfunction deriveAxes(coverage: CoverageCell[]): Array<{ name: string; values: string[] }> {\n const order: string[] = []\n const seen = new Map<string, Set<string>>()\n for (const c of coverage) {\n for (const [k, v] of Object.entries(c.cell.coords)) {\n if (!seen.has(k)) {\n seen.set(k, new Set())\n order.push(k)\n }\n seen.get(k)?.add(v)\n }\n }\n return order.map((name) => ({ name, values: [...(seen.get(name) ?? [])] }))\n}\n\n/** The weakest dimension chip for a cell, e.g. `safety 32%` — shown when scores exist. */\nfunction weakestDim(c: CoverageCell): string {\n const entries = Object.entries(c.dimensions).map(([k, d]) => [k, d.mean] as [string, number])\n if (entries.length === 0) return ''\n const sorted = entries.sort((a, b) => a[1] - b[1])\n const w = sorted[0]\n if (!w) return ''\n return `<span class=\"dim\">${esc(w[0])} ${pct(w[1])}</span>`\n}\n\nfunction heatmapHtml(coverage: CoverageCell[]): string {\n const axes = deriveAxes(coverage)\n const byId = new Map(coverage.map((c) => [c.cell.id, c]))\n const tile = (c: CoverageCell | undefined, label: string): string => {\n const r = c?.score?.mean ?? null\n const title = c?.score\n ? `${pct(c.score.mean)} robust (median ${pct(c.score.median)}, min ${pct(c.score.min)}) · ${c.runs} runs · ${pct(c.findingRate)} flagged${c.latencyMs ? ` · ${(c.latencyMs.median / 1000).toFixed(1)}s median` : ''}${c.costUsd !== undefined ? ` · $${c.costUsd.toFixed(2)}` : ''}`\n : 'not covered'\n return `<div class=\"tile\" style=\"background:${robustnessColor(r)}\" title=\"${esc(title)}\">${label ? `<span class=\"tl\">${esc(label)}</span>` : ''}<span class=\"tv\">${c && r != null ? pct(r) : '—'}</span>${c ? weakestDim(c) : ''}</div>`\n }\n\n const rowAxis = axes[0]\n const colAxis = axes[1]\n if (axes.length === 2 && rowAxis && colAxis) {\n const head = `<tr><th></th>${colAxis.values.map((v) => `<th>${esc(v)}</th>`).join('')}</tr>`\n const rows = rowAxis.values\n .map((rv) => {\n const cells = colAxis.values\n .map((cv) => {\n const id = `${rowAxis.name}=${rv}|${colAxis.name}=${cv}`\n return `<td>${tile(byId.get(id), '')}</td>`\n })\n .join('')\n return `<tr><th class=\"rh\">${esc(rv)}</th>${cells}</tr>`\n })\n .join('')\n return `<div class=\"axis-label\">rows: <b>${esc(rowAxis.name)}</b> · cols: <b>${esc(colAxis.name)}</b></div><table class=\"heat\">${head}${rows}</table>`\n }\n\n const sorted = [...coverage].sort((a, b) => (a.score?.mean ?? 2) - (b.score?.mean ?? 2))\n return `<div class=\"grid\">${sorted.map((c) => tile(c, Object.values(c.cell.coords).join(' · '))).join('')}</div>`\n}\n\nfunction findingsHtml<S>(findings: Finding<S>[], limit: number): string {\n if (findings.length === 0)\n return `<p class=\"none\">No verified findings — the target held across every covered cell.</p>`\n return findings\n .slice(0, limit)\n .map((f) => {\n const coords = Object.entries(f.cell.coords)\n .map(([k, v]) => `${k}:${v}`)\n .join(' · ')\n const labels = (f.evaluation.labels ?? [])\n .map((l) => `<span class=\"fclass\">${esc(l)}</span>`)\n .join('')\n const dims = Object.entries(f.evaluation.scores ?? {})\n .sort((a, b) => a[1] - b[1])\n .slice(0, 3)\n .map(([k, v]) => `<span class=\"dim\">${esc(k)} ${pct(v)}</span>`)\n .join('')\n return `<div class=\"fail\"><div class=\"fmeta\"><span class=\"fcell\">${esc(coords)}</span>${labels}<span class=\"fsev\">interest ${pct(f.interest)}</span></div><div class=\"sevbar\"><div class=\"sevfill\" style=\"width:${pct(f.interest)}\"></div></div><div class=\"ftext\">${esc(f.text ?? '(scenario text not captured)')}</div>${dims ? `<div class=\"fdims\">${dims}</div>` : ''}</div>`\n })\n .join('')\n}\n\nexport interface RenderCapsuleOptions {\n /** Max finding exemplars to show. Default 8. */\n maxFindings?: number\n /** ISO timestamp to stamp into the page (keeps the pure capsule clock-free). */\n generatedAt?: string\n}\n\n/** Render a self-contained HTML capsule — heat-map + per-dimension chips + verified findings. */\nexport function renderCapsuleHtml<S>(\n capsule: CapsuleData<S>,\n opts: RenderCapsuleOptions = {},\n): string {\n const s = capsule.stats\n const kpi = (label: string, value: string, accent = '#e6e6e6'): string =>\n `<div class=\"kpi\"><div class=\"kv\" style=\"color:${accent}\">${esc(value)}</div><div class=\"kl\">${esc(label)}</div></div>`\n const lift = capsule.lift\n ? kpi(\n 'hardening lift',\n `${capsule.lift.before.toFixed(2)} → ${capsule.lift.after.toFixed(2)}`,\n '#5ad17a',\n )\n : ''\n // Cost KPI only when tracking was wired — an untracked run never shows $0.\n // Unpriced runs are named in the label (amber): the total is a lower bound.\n const cost =\n s.costUsd !== undefined\n ? kpi(\n s.costUnknownRuns ? `cost · ${s.costUnknownRuns} runs unpriced` : 'cost',\n `$${s.costUsd.toFixed(2)}`,\n s.costUnknownRuns ? '#e5c07b' : '#e6e6e6',\n )\n : ''\n const stamp = opts.generatedAt ?? capsule.generatedAt ?? ''\n\n return `<!doctype html><html><head><meta charset=\"utf-8\"><meta name=\"viewport\" content=\"width=device-width,initial-scale=1\">\n<title>${esc(capsule.objective)} capsule — ${esc(capsule.target)}</title>\n<style>\n:root{color-scheme:dark}\n*{box-sizing:border-box}\nbody{margin:0;background:#0c0c0f;color:#e6e6e6;font:14px/1.5 ui-sans-serif,system-ui,-apple-system,Segoe UI,Roboto,sans-serif}\n.wrap{max-width:980px;margin:0 auto;padding:40px 24px}\nh1{font-size:22px;margin:0 0 2px;letter-spacing:-.01em}\n.sub{color:#8a8a93;font-size:13px;margin-bottom:28px}\n.kpis{display:flex;gap:14px;flex-wrap:wrap;margin-bottom:32px}\n.kpi{background:#16161b;border:1px solid #24242b;border-radius:12px;padding:14px 18px;min-width:120px}\n.kv{font-size:24px;font-weight:650;letter-spacing:-.02em}\n.kl{color:#8a8a93;font-size:12px;margin-top:2px}\nh2{font-size:13px;text-transform:uppercase;letter-spacing:.08em;color:#8a8a93;margin:34px 0 14px}\n.axis-label{color:#8a8a93;font-size:12px;margin-bottom:10px}\ntable.heat{border-collapse:separate;border-spacing:6px}\ntable.heat th{font-weight:500;color:#a8a8b0;font-size:12px;padding:4px 8px;text-align:center}\ntable.heat th.rh{text-align:right}\n.tile{position:relative;width:104px;height:66px;border-radius:10px;display:flex;flex-direction:column;justify-content:center;align-items:center;gap:1px;color:#06120a;font-weight:600}\n.tile .tv{font-size:17px}\n.tile .tl{font-size:10px;font-weight:600;opacity:.9;padding:0 6px;text-align:center}\n.tile .dim{font-size:9px;font-weight:600;opacity:.85;background:rgba(0,0,0,.22);padding:0 6px;border-radius:999px}\n.grid{display:flex;flex-wrap:wrap;gap:8px}\n.grid .tile{width:138px;height:74px}\n.fail{background:#16161b;border:1px solid #24242b;border-left:3px solid #d1495b;border-radius:10px;padding:12px 14px;margin-bottom:10px}\n.fmeta{display:flex;gap:10px;align-items:center;font-size:12px;color:#a8a8b0;margin-bottom:8px}\n.fcell{color:#cfcfd6}\n.fclass{background:#2a1a1d;color:#e58a96;padding:1px 8px;border-radius:999px;font-size:11px}\n.fsev{margin-left:auto;color:#e58a96}\n.sevbar{height:4px;background:#24242b;border-radius:2px;overflow:hidden;margin-bottom:10px}\n.sevfill{height:100%;background:#d1495b}\n.ftext{font-size:13px;color:#d8d8df;white-space:pre-wrap}\n.fdims{display:flex;gap:6px;margin-top:8px}\n.fdims .dim{background:#1d2530;color:#9fc1e8;padding:1px 8px;border-radius:999px;font-size:11px}\n.none{color:#5ad17a}\n.foot{color:#5a5a63;font-size:11px;margin-top:36px}\n</style></head><body><div class=\"wrap\">\n<h1>${esc(capsule.objective)} exploration · ${esc(capsule.target)}</h1>\n<div class=\"sub\">${s.totalRuns} scenarios across ${s.cellsCovered}/${s.cellsTotal} planned cells${s.behaviorBinsObserved > 0 ? ` · ${s.behaviorBinsObserved} measured behavior bins` : ''}${stamp ? ` · ${esc(stamp)}` : ''}</div>\n<div class=\"kpis\">\n${s.robustness ? kpi('robustness', `${pct(s.robustness.mean)}`, s.robustness.mean < 0.6 ? '#e58a96' : '#5ad17a') : ''}\n${s.robustness ? kpi('cell spread', `${pct(s.robustness.min)}–${pct(s.robustness.max)}`) : ''}\n${s.latencyMs ? kpi('median latency', `${(s.latencyMs.median / 1000).toFixed(1)}s`, s.latencyMs.p90 > 4 * s.latencyMs.median ? '#e5b566' : '#e6e6e6') : ''}\n${kpi('verified findings', String(s.verifiedFindings), s.verifiedFindings > 0 ? '#e58a96' : '#5ad17a')}\n${kpi('cells covered', `${s.cellsCovered}/${s.cellsTotal}`)}\n${kpi('scenarios run', String(s.totalRuns))}\n${cost}\n${s.evalErrors > 0 ? kpi('eval errors', String(s.evalErrors), '#e5b566') : ''}\n${lift}\n</div>\n${s.stoppedEarly ? `<div class=\"sub\" style=\"color:#e5b566\">stopped early: ${esc(s.stoppedEarly.detail)} — coverage below reflects the completed portion only</div>` : ''}\n<h2>Coverage map</h2>\n${heatmapHtml(capsule.coverage)}\n<h2>Verified findings${s.candidateFindings > s.verifiedFindings ? ` · ${s.verifiedFindings} of ${s.candidateFindings} candidates passed the validity gates` : ''}</h2>\n${findingsHtml(capsule.findings, opts.maxFindings ?? 8)}\n<div class=\"foot\">Findings are gate-verified: each is a fair, answerable task that reproduces under a meaning-preserving rephrase, minimized to the smallest trigger.</div>\n</div></body></html>`\n}\n","/**\n * Shipped policies for the exploration engine.\n *\n * `MutationProposer` is a plain function type — an agent running a generator skill\n * IS a proposer (`(ctx) => dispatchToSkill(ctx)`), no wrapper needed. `mutationProposer`\n * builds the deterministic, LLM-free one from mutation operators. Objectives are\n * interfaces because the engine reads `kind` + `threshold` off them.\n */\n\nimport type { AdversarialMutation } from '../rl/adversarial'\nimport type {\n Cell,\n Evaluation,\n MutationProposer,\n Objective,\n ObjectiveContext,\n ProposeContext,\n} from './types'\n\nconst clamp01 = (x: number): number => Math.max(0, Math.min(1, x))\n\n// ── proposers ─────────────────────────────────────────────────────────────────\n\n/**\n * Perturbation-based search: apply the cell's mutation operators to the current\n * elites + seeds, deduping by id. Elites first — mutating the most interesting\n * scenario found so far is what makes the search deepen across rounds.\n */\nexport function mutationProposer<S>(opts: {\n mutationsFor: (cell: Cell) => AdversarialMutation<S>[]\n scenarioId: (s: S) => string\n}): MutationProposer<S> {\n return async (ctx: ProposeContext<S>): Promise<S[]> => {\n const mutations = opts.mutationsFor(ctx.cell)\n const parents = [...ctx.elites, ...ctx.seeds]\n const seen = new Set(parents.map(opts.scenarioId))\n const out: S[] = []\n for (const parent of parents) {\n if (out.length >= ctx.count) break\n for (const m of mutations) {\n const children = await m.mutate(parent, ctx.rng)\n for (const child of children) {\n const id = opts.scenarioId(child)\n if (seen.has(id)) continue\n seen.add(id)\n out.push(child)\n if (out.length >= ctx.count) break\n }\n if (out.length >= ctx.count) break\n }\n }\n return out\n }\n}\n\n// ── objectives ────────────────────────────────────────────────────────────────\n\n/** Adversarial: a low headline score is interesting — find where the agent fails. */\nexport function adversarialObjective(threshold = 0.5): Objective {\n return { kind: 'adversarial', threshold, interest: (ev) => clamp01(1 - ev.score) }\n}\n\nfunction hamming(\n a: Record<string, string> | undefined,\n b: Record<string, string> | undefined,\n): number {\n if (!a || !b) return 1\n const keys = new Set([...Object.keys(a), ...Object.keys(b)])\n if (keys.size === 0) return 0\n let diff = 0\n for (const k of keys) if (a[k] !== b[k]) diff++\n return diff / keys.size\n}\n\n/**\n * Novelty: interesting when far from the archive in score AND measured behavior\n * descriptor — quality-diversity's diversity pressure; drives corpus growth\n * rather than re-finding the same hole.\n */\nexport function noveltyObjective(threshold = 0.3): Objective {\n return {\n kind: 'novelty',\n threshold,\n interest: (ev: Evaluation, ctx: ObjectiveContext) => {\n const scoreNovelty =\n ctx.archiveScores.length === 0\n ? 1\n : Math.min(...ctx.archiveScores.map((s) => Math.abs(s - ev.score)))\n const descNovelty =\n ctx.archiveDescriptors.length === 0\n ? 1\n : Math.min(...ctx.archiveDescriptors.map((d) => hamming(d, ev.descriptor)))\n return clamp01(0.5 * scoreNovelty + 0.5 * descNovelty)\n },\n }\n}\n","/**\n * The exploration engine — a stateful session over a behavior space.\n *\n * Each `step()`: allocate budget across INPUT cells (floor first, then variance\n * steering toward the least-certain cells), propose candidates (the proposer\n * reads current elites + findings, so the search deepens generationally),\n * evaluate with bounded concurrency, archive the most interesting scenario per\n * input×measured bin, and admit notable candidates that pass the validity gates.\n * `run()` loops to budget. `coverage()`/`findings()`/`capsule()` read live state —\n * the surface `makeExploreTools` exposes so an agent can drive the session.\n *\n * One evaluation log (`EvalRecord[]`) is the source of truth; allocation\n * observations and coverage are projections of it.\n */\n\nimport { CostLedger, type CostLedgerHandle, CostReceiptCaptureError } from '../cost-ledger'\nimport { ValidationError } from '../errors'\nimport { varianceBasedCurriculum } from '../rl/active-curriculum'\nimport { buildCapsule } from './capsule'\nimport type { EvalRecord } from './cube'\nimport { enumerateCells } from './cube'\nimport { adversarialObjective } from './policies'\nimport type {\n ArchiveEntry,\n CapsuleData,\n Cell,\n CoverageCell,\n ExploreOptions,\n Finding,\n Objective,\n} from './types'\n\nasync function pMap<T>(\n items: T[],\n fn: (item: T) => Promise<void>,\n concurrency: number,\n signal?: AbortSignal,\n): Promise<void> {\n let i = 0\n const workers = Array.from(\n { length: Math.max(1, Math.min(concurrency, items.length)) },\n async () => {\n while (i < items.length) {\n if (signal?.aborted) return\n const item = items[i++]\n if (item === undefined) return\n await fn(item)\n }\n },\n )\n await Promise.all(workers)\n}\n\nexport class BehaviorExplorer<S> {\n private readonly cells: Cell[]\n private readonly cellById: Map<string, Cell>\n private readonly objective: Objective\n private readonly threshold: number\n private readonly floorPerCell: number\n private readonly perRoundBudget: number\n\n /** The single evaluation log — coverage + allocation are projections of it. */\n private readonly log: Array<EvalRecord & { scenarioId: string }> = []\n /** binId (input × measured coords) → the most interesting entry seen. */\n private readonly archiveByBin = new Map<string, ArchiveEntry<S>>()\n private readonly _findings: Finding<S>[] = []\n private runsUsed = 0\n private candidateFindings = 0\n private evalErrors = 0\n private consecutiveEvalErrors = 0\n private stoppedEarly: { reason: 'eval-errors'; detail: string } | undefined\n private rngState: number\n private readonly costLedger?: CostLedgerHandle\n private readonly costPhase = 'fuzz.explore'\n\n constructor(private readonly opts: ExploreOptions<S>) {\n this.cells = enumerateCells(opts.space)\n if (this.cells.length === 0)\n throw new Error('BehaviorExplorer: space has no cells — every axis needs ≥1 value')\n if (opts.costBudgetUsd !== undefined) {\n if (\n typeof opts.costBudgetUsd !== 'number' ||\n !Number.isFinite(opts.costBudgetUsd) ||\n opts.costBudgetUsd < 0\n ) {\n throw new RangeError(\n `BehaviorExplorer: costBudgetUsd must be a nonnegative finite number, got ${String(opts.costBudgetUsd)}`,\n )\n }\n }\n if (!opts.costOf && (opts.costBudgetUsd !== undefined || opts.ledger || opts.onCost)) {\n throw new ValidationError(\n 'BehaviorExplorer: costBudgetUsd/ledger/onCost require costOf — the explorer ' +\n 'cannot know run cost without it; supply costOf or drop the cost options',\n )\n }\n if (\n (opts.costBudgetUsd !== undefined || opts.ledger?.costCeilingUsd !== undefined) &&\n !opts.maximumChargeOf\n ) {\n throw new ValidationError(\n 'BehaviorExplorer: capped cost tracking requires maximumChargeOf before evaluation',\n )\n }\n if (\n opts.ledger &&\n opts.costBudgetUsd !== undefined &&\n opts.ledger.costCeilingUsd !== opts.costBudgetUsd\n ) {\n throw new ValidationError(\n 'BehaviorExplorer: costBudgetUsd must match the supplied CostLedger ceiling',\n )\n }\n if (opts.costOf) this.costLedger = opts.ledger ?? new CostLedger(opts.costBudgetUsd)\n this.cellById = new Map(this.cells.map((c) => [c.id, c]))\n this.objective = opts.objective ?? adversarialObjective(0.5)\n this.threshold = this.objective.threshold ?? 0.5\n this.floorPerCell = opts.floorPerCell ?? 2\n this.perRoundBudget = Math.max(\n this.cells.length * this.floorPerCell,\n Math.ceil(opts.budget / 4),\n )\n this.rngState = (opts.seed ?? 1) >>> 0\n }\n\n // mulberry32 — deterministic per session.\n private rng = (): number => {\n this.rngState = (this.rngState + 0x6d2b79f5) | 0\n let t = this.rngState\n t = Math.imul(t ^ (t >>> 15), t | 1)\n t ^= t + Math.imul(t ^ (t >>> 7), t | 61)\n return ((t ^ (t >>> 14)) >>> 0) / 4294967296\n }\n\n private binId(cell: Cell, descriptor: Record<string, string> | undefined): string {\n if (!descriptor || Object.keys(descriptor).length === 0) return cell.id\n const measured = Object.keys(descriptor)\n .sort()\n .map((k) => `${k}=${descriptor[k]}`)\n .join('|')\n return `${cell.id}|${measured}`\n }\n\n private allocate(budget: number): Array<{ cellId: string; count: number }> {\n if ((this.opts.allocation ?? 'variance') === 'uniform') {\n const per = Math.max(this.floorPerCell, Math.floor(budget / this.cells.length))\n return this.cells.map((c) => ({ cellId: c.id, count: per }))\n }\n return varianceBasedCurriculum(\n this.log.map((r) => ({\n variantId: r.cell.id,\n scenarioId: r.scenarioId,\n score: r.ev.score,\n pass: r.ev.valid && r.ev.score >= 0.5,\n })),\n this.cells.map((c) => ({ variantId: c.id, scenarioId: '*' })),\n { budget, floorPerCell: this.floorPerCell },\n ).map((a) => ({ cellId: a.variantId, count: a.count }))\n }\n\n private objectiveContext() {\n const entries = [...this.archiveByBin.values()]\n return {\n archiveScores: entries.map((e) => e.evaluation.score),\n archiveDescriptors: entries.map((e) => e.evaluation.descriptor),\n }\n }\n\n /** Elites whose INPUT cell matches — what the proposer mutates/deepens from. */\n private elitesFor(cellId: string): S[] {\n const out: S[] = []\n for (const e of this.archiveByBin.values()) if (e.cell.id === cellId) out.push(e.scenario)\n return out\n }\n\n /** One allocate → propose → evaluate → gate → archive round. */\n async step(): Promise<{ runs: number; findings: Finding<S>[] }> {\n const remaining = this.opts.budget - this.runsUsed\n if (remaining <= 0 || this.opts.signal?.aborted) return { runs: 0, findings: [] }\n\n const allocations = this.allocate(Math.min(this.perRoundBudget, remaining))\n const newFindings: Finding<S>[] = []\n let runsThisStep = 0\n\n for (const alloc of allocations) {\n if (\n this.runsUsed >= this.opts.budget ||\n this.stoppedEarly !== undefined ||\n this.opts.signal?.aborted\n )\n break\n const cell = this.cellById.get(alloc.cellId)\n if (!cell) continue\n const cap = Math.min(alloc.count, this.opts.budget - this.runsUsed)\n if (cap <= 0) continue\n this.opts.onProgress?.({ type: 'cell-allocated', cell, count: cap })\n\n const seeds = await this.opts.seedsFor(cell)\n const elites = this.elitesFor(cell.id)\n const proposed = await this.opts.proposer({\n cell,\n seeds,\n elites,\n findings: this._findings,\n count: cap,\n rng: this.rng,\n })\n // Cold cells evaluate their seeds first (the coverage floor); warm cells\n // trust the proposer, which already saw the elites.\n const cold = this.log.every((r) => r.cell.id !== cell.id)\n const toEval = [...(cold ? seeds : []), ...proposed].slice(0, cap)\n\n await pMap(\n toEval,\n async (scenario) => {\n if (\n this.runsUsed >= this.opts.budget ||\n this.stoppedEarly !== undefined ||\n this.opts.signal?.aborted\n )\n return\n // evaluate/gates/minimize cross an external boundary (router, backend,\n // judge). A throw there is an infra outcome: record it as a typed\n // eval-error and keep exploring — one 5xx must not kill a campaign.\n // Consecutive failures trip the circuit breaker instead, so a dead\n // backend stops the run rather than burning the remaining budget.\n try {\n const startedAt = performance.now()\n const paid = this.costLedger\n ? await this.costLedger.runPaidCall({\n channel: 'agent',\n phase: this.costPhase,\n actor: 'fuzz.evaluate',\n tags: { target: this.opts.target, cell: cell.id },\n signal: this.opts.signal,\n maximumCharge: this.opts.maximumChargeOf?.(scenario, cell),\n execute: () => this.opts.evaluate(scenario, cell),\n receipt: (evaluation) => {\n const cost = this.opts.costOf!(scenario, cell, evaluation)\n if (cost === null) {\n return {\n model: 'unattributed',\n inputTokens: 0,\n outputTokens: 0,\n costUnknown: true,\n }\n }\n if (!Number.isFinite(cost.usd) || cost.usd < 0) {\n throw new RangeError(\n `BehaviorExplorer: costOf returned an invalid usd (${String(cost.usd)}) — ` +\n 'return null when cost is unknown, never a fabricated number',\n )\n }\n return {\n model: cost.model ?? 'unattributed',\n inputTokens: 0,\n outputTokens: 0,\n actualCostUsd: cost.usd,\n }\n },\n })\n : undefined\n if (paid && !paid.succeeded) {\n if (\n paid.error instanceof CostReceiptCaptureError &&\n paid.error.receiptError instanceof RangeError\n ) {\n throw paid.error.receiptError\n }\n throw paid.error\n }\n const ev = paid ? paid.value : await this.opts.evaluate(scenario, cell)\n // Consumer-measured latency wins (it can exclude judge time); the\n // engine's wall-clock is the default so latency is never missing.\n const latencyMs = ev.latencyMs ?? performance.now() - startedAt\n this.runsUsed++\n runsThisStep++\n this.consecutiveEvalErrors = 0\n const costUsd = paid\n ? paid.receipt.costUnknown\n ? null\n : paid.receipt.costUsd\n : undefined\n if (paid?.receipt.actualCostUsd !== undefined) {\n this.opts.onCost?.({ usd: paid.receipt.actualCostUsd, channel: 'agent' })\n }\n const interest = this.objective.interest(ev, this.objectiveContext())\n this.log.push({\n cell,\n ev,\n interest,\n latencyMs,\n ...(costUsd !== undefined ? { costUsd } : {}),\n scenarioId: this.opts.scenarioId(scenario),\n })\n this.opts.onProgress?.({ type: 'evaluated', cell, scenario, evaluation: ev })\n\n const bin = this.binId(cell, ev.descriptor)\n const cur = this.archiveByBin.get(bin)\n if (!cur || interest > cur.interest)\n this.archiveByBin.set(bin, { binId: bin, cell, scenario, evaluation: ev, interest })\n\n if (interest < this.threshold) return\n this.candidateFindings++\n if (this.opts.gates?.isValid && !(await this.opts.gates.isValid(scenario, ev, cell)))\n return\n if (\n this.opts.gates?.isUncontaminated &&\n !(await this.opts.gates.isUncontaminated(scenario, ev, cell))\n )\n return\n\n const minimized = this.opts.minimize\n ? await this.opts.minimize(scenario, this.opts.evaluate, cell)\n : scenario\n const finding: Finding<S> = {\n id: this.opts.scenarioId(scenario),\n cell,\n scenario,\n minimized,\n text: this.opts.scenarioText?.(minimized),\n evaluation: ev,\n interest,\n objective: this.objective.kind,\n }\n this._findings.push(finding)\n newFindings.push(finding)\n this.opts.onProgress?.({ type: 'finding', finding })\n } catch (err) {\n // Internal validation errors (e.g. a fabricated costOf number) are\n // programming mistakes, not backend outcomes — they stay loud.\n if (err instanceof RangeError) throw err\n this.evalErrors++\n this.consecutiveEvalErrors++\n const message = err instanceof Error ? err.message : String(err)\n this.opts.onProgress?.({\n type: 'eval-error',\n cell,\n scenarioId: this.opts.scenarioId(scenario),\n message,\n })\n const limit = this.opts.maxConsecutiveEvalErrors ?? 5\n if (this.consecutiveEvalErrors >= limit) {\n this.stoppedEarly = {\n reason: 'eval-errors',\n detail: `${this.consecutiveEvalErrors} consecutive eval errors (last: ${message})`,\n }\n }\n }\n },\n this.opts.concurrency ?? 1,\n this.opts.signal,\n )\n }\n\n this.opts.onProgress?.({ type: 'round', runsUsed: this.runsUsed, budget: this.opts.budget })\n return { runs: runsThisStep, findings: newFindings }\n }\n\n /** Loop `step()` until the run or dollar budget is spent, the signal aborts,\n * or no progress is made. */\n async run(): Promise<CapsuleData<S>> {\n while (\n this.runsUsed < this.opts.budget &&\n this.stoppedEarly === undefined &&\n !this.opts.signal?.aborted\n ) {\n const { runs } = await this.step()\n if (runs === 0 && this.stoppedEarly === undefined) break\n }\n return this.capsule()\n }\n\n coverage(): CoverageCell[] {\n return this.capsule().coverage\n }\n\n findings(): Finding<S>[] {\n return [...this._findings].sort((a, b) => b.interest - a.interest)\n }\n\n capsule(): CapsuleData<S> {\n const cost = this.costLedger?.summary({\n phase: this.costPhase,\n tags: { target: this.opts.target },\n })\n return buildCapsule({\n target: this.opts.target,\n objective: this.objective.kind,\n cells: this.cells,\n log: this.log,\n threshold: this.threshold,\n archive: [...this.archiveByBin.values()],\n findings: this._findings,\n candidateFindings: this.candidateFindings,\n runsUsed: this.runsUsed,\n cost: cost\n ? {\n costUsd: cost.totalCostUsd,\n costUnknownRuns: cost.byChannel.reduce((sum, row) => sum + row.unpricedCalls, 0),\n }\n : undefined,\n evalErrors: this.evalErrors,\n stoppedEarly: this.stoppedEarly,\n })\n }\n}\n","/**\n * `fuzzAgent` — the adversarial batch preset over `BehaviorExplorer`.\n *\n * One call: explore the space to budget with the adversarial objective and\n * return the capsule. For agent-driven, incremental, or multi-objective use,\n * construct a `BehaviorExplorer` and drive it via `makeExploreTools`.\n */\n\nimport { BehaviorExplorer } from './explorer'\nimport { adversarialObjective } from './policies'\nimport type { CapsuleData, ExploreOptions } from './types'\n\nexport type FuzzAgentOptions<S> = Omit<ExploreOptions<S>, 'objective'> & {\n /** Score strictly below this is a candidate failure. Default 0.5. */\n failureThreshold?: number\n}\n\nexport async function fuzzAgent<S>(\n opts: FuzzAgentOptions<S>,\n): Promise<{ capsule: CapsuleData<S> }> {\n const { failureThreshold, ...rest } = opts\n const explorer = new BehaviorExplorer<S>({\n ...rest,\n objective: adversarialObjective(failureThreshold ?? 0.5),\n })\n return { capsule: await explorer.run() }\n}\n","/**\n * Validity gates — what separates a fuzzer from a slop generator.\n *\n * A notable candidate is admitted only when it is fair and reproducible. None of\n * these are on by default: the live wiring opts in, so reported findings carry\n * their proof.\n */\n\nimport type { Cell, Evaluation, Evaluator, ValidityGates } from './types'\n\n/** Combine gate sets; a candidate must pass every gate in every set. */\nexport function composeGates<S>(...sets: Array<ValidityGates<S> | undefined>): ValidityGates<S> {\n const present = sets.filter((s): s is ValidityGates<S> => s != null)\n return {\n isValid: async (scenario, ev, cell) => {\n for (const g of present) {\n if (g.isValid && !(await g.isValid(scenario, ev, cell))) return false\n }\n return true\n },\n isUncontaminated: async (scenario, ev, cell) => {\n for (const g of present) {\n if (g.isUncontaminated && !(await g.isUncontaminated(scenario, ev, cell))) return false\n }\n return true\n },\n }\n}\n\n/**\n * Reproducibility gate. Re-run the target on a meaning-preserving rephrase of the\n * flagged scenario; keep the finding only when the rephrase ALSO scores below the\n * threshold. A finding that flips under a cosmetic rewrite was keyed to surface\n * form, not the task — a false signal we must not report. Costs one extra\n * evaluation per candidate (candidates are rare, so cheap).\n */\nexport function perturbationStabilityGate<S>(opts: {\n evaluate: Evaluator<S>\n /** Produce a semantic-preserving rephrase. Return null to skip (treated as pass). */\n perturb: (scenario: S) => S | null\n /** Score strictly below this still counts as failing. Default 0.5. */\n failureThreshold?: number\n}): ValidityGates<S> {\n const threshold = opts.failureThreshold ?? 0.5\n return {\n isUncontaminated: async (scenario: S, _ev: Evaluation, cell: Cell) => {\n const rephrased = opts.perturb(scenario)\n if (rephrased == null) return true\n const re = await opts.evaluate(rephrased, cell)\n return re.score < threshold\n },\n }\n}\n\n/**\n * Severity-floor gate. Reject borderline candidates whose score sits in a band\n * just under the threshold — judge noise, not a real defect.\n */\nexport function severityFloorGate<S>(opts: {\n failureThreshold?: number\n margin?: number\n}): ValidityGates<S> {\n const threshold = opts.failureThreshold ?? 0.5\n const margin = opts.margin ?? 0.1\n return {\n isValid: (_scenario, ev: Evaluation) => ev.score <= threshold - margin,\n }\n}\n","/**\n * Agent-drivable surface over a live exploration session.\n *\n * Framework-neutral tool defs ({name, description, parameters: JSON Schema,\n * handler}) so the on-demand agent — not a batch script — drives the search:\n * step it, read coverage, inspect findings, render the capsule. Transport\n * encodings (OpenAI function shape, MCP) are one-line mappings the host owns.\n */\n\nimport { renderCapsuleHtml } from './capsule'\nimport type { BehaviorExplorer } from './explorer'\n\nexport interface ExploreToolDef {\n name: string\n description: string\n /** JSON Schema (draft-07+) for the arguments. */\n parameters: Record<string, unknown>\n handler: (args: unknown, ctx?: { signal?: AbortSignal }) => Promise<unknown>\n}\n\nconst NO_ARGS = { type: 'object', properties: {}, additionalProperties: false }\n\nexport function makeExploreTools<S>(explorer: BehaviorExplorer<S>): ExploreToolDef[] {\n return [\n {\n name: 'explore_step',\n description:\n 'Run one exploration round: allocate budget across cells, propose + evaluate scenarios, archive elites, admit gate-verified findings. Returns runs spent and new findings.',\n parameters: NO_ARGS,\n handler: async () => {\n const { runs, findings } = await explorer.step()\n return { runs, newFindings: findings.length, findings }\n },\n },\n {\n name: 'explore_coverage',\n description:\n 'Read the live coverage map: per planned cell — runs, robustness (null = uncovered), finding rate, per-dimension means.',\n parameters: NO_ARGS,\n handler: async () => explorer.coverage(),\n },\n {\n name: 'explore_findings',\n description: 'List gate-verified findings so far, sorted by descending interest.',\n parameters: {\n type: 'object',\n properties: { limit: { type: 'number', description: 'max findings to return' } },\n additionalProperties: false,\n },\n handler: async (args) => {\n const limit = (args as { limit?: number } | undefined)?.limit\n const all = explorer.findings()\n return typeof limit === 'number' ? all.slice(0, limit) : all\n },\n },\n {\n name: 'explore_capsule',\n description:\n 'Build the capsule artifact from the current session state. format \"data\" returns the structured CapsuleData; \"html\" returns the standalone page.',\n parameters: {\n type: 'object',\n properties: {\n format: { type: 'string', enum: ['data', 'html'] },\n generatedAt: { type: 'string', description: 'ISO timestamp to stamp into the page' },\n },\n additionalProperties: false,\n },\n handler: async (args) => {\n const a = (args ?? {}) as { format?: string; generatedAt?: string }\n const capsule = explorer.capsule()\n if (a.format === 'html') return renderCapsuleHtml(capsule, { generatedAt: a.generatedAt })\n return capsule\n },\n },\n ]\n}\n"],"mappings":";;;;;;;;;;;;;;AA2BO,SAAS,eAAe,OAA8B;AAC3D,MAAI,MAAM,KAAK,WAAW,EAAG,QAAO,CAAC;AACrC,MAAI,WAA0C,CAAC,CAAC,CAAC;AACjD,aAAW,QAAQ,MAAM,MAAM;AAC7B,UAAM,OAAsC,CAAC;AAC7C,eAAW,WAAW,UAAU;AAC9B,iBAAW,SAAS,KAAK,OAAQ,MAAK,KAAK,EAAE,GAAG,SAAS,CAAC,KAAK,IAAI,GAAG,MAAM,CAAC;AAAA,IAC/E;AACA,eAAW;AAAA,EACb;AACA,SAAO,SAAS,IAAI,CAAC,YAAY,EAAE,IAAI,OAAO,OAAO,MAAM,GAAG,OAAO,EAAE;AACzE;AAGO,SAAS,OAAO,OAAsB,QAAwC;AACnF,SAAO,MAAM,KAAK,IAAI,CAAC,MAAM,GAAG,EAAE,IAAI,IAAI,OAAO,EAAE,IAAI,CAAC,EAAE,EAAE,KAAK,GAAG;AACtE;AAGA,SAAS,WAAW,QAAkB,GAAmB;AACvD,QAAM,MAAM,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,OAAO,MAAM,IAAI,CAAC,CAAC;AACrF,SAAO,OAAO,GAAG;AACnB;AAIO,SAAS,aAAa,QAAgC;AAC3D,MAAI,OAAO,WAAW;AACpB,UAAM,IAAI,MAAM,6EAAwE;AAC1F,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC/C,QAAM,OAAO,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AACxD,SAAO;AAAA,IACL;AAAA,IACA,QAAQ,WAAW,QAAQ,GAAG;AAAA,IAC9B,KAAK,WAAW,QAAQ,GAAG;AAAA,IAC3B,KAAK,OAAO,CAAC;AAAA,IACb,KAAK,OAAO,OAAO,SAAS,CAAC;AAAA,IAC7B,GAAG,OAAO;AAAA,EACZ;AACF;AAMO,SAAS,cAAc,OAAe,KAAmB,WAAmC;AACjG,QAAM,SAAS,oBAAI,IAA0B;AAC7C,aAAW,KAAK,KAAK;AACnB,UAAM,MAAM,OAAO,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC;AACtC,QAAI,KAAK,CAAC;AACV,WAAO,IAAI,EAAE,KAAK,IAAI,GAAG;AAAA,EAC3B;AACA,SAAO,MAAM,IAAI,CAAC,SAAS;AACzB,UAAM,OAAO,OAAO,IAAI,KAAK,EAAE,KAAK,CAAC;AACrC,UAAM,OAAO,KAAK;AAClB,QAAI,SAAS;AACX,aAAO,EAAE,MAAM,MAAM,GAAG,OAAO,MAAM,aAAa,GAAG,YAAY,CAAC,GAAG,WAAW,KAAK;AAEvF,UAAM,QAAQ,aAAa,KAAK,IAAI,CAAC,MAAM,EAAE,GAAG,KAAK,CAAC;AACtD,UAAM,YAAY,aAAa,KAAK,IAAI,CAAC,MAAM,EAAE,SAAS,CAAC;AAC3D,UAAM,cAAc,KAAK,OAAO,CAAC,MAAM,EAAE,YAAY,SAAS,EAAE,SAAS;AAEzE,UAAM,aAAuC,CAAC;AAC9C,eAAW,KAAK,MAAM;AACpB,iBAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,EAAE,GAAG,UAAU,CAAC,CAAC,GAAG;AACtD;AAAC,SAAC,WAAW,CAAC,MAAM,CAAC,GAAG,KAAK,CAAC;AAAA,MAChC;AAAA,IACF;AACA,UAAM,aAA2C,CAAC;AAClD,eAAW,CAAC,GAAG,EAAE,KAAK,OAAO,QAAQ,UAAU,EAAG,YAAW,CAAC,IAAI,aAAa,EAAE;AAIjF,UAAM,UAAU,KAAK,OAAO,CAAC,MAAM,EAAE,YAAY,MAAS;AAC1D,UAAM,QAAQ,QAAQ,OAAO,CAAC,MAAM,EAAE,YAAY,IAAI;AACtD,UAAM,OACJ,QAAQ,SAAS,IACb;AAAA,MACE,SAAS,MAAM,OAAO,CAAC,GAAG,MAAM,IAAK,EAAE,SAAoB,CAAC;AAAA,MAC5D,GAAI,QAAQ,SAAS,MAAM,SACvB,EAAE,iBAAiB,QAAQ,SAAS,MAAM,OAAO,IACjD,CAAC;AAAA,IACP,IACA,CAAC;AAEP,WAAO,EAAE,MAAM,MAAM,OAAO,aAAa,YAAY,WAAW,GAAG,KAAK;AAAA,EAC1E,CAAC;AACH;;;AChFO,SAAS,aAAgB,OAA6C;AAC3E,QAAM,WAAW,cAAc,MAAM,OAAO,MAAM,KAAK,MAAM,SAAS;AACtE,QAAM,UAAU,SAAS,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC;AAGjD,QAAM,aACJ,QAAQ,WAAW,IAAI,OAAO,aAAa,QAAQ,IAAI,CAAC,MAAO,EAAE,MAAuB,IAAI,CAAC;AAC/F,QAAM,YAAY,MAAM,IAAI,WAAW,IAAI,OAAO,aAAa,MAAM,IAAI,IAAI,CAAC,MAAM,EAAE,SAAS,CAAC;AAEhG,QAAM,uBAAuB,MAAM,QAAQ,OAAO,CAAC,MAAM,EAAE,UAAU,EAAE,KAAK,EAAE,EAAE;AAEhF,SAAO;AAAA,IACL,QAAQ,MAAM;AAAA,IACd,WAAW,MAAM;AAAA,IACjB;AAAA,IACA,UAAU,CAAC,GAAG,MAAM,QAAQ,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,IACpE,SAAS,CAAC,GAAG,MAAM,OAAO,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,IAClE,OAAO;AAAA,MACL,WAAW,MAAM;AAAA,MACjB,YAAY,MAAM,MAAM;AAAA,MACxB,cAAc,QAAQ;AAAA,MACtB;AAAA,MACA,mBAAmB,MAAM;AAAA,MACzB,kBAAkB,MAAM,SAAS;AAAA,MACjC;AAAA,MACA;AAAA,MACA,GAAI,MAAM,OACN,EAAE,SAAS,MAAM,KAAK,SAAS,iBAAiB,MAAM,KAAK,gBAAgB,IAC3E,CAAC;AAAA,MACL,YAAY,MAAM;AAAA,MAClB,GAAI,MAAM,eAAe,EAAE,cAAc,MAAM,aAAa,IAAI,CAAC;AAAA,IACnE;AAAA,EACF;AACF;AAIA,SAAS,IAAI,GAAmB;AAC9B,SAAO,EAAE;AAAA,IACP;AAAA,IACA,CAAC,OAAO,EAAE,KAAK,SAAS,KAAK,QAAQ,KAAK,QAAQ,KAAK,SAAS,GAAG,CAAC;AAAA,EACtE;AACF;AAGA,SAAS,gBAAgB,GAA0B;AACjD,MAAI,KAAK,KAAM,QAAO;AACtB,SAAO,OAAO,KAAK,MAAM,IAAI,GAAG,CAAC;AACnC;AAEA,SAAS,IAAI,GAAmB;AAC9B,SAAO,GAAG,KAAK,MAAM,IAAI,GAAG,CAAC;AAC/B;AAEA,SAAS,WAAW,UAAqE;AACvF,QAAM,QAAkB,CAAC;AACzB,QAAM,OAAO,oBAAI,IAAyB;AAC1C,aAAW,KAAK,UAAU;AACxB,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,EAAE,KAAK,MAAM,GAAG;AAClD,UAAI,CAAC,KAAK,IAAI,CAAC,GAAG;AAChB,aAAK,IAAI,GAAG,oBAAI,IAAI,CAAC;AACrB,cAAM,KAAK,CAAC;AAAA,MACd;AACA,WAAK,IAAI,CAAC,GAAG,IAAI,CAAC;AAAA,IACpB;AAAA,EACF;AACA,SAAO,MAAM,IAAI,CAAC,UAAU,EAAE,MAAM,QAAQ,CAAC,GAAI,KAAK,IAAI,IAAI,KAAK,CAAC,CAAE,EAAE,EAAE;AAC5E;AAGA,SAAS,WAAW,GAAyB;AAC3C,QAAM,UAAU,OAAO,QAAQ,EAAE,UAAU,EAAE,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,EAAE,IAAI,CAAqB;AAC5F,MAAI,QAAQ,WAAW,EAAG,QAAO;AACjC,QAAM,SAAS,QAAQ,KAAK,CAAC,GAAG,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC;AACjD,QAAM,IAAI,OAAO,CAAC;AAClB,MAAI,CAAC,EAAG,QAAO;AACf,SAAO,qBAAqB,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,CAAC;AACpD;AAEA,SAAS,YAAY,UAAkC;AACrD,QAAM,OAAO,WAAW,QAAQ;AAChC,QAAM,OAAO,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,CAAC,EAAE,KAAK,IAAI,CAAC,CAAC,CAAC;AACxD,QAAM,OAAO,CAAC,GAA6B,UAA0B;AACnE,UAAM,IAAI,GAAG,OAAO,QAAQ;AAC5B,UAAM,QAAQ,GAAG,QACb,GAAG,IAAI,EAAE,MAAM,IAAI,CAAC,mBAAmB,IAAI,EAAE,MAAM,MAAM,CAAC,SAAS,IAAI,EAAE,MAAM,GAAG,CAAC,UAAO,EAAE,IAAI,cAAW,IAAI,EAAE,WAAW,CAAC,WAAW,EAAE,YAAY,UAAO,EAAE,UAAU,SAAS,KAAM,QAAQ,CAAC,CAAC,aAAa,EAAE,GAAG,EAAE,YAAY,SAAY,UAAO,EAAE,QAAQ,QAAQ,CAAC,CAAC,KAAK,EAAE,KAChR;AACJ,WAAO,uCAAuC,gBAAgB,CAAC,CAAC,YAAY,IAAI,KAAK,CAAC,KAAK,QAAQ,oBAAoB,IAAI,KAAK,CAAC,YAAY,EAAE,oBAAoB,KAAK,KAAK,OAAO,IAAI,CAAC,IAAI,QAAG,UAAU,IAAI,WAAW,CAAC,IAAI,EAAE;AAAA,EAClO;AAEA,QAAM,UAAU,KAAK,CAAC;AACtB,QAAM,UAAU,KAAK,CAAC;AACtB,MAAI,KAAK,WAAW,KAAK,WAAW,SAAS;AAC3C,UAAM,OAAO,gBAAgB,QAAQ,OAAO,IAAI,CAAC,MAAM,OAAO,IAAI,CAAC,CAAC,OAAO,EAAE,KAAK,EAAE,CAAC;AACrF,UAAM,OAAO,QAAQ,OAClB,IAAI,CAAC,OAAO;AACX,YAAM,QAAQ,QAAQ,OACnB,IAAI,CAAC,OAAO;AACX,cAAM,KAAK,GAAG,QAAQ,IAAI,IAAI,EAAE,IAAI,QAAQ,IAAI,IAAI,EAAE;AACtD,eAAO,OAAO,KAAK,KAAK,IAAI,EAAE,GAAG,EAAE,CAAC;AAAA,MACtC,CAAC,EACA,KAAK,EAAE;AACV,aAAO,sBAAsB,IAAI,EAAE,CAAC,QAAQ,KAAK;AAAA,IACnD,CAAC,EACA,KAAK,EAAE;AACV,WAAO,oCAAoC,IAAI,QAAQ,IAAI,CAAC,sBAAmB,IAAI,QAAQ,IAAI,CAAC,iCAAiC,IAAI,GAAG,IAAI;AAAA,EAC9I;AAEA,QAAM,SAAS,CAAC,GAAG,QAAQ,EAAE,KAAK,CAAC,GAAG,OAAO,EAAE,OAAO,QAAQ,MAAM,EAAE,OAAO,QAAQ,EAAE;AACvF,SAAO,qBAAqB,OAAO,IAAI,CAAC,MAAM,KAAK,GAAG,OAAO,OAAO,EAAE,KAAK,MAAM,EAAE,KAAK,QAAK,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC;AAC3G;AAEA,SAAS,aAAgB,UAAwB,OAAuB;AACtE,MAAI,SAAS,WAAW;AACtB,WAAO;AACT,SAAO,SACJ,MAAM,GAAG,KAAK,EACd,IAAI,CAAC,MAAM;AACV,UAAM,SAAS,OAAO,QAAQ,EAAE,KAAK,MAAM,EACxC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,EAAE,EAC3B,KAAK,QAAK;AACb,UAAM,UAAU,EAAE,WAAW,UAAU,CAAC,GACrC,IAAI,CAAC,MAAM,wBAAwB,IAAI,CAAC,CAAC,SAAS,EAClD,KAAK,EAAE;AACV,UAAM,OAAO,OAAO,QAAQ,EAAE,WAAW,UAAU,CAAC,CAAC,EAClD,KAAK,CAAC,GAAG,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,EAC1B,MAAM,GAAG,CAAC,EACV,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,qBAAqB,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,SAAS,EAC9D,KAAK,EAAE;AACV,WAAO,4DAA4D,IAAI,MAAM,CAAC,UAAU,MAAM,+BAA+B,IAAI,EAAE,QAAQ,CAAC,sEAAsE,IAAI,EAAE,QAAQ,CAAC,oCAAoC,IAAI,EAAE,QAAQ,8BAA8B,CAAC,SAAS,OAAO,sBAAsB,IAAI,WAAW,EAAE;AAAA,EAC3W,CAAC,EACA,KAAK,EAAE;AACZ;AAUO,SAAS,kBACd,SACA,OAA6B,CAAC,GACtB;AACR,QAAM,IAAI,QAAQ;AAClB,QAAM,MAAM,CAAC,OAAe,OAAe,SAAS,cAClD,iDAAiD,MAAM,KAAK,IAAI,KAAK,CAAC,yBAAyB,IAAI,KAAK,CAAC;AAC3G,QAAM,OAAO,QAAQ,OACjB;AAAA,IACE;AAAA,IACA,GAAG,QAAQ,KAAK,OAAO,QAAQ,CAAC,CAAC,WAAM,QAAQ,KAAK,MAAM,QAAQ,CAAC,CAAC;AAAA,IACpE;AAAA,EACF,IACA;AAGJ,QAAM,OACJ,EAAE,YAAY,SACV;AAAA,IACE,EAAE,kBAAkB,aAAU,EAAE,eAAe,mBAAmB;AAAA,IAClE,IAAI,EAAE,QAAQ,QAAQ,CAAC,CAAC;AAAA,IACxB,EAAE,kBAAkB,YAAY;AAAA,EAClC,IACA;AACN,QAAM,QAAQ,KAAK,eAAe,QAAQ,eAAe;AAEzD,SAAO;AAAA,SACA,IAAI,QAAQ,SAAS,CAAC,mBAAc,IAAI,QAAQ,MAAM,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,MAoC1D,IAAI,QAAQ,SAAS,CAAC,qBAAkB,IAAI,QAAQ,MAAM,CAAC;AAAA,mBAC9C,EAAE,SAAS,qBAAqB,EAAE,YAAY,IAAI,EAAE,UAAU,iBAAiB,EAAE,uBAAuB,IAAI,SAAM,EAAE,oBAAoB,4BAA4B,EAAE,GAAG,QAAQ,SAAM,IAAI,KAAK,CAAC,KAAK,EAAE;AAAA;AAAA,EAEzN,EAAE,aAAa,IAAI,cAAc,GAAG,IAAI,EAAE,WAAW,IAAI,CAAC,IAAI,EAAE,WAAW,OAAO,MAAM,YAAY,SAAS,IAAI,EAAE;AAAA,EACnH,EAAE,aAAa,IAAI,eAAe,GAAG,IAAI,EAAE,WAAW,GAAG,CAAC,SAAI,IAAI,EAAE,WAAW,GAAG,CAAC,EAAE,IAAI,EAAE;AAAA,EAC3F,EAAE,YAAY,IAAI,kBAAkB,IAAI,EAAE,UAAU,SAAS,KAAM,QAAQ,CAAC,CAAC,KAAK,EAAE,UAAU,MAAM,IAAI,EAAE,UAAU,SAAS,YAAY,SAAS,IAAI,EAAE;AAAA,EACxJ,IAAI,qBAAqB,OAAO,EAAE,gBAAgB,GAAG,EAAE,mBAAmB,IAAI,YAAY,SAAS,CAAC;AAAA,EACpG,IAAI,iBAAiB,GAAG,EAAE,YAAY,IAAI,EAAE,UAAU,EAAE,CAAC;AAAA,EACzD,IAAI,iBAAiB,OAAO,EAAE,SAAS,CAAC,CAAC;AAAA,EACzC,IAAI;AAAA,EACJ,EAAE,aAAa,IAAI,IAAI,eAAe,OAAO,EAAE,UAAU,GAAG,SAAS,IAAI,EAAE;AAAA,EAC3E,IAAI;AAAA;AAAA,EAEJ,EAAE,eAAe,yDAAyD,IAAI,EAAE,aAAa,MAAM,CAAC,qEAAgE,EAAE;AAAA;AAAA,EAEtK,YAAY,QAAQ,QAAQ,CAAC;AAAA,uBACR,EAAE,oBAAoB,EAAE,mBAAmB,SAAM,EAAE,gBAAgB,OAAO,EAAE,iBAAiB,0CAA0C,EAAE;AAAA,EAC9J,aAAa,QAAQ,UAAU,KAAK,eAAe,CAAC,CAAC;AAAA;AAAA;AAGvD;;;AChPA,IAAM,UAAU,CAAC,MAAsB,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,CAAC,CAAC;AAS1D,SAAS,iBAAoB,MAGZ;AACtB,SAAO,OAAO,QAAyC;AACrD,UAAM,YAAY,KAAK,aAAa,IAAI,IAAI;AAC5C,UAAM,UAAU,CAAC,GAAG,IAAI,QAAQ,GAAG,IAAI,KAAK;AAC5C,UAAM,OAAO,IAAI,IAAI,QAAQ,IAAI,KAAK,UAAU,CAAC;AACjD,UAAM,MAAW,CAAC;AAClB,eAAW,UAAU,SAAS;AAC5B,UAAI,IAAI,UAAU,IAAI,MAAO;AAC7B,iBAAW,KAAK,WAAW;AACzB,cAAM,WAAW,MAAM,EAAE,OAAO,QAAQ,IAAI,GAAG;AAC/C,mBAAW,SAAS,UAAU;AAC5B,gBAAM,KAAK,KAAK,WAAW,KAAK;AAChC,cAAI,KAAK,IAAI,EAAE,EAAG;AAClB,eAAK,IAAI,EAAE;AACX,cAAI,KAAK,KAAK;AACd,cAAI,IAAI,UAAU,IAAI,MAAO;AAAA,QAC/B;AACA,YAAI,IAAI,UAAU,IAAI,MAAO;AAAA,MAC/B;AAAA,IACF;AACA,WAAO;AAAA,EACT;AACF;AAKO,SAAS,qBAAqB,YAAY,KAAgB;AAC/D,SAAO,EAAE,MAAM,eAAe,WAAW,UAAU,CAAC,OAAO,QAAQ,IAAI,GAAG,KAAK,EAAE;AACnF;AAEA,SAAS,QACP,GACA,GACQ;AACR,MAAI,CAAC,KAAK,CAAC,EAAG,QAAO;AACrB,QAAM,OAAO,oBAAI,IAAI,CAAC,GAAG,OAAO,KAAK,CAAC,GAAG,GAAG,OAAO,KAAK,CAAC,CAAC,CAAC;AAC3D,MAAI,KAAK,SAAS,EAAG,QAAO;AAC5B,MAAI,OAAO;AACX,aAAW,KAAK,KAAM,KAAI,EAAE,CAAC,MAAM,EAAE,CAAC,EAAG;AACzC,SAAO,OAAO,KAAK;AACrB;AAOO,SAAS,iBAAiB,YAAY,KAAgB;AAC3D,SAAO;AAAA,IACL,MAAM;AAAA,IACN;AAAA,IACA,UAAU,CAAC,IAAgB,QAA0B;AACnD,YAAM,eACJ,IAAI,cAAc,WAAW,IACzB,IACA,KAAK,IAAI,GAAG,IAAI,cAAc,IAAI,CAAC,MAAM,KAAK,IAAI,IAAI,GAAG,KAAK,CAAC,CAAC;AACtE,YAAM,cACJ,IAAI,mBAAmB,WAAW,IAC9B,IACA,KAAK,IAAI,GAAG,IAAI,mBAAmB,IAAI,CAAC,MAAM,QAAQ,GAAG,GAAG,UAAU,CAAC,CAAC;AAC9E,aAAO,QAAQ,MAAM,eAAe,MAAM,WAAW;AAAA,IACvD;AAAA,EACF;AACF;;;AC/DA,eAAe,KACb,OACA,IACA,aACA,QACe;AACf,MAAI,IAAI;AACR,QAAM,UAAU,MAAM;AAAA,IACpB,EAAE,QAAQ,KAAK,IAAI,GAAG,KAAK,IAAI,aAAa,MAAM,MAAM,CAAC,EAAE;AAAA,IAC3D,YAAY;AACV,aAAO,IAAI,MAAM,QAAQ;AACvB,YAAI,QAAQ,QAAS;AACrB,cAAM,OAAO,MAAM,GAAG;AACtB,YAAI,SAAS,OAAW;AACxB,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF;AAAA,EACF;AACA,QAAM,QAAQ,IAAI,OAAO;AAC3B;AAEO,IAAM,mBAAN,MAA0B;AAAA,EAsB/B,YAA6B,MAAyB;AAAzB;AAC3B,SAAK,QAAQ,eAAe,KAAK,KAAK;AACtC,QAAI,KAAK,MAAM,WAAW;AACxB,YAAM,IAAI,MAAM,4EAAkE;AACpF,QAAI,KAAK,kBAAkB,QAAW;AACpC,UACE,OAAO,KAAK,kBAAkB,YAC9B,CAAC,OAAO,SAAS,KAAK,aAAa,KACnC,KAAK,gBAAgB,GACrB;AACA,cAAM,IAAI;AAAA,UACR,4EAA4E,OAAO,KAAK,aAAa,CAAC;AAAA,QACxG;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,KAAK,WAAW,KAAK,kBAAkB,UAAa,KAAK,UAAU,KAAK,SAAS;AACpF,YAAM,IAAI;AAAA,QACR;AAAA,MAEF;AAAA,IACF;AACA,SACG,KAAK,kBAAkB,UAAa,KAAK,QAAQ,mBAAmB,WACrE,CAAC,KAAK,iBACN;AACA,YAAM,IAAI;AAAA,QACR;AAAA,MACF;AAAA,IACF;AACA,QACE,KAAK,UACL,KAAK,kBAAkB,UACvB,KAAK,OAAO,mBAAmB,KAAK,eACpC;AACA,YAAM,IAAI;AAAA,QACR;AAAA,MACF;AAAA,IACF;AACA,QAAI,KAAK,OAAQ,MAAK,aAAa,KAAK,UAAU,IAAI,WAAW,KAAK,aAAa;AACnF,SAAK,WAAW,IAAI,IAAI,KAAK,MAAM,IAAI,CAAC,MAAM,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;AACxD,SAAK,YAAY,KAAK,aAAa,qBAAqB,GAAG;AAC3D,SAAK,YAAY,KAAK,UAAU,aAAa;AAC7C,SAAK,eAAe,KAAK,gBAAgB;AACzC,SAAK,iBAAiB,KAAK;AAAA,MACzB,KAAK,MAAM,SAAS,KAAK;AAAA,MACzB,KAAK,KAAK,KAAK,SAAS,CAAC;AAAA,IAC3B;AACA,SAAK,YAAY,KAAK,QAAQ,OAAO;AAAA,EACvC;AAAA,EAhD6B;AAAA,EArBZ;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA;AAAA,EAGA,MAAkD,CAAC;AAAA;AAAA,EAEnD,eAAe,oBAAI,IAA6B;AAAA,EAChD,YAA0B,CAAC;AAAA,EACpC,WAAW;AAAA,EACX,oBAAoB;AAAA,EACpB,aAAa;AAAA,EACb,wBAAwB;AAAA,EACxB;AAAA,EACA;AAAA,EACS;AAAA,EACA,YAAY;AAAA;AAAA,EAqDrB,MAAM,MAAc;AAC1B,SAAK,WAAY,KAAK,WAAW,aAAc;AAC/C,QAAI,IAAI,KAAK;AACb,QAAI,KAAK,KAAK,IAAK,MAAM,IAAK,IAAI,CAAC;AACnC,SAAK,IAAI,KAAK,KAAK,IAAK,MAAM,GAAI,IAAI,EAAE;AACxC,aAAS,IAAK,MAAM,QAAS,KAAK;AAAA,EACpC;AAAA,EAEQ,MAAM,MAAY,YAAwD;AAChF,QAAI,CAAC,cAAc,OAAO,KAAK,UAAU,EAAE,WAAW,EAAG,QAAO,KAAK;AACrE,UAAM,WAAW,OAAO,KAAK,UAAU,EACpC,KAAK,EACL,IAAI,CAAC,MAAM,GAAG,CAAC,IAAI,WAAW,CAAC,CAAC,EAAE,EAClC,KAAK,GAAG;AACX,WAAO,GAAG,KAAK,EAAE,IAAI,QAAQ;AAAA,EAC/B;AAAA,EAEQ,SAAS,QAA0D;AACzE,SAAK,KAAK,KAAK,cAAc,gBAAgB,WAAW;AACtD,YAAM,MAAM,KAAK,IAAI,KAAK,cAAc,KAAK,MAAM,SAAS,KAAK,MAAM,MAAM,CAAC;AAC9E,aAAO,KAAK,MAAM,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,IAAI,OAAO,IAAI,EAAE;AAAA,IAC7D;AACA,WAAO;AAAA,MACL,KAAK,IAAI,IAAI,CAAC,OAAO;AAAA,QACnB,WAAW,EAAE,KAAK;AAAA,QAClB,YAAY,EAAE;AAAA,QACd,OAAO,EAAE,GAAG;AAAA,QACZ,MAAM,EAAE,GAAG,SAAS,EAAE,GAAG,SAAS;AAAA,MACpC,EAAE;AAAA,MACF,KAAK,MAAM,IAAI,CAAC,OAAO,EAAE,WAAW,EAAE,IAAI,YAAY,IAAI,EAAE;AAAA,MAC5D,EAAE,QAAQ,cAAc,KAAK,aAAa;AAAA,IAC5C,EAAE,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,OAAO,EAAE,MAAM,EAAE;AAAA,EACxD;AAAA,EAEQ,mBAAmB;AACzB,UAAM,UAAU,CAAC,GAAG,KAAK,aAAa,OAAO,CAAC;AAC9C,WAAO;AAAA,MACL,eAAe,QAAQ,IAAI,CAAC,MAAM,EAAE,WAAW,KAAK;AAAA,MACpD,oBAAoB,QAAQ,IAAI,CAAC,MAAM,EAAE,WAAW,UAAU;AAAA,IAChE;AAAA,EACF;AAAA;AAAA,EAGQ,UAAUA,SAAqB;AACrC,UAAM,MAAW,CAAC;AAClB,eAAW,KAAK,KAAK,aAAa,OAAO,EAAG,KAAI,EAAE,KAAK,OAAOA,QAAQ,KAAI,KAAK,EAAE,QAAQ;AACzF,WAAO;AAAA,EACT;AAAA;AAAA,EAGA,MAAM,OAA0D;AAC9D,UAAM,YAAY,KAAK,KAAK,SAAS,KAAK;AAC1C,QAAI,aAAa,KAAK,KAAK,KAAK,QAAQ,QAAS,QAAO,EAAE,MAAM,GAAG,UAAU,CAAC,EAAE;AAEhF,UAAM,cAAc,KAAK,SAAS,KAAK,IAAI,KAAK,gBAAgB,SAAS,CAAC;AAC1E,UAAM,cAA4B,CAAC;AACnC,QAAI,eAAe;AAEnB,eAAW,SAAS,aAAa;AAC/B,UACE,KAAK,YAAY,KAAK,KAAK,UAC3B,KAAK,iBAAiB,UACtB,KAAK,KAAK,QAAQ;AAElB;AACF,YAAM,OAAO,KAAK,SAAS,IAAI,MAAM,MAAM;AAC3C,UAAI,CAAC,KAAM;AACX,YAAM,MAAM,KAAK,IAAI,MAAM,OAAO,KAAK,KAAK,SAAS,KAAK,QAAQ;AAClE,UAAI,OAAO,EAAG;AACd,WAAK,KAAK,aAAa,EAAE,MAAM,kBAAkB,MAAM,OAAO,IAAI,CAAC;AAEnE,YAAM,QAAQ,MAAM,KAAK,KAAK,SAAS,IAAI;AAC3C,YAAM,SAAS,KAAK,UAAU,KAAK,EAAE;AACrC,YAAM,WAAW,MAAM,KAAK,KAAK,SAAS;AAAA,QACxC;AAAA,QACA;AAAA,QACA;AAAA,QACA,UAAU,KAAK;AAAA,QACf,OAAO;AAAA,QACP,KAAK,KAAK;AAAA,MACZ,CAAC;AAGD,YAAM,OAAO,KAAK,IAAI,MAAM,CAAC,MAAM,EAAE,KAAK,OAAO,KAAK,EAAE;AACxD,YAAM,SAAS,CAAC,GAAI,OAAO,QAAQ,CAAC,GAAI,GAAG,QAAQ,EAAE,MAAM,GAAG,GAAG;AAEjE,YAAM;AAAA,QACJ;AAAA,QACA,OAAO,aAAa;AAClB,cACE,KAAK,YAAY,KAAK,KAAK,UAC3B,KAAK,iBAAiB,UACtB,KAAK,KAAK,QAAQ;AAElB;AAMF,cAAI;AACF,kBAAM,YAAY,YAAY,IAAI;AAClC,kBAAM,OAAO,KAAK,aACd,MAAM,KAAK,WAAW,YAAY;AAAA,cAChC,SAAS;AAAA,cACT,OAAO,KAAK;AAAA,cACZ,OAAO;AAAA,cACP,MAAM,EAAE,QAAQ,KAAK,KAAK,QAAQ,MAAM,KAAK,GAAG;AAAA,cAChD,QAAQ,KAAK,KAAK;AAAA,cAClB,eAAe,KAAK,KAAK,kBAAkB,UAAU,IAAI;AAAA,cACzD,SAAS,MAAM,KAAK,KAAK,SAAS,UAAU,IAAI;AAAA,cAChD,SAAS,CAAC,eAAe;AACvB,sBAAM,OAAO,KAAK,KAAK,OAAQ,UAAU,MAAM,UAAU;AACzD,oBAAI,SAAS,MAAM;AACjB,yBAAO;AAAA,oBACL,OAAO;AAAA,oBACP,aAAa;AAAA,oBACb,cAAc;AAAA,oBACd,aAAa;AAAA,kBACf;AAAA,gBACF;AACA,oBAAI,CAAC,OAAO,SAAS,KAAK,GAAG,KAAK,KAAK,MAAM,GAAG;AAC9C,wBAAM,IAAI;AAAA,oBACR,qDAAqD,OAAO,KAAK,GAAG,CAAC;AAAA,kBAEvE;AAAA,gBACF;AACA,uBAAO;AAAA,kBACL,OAAO,KAAK,SAAS;AAAA,kBACrB,aAAa;AAAA,kBACb,cAAc;AAAA,kBACd,eAAe,KAAK;AAAA,gBACtB;AAAA,cACF;AAAA,YACF,CAAC,IACD;AACJ,gBAAI,QAAQ,CAAC,KAAK,WAAW;AAC3B,kBACE,KAAK,iBAAiB,2BACtB,KAAK,MAAM,wBAAwB,YACnC;AACA,sBAAM,KAAK,MAAM;AAAA,cACnB;AACA,oBAAM,KAAK;AAAA,YACb;AACA,kBAAM,KAAK,OAAO,KAAK,QAAQ,MAAM,KAAK,KAAK,SAAS,UAAU,IAAI;AAGtE,kBAAM,YAAY,GAAG,aAAa,YAAY,IAAI,IAAI;AACtD,iBAAK;AACL;AACA,iBAAK,wBAAwB;AAC7B,kBAAM,UAAU,OACZ,KAAK,QAAQ,cACX,OACA,KAAK,QAAQ,UACf;AACJ,gBAAI,MAAM,QAAQ,kBAAkB,QAAW;AAC7C,mBAAK,KAAK,SAAS,EAAE,KAAK,KAAK,QAAQ,eAAe,SAAS,QAAQ,CAAC;AAAA,YAC1E;AACA,kBAAM,WAAW,KAAK,UAAU,SAAS,IAAI,KAAK,iBAAiB,CAAC;AACpE,iBAAK,IAAI,KAAK;AAAA,cACZ;AAAA,cACA;AAAA,cACA;AAAA,cACA;AAAA,cACA,GAAI,YAAY,SAAY,EAAE,QAAQ,IAAI,CAAC;AAAA,cAC3C,YAAY,KAAK,KAAK,WAAW,QAAQ;AAAA,YAC3C,CAAC;AACD,iBAAK,KAAK,aAAa,EAAE,MAAM,aAAa,MAAM,UAAU,YAAY,GAAG,CAAC;AAE5E,kBAAM,MAAM,KAAK,MAAM,MAAM,GAAG,UAAU;AAC1C,kBAAM,MAAM,KAAK,aAAa,IAAI,GAAG;AACrC,gBAAI,CAAC,OAAO,WAAW,IAAI;AACzB,mBAAK,aAAa,IAAI,KAAK,EAAE,OAAO,KAAK,MAAM,UAAU,YAAY,IAAI,SAAS,CAAC;AAErF,gBAAI,WAAW,KAAK,UAAW;AAC/B,iBAAK;AACL,gBAAI,KAAK,KAAK,OAAO,WAAW,CAAE,MAAM,KAAK,KAAK,MAAM,QAAQ,UAAU,IAAI,IAAI;AAChF;AACF,gBACE,KAAK,KAAK,OAAO,oBACjB,CAAE,MAAM,KAAK,KAAK,MAAM,iBAAiB,UAAU,IAAI,IAAI;AAE3D;AAEF,kBAAM,YAAY,KAAK,KAAK,WACxB,MAAM,KAAK,KAAK,SAAS,UAAU,KAAK,KAAK,UAAU,IAAI,IAC3D;AACJ,kBAAM,UAAsB;AAAA,cAC1B,IAAI,KAAK,KAAK,WAAW,QAAQ;AAAA,cACjC;AAAA,cACA;AAAA,cACA;AAAA,cACA,MAAM,KAAK,KAAK,eAAe,SAAS;AAAA,cACxC,YAAY;AAAA,cACZ;AAAA,cACA,WAAW,KAAK,UAAU;AAAA,YAC5B;AACA,iBAAK,UAAU,KAAK,OAAO;AAC3B,wBAAY,KAAK,OAAO;AACxB,iBAAK,KAAK,aAAa,EAAE,MAAM,WAAW,QAAQ,CAAC;AAAA,UACrD,SAAS,KAAK;AAGZ,gBAAI,eAAe,WAAY,OAAM;AACrC,iBAAK;AACL,iBAAK;AACL,kBAAM,UAAU,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC/D,iBAAK,KAAK,aAAa;AAAA,cACrB,MAAM;AAAA,cACN;AAAA,cACA,YAAY,KAAK,KAAK,WAAW,QAAQ;AAAA,cACzC;AAAA,YACF,CAAC;AACD,kBAAM,QAAQ,KAAK,KAAK,4BAA4B;AACpD,gBAAI,KAAK,yBAAyB,OAAO;AACvC,mBAAK,eAAe;AAAA,gBAClB,QAAQ;AAAA,gBACR,QAAQ,GAAG,KAAK,qBAAqB,mCAAmC,OAAO;AAAA,cACjF;AAAA,YACF;AAAA,UACF;AAAA,QACF;AAAA,QACA,KAAK,KAAK,eAAe;AAAA,QACzB,KAAK,KAAK;AAAA,MACZ;AAAA,IACF;AAEA,SAAK,KAAK,aAAa,EAAE,MAAM,SAAS,UAAU,KAAK,UAAU,QAAQ,KAAK,KAAK,OAAO,CAAC;AAC3F,WAAO,EAAE,MAAM,cAAc,UAAU,YAAY;AAAA,EACrD;AAAA;AAAA;AAAA,EAIA,MAAM,MAA+B;AACnC,WACE,KAAK,WAAW,KAAK,KAAK,UAC1B,KAAK,iBAAiB,UACtB,CAAC,KAAK,KAAK,QAAQ,SACnB;AACA,YAAM,EAAE,KAAK,IAAI,MAAM,KAAK,KAAK;AACjC,UAAI,SAAS,KAAK,KAAK,iBAAiB,OAAW;AAAA,IACrD;AACA,WAAO,KAAK,QAAQ;AAAA,EACtB;AAAA,EAEA,WAA2B;AACzB,WAAO,KAAK,QAAQ,EAAE;AAAA,EACxB;AAAA,EAEA,WAAyB;AACvB,WAAO,CAAC,GAAG,KAAK,SAAS,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,EACnE;AAAA,EAEA,UAA0B;AACxB,UAAM,OAAO,KAAK,YAAY,QAAQ;AAAA,MACpC,OAAO,KAAK;AAAA,MACZ,MAAM,EAAE,QAAQ,KAAK,KAAK,OAAO;AAAA,IACnC,CAAC;AACD,WAAO,aAAa;AAAA,MAClB,QAAQ,KAAK,KAAK;AAAA,MAClB,WAAW,KAAK,UAAU;AAAA,MAC1B,OAAO,KAAK;AAAA,MACZ,KAAK,KAAK;AAAA,MACV,WAAW,KAAK;AAAA,MAChB,SAAS,CAAC,GAAG,KAAK,aAAa,OAAO,CAAC;AAAA,MACvC,UAAU,KAAK;AAAA,MACf,mBAAmB,KAAK;AAAA,MACxB,UAAU,KAAK;AAAA,MACf,MAAM,OACF;AAAA,QACE,SAAS,KAAK;AAAA,QACd,iBAAiB,KAAK,UAAU,OAAO,CAAC,KAAK,QAAQ,MAAM,IAAI,eAAe,CAAC;AAAA,MACjF,IACA;AAAA,MACJ,YAAY,KAAK;AAAA,MACjB,cAAc,KAAK;AAAA,IACrB,CAAC;AAAA,EACH;AACF;;;ACrYA,eAAsB,UACpB,MACsC;AACtC,QAAM,EAAE,kBAAkB,GAAG,KAAK,IAAI;AACtC,QAAM,WAAW,IAAI,iBAAoB;AAAA,IACvC,GAAG;AAAA,IACH,WAAW,qBAAqB,oBAAoB,GAAG;AAAA,EACzD,CAAC;AACD,SAAO,EAAE,SAAS,MAAM,SAAS,IAAI,EAAE;AACzC;;;ACfO,SAAS,gBAAmB,MAA6D;AAC9F,QAAM,UAAU,KAAK,OAAO,CAAC,MAA6B,KAAK,IAAI;AACnE,SAAO;AAAA,IACL,SAAS,OAAO,UAAU,IAAI,SAAS;AACrC,iBAAW,KAAK,SAAS;AACvB,YAAI,EAAE,WAAW,CAAE,MAAM,EAAE,QAAQ,UAAU,IAAI,IAAI,EAAI,QAAO;AAAA,MAClE;AACA,aAAO;AAAA,IACT;AAAA,IACA,kBAAkB,OAAO,UAAU,IAAI,SAAS;AAC9C,iBAAW,KAAK,SAAS;AACvB,YAAI,EAAE,oBAAoB,CAAE,MAAM,EAAE,iBAAiB,UAAU,IAAI,IAAI,EAAI,QAAO;AAAA,MACpF;AACA,aAAO;AAAA,IACT;AAAA,EACF;AACF;AASO,SAAS,0BAA6B,MAMxB;AACnB,QAAM,YAAY,KAAK,oBAAoB;AAC3C,SAAO;AAAA,IACL,kBAAkB,OAAO,UAAa,KAAiB,SAAe;AACpE,YAAM,YAAY,KAAK,QAAQ,QAAQ;AACvC,UAAI,aAAa,KAAM,QAAO;AAC9B,YAAM,KAAK,MAAM,KAAK,SAAS,WAAW,IAAI;AAC9C,aAAO,GAAG,QAAQ;AAAA,IACpB;AAAA,EACF;AACF;AAMO,SAAS,kBAAqB,MAGhB;AACnB,QAAM,YAAY,KAAK,oBAAoB;AAC3C,QAAM,SAAS,KAAK,UAAU;AAC9B,SAAO;AAAA,IACL,SAAS,CAAC,WAAW,OAAmB,GAAG,SAAS,YAAY;AAAA,EAClE;AACF;;;AC/CA,IAAM,UAAU,EAAE,MAAM,UAAU,YAAY,CAAC,GAAG,sBAAsB,MAAM;AAEvE,SAAS,iBAAoB,UAAiD;AACnF,SAAO;AAAA,IACL;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,MACZ,SAAS,YAAY;AACnB,cAAM,EAAE,MAAM,SAAS,IAAI,MAAM,SAAS,KAAK;AAC/C,eAAO,EAAE,MAAM,aAAa,SAAS,QAAQ,SAAS;AAAA,MACxD;AAAA,IACF;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,MACZ,SAAS,YAAY,SAAS,SAAS;AAAA,IACzC;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aAAa;AAAA,MACb,YAAY;AAAA,QACV,MAAM;AAAA,QACN,YAAY,EAAE,OAAO,EAAE,MAAM,UAAU,aAAa,yBAAyB,EAAE;AAAA,QAC/E,sBAAsB;AAAA,MACxB;AAAA,MACA,SAAS,OAAO,SAAS;AACvB,cAAM,QAAS,MAAyC;AACxD,cAAM,MAAM,SAAS,SAAS;AAC9B,eAAO,OAAO,UAAU,WAAW,IAAI,MAAM,GAAG,KAAK,IAAI;AAAA,MAC3D;AAAA,IACF;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,QACV,MAAM;AAAA,QACN,YAAY;AAAA,UACV,QAAQ,EAAE,MAAM,UAAU,MAAM,CAAC,QAAQ,MAAM,EAAE;AAAA,UACjD,aAAa,EAAE,MAAM,UAAU,aAAa,uCAAuC;AAAA,QACrF;AAAA,QACA,sBAAsB;AAAA,MACxB;AAAA,MACA,SAAS,OAAO,SAAS;AACvB,cAAM,IAAK,QAAQ,CAAC;AACpB,cAAM,UAAU,SAAS,QAAQ;AACjC,YAAI,EAAE,WAAW,OAAQ,QAAO,kBAAkB,SAAS,EAAE,aAAa,EAAE,YAAY,CAAC;AACzF,eAAO;AAAA,MACT;AAAA,IACF;AAAA,EACF;AACF;","names":["cellId"]}
|
|
1
|
+
{"version":3,"sources":["../src/fuzz/cube.ts","../src/fuzz/capsule.ts","../src/fuzz/policies.ts","../src/fuzz/explorer.ts","../src/fuzz/fuzz-agent.ts","../src/fuzz/gates.ts","../src/fuzz/tools.ts"],"sourcesContent":["/**\n * Input-space tiling + coverage projection.\n *\n * Cells are the cartesian product of the input axes — the stratification plan,\n * enumerable up front so the planned-vs-covered denominator is honest. Coverage\n * is projected from the evaluation log: per cell, the full DISTRIBUTION of the\n * headline score, of each scored dimension, and of evaluation latency — a bare\n * mean hides outliers, so every aggregate carries its spread. Per-cell cost is\n * split known-dollars vs unknown-runs, never folded into a fabricated $0.\n */\n\nimport type { BehaviorSpace, Cell, CoverageCell, Distribution, Evaluation } from './types'\n\n/** One recorded evaluation — the unit coverage and the capsule are built from. */\nexport interface EvalRecord {\n cell: Cell\n ev: Evaluation\n /** The objective's interest score for this evaluation. */\n interest: number\n /** Evaluation wall-clock — engine-measured unless `ev.latencyMs` overrode it. */\n latencyMs: number\n /** Known dollars for this run. `null` = cost tracking was wired but this\n * run's cost was unknowable (counted apart). Absent = not tracked at all. */\n costUsd?: number | null\n}\n\n/** Enumerate every input cell (cartesian product of the axes), in stable order. */\nexport function enumerateCells(space: BehaviorSpace): Cell[] {\n if (space.axes.length === 0) return []\n let partials: Array<Record<string, string>> = [{}]\n for (const axis of space.axes) {\n const next: Array<Record<string, string>> = []\n for (const partial of partials) {\n for (const value of axis.values) next.push({ ...partial, [axis.name]: value })\n }\n partials = next\n }\n return partials.map((coords) => ({ id: cellId(space, coords), coords }))\n}\n\n/** Deterministic id for a coordinate map, e.g. `matterType=nda|difficulty=hard`. */\nexport function cellId(space: BehaviorSpace, coords: Record<string, string>): string {\n return space.axes.map((a) => `${a.name}=${coords[a.name]}`).join('|')\n}\n\n/** Nearest-rank percentile on a pre-sorted ascending sample. */\nfunction percentile(sorted: number[], p: number): number {\n const idx = Math.min(sorted.length - 1, Math.max(0, Math.ceil(p * sorted.length) - 1))\n return sorted[idx] as number\n}\n\n/** Summarize a sample. Throws on an empty sample — callers represent \"no data\"\n * as `null`, never as a zeroed distribution. */\nexport function distribution(values: number[]): Distribution {\n if (values.length === 0)\n throw new Error('distribution: empty sample — represent missing data as null, not zeros')\n const sorted = [...values].sort((a, b) => a - b)\n const mean = sorted.reduce((a, b) => a + b, 0) / sorted.length\n return {\n mean,\n median: percentile(sorted, 0.5),\n p90: percentile(sorted, 0.9),\n min: sorted[0] as number,\n max: sorted[sorted.length - 1] as number,\n n: sorted.length,\n }\n}\n\n/**\n * Project the evaluation log into the per-input-cell coverage map. A cell with\n * no evaluations reports `score: null` (honestly uncovered), never zeros.\n */\nexport function buildCoverage(cells: Cell[], log: EvalRecord[], threshold: number): CoverageCell[] {\n const byCell = new Map<string, EvalRecord[]>()\n for (const r of log) {\n const arr = byCell.get(r.cell.id) ?? []\n arr.push(r)\n byCell.set(r.cell.id, arr)\n }\n return cells.map((cell) => {\n const recs = byCell.get(cell.id) ?? []\n const runs = recs.length\n if (runs === 0)\n return { cell, runs: 0, score: null, findingRate: 0, dimensions: {}, latencyMs: null }\n\n const score = distribution(recs.map((r) => r.ev.score))\n const latencyMs = distribution(recs.map((r) => r.latencyMs))\n const findingRate = recs.filter((r) => r.interest >= threshold).length / runs\n\n const dimSamples: Record<string, number[]> = {}\n for (const r of recs) {\n for (const [k, v] of Object.entries(r.ev.scores ?? {})) {\n const samples = dimSamples[k] ?? []\n samples.push(v)\n dimSamples[k] = samples\n }\n }\n const dimensions: Record<string, Distribution> = {}\n for (const [k, xs] of Object.entries(dimSamples)) dimensions[k] = distribution(xs)\n\n // Cost fields appear only when tracking was wired: known dollars sum, and\n // tracked-but-unknown runs counted apart — never folded in as $0.\n const tracked = recs.filter((r) => r.costUsd !== undefined)\n const known = tracked.filter((r) => r.costUsd !== null)\n const cost =\n tracked.length > 0\n ? {\n costUsd: known.reduce((a, r) => a + (r.costUsd as number), 0),\n ...(tracked.length > known.length\n ? { costUnknownRuns: tracked.length - known.length }\n : {}),\n }\n : {}\n\n return { cell, runs, score, findingRate, dimensions, latencyMs, ...cost }\n })\n}\n","/**\n * The capsule — the artifact every exploration produces.\n *\n * `buildCapsule` assembles coverage + verified findings + the QD archive into a\n * pure `CapsuleData` (no clock, no I/O — deterministic and snapshot-testable).\n * `renderCapsuleHtml` turns it into a standalone page: the input-cell heat-map\n * (planned vs covered), per-dimension weakness chips, and the minimized finding\n * exemplars. One artifact — the hardening map and the shareable proof object.\n */\n\nimport type { EvalRecord } from './cube'\nimport { buildCoverage, distribution } from './cube'\nimport type { ArchiveEntry, CapsuleData, Cell, CoverageCell, Distribution, Finding } from './types'\n\nexport interface BuildCapsuleInput<S> {\n target: string\n objective: string\n cells: Cell[]\n log: EvalRecord[]\n /** The objective's notable threshold — drives findingRate. */\n threshold: number\n archive: ArchiveEntry<S>[]\n findings: Finding<S>[]\n candidateFindings: number\n runsUsed: number\n /** Known-dollar / unknown-run split — present only when cost tracking was\n * wired; the capsule never fabricates a $0 total. */\n cost?: { costUsd: number; costUnknownRuns: number }\n /** Evaluations that threw — infra outcomes, never folded into robustness. */\n evalErrors: number\n /** Set when the consecutive-error circuit breaker stopped the run early. */\n stoppedEarly?: { reason: 'eval-errors'; detail: string }\n}\n\nexport function buildCapsule<S>(input: BuildCapsuleInput<S>): CapsuleData<S> {\n const coverage = buildCoverage(input.cells, input.log, input.threshold)\n const covered = coverage.filter((c) => c.runs > 0)\n // Cells weigh equally: variance steering sends more runs to weak cells, so a\n // run-weighted average would bias the headline low.\n const robustness =\n covered.length === 0 ? null : distribution(covered.map((c) => (c.score as Distribution).mean))\n const latencyMs = input.log.length === 0 ? null : distribution(input.log.map((r) => r.latencyMs))\n // Measured-descriptor bins beyond the bare input cell — observed, never planned.\n const behaviorBinsObserved = input.archive.filter((e) => e.binId !== e.cell.id).length\n\n return {\n target: input.target,\n objective: input.objective,\n coverage,\n findings: [...input.findings].sort((a, b) => b.interest - a.interest),\n archive: [...input.archive].sort((a, b) => b.interest - a.interest),\n stats: {\n totalRuns: input.runsUsed,\n cellsTotal: input.cells.length,\n cellsCovered: covered.length,\n behaviorBinsObserved,\n candidateFindings: input.candidateFindings,\n verifiedFindings: input.findings.length,\n robustness,\n latencyMs,\n ...(input.cost\n ? { costUsd: input.cost.costUsd, costUnknownRuns: input.cost.costUnknownRuns }\n : {}),\n evalErrors: input.evalErrors,\n ...(input.stoppedEarly ? { stoppedEarly: input.stoppedEarly } : {}),\n },\n }\n}\n\n// ── HTML capsule ──────────────────────────────────────────────────────────────\n\nfunction esc(s: string): string {\n return s.replace(\n /[&<>\"]/g,\n (c) => ({ '&': '&', '<': '<', '>': '>', '\"': '"' })[c] as string,\n )\n}\n\n/** red (0) → amber (.5) → green (1); uncovered cells render gray. */\nfunction robustnessColor(r: number | null): string {\n if (r == null) return '#2a2a2e'\n return `hsl(${Math.round(r * 120)} 70% 42%)`\n}\n\nfunction pct(x: number): string {\n return `${Math.round(x * 100)}%`\n}\n\nfunction deriveAxes(coverage: CoverageCell[]): Array<{ name: string; values: string[] }> {\n const order: string[] = []\n const seen = new Map<string, Set<string>>()\n for (const c of coverage) {\n for (const [k, v] of Object.entries(c.cell.coords)) {\n if (!seen.has(k)) {\n seen.set(k, new Set())\n order.push(k)\n }\n seen.get(k)?.add(v)\n }\n }\n return order.map((name) => ({ name, values: [...(seen.get(name) ?? [])] }))\n}\n\n/** The weakest dimension chip for a cell, e.g. `safety 32%` — shown when scores exist. */\nfunction weakestDim(c: CoverageCell): string {\n const entries = Object.entries(c.dimensions).map(([k, d]) => [k, d.mean] as [string, number])\n if (entries.length === 0) return ''\n const sorted = entries.sort((a, b) => a[1] - b[1])\n const w = sorted[0]\n if (!w) return ''\n return `<span class=\"dim\">${esc(w[0])} ${pct(w[1])}</span>`\n}\n\nfunction heatmapHtml(coverage: CoverageCell[]): string {\n const axes = deriveAxes(coverage)\n const byId = new Map(coverage.map((c) => [c.cell.id, c]))\n const tile = (c: CoverageCell | undefined, label: string): string => {\n const r = c?.score?.mean ?? null\n const title = c?.score\n ? `${pct(c.score.mean)} robust (median ${pct(c.score.median)}, min ${pct(c.score.min)}) · ${c.runs} runs · ${pct(c.findingRate)} flagged${c.latencyMs ? ` · ${(c.latencyMs.median / 1000).toFixed(1)}s median` : ''}${c.costUsd !== undefined ? ` · $${c.costUsd.toFixed(2)}` : ''}`\n : 'not covered'\n return `<div class=\"tile\" style=\"background:${robustnessColor(r)}\" title=\"${esc(title)}\">${label ? `<span class=\"tl\">${esc(label)}</span>` : ''}<span class=\"tv\">${c && r != null ? pct(r) : '—'}</span>${c ? weakestDim(c) : ''}</div>`\n }\n\n const rowAxis = axes[0]\n const colAxis = axes[1]\n if (axes.length === 2 && rowAxis && colAxis) {\n const head = `<tr><th></th>${colAxis.values.map((v) => `<th>${esc(v)}</th>`).join('')}</tr>`\n const rows = rowAxis.values\n .map((rv) => {\n const cells = colAxis.values\n .map((cv) => {\n const id = `${rowAxis.name}=${rv}|${colAxis.name}=${cv}`\n return `<td>${tile(byId.get(id), '')}</td>`\n })\n .join('')\n return `<tr><th class=\"rh\">${esc(rv)}</th>${cells}</tr>`\n })\n .join('')\n return `<div class=\"axis-label\">rows: <b>${esc(rowAxis.name)}</b> · cols: <b>${esc(colAxis.name)}</b></div><table class=\"heat\">${head}${rows}</table>`\n }\n\n const sorted = [...coverage].sort((a, b) => (a.score?.mean ?? 2) - (b.score?.mean ?? 2))\n return `<div class=\"grid\">${sorted.map((c) => tile(c, Object.values(c.cell.coords).join(' · '))).join('')}</div>`\n}\n\nfunction findingsHtml<S>(findings: Finding<S>[], limit: number): string {\n if (findings.length === 0)\n return `<p class=\"none\">No verified findings — the target held across every covered cell.</p>`\n return findings\n .slice(0, limit)\n .map((f) => {\n const coords = Object.entries(f.cell.coords)\n .map(([k, v]) => `${k}:${v}`)\n .join(' · ')\n const labels = (f.evaluation.labels ?? [])\n .map((l) => `<span class=\"fclass\">${esc(l)}</span>`)\n .join('')\n const dims = Object.entries(f.evaluation.scores ?? {})\n .sort((a, b) => a[1] - b[1])\n .slice(0, 3)\n .map(([k, v]) => `<span class=\"dim\">${esc(k)} ${pct(v)}</span>`)\n .join('')\n return `<div class=\"fail\"><div class=\"fmeta\"><span class=\"fcell\">${esc(coords)}</span>${labels}<span class=\"fsev\">interest ${pct(f.interest)}</span></div><div class=\"sevbar\"><div class=\"sevfill\" style=\"width:${pct(f.interest)}\"></div></div><div class=\"ftext\">${esc(f.text ?? '(scenario text not captured)')}</div>${dims ? `<div class=\"fdims\">${dims}</div>` : ''}</div>`\n })\n .join('')\n}\n\nexport interface RenderCapsuleOptions {\n /** Max finding exemplars to show. Default 8. */\n maxFindings?: number\n /** ISO timestamp to stamp into the page (keeps the pure capsule clock-free). */\n generatedAt?: string\n}\n\n/** Render a self-contained HTML capsule — heat-map + per-dimension chips + verified findings. */\nexport function renderCapsuleHtml<S>(\n capsule: CapsuleData<S>,\n opts: RenderCapsuleOptions = {},\n): string {\n const s = capsule.stats\n const kpi = (label: string, value: string, accent = '#e6e6e6'): string =>\n `<div class=\"kpi\"><div class=\"kv\" style=\"color:${accent}\">${esc(value)}</div><div class=\"kl\">${esc(label)}</div></div>`\n const lift = capsule.lift\n ? kpi(\n 'hardening lift',\n `${capsule.lift.before.toFixed(2)} → ${capsule.lift.after.toFixed(2)}`,\n '#5ad17a',\n )\n : ''\n // Cost KPI only when tracking was wired — an untracked run never shows $0.\n // Unpriced runs are named in the label (amber): the total is a lower bound.\n const cost =\n s.costUsd !== undefined\n ? kpi(\n s.costUnknownRuns ? `cost · ${s.costUnknownRuns} runs unpriced` : 'cost',\n `$${s.costUsd.toFixed(2)}`,\n s.costUnknownRuns ? '#e5c07b' : '#e6e6e6',\n )\n : ''\n const stamp = opts.generatedAt ?? capsule.generatedAt ?? ''\n\n return `<!doctype html><html><head><meta charset=\"utf-8\"><meta name=\"viewport\" content=\"width=device-width,initial-scale=1\">\n<title>${esc(capsule.objective)} capsule — ${esc(capsule.target)}</title>\n<style>\n:root{color-scheme:dark}\n*{box-sizing:border-box}\nbody{margin:0;background:#0c0c0f;color:#e6e6e6;font:14px/1.5 ui-sans-serif,system-ui,-apple-system,Segoe UI,Roboto,sans-serif}\n.wrap{max-width:980px;margin:0 auto;padding:40px 24px}\nh1{font-size:22px;margin:0 0 2px;letter-spacing:-.01em}\n.sub{color:#8a8a93;font-size:13px;margin-bottom:28px}\n.kpis{display:flex;gap:14px;flex-wrap:wrap;margin-bottom:32px}\n.kpi{background:#16161b;border:1px solid #24242b;border-radius:12px;padding:14px 18px;min-width:120px}\n.kv{font-size:24px;font-weight:650;letter-spacing:-.02em}\n.kl{color:#8a8a93;font-size:12px;margin-top:2px}\nh2{font-size:13px;text-transform:uppercase;letter-spacing:.08em;color:#8a8a93;margin:34px 0 14px}\n.axis-label{color:#8a8a93;font-size:12px;margin-bottom:10px}\ntable.heat{border-collapse:separate;border-spacing:6px}\ntable.heat th{font-weight:500;color:#a8a8b0;font-size:12px;padding:4px 8px;text-align:center}\ntable.heat th.rh{text-align:right}\n.tile{position:relative;width:104px;height:66px;border-radius:10px;display:flex;flex-direction:column;justify-content:center;align-items:center;gap:1px;color:#06120a;font-weight:600}\n.tile .tv{font-size:17px}\n.tile .tl{font-size:10px;font-weight:600;opacity:.9;padding:0 6px;text-align:center}\n.tile .dim{font-size:9px;font-weight:600;opacity:.85;background:rgba(0,0,0,.22);padding:0 6px;border-radius:999px}\n.grid{display:flex;flex-wrap:wrap;gap:8px}\n.grid .tile{width:138px;height:74px}\n.fail{background:#16161b;border:1px solid #24242b;border-left:3px solid #d1495b;border-radius:10px;padding:12px 14px;margin-bottom:10px}\n.fmeta{display:flex;gap:10px;align-items:center;font-size:12px;color:#a8a8b0;margin-bottom:8px}\n.fcell{color:#cfcfd6}\n.fclass{background:#2a1a1d;color:#e58a96;padding:1px 8px;border-radius:999px;font-size:11px}\n.fsev{margin-left:auto;color:#e58a96}\n.sevbar{height:4px;background:#24242b;border-radius:2px;overflow:hidden;margin-bottom:10px}\n.sevfill{height:100%;background:#d1495b}\n.ftext{font-size:13px;color:#d8d8df;white-space:pre-wrap}\n.fdims{display:flex;gap:6px;margin-top:8px}\n.fdims .dim{background:#1d2530;color:#9fc1e8;padding:1px 8px;border-radius:999px;font-size:11px}\n.none{color:#5ad17a}\n.foot{color:#5a5a63;font-size:11px;margin-top:36px}\n</style></head><body><div class=\"wrap\">\n<h1>${esc(capsule.objective)} exploration · ${esc(capsule.target)}</h1>\n<div class=\"sub\">${s.totalRuns} scenarios across ${s.cellsCovered}/${s.cellsTotal} planned cells${s.behaviorBinsObserved > 0 ? ` · ${s.behaviorBinsObserved} measured behavior bins` : ''}${stamp ? ` · ${esc(stamp)}` : ''}</div>\n<div class=\"kpis\">\n${s.robustness ? kpi('robustness', `${pct(s.robustness.mean)}`, s.robustness.mean < 0.6 ? '#e58a96' : '#5ad17a') : ''}\n${s.robustness ? kpi('cell spread', `${pct(s.robustness.min)}–${pct(s.robustness.max)}`) : ''}\n${s.latencyMs ? kpi('median latency', `${(s.latencyMs.median / 1000).toFixed(1)}s`, s.latencyMs.p90 > 4 * s.latencyMs.median ? '#e5b566' : '#e6e6e6') : ''}\n${kpi('verified findings', String(s.verifiedFindings), s.verifiedFindings > 0 ? '#e58a96' : '#5ad17a')}\n${kpi('cells covered', `${s.cellsCovered}/${s.cellsTotal}`)}\n${kpi('scenarios run', String(s.totalRuns))}\n${cost}\n${s.evalErrors > 0 ? kpi('eval errors', String(s.evalErrors), '#e5b566') : ''}\n${lift}\n</div>\n${s.stoppedEarly ? `<div class=\"sub\" style=\"color:#e5b566\">stopped early: ${esc(s.stoppedEarly.detail)} — coverage below reflects the completed portion only</div>` : ''}\n<h2>Coverage map</h2>\n${heatmapHtml(capsule.coverage)}\n<h2>Verified findings${s.candidateFindings > s.verifiedFindings ? ` · ${s.verifiedFindings} of ${s.candidateFindings} candidates passed the validity gates` : ''}</h2>\n${findingsHtml(capsule.findings, opts.maxFindings ?? 8)}\n<div class=\"foot\">Findings are gate-verified: each is a fair, answerable task that reproduces under a meaning-preserving rephrase, minimized to the smallest trigger.</div>\n</div></body></html>`\n}\n","/**\n * Shipped policies for the exploration engine.\n *\n * `MutationProposer` is a plain function type — an agent running a generator skill\n * IS a proposer (`(ctx) => dispatchToSkill(ctx)`), no wrapper needed. `mutationProposer`\n * builds the deterministic, LLM-free one from mutation operators. Objectives are\n * interfaces because the engine reads `kind` + `threshold` off them.\n */\n\nimport type { AdversarialMutation } from '../rl/adversarial'\nimport type {\n Cell,\n Evaluation,\n MutationProposer,\n Objective,\n ObjectiveContext,\n ProposeContext,\n} from './types'\n\nconst clamp01 = (x: number): number => Math.max(0, Math.min(1, x))\n\n// ── proposers ─────────────────────────────────────────────────────────────────\n\n/**\n * Perturbation-based search: apply the cell's mutation operators to the current\n * elites + seeds, deduping by id. Elites first — mutating the most interesting\n * scenario found so far is what makes the search deepen across rounds.\n */\nexport function mutationProposer<S>(opts: {\n mutationsFor: (cell: Cell) => AdversarialMutation<S>[]\n scenarioId: (s: S) => string\n}): MutationProposer<S> {\n return async (ctx: ProposeContext<S>): Promise<S[]> => {\n const mutations = opts.mutationsFor(ctx.cell)\n const parents = [...ctx.elites, ...ctx.seeds]\n const seen = new Set(parents.map(opts.scenarioId))\n const out: S[] = []\n for (const parent of parents) {\n if (out.length >= ctx.count) break\n for (const m of mutations) {\n const children = await m.mutate(parent, ctx.rng)\n for (const child of children) {\n const id = opts.scenarioId(child)\n if (seen.has(id)) continue\n seen.add(id)\n out.push(child)\n if (out.length >= ctx.count) break\n }\n if (out.length >= ctx.count) break\n }\n }\n return out\n }\n}\n\n// ── objectives ────────────────────────────────────────────────────────────────\n\n/** Adversarial: a low headline score is interesting — find where the agent fails. */\nexport function adversarialObjective(threshold = 0.5): Objective {\n return { kind: 'adversarial', threshold, interest: (ev) => clamp01(1 - ev.score) }\n}\n\nfunction hamming(\n a: Record<string, string> | undefined,\n b: Record<string, string> | undefined,\n): number {\n if (!a || !b) return 1\n const keys = new Set([...Object.keys(a), ...Object.keys(b)])\n if (keys.size === 0) return 0\n let diff = 0\n for (const k of keys) if (a[k] !== b[k]) diff++\n return diff / keys.size\n}\n\n/**\n * Novelty: interesting when far from the archive in score AND measured behavior\n * descriptor — quality-diversity's diversity pressure; drives corpus growth\n * rather than re-finding the same hole.\n */\nexport function noveltyObjective(threshold = 0.3): Objective {\n return {\n kind: 'novelty',\n threshold,\n interest: (ev: Evaluation, ctx: ObjectiveContext) => {\n const scoreNovelty =\n ctx.archiveScores.length === 0\n ? 1\n : Math.min(...ctx.archiveScores.map((s) => Math.abs(s - ev.score)))\n const descNovelty =\n ctx.archiveDescriptors.length === 0\n ? 1\n : Math.min(...ctx.archiveDescriptors.map((d) => hamming(d, ev.descriptor)))\n return clamp01(0.5 * scoreNovelty + 0.5 * descNovelty)\n },\n }\n}\n","/**\n * The exploration engine — a stateful session over a behavior space.\n *\n * Each `step()`: allocate budget across INPUT cells (floor first, then variance\n * steering toward the least-certain cells), propose candidates (the proposer\n * reads current elites + findings, so the search deepens generationally),\n * evaluate with bounded concurrency, archive the most interesting scenario per\n * input×measured bin, and admit notable candidates that pass the validity gates.\n * `run()` loops to budget. `coverage()`/`findings()`/`capsule()` read live state —\n * the surface `makeExploreTools` exposes so an agent can drive the session.\n *\n * One evaluation log (`EvalRecord[]`) is the source of truth; allocation\n * observations and coverage are projections of it.\n */\n\nimport { CostLedger, type CostLedgerHandle, CostReceiptCaptureError } from '../cost-ledger'\nimport { ValidationError } from '../errors'\nimport { varianceBasedCurriculum } from '../rl/active-curriculum'\nimport { buildCapsule } from './capsule'\nimport type { EvalRecord } from './cube'\nimport { enumerateCells } from './cube'\nimport { adversarialObjective } from './policies'\nimport type {\n ArchiveEntry,\n CapsuleData,\n Cell,\n CoverageCell,\n ExploreOptions,\n Finding,\n Objective,\n} from './types'\n\nasync function pMap<T>(\n items: T[],\n fn: (item: T) => Promise<void>,\n concurrency: number,\n signal?: AbortSignal,\n): Promise<void> {\n let i = 0\n const workers = Array.from(\n { length: Math.max(1, Math.min(concurrency, items.length)) },\n async () => {\n while (i < items.length) {\n if (signal?.aborted) return\n const item = items[i++]\n if (item === undefined) return\n await fn(item)\n }\n },\n )\n await Promise.all(workers)\n}\n\nexport class BehaviorExplorer<S> {\n private readonly cells: Cell[]\n private readonly cellById: Map<string, Cell>\n private readonly objective: Objective\n private readonly threshold: number\n private readonly floorPerCell: number\n private readonly perRoundBudget: number\n\n /** The single evaluation log — coverage + allocation are projections of it. */\n private readonly log: Array<EvalRecord & { scenarioId: string }> = []\n /** binId (input × measured coords) → the most interesting entry seen. */\n private readonly archiveByBin = new Map<string, ArchiveEntry<S>>()\n private readonly _findings: Finding<S>[] = []\n private runsUsed = 0\n private candidateFindings = 0\n private evalErrors = 0\n private consecutiveEvalErrors = 0\n private stoppedEarly: { reason: 'eval-errors'; detail: string } | undefined\n private rngState: number\n private readonly costLedger?: CostLedgerHandle\n private readonly costPhase = 'fuzz.explore'\n\n constructor(private readonly opts: ExploreOptions<S>) {\n this.cells = enumerateCells(opts.space)\n if (this.cells.length === 0)\n throw new Error('BehaviorExplorer: space has no cells — every axis needs ≥1 value')\n if (opts.costBudgetUsd !== undefined) {\n if (\n typeof opts.costBudgetUsd !== 'number' ||\n !Number.isFinite(opts.costBudgetUsd) ||\n opts.costBudgetUsd < 0\n ) {\n throw new RangeError(\n `BehaviorExplorer: costBudgetUsd must be a nonnegative finite number, got ${String(opts.costBudgetUsd)}`,\n )\n }\n }\n if (!opts.costOf && (opts.costBudgetUsd !== undefined || opts.ledger || opts.onCost)) {\n throw new ValidationError(\n 'BehaviorExplorer: costBudgetUsd/ledger/onCost require costOf — the explorer ' +\n 'cannot know run cost without it; supply costOf or drop the cost options',\n )\n }\n if (\n (opts.costBudgetUsd !== undefined || opts.ledger?.costCeilingUsd !== undefined) &&\n !opts.maximumChargeOf\n ) {\n throw new ValidationError(\n 'BehaviorExplorer: capped cost tracking requires maximumChargeOf before evaluation',\n )\n }\n if (\n opts.ledger &&\n opts.costBudgetUsd !== undefined &&\n opts.ledger.costCeilingUsd !== opts.costBudgetUsd\n ) {\n throw new ValidationError(\n 'BehaviorExplorer: costBudgetUsd must match the supplied CostLedger ceiling',\n )\n }\n if (opts.costOf) this.costLedger = opts.ledger ?? new CostLedger(opts.costBudgetUsd)\n this.cellById = new Map(this.cells.map((c) => [c.id, c]))\n this.objective = opts.objective ?? adversarialObjective(0.5)\n this.threshold = this.objective.threshold ?? 0.5\n this.floorPerCell = opts.floorPerCell ?? 2\n this.perRoundBudget = Math.max(\n this.cells.length * this.floorPerCell,\n Math.ceil(opts.budget / 4),\n )\n this.rngState = (opts.seed ?? 1) >>> 0\n }\n\n // mulberry32 — deterministic per session.\n private rng = (): number => {\n this.rngState = (this.rngState + 0x6d2b79f5) | 0\n let t = this.rngState\n t = Math.imul(t ^ (t >>> 15), t | 1)\n t ^= t + Math.imul(t ^ (t >>> 7), t | 61)\n return ((t ^ (t >>> 14)) >>> 0) / 4294967296\n }\n\n private binId(cell: Cell, descriptor: Record<string, string> | undefined): string {\n if (!descriptor || Object.keys(descriptor).length === 0) return cell.id\n const measured = Object.keys(descriptor)\n .sort()\n .map((k) => `${k}=${descriptor[k]}`)\n .join('|')\n return `${cell.id}|${measured}`\n }\n\n private allocate(budget: number): Array<{ cellId: string; count: number }> {\n if ((this.opts.allocation ?? 'variance') === 'uniform') {\n const per = Math.max(this.floorPerCell, Math.floor(budget / this.cells.length))\n return this.cells.map((c) => ({ cellId: c.id, count: per }))\n }\n return varianceBasedCurriculum(\n this.log.map((r) => ({\n variantId: r.cell.id,\n scenarioId: r.scenarioId,\n score: r.ev.score,\n pass: r.ev.valid && r.ev.score >= 0.5,\n })),\n this.cells.map((c) => ({ variantId: c.id, scenarioId: '*' })),\n { budget, floorPerCell: this.floorPerCell },\n ).map((a) => ({ cellId: a.variantId, count: a.count }))\n }\n\n private objectiveContext() {\n const entries = [...this.archiveByBin.values()]\n return {\n archiveScores: entries.map((e) => e.evaluation.score),\n archiveDescriptors: entries.map((e) => e.evaluation.descriptor),\n }\n }\n\n /** Elites whose INPUT cell matches — what the proposer mutates/deepens from. */\n private elitesFor(cellId: string): S[] {\n const out: S[] = []\n for (const e of this.archiveByBin.values()) if (e.cell.id === cellId) out.push(e.scenario)\n return out\n }\n\n /** One allocate → propose → evaluate → gate → archive round. */\n async step(): Promise<{ runs: number; findings: Finding<S>[] }> {\n const remaining = this.opts.budget - this.runsUsed\n if (remaining <= 0 || this.opts.signal?.aborted) return { runs: 0, findings: [] }\n\n const allocations = this.allocate(Math.min(this.perRoundBudget, remaining))\n const newFindings: Finding<S>[] = []\n let runsThisStep = 0\n\n for (const alloc of allocations) {\n if (\n this.runsUsed >= this.opts.budget ||\n this.stoppedEarly !== undefined ||\n this.opts.signal?.aborted\n )\n break\n const cell = this.cellById.get(alloc.cellId)\n if (!cell) continue\n const cap = Math.min(alloc.count, this.opts.budget - this.runsUsed)\n if (cap <= 0) continue\n this.opts.onProgress?.({ type: 'cell-allocated', cell, count: cap })\n\n const seeds = await this.opts.seedsFor(cell)\n const elites = this.elitesFor(cell.id)\n const proposed = await this.opts.proposer({\n cell,\n seeds,\n elites,\n findings: this._findings,\n count: cap,\n rng: this.rng,\n })\n // Cold cells evaluate their seeds first (the coverage floor); warm cells\n // trust the proposer, which already saw the elites.\n const cold = this.log.every((r) => r.cell.id !== cell.id)\n const toEval = [...(cold ? seeds : []), ...proposed].slice(0, cap)\n\n await pMap(\n toEval,\n async (scenario) => {\n if (\n this.runsUsed >= this.opts.budget ||\n this.stoppedEarly !== undefined ||\n this.opts.signal?.aborted\n )\n return\n // evaluate/gates/minimize cross an external boundary (router, backend,\n // judge). A throw there is an infra outcome: record it as a typed\n // eval-error and keep exploring — one 5xx must not kill a campaign.\n // Consecutive failures trip the circuit breaker instead, so a dead\n // backend stops the run rather than burning the remaining budget.\n try {\n const startedAt = performance.now()\n const paid = this.costLedger\n ? await this.costLedger.runPaidCall({\n channel: 'agent',\n phase: this.costPhase,\n actor: 'fuzz.evaluate',\n tags: { target: this.opts.target, cell: cell.id },\n signal: this.opts.signal,\n maximumCharge: this.opts.maximumChargeOf?.(scenario, cell),\n execute: () => this.opts.evaluate(scenario, cell),\n receipt: (evaluation) => {\n const cost = this.opts.costOf!(scenario, cell, evaluation)\n if (cost === null) {\n return {\n model: 'unattributed',\n inputTokens: 0,\n outputTokens: 0,\n costUnknown: true,\n }\n }\n if (!Number.isFinite(cost.usd) || cost.usd < 0) {\n throw new RangeError(\n `BehaviorExplorer: costOf returned an invalid usd (${String(cost.usd)}) — ` +\n 'return null when cost is unknown, never a fabricated number',\n )\n }\n return {\n model: cost.model ?? 'unattributed',\n inputTokens: 0,\n outputTokens: 0,\n actualCostUsd: cost.usd,\n }\n },\n })\n : undefined\n if (paid && !paid.succeeded) {\n if (\n paid.error instanceof CostReceiptCaptureError &&\n paid.error.receiptError instanceof RangeError\n ) {\n throw paid.error.receiptError\n }\n throw paid.error\n }\n const ev = paid ? paid.value : await this.opts.evaluate(scenario, cell)\n // Consumer-measured latency wins (it can exclude judge time); the\n // engine's wall-clock is the default so latency is never missing.\n const latencyMs = ev.latencyMs ?? performance.now() - startedAt\n this.runsUsed++\n runsThisStep++\n this.consecutiveEvalErrors = 0\n const costUsd = paid\n ? paid.receipt.costUnknown\n ? null\n : paid.receipt.costUsd\n : undefined\n if (paid?.receipt.actualCostUsd !== undefined) {\n this.opts.onCost?.({ usd: paid.receipt.actualCostUsd, channel: 'agent' })\n }\n const interest = this.objective.interest(ev, this.objectiveContext())\n this.log.push({\n cell,\n ev,\n interest,\n latencyMs,\n ...(costUsd !== undefined ? { costUsd } : {}),\n scenarioId: this.opts.scenarioId(scenario),\n })\n this.opts.onProgress?.({ type: 'evaluated', cell, scenario, evaluation: ev })\n\n const bin = this.binId(cell, ev.descriptor)\n const cur = this.archiveByBin.get(bin)\n if (!cur || interest > cur.interest)\n this.archiveByBin.set(bin, { binId: bin, cell, scenario, evaluation: ev, interest })\n\n if (interest < this.threshold) return\n this.candidateFindings++\n if (this.opts.gates?.isValid && !(await this.opts.gates.isValid(scenario, ev, cell)))\n return\n if (\n this.opts.gates?.isUncontaminated &&\n !(await this.opts.gates.isUncontaminated(scenario, ev, cell))\n )\n return\n\n const minimized = this.opts.minimize\n ? await this.opts.minimize(scenario, this.opts.evaluate, cell)\n : scenario\n const finding: Finding<S> = {\n id: this.opts.scenarioId(scenario),\n cell,\n scenario,\n minimized,\n text: this.opts.scenarioText?.(minimized),\n evaluation: ev,\n interest,\n objective: this.objective.kind,\n }\n this._findings.push(finding)\n newFindings.push(finding)\n this.opts.onProgress?.({ type: 'finding', finding })\n } catch (err) {\n // Internal validation errors (e.g. a fabricated costOf number) are\n // programming mistakes, not backend outcomes — they stay loud.\n if (err instanceof RangeError) throw err\n this.evalErrors++\n this.consecutiveEvalErrors++\n const message = err instanceof Error ? err.message : String(err)\n this.opts.onProgress?.({\n type: 'eval-error',\n cell,\n scenarioId: this.opts.scenarioId(scenario),\n message,\n })\n const limit = this.opts.maxConsecutiveEvalErrors ?? 5\n if (this.consecutiveEvalErrors >= limit) {\n this.stoppedEarly = {\n reason: 'eval-errors',\n detail: `${this.consecutiveEvalErrors} consecutive eval errors (last: ${message})`,\n }\n }\n }\n },\n this.opts.concurrency ?? 1,\n this.opts.signal,\n )\n }\n\n this.opts.onProgress?.({ type: 'round', runsUsed: this.runsUsed, budget: this.opts.budget })\n return { runs: runsThisStep, findings: newFindings }\n }\n\n /** Loop `step()` until the run or dollar budget is spent, the signal aborts,\n * or no progress is made. */\n async run(): Promise<CapsuleData<S>> {\n while (\n this.runsUsed < this.opts.budget &&\n this.stoppedEarly === undefined &&\n !this.opts.signal?.aborted\n ) {\n const { runs } = await this.step()\n if (runs === 0 && this.stoppedEarly === undefined) break\n }\n return this.capsule()\n }\n\n coverage(): CoverageCell[] {\n return this.capsule().coverage\n }\n\n findings(): Finding<S>[] {\n return [...this._findings].sort((a, b) => b.interest - a.interest)\n }\n\n capsule(): CapsuleData<S> {\n const cost = this.costLedger?.summary({\n phase: this.costPhase,\n tags: { target: this.opts.target },\n })\n return buildCapsule({\n target: this.opts.target,\n objective: this.objective.kind,\n cells: this.cells,\n log: this.log,\n threshold: this.threshold,\n archive: [...this.archiveByBin.values()],\n findings: this._findings,\n candidateFindings: this.candidateFindings,\n runsUsed: this.runsUsed,\n cost: cost\n ? {\n costUsd: cost.totalCostUsd,\n costUnknownRuns: cost.byChannel.reduce((sum, row) => sum + row.unpricedCalls, 0),\n }\n : undefined,\n evalErrors: this.evalErrors,\n stoppedEarly: this.stoppedEarly,\n })\n }\n}\n","/**\n * `fuzzAgent` — the adversarial batch preset over `BehaviorExplorer`.\n *\n * One call: explore the space to budget with the adversarial objective and\n * return the capsule. For agent-driven, incremental, or multi-objective use,\n * construct a `BehaviorExplorer` and drive it via `makeExploreTools`.\n */\n\nimport { BehaviorExplorer } from './explorer'\nimport { adversarialObjective } from './policies'\nimport type { CapsuleData, ExploreOptions } from './types'\n\nexport type FuzzAgentOptions<S> = Omit<ExploreOptions<S>, 'objective'> & {\n /** Score strictly below this is a candidate failure. Default 0.5. */\n failureThreshold?: number\n}\n\nexport async function fuzzAgent<S>(\n opts: FuzzAgentOptions<S>,\n): Promise<{ capsule: CapsuleData<S> }> {\n const { failureThreshold, ...rest } = opts\n const explorer = new BehaviorExplorer<S>({\n ...rest,\n objective: adversarialObjective(failureThreshold ?? 0.5),\n })\n return { capsule: await explorer.run() }\n}\n","/**\n * Validity gates — what separates a fuzzer from a slop generator.\n *\n * A notable candidate is admitted only when it is fair and reproducible. None of\n * these are on by default: the live wiring opts in, so reported findings carry\n * their proof.\n */\n\nimport type { Cell, Evaluation, Evaluator, ValidityGates } from './types'\n\n/** Combine gate sets; a candidate must pass every gate in every set. */\nexport function composeGates<S>(...sets: Array<ValidityGates<S> | undefined>): ValidityGates<S> {\n const present = sets.filter((s): s is ValidityGates<S> => s != null)\n return {\n isValid: async (scenario, ev, cell) => {\n for (const g of present) {\n if (g.isValid && !(await g.isValid(scenario, ev, cell))) return false\n }\n return true\n },\n isUncontaminated: async (scenario, ev, cell) => {\n for (const g of present) {\n if (g.isUncontaminated && !(await g.isUncontaminated(scenario, ev, cell))) return false\n }\n return true\n },\n }\n}\n\n/**\n * Reproducibility gate. Re-run the target on a meaning-preserving rephrase of the\n * flagged scenario; keep the finding only when the rephrase ALSO scores below the\n * threshold. A finding that flips under a cosmetic rewrite was keyed to surface\n * form, not the task — a false signal we must not report. Costs one extra\n * evaluation per candidate (candidates are rare, so cheap).\n */\nexport function perturbationStabilityGate<S>(opts: {\n evaluate: Evaluator<S>\n /** Produce a semantic-preserving rephrase. Return null to skip (treated as pass). */\n perturb: (scenario: S) => S | null\n /** Score strictly below this still counts as failing. Default 0.5. */\n failureThreshold?: number\n}): ValidityGates<S> {\n const threshold = opts.failureThreshold ?? 0.5\n return {\n isUncontaminated: async (scenario: S, _ev: Evaluation, cell: Cell) => {\n const rephrased = opts.perturb(scenario)\n if (rephrased == null) return true\n const re = await opts.evaluate(rephrased, cell)\n return re.score < threshold\n },\n }\n}\n\n/**\n * Severity-floor gate. Reject borderline candidates whose score sits in a band\n * just under the threshold — judge noise, not a real defect.\n */\nexport function severityFloorGate<S>(opts: {\n failureThreshold?: number\n margin?: number\n}): ValidityGates<S> {\n const threshold = opts.failureThreshold ?? 0.5\n const margin = opts.margin ?? 0.1\n return {\n isValid: (_scenario, ev: Evaluation) => ev.score <= threshold - margin,\n }\n}\n","/**\n * Agent-drivable surface over a live exploration session.\n *\n * Framework-neutral tool defs ({name, description, parameters: JSON Schema,\n * handler}) so the on-demand agent — not a batch script — drives the search:\n * step it, read coverage, inspect findings, render the capsule. Transport\n * encodings (OpenAI function shape, MCP) are one-line mappings the host owns.\n */\n\nimport { renderCapsuleHtml } from './capsule'\nimport type { BehaviorExplorer } from './explorer'\n\nexport interface ExploreToolDef {\n name: string\n description: string\n /** JSON Schema (draft-07+) for the arguments. */\n parameters: Record<string, unknown>\n handler: (args: unknown, ctx?: { signal?: AbortSignal }) => Promise<unknown>\n}\n\nconst NO_ARGS = { type: 'object', properties: {}, additionalProperties: false }\n\nexport function makeExploreTools<S>(explorer: BehaviorExplorer<S>): ExploreToolDef[] {\n return [\n {\n name: 'explore_step',\n description:\n 'Run one exploration round: allocate budget across cells, propose + evaluate scenarios, archive elites, admit gate-verified findings. Returns runs spent and new findings.',\n parameters: NO_ARGS,\n handler: async () => {\n const { runs, findings } = await explorer.step()\n return { runs, newFindings: findings.length, findings }\n },\n },\n {\n name: 'explore_coverage',\n description:\n 'Read the live coverage map: per planned cell — runs, robustness (null = uncovered), finding rate, per-dimension means.',\n parameters: NO_ARGS,\n handler: async () => explorer.coverage(),\n },\n {\n name: 'explore_findings',\n description: 'List gate-verified findings so far, sorted by descending interest.',\n parameters: {\n type: 'object',\n properties: { limit: { type: 'number', description: 'max findings to return' } },\n additionalProperties: false,\n },\n handler: async (args) => {\n const limit = (args as { limit?: number } | undefined)?.limit\n const all = explorer.findings()\n return typeof limit === 'number' ? all.slice(0, limit) : all\n },\n },\n {\n name: 'explore_capsule',\n description:\n 'Build the capsule artifact from the current session state. format \"data\" returns the structured CapsuleData; \"html\" returns the standalone page.',\n parameters: {\n type: 'object',\n properties: {\n format: { type: 'string', enum: ['data', 'html'] },\n generatedAt: { type: 'string', description: 'ISO timestamp to stamp into the page' },\n },\n additionalProperties: false,\n },\n handler: async (args) => {\n const a = (args ?? {}) as { format?: string; generatedAt?: string }\n const capsule = explorer.capsule()\n if (a.format === 'html') return renderCapsuleHtml(capsule, { generatedAt: a.generatedAt })\n return capsule\n },\n },\n ]\n}\n"],"mappings":";;;;;;;;;;;;;;AA2BO,SAAS,eAAe,OAA8B;AAC3D,MAAI,MAAM,KAAK,WAAW,EAAG,QAAO,CAAC;AACrC,MAAI,WAA0C,CAAC,CAAC,CAAC;AACjD,aAAW,QAAQ,MAAM,MAAM;AAC7B,UAAM,OAAsC,CAAC;AAC7C,eAAW,WAAW,UAAU;AAC9B,iBAAW,SAAS,KAAK,OAAQ,MAAK,KAAK,EAAE,GAAG,SAAS,CAAC,KAAK,IAAI,GAAG,MAAM,CAAC;AAAA,IAC/E;AACA,eAAW;AAAA,EACb;AACA,SAAO,SAAS,IAAI,CAAC,YAAY,EAAE,IAAI,OAAO,OAAO,MAAM,GAAG,OAAO,EAAE;AACzE;AAGO,SAAS,OAAO,OAAsB,QAAwC;AACnF,SAAO,MAAM,KAAK,IAAI,CAAC,MAAM,GAAG,EAAE,IAAI,IAAI,OAAO,EAAE,IAAI,CAAC,EAAE,EAAE,KAAK,GAAG;AACtE;AAGA,SAAS,WAAW,QAAkB,GAAmB;AACvD,QAAM,MAAM,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,OAAO,MAAM,IAAI,CAAC,CAAC;AACrF,SAAO,OAAO,GAAG;AACnB;AAIO,SAAS,aAAa,QAAgC;AAC3D,MAAI,OAAO,WAAW;AACpB,UAAM,IAAI,MAAM,6EAAwE;AAC1F,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC/C,QAAM,OAAO,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AACxD,SAAO;AAAA,IACL;AAAA,IACA,QAAQ,WAAW,QAAQ,GAAG;AAAA,IAC9B,KAAK,WAAW,QAAQ,GAAG;AAAA,IAC3B,KAAK,OAAO,CAAC;AAAA,IACb,KAAK,OAAO,OAAO,SAAS,CAAC;AAAA,IAC7B,GAAG,OAAO;AAAA,EACZ;AACF;AAMO,SAAS,cAAc,OAAe,KAAmB,WAAmC;AACjG,QAAM,SAAS,oBAAI,IAA0B;AAC7C,aAAW,KAAK,KAAK;AACnB,UAAM,MAAM,OAAO,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC;AACtC,QAAI,KAAK,CAAC;AACV,WAAO,IAAI,EAAE,KAAK,IAAI,GAAG;AAAA,EAC3B;AACA,SAAO,MAAM,IAAI,CAAC,SAAS;AACzB,UAAM,OAAO,OAAO,IAAI,KAAK,EAAE,KAAK,CAAC;AACrC,UAAM,OAAO,KAAK;AAClB,QAAI,SAAS;AACX,aAAO,EAAE,MAAM,MAAM,GAAG,OAAO,MAAM,aAAa,GAAG,YAAY,CAAC,GAAG,WAAW,KAAK;AAEvF,UAAM,QAAQ,aAAa,KAAK,IAAI,CAAC,MAAM,EAAE,GAAG,KAAK,CAAC;AACtD,UAAM,YAAY,aAAa,KAAK,IAAI,CAAC,MAAM,EAAE,SAAS,CAAC;AAC3D,UAAM,cAAc,KAAK,OAAO,CAAC,MAAM,EAAE,YAAY,SAAS,EAAE,SAAS;AAEzE,UAAM,aAAuC,CAAC;AAC9C,eAAW,KAAK,MAAM;AACpB,iBAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,EAAE,GAAG,UAAU,CAAC,CAAC,GAAG;AACtD,cAAM,UAAU,WAAW,CAAC,KAAK,CAAC;AAClC,gBAAQ,KAAK,CAAC;AACd,mBAAW,CAAC,IAAI;AAAA,MAClB;AAAA,IACF;AACA,UAAM,aAA2C,CAAC;AAClD,eAAW,CAAC,GAAG,EAAE,KAAK,OAAO,QAAQ,UAAU,EAAG,YAAW,CAAC,IAAI,aAAa,EAAE;AAIjF,UAAM,UAAU,KAAK,OAAO,CAAC,MAAM,EAAE,YAAY,MAAS;AAC1D,UAAM,QAAQ,QAAQ,OAAO,CAAC,MAAM,EAAE,YAAY,IAAI;AACtD,UAAM,OACJ,QAAQ,SAAS,IACb;AAAA,MACE,SAAS,MAAM,OAAO,CAAC,GAAG,MAAM,IAAK,EAAE,SAAoB,CAAC;AAAA,MAC5D,GAAI,QAAQ,SAAS,MAAM,SACvB,EAAE,iBAAiB,QAAQ,SAAS,MAAM,OAAO,IACjD,CAAC;AAAA,IACP,IACA,CAAC;AAEP,WAAO,EAAE,MAAM,MAAM,OAAO,aAAa,YAAY,WAAW,GAAG,KAAK;AAAA,EAC1E,CAAC;AACH;;;AClFO,SAAS,aAAgB,OAA6C;AAC3E,QAAM,WAAW,cAAc,MAAM,OAAO,MAAM,KAAK,MAAM,SAAS;AACtE,QAAM,UAAU,SAAS,OAAO,CAAC,MAAM,EAAE,OAAO,CAAC;AAGjD,QAAM,aACJ,QAAQ,WAAW,IAAI,OAAO,aAAa,QAAQ,IAAI,CAAC,MAAO,EAAE,MAAuB,IAAI,CAAC;AAC/F,QAAM,YAAY,MAAM,IAAI,WAAW,IAAI,OAAO,aAAa,MAAM,IAAI,IAAI,CAAC,MAAM,EAAE,SAAS,CAAC;AAEhG,QAAM,uBAAuB,MAAM,QAAQ,OAAO,CAAC,MAAM,EAAE,UAAU,EAAE,KAAK,EAAE,EAAE;AAEhF,SAAO;AAAA,IACL,QAAQ,MAAM;AAAA,IACd,WAAW,MAAM;AAAA,IACjB;AAAA,IACA,UAAU,CAAC,GAAG,MAAM,QAAQ,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,IACpE,SAAS,CAAC,GAAG,MAAM,OAAO,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,IAClE,OAAO;AAAA,MACL,WAAW,MAAM;AAAA,MACjB,YAAY,MAAM,MAAM;AAAA,MACxB,cAAc,QAAQ;AAAA,MACtB;AAAA,MACA,mBAAmB,MAAM;AAAA,MACzB,kBAAkB,MAAM,SAAS;AAAA,MACjC;AAAA,MACA;AAAA,MACA,GAAI,MAAM,OACN,EAAE,SAAS,MAAM,KAAK,SAAS,iBAAiB,MAAM,KAAK,gBAAgB,IAC3E,CAAC;AAAA,MACL,YAAY,MAAM;AAAA,MAClB,GAAI,MAAM,eAAe,EAAE,cAAc,MAAM,aAAa,IAAI,CAAC;AAAA,IACnE;AAAA,EACF;AACF;AAIA,SAAS,IAAI,GAAmB;AAC9B,SAAO,EAAE;AAAA,IACP;AAAA,IACA,CAAC,OAAO,EAAE,KAAK,SAAS,KAAK,QAAQ,KAAK,QAAQ,KAAK,SAAS,GAAG,CAAC;AAAA,EACtE;AACF;AAGA,SAAS,gBAAgB,GAA0B;AACjD,MAAI,KAAK,KAAM,QAAO;AACtB,SAAO,OAAO,KAAK,MAAM,IAAI,GAAG,CAAC;AACnC;AAEA,SAAS,IAAI,GAAmB;AAC9B,SAAO,GAAG,KAAK,MAAM,IAAI,GAAG,CAAC;AAC/B;AAEA,SAAS,WAAW,UAAqE;AACvF,QAAM,QAAkB,CAAC;AACzB,QAAM,OAAO,oBAAI,IAAyB;AAC1C,aAAW,KAAK,UAAU;AACxB,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,EAAE,KAAK,MAAM,GAAG;AAClD,UAAI,CAAC,KAAK,IAAI,CAAC,GAAG;AAChB,aAAK,IAAI,GAAG,oBAAI,IAAI,CAAC;AACrB,cAAM,KAAK,CAAC;AAAA,MACd;AACA,WAAK,IAAI,CAAC,GAAG,IAAI,CAAC;AAAA,IACpB;AAAA,EACF;AACA,SAAO,MAAM,IAAI,CAAC,UAAU,EAAE,MAAM,QAAQ,CAAC,GAAI,KAAK,IAAI,IAAI,KAAK,CAAC,CAAE,EAAE,EAAE;AAC5E;AAGA,SAAS,WAAW,GAAyB;AAC3C,QAAM,UAAU,OAAO,QAAQ,EAAE,UAAU,EAAE,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,EAAE,IAAI,CAAqB;AAC5F,MAAI,QAAQ,WAAW,EAAG,QAAO;AACjC,QAAM,SAAS,QAAQ,KAAK,CAAC,GAAG,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC;AACjD,QAAM,IAAI,OAAO,CAAC;AAClB,MAAI,CAAC,EAAG,QAAO;AACf,SAAO,qBAAqB,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,CAAC;AACpD;AAEA,SAAS,YAAY,UAAkC;AACrD,QAAM,OAAO,WAAW,QAAQ;AAChC,QAAM,OAAO,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,CAAC,EAAE,KAAK,IAAI,CAAC,CAAC,CAAC;AACxD,QAAM,OAAO,CAAC,GAA6B,UAA0B;AACnE,UAAM,IAAI,GAAG,OAAO,QAAQ;AAC5B,UAAM,QAAQ,GAAG,QACb,GAAG,IAAI,EAAE,MAAM,IAAI,CAAC,mBAAmB,IAAI,EAAE,MAAM,MAAM,CAAC,SAAS,IAAI,EAAE,MAAM,GAAG,CAAC,UAAO,EAAE,IAAI,cAAW,IAAI,EAAE,WAAW,CAAC,WAAW,EAAE,YAAY,UAAO,EAAE,UAAU,SAAS,KAAM,QAAQ,CAAC,CAAC,aAAa,EAAE,GAAG,EAAE,YAAY,SAAY,UAAO,EAAE,QAAQ,QAAQ,CAAC,CAAC,KAAK,EAAE,KAChR;AACJ,WAAO,uCAAuC,gBAAgB,CAAC,CAAC,YAAY,IAAI,KAAK,CAAC,KAAK,QAAQ,oBAAoB,IAAI,KAAK,CAAC,YAAY,EAAE,oBAAoB,KAAK,KAAK,OAAO,IAAI,CAAC,IAAI,QAAG,UAAU,IAAI,WAAW,CAAC,IAAI,EAAE;AAAA,EAClO;AAEA,QAAM,UAAU,KAAK,CAAC;AACtB,QAAM,UAAU,KAAK,CAAC;AACtB,MAAI,KAAK,WAAW,KAAK,WAAW,SAAS;AAC3C,UAAM,OAAO,gBAAgB,QAAQ,OAAO,IAAI,CAAC,MAAM,OAAO,IAAI,CAAC,CAAC,OAAO,EAAE,KAAK,EAAE,CAAC;AACrF,UAAM,OAAO,QAAQ,OAClB,IAAI,CAAC,OAAO;AACX,YAAM,QAAQ,QAAQ,OACnB,IAAI,CAAC,OAAO;AACX,cAAM,KAAK,GAAG,QAAQ,IAAI,IAAI,EAAE,IAAI,QAAQ,IAAI,IAAI,EAAE;AACtD,eAAO,OAAO,KAAK,KAAK,IAAI,EAAE,GAAG,EAAE,CAAC;AAAA,MACtC,CAAC,EACA,KAAK,EAAE;AACV,aAAO,sBAAsB,IAAI,EAAE,CAAC,QAAQ,KAAK;AAAA,IACnD,CAAC,EACA,KAAK,EAAE;AACV,WAAO,oCAAoC,IAAI,QAAQ,IAAI,CAAC,sBAAmB,IAAI,QAAQ,IAAI,CAAC,iCAAiC,IAAI,GAAG,IAAI;AAAA,EAC9I;AAEA,QAAM,SAAS,CAAC,GAAG,QAAQ,EAAE,KAAK,CAAC,GAAG,OAAO,EAAE,OAAO,QAAQ,MAAM,EAAE,OAAO,QAAQ,EAAE;AACvF,SAAO,qBAAqB,OAAO,IAAI,CAAC,MAAM,KAAK,GAAG,OAAO,OAAO,EAAE,KAAK,MAAM,EAAE,KAAK,QAAK,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC;AAC3G;AAEA,SAAS,aAAgB,UAAwB,OAAuB;AACtE,MAAI,SAAS,WAAW;AACtB,WAAO;AACT,SAAO,SACJ,MAAM,GAAG,KAAK,EACd,IAAI,CAAC,MAAM;AACV,UAAM,SAAS,OAAO,QAAQ,EAAE,KAAK,MAAM,EACxC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,EAAE,EAC3B,KAAK,QAAK;AACb,UAAM,UAAU,EAAE,WAAW,UAAU,CAAC,GACrC,IAAI,CAAC,MAAM,wBAAwB,IAAI,CAAC,CAAC,SAAS,EAClD,KAAK,EAAE;AACV,UAAM,OAAO,OAAO,QAAQ,EAAE,WAAW,UAAU,CAAC,CAAC,EAClD,KAAK,CAAC,GAAG,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,EAC1B,MAAM,GAAG,CAAC,EACV,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,qBAAqB,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,SAAS,EAC9D,KAAK,EAAE;AACV,WAAO,4DAA4D,IAAI,MAAM,CAAC,UAAU,MAAM,+BAA+B,IAAI,EAAE,QAAQ,CAAC,sEAAsE,IAAI,EAAE,QAAQ,CAAC,oCAAoC,IAAI,EAAE,QAAQ,8BAA8B,CAAC,SAAS,OAAO,sBAAsB,IAAI,WAAW,EAAE;AAAA,EAC3W,CAAC,EACA,KAAK,EAAE;AACZ;AAUO,SAAS,kBACd,SACA,OAA6B,CAAC,GACtB;AACR,QAAM,IAAI,QAAQ;AAClB,QAAM,MAAM,CAAC,OAAe,OAAe,SAAS,cAClD,iDAAiD,MAAM,KAAK,IAAI,KAAK,CAAC,yBAAyB,IAAI,KAAK,CAAC;AAC3G,QAAM,OAAO,QAAQ,OACjB;AAAA,IACE;AAAA,IACA,GAAG,QAAQ,KAAK,OAAO,QAAQ,CAAC,CAAC,WAAM,QAAQ,KAAK,MAAM,QAAQ,CAAC,CAAC;AAAA,IACpE;AAAA,EACF,IACA;AAGJ,QAAM,OACJ,EAAE,YAAY,SACV;AAAA,IACE,EAAE,kBAAkB,aAAU,EAAE,eAAe,mBAAmB;AAAA,IAClE,IAAI,EAAE,QAAQ,QAAQ,CAAC,CAAC;AAAA,IACxB,EAAE,kBAAkB,YAAY;AAAA,EAClC,IACA;AACN,QAAM,QAAQ,KAAK,eAAe,QAAQ,eAAe;AAEzD,SAAO;AAAA,SACA,IAAI,QAAQ,SAAS,CAAC,mBAAc,IAAI,QAAQ,MAAM,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,MAoC1D,IAAI,QAAQ,SAAS,CAAC,qBAAkB,IAAI,QAAQ,MAAM,CAAC;AAAA,mBAC9C,EAAE,SAAS,qBAAqB,EAAE,YAAY,IAAI,EAAE,UAAU,iBAAiB,EAAE,uBAAuB,IAAI,SAAM,EAAE,oBAAoB,4BAA4B,EAAE,GAAG,QAAQ,SAAM,IAAI,KAAK,CAAC,KAAK,EAAE;AAAA;AAAA,EAEzN,EAAE,aAAa,IAAI,cAAc,GAAG,IAAI,EAAE,WAAW,IAAI,CAAC,IAAI,EAAE,WAAW,OAAO,MAAM,YAAY,SAAS,IAAI,EAAE;AAAA,EACnH,EAAE,aAAa,IAAI,eAAe,GAAG,IAAI,EAAE,WAAW,GAAG,CAAC,SAAI,IAAI,EAAE,WAAW,GAAG,CAAC,EAAE,IAAI,EAAE;AAAA,EAC3F,EAAE,YAAY,IAAI,kBAAkB,IAAI,EAAE,UAAU,SAAS,KAAM,QAAQ,CAAC,CAAC,KAAK,EAAE,UAAU,MAAM,IAAI,EAAE,UAAU,SAAS,YAAY,SAAS,IAAI,EAAE;AAAA,EACxJ,IAAI,qBAAqB,OAAO,EAAE,gBAAgB,GAAG,EAAE,mBAAmB,IAAI,YAAY,SAAS,CAAC;AAAA,EACpG,IAAI,iBAAiB,GAAG,EAAE,YAAY,IAAI,EAAE,UAAU,EAAE,CAAC;AAAA,EACzD,IAAI,iBAAiB,OAAO,EAAE,SAAS,CAAC,CAAC;AAAA,EACzC,IAAI;AAAA,EACJ,EAAE,aAAa,IAAI,IAAI,eAAe,OAAO,EAAE,UAAU,GAAG,SAAS,IAAI,EAAE;AAAA,EAC3E,IAAI;AAAA;AAAA,EAEJ,EAAE,eAAe,yDAAyD,IAAI,EAAE,aAAa,MAAM,CAAC,qEAAgE,EAAE;AAAA;AAAA,EAEtK,YAAY,QAAQ,QAAQ,CAAC;AAAA,uBACR,EAAE,oBAAoB,EAAE,mBAAmB,SAAM,EAAE,gBAAgB,OAAO,EAAE,iBAAiB,0CAA0C,EAAE;AAAA,EAC9J,aAAa,QAAQ,UAAU,KAAK,eAAe,CAAC,CAAC;AAAA;AAAA;AAGvD;;;AChPA,IAAM,UAAU,CAAC,MAAsB,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,CAAC,CAAC;AAS1D,SAAS,iBAAoB,MAGZ;AACtB,SAAO,OAAO,QAAyC;AACrD,UAAM,YAAY,KAAK,aAAa,IAAI,IAAI;AAC5C,UAAM,UAAU,CAAC,GAAG,IAAI,QAAQ,GAAG,IAAI,KAAK;AAC5C,UAAM,OAAO,IAAI,IAAI,QAAQ,IAAI,KAAK,UAAU,CAAC;AACjD,UAAM,MAAW,CAAC;AAClB,eAAW,UAAU,SAAS;AAC5B,UAAI,IAAI,UAAU,IAAI,MAAO;AAC7B,iBAAW,KAAK,WAAW;AACzB,cAAM,WAAW,MAAM,EAAE,OAAO,QAAQ,IAAI,GAAG;AAC/C,mBAAW,SAAS,UAAU;AAC5B,gBAAM,KAAK,KAAK,WAAW,KAAK;AAChC,cAAI,KAAK,IAAI,EAAE,EAAG;AAClB,eAAK,IAAI,EAAE;AACX,cAAI,KAAK,KAAK;AACd,cAAI,IAAI,UAAU,IAAI,MAAO;AAAA,QAC/B;AACA,YAAI,IAAI,UAAU,IAAI,MAAO;AAAA,MAC/B;AAAA,IACF;AACA,WAAO;AAAA,EACT;AACF;AAKO,SAAS,qBAAqB,YAAY,KAAgB;AAC/D,SAAO,EAAE,MAAM,eAAe,WAAW,UAAU,CAAC,OAAO,QAAQ,IAAI,GAAG,KAAK,EAAE;AACnF;AAEA,SAAS,QACP,GACA,GACQ;AACR,MAAI,CAAC,KAAK,CAAC,EAAG,QAAO;AACrB,QAAM,OAAO,oBAAI,IAAI,CAAC,GAAG,OAAO,KAAK,CAAC,GAAG,GAAG,OAAO,KAAK,CAAC,CAAC,CAAC;AAC3D,MAAI,KAAK,SAAS,EAAG,QAAO;AAC5B,MAAI,OAAO;AACX,aAAW,KAAK,KAAM,KAAI,EAAE,CAAC,MAAM,EAAE,CAAC,EAAG;AACzC,SAAO,OAAO,KAAK;AACrB;AAOO,SAAS,iBAAiB,YAAY,KAAgB;AAC3D,SAAO;AAAA,IACL,MAAM;AAAA,IACN;AAAA,IACA,UAAU,CAAC,IAAgB,QAA0B;AACnD,YAAM,eACJ,IAAI,cAAc,WAAW,IACzB,IACA,KAAK,IAAI,GAAG,IAAI,cAAc,IAAI,CAAC,MAAM,KAAK,IAAI,IAAI,GAAG,KAAK,CAAC,CAAC;AACtE,YAAM,cACJ,IAAI,mBAAmB,WAAW,IAC9B,IACA,KAAK,IAAI,GAAG,IAAI,mBAAmB,IAAI,CAAC,MAAM,QAAQ,GAAG,GAAG,UAAU,CAAC,CAAC;AAC9E,aAAO,QAAQ,MAAM,eAAe,MAAM,WAAW;AAAA,IACvD;AAAA,EACF;AACF;;;AC/DA,eAAe,KACb,OACA,IACA,aACA,QACe;AACf,MAAI,IAAI;AACR,QAAM,UAAU,MAAM;AAAA,IACpB,EAAE,QAAQ,KAAK,IAAI,GAAG,KAAK,IAAI,aAAa,MAAM,MAAM,CAAC,EAAE;AAAA,IAC3D,YAAY;AACV,aAAO,IAAI,MAAM,QAAQ;AACvB,YAAI,QAAQ,QAAS;AACrB,cAAM,OAAO,MAAM,GAAG;AACtB,YAAI,SAAS,OAAW;AACxB,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF;AAAA,EACF;AACA,QAAM,QAAQ,IAAI,OAAO;AAC3B;AAEO,IAAM,mBAAN,MAA0B;AAAA,EAsB/B,YAA6B,MAAyB;AAAzB;AAC3B,SAAK,QAAQ,eAAe,KAAK,KAAK;AACtC,QAAI,KAAK,MAAM,WAAW;AACxB,YAAM,IAAI,MAAM,4EAAkE;AACpF,QAAI,KAAK,kBAAkB,QAAW;AACpC,UACE,OAAO,KAAK,kBAAkB,YAC9B,CAAC,OAAO,SAAS,KAAK,aAAa,KACnC,KAAK,gBAAgB,GACrB;AACA,cAAM,IAAI;AAAA,UACR,4EAA4E,OAAO,KAAK,aAAa,CAAC;AAAA,QACxG;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,KAAK,WAAW,KAAK,kBAAkB,UAAa,KAAK,UAAU,KAAK,SAAS;AACpF,YAAM,IAAI;AAAA,QACR;AAAA,MAEF;AAAA,IACF;AACA,SACG,KAAK,kBAAkB,UAAa,KAAK,QAAQ,mBAAmB,WACrE,CAAC,KAAK,iBACN;AACA,YAAM,IAAI;AAAA,QACR;AAAA,MACF;AAAA,IACF;AACA,QACE,KAAK,UACL,KAAK,kBAAkB,UACvB,KAAK,OAAO,mBAAmB,KAAK,eACpC;AACA,YAAM,IAAI;AAAA,QACR;AAAA,MACF;AAAA,IACF;AACA,QAAI,KAAK,OAAQ,MAAK,aAAa,KAAK,UAAU,IAAI,WAAW,KAAK,aAAa;AACnF,SAAK,WAAW,IAAI,IAAI,KAAK,MAAM,IAAI,CAAC,MAAM,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;AACxD,SAAK,YAAY,KAAK,aAAa,qBAAqB,GAAG;AAC3D,SAAK,YAAY,KAAK,UAAU,aAAa;AAC7C,SAAK,eAAe,KAAK,gBAAgB;AACzC,SAAK,iBAAiB,KAAK;AAAA,MACzB,KAAK,MAAM,SAAS,KAAK;AAAA,MACzB,KAAK,KAAK,KAAK,SAAS,CAAC;AAAA,IAC3B;AACA,SAAK,YAAY,KAAK,QAAQ,OAAO;AAAA,EACvC;AAAA,EAhD6B;AAAA,EArBZ;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA;AAAA,EAGA,MAAkD,CAAC;AAAA;AAAA,EAEnD,eAAe,oBAAI,IAA6B;AAAA,EAChD,YAA0B,CAAC;AAAA,EACpC,WAAW;AAAA,EACX,oBAAoB;AAAA,EACpB,aAAa;AAAA,EACb,wBAAwB;AAAA,EACxB;AAAA,EACA;AAAA,EACS;AAAA,EACA,YAAY;AAAA;AAAA,EAqDrB,MAAM,MAAc;AAC1B,SAAK,WAAY,KAAK,WAAW,aAAc;AAC/C,QAAI,IAAI,KAAK;AACb,QAAI,KAAK,KAAK,IAAK,MAAM,IAAK,IAAI,CAAC;AACnC,SAAK,IAAI,KAAK,KAAK,IAAK,MAAM,GAAI,IAAI,EAAE;AACxC,aAAS,IAAK,MAAM,QAAS,KAAK;AAAA,EACpC;AAAA,EAEQ,MAAM,MAAY,YAAwD;AAChF,QAAI,CAAC,cAAc,OAAO,KAAK,UAAU,EAAE,WAAW,EAAG,QAAO,KAAK;AACrE,UAAM,WAAW,OAAO,KAAK,UAAU,EACpC,KAAK,EACL,IAAI,CAAC,MAAM,GAAG,CAAC,IAAI,WAAW,CAAC,CAAC,EAAE,EAClC,KAAK,GAAG;AACX,WAAO,GAAG,KAAK,EAAE,IAAI,QAAQ;AAAA,EAC/B;AAAA,EAEQ,SAAS,QAA0D;AACzE,SAAK,KAAK,KAAK,cAAc,gBAAgB,WAAW;AACtD,YAAM,MAAM,KAAK,IAAI,KAAK,cAAc,KAAK,MAAM,SAAS,KAAK,MAAM,MAAM,CAAC;AAC9E,aAAO,KAAK,MAAM,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,IAAI,OAAO,IAAI,EAAE;AAAA,IAC7D;AACA,WAAO;AAAA,MACL,KAAK,IAAI,IAAI,CAAC,OAAO;AAAA,QACnB,WAAW,EAAE,KAAK;AAAA,QAClB,YAAY,EAAE;AAAA,QACd,OAAO,EAAE,GAAG;AAAA,QACZ,MAAM,EAAE,GAAG,SAAS,EAAE,GAAG,SAAS;AAAA,MACpC,EAAE;AAAA,MACF,KAAK,MAAM,IAAI,CAAC,OAAO,EAAE,WAAW,EAAE,IAAI,YAAY,IAAI,EAAE;AAAA,MAC5D,EAAE,QAAQ,cAAc,KAAK,aAAa;AAAA,IAC5C,EAAE,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,OAAO,EAAE,MAAM,EAAE;AAAA,EACxD;AAAA,EAEQ,mBAAmB;AACzB,UAAM,UAAU,CAAC,GAAG,KAAK,aAAa,OAAO,CAAC;AAC9C,WAAO;AAAA,MACL,eAAe,QAAQ,IAAI,CAAC,MAAM,EAAE,WAAW,KAAK;AAAA,MACpD,oBAAoB,QAAQ,IAAI,CAAC,MAAM,EAAE,WAAW,UAAU;AAAA,IAChE;AAAA,EACF;AAAA;AAAA,EAGQ,UAAUA,SAAqB;AACrC,UAAM,MAAW,CAAC;AAClB,eAAW,KAAK,KAAK,aAAa,OAAO,EAAG,KAAI,EAAE,KAAK,OAAOA,QAAQ,KAAI,KAAK,EAAE,QAAQ;AACzF,WAAO;AAAA,EACT;AAAA;AAAA,EAGA,MAAM,OAA0D;AAC9D,UAAM,YAAY,KAAK,KAAK,SAAS,KAAK;AAC1C,QAAI,aAAa,KAAK,KAAK,KAAK,QAAQ,QAAS,QAAO,EAAE,MAAM,GAAG,UAAU,CAAC,EAAE;AAEhF,UAAM,cAAc,KAAK,SAAS,KAAK,IAAI,KAAK,gBAAgB,SAAS,CAAC;AAC1E,UAAM,cAA4B,CAAC;AACnC,QAAI,eAAe;AAEnB,eAAW,SAAS,aAAa;AAC/B,UACE,KAAK,YAAY,KAAK,KAAK,UAC3B,KAAK,iBAAiB,UACtB,KAAK,KAAK,QAAQ;AAElB;AACF,YAAM,OAAO,KAAK,SAAS,IAAI,MAAM,MAAM;AAC3C,UAAI,CAAC,KAAM;AACX,YAAM,MAAM,KAAK,IAAI,MAAM,OAAO,KAAK,KAAK,SAAS,KAAK,QAAQ;AAClE,UAAI,OAAO,EAAG;AACd,WAAK,KAAK,aAAa,EAAE,MAAM,kBAAkB,MAAM,OAAO,IAAI,CAAC;AAEnE,YAAM,QAAQ,MAAM,KAAK,KAAK,SAAS,IAAI;AAC3C,YAAM,SAAS,KAAK,UAAU,KAAK,EAAE;AACrC,YAAM,WAAW,MAAM,KAAK,KAAK,SAAS;AAAA,QACxC;AAAA,QACA;AAAA,QACA;AAAA,QACA,UAAU,KAAK;AAAA,QACf,OAAO;AAAA,QACP,KAAK,KAAK;AAAA,MACZ,CAAC;AAGD,YAAM,OAAO,KAAK,IAAI,MAAM,CAAC,MAAM,EAAE,KAAK,OAAO,KAAK,EAAE;AACxD,YAAM,SAAS,CAAC,GAAI,OAAO,QAAQ,CAAC,GAAI,GAAG,QAAQ,EAAE,MAAM,GAAG,GAAG;AAEjE,YAAM;AAAA,QACJ;AAAA,QACA,OAAO,aAAa;AAClB,cACE,KAAK,YAAY,KAAK,KAAK,UAC3B,KAAK,iBAAiB,UACtB,KAAK,KAAK,QAAQ;AAElB;AAMF,cAAI;AACF,kBAAM,YAAY,YAAY,IAAI;AAClC,kBAAM,OAAO,KAAK,aACd,MAAM,KAAK,WAAW,YAAY;AAAA,cAChC,SAAS;AAAA,cACT,OAAO,KAAK;AAAA,cACZ,OAAO;AAAA,cACP,MAAM,EAAE,QAAQ,KAAK,KAAK,QAAQ,MAAM,KAAK,GAAG;AAAA,cAChD,QAAQ,KAAK,KAAK;AAAA,cAClB,eAAe,KAAK,KAAK,kBAAkB,UAAU,IAAI;AAAA,cACzD,SAAS,MAAM,KAAK,KAAK,SAAS,UAAU,IAAI;AAAA,cAChD,SAAS,CAAC,eAAe;AACvB,sBAAM,OAAO,KAAK,KAAK,OAAQ,UAAU,MAAM,UAAU;AACzD,oBAAI,SAAS,MAAM;AACjB,yBAAO;AAAA,oBACL,OAAO;AAAA,oBACP,aAAa;AAAA,oBACb,cAAc;AAAA,oBACd,aAAa;AAAA,kBACf;AAAA,gBACF;AACA,oBAAI,CAAC,OAAO,SAAS,KAAK,GAAG,KAAK,KAAK,MAAM,GAAG;AAC9C,wBAAM,IAAI;AAAA,oBACR,qDAAqD,OAAO,KAAK,GAAG,CAAC;AAAA,kBAEvE;AAAA,gBACF;AACA,uBAAO;AAAA,kBACL,OAAO,KAAK,SAAS;AAAA,kBACrB,aAAa;AAAA,kBACb,cAAc;AAAA,kBACd,eAAe,KAAK;AAAA,gBACtB;AAAA,cACF;AAAA,YACF,CAAC,IACD;AACJ,gBAAI,QAAQ,CAAC,KAAK,WAAW;AAC3B,kBACE,KAAK,iBAAiB,2BACtB,KAAK,MAAM,wBAAwB,YACnC;AACA,sBAAM,KAAK,MAAM;AAAA,cACnB;AACA,oBAAM,KAAK;AAAA,YACb;AACA,kBAAM,KAAK,OAAO,KAAK,QAAQ,MAAM,KAAK,KAAK,SAAS,UAAU,IAAI;AAGtE,kBAAM,YAAY,GAAG,aAAa,YAAY,IAAI,IAAI;AACtD,iBAAK;AACL;AACA,iBAAK,wBAAwB;AAC7B,kBAAM,UAAU,OACZ,KAAK,QAAQ,cACX,OACA,KAAK,QAAQ,UACf;AACJ,gBAAI,MAAM,QAAQ,kBAAkB,QAAW;AAC7C,mBAAK,KAAK,SAAS,EAAE,KAAK,KAAK,QAAQ,eAAe,SAAS,QAAQ,CAAC;AAAA,YAC1E;AACA,kBAAM,WAAW,KAAK,UAAU,SAAS,IAAI,KAAK,iBAAiB,CAAC;AACpE,iBAAK,IAAI,KAAK;AAAA,cACZ;AAAA,cACA;AAAA,cACA;AAAA,cACA;AAAA,cACA,GAAI,YAAY,SAAY,EAAE,QAAQ,IAAI,CAAC;AAAA,cAC3C,YAAY,KAAK,KAAK,WAAW,QAAQ;AAAA,YAC3C,CAAC;AACD,iBAAK,KAAK,aAAa,EAAE,MAAM,aAAa,MAAM,UAAU,YAAY,GAAG,CAAC;AAE5E,kBAAM,MAAM,KAAK,MAAM,MAAM,GAAG,UAAU;AAC1C,kBAAM,MAAM,KAAK,aAAa,IAAI,GAAG;AACrC,gBAAI,CAAC,OAAO,WAAW,IAAI;AACzB,mBAAK,aAAa,IAAI,KAAK,EAAE,OAAO,KAAK,MAAM,UAAU,YAAY,IAAI,SAAS,CAAC;AAErF,gBAAI,WAAW,KAAK,UAAW;AAC/B,iBAAK;AACL,gBAAI,KAAK,KAAK,OAAO,WAAW,CAAE,MAAM,KAAK,KAAK,MAAM,QAAQ,UAAU,IAAI,IAAI;AAChF;AACF,gBACE,KAAK,KAAK,OAAO,oBACjB,CAAE,MAAM,KAAK,KAAK,MAAM,iBAAiB,UAAU,IAAI,IAAI;AAE3D;AAEF,kBAAM,YAAY,KAAK,KAAK,WACxB,MAAM,KAAK,KAAK,SAAS,UAAU,KAAK,KAAK,UAAU,IAAI,IAC3D;AACJ,kBAAM,UAAsB;AAAA,cAC1B,IAAI,KAAK,KAAK,WAAW,QAAQ;AAAA,cACjC;AAAA,cACA;AAAA,cACA;AAAA,cACA,MAAM,KAAK,KAAK,eAAe,SAAS;AAAA,cACxC,YAAY;AAAA,cACZ;AAAA,cACA,WAAW,KAAK,UAAU;AAAA,YAC5B;AACA,iBAAK,UAAU,KAAK,OAAO;AAC3B,wBAAY,KAAK,OAAO;AACxB,iBAAK,KAAK,aAAa,EAAE,MAAM,WAAW,QAAQ,CAAC;AAAA,UACrD,SAAS,KAAK;AAGZ,gBAAI,eAAe,WAAY,OAAM;AACrC,iBAAK;AACL,iBAAK;AACL,kBAAM,UAAU,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC/D,iBAAK,KAAK,aAAa;AAAA,cACrB,MAAM;AAAA,cACN;AAAA,cACA,YAAY,KAAK,KAAK,WAAW,QAAQ;AAAA,cACzC;AAAA,YACF,CAAC;AACD,kBAAM,QAAQ,KAAK,KAAK,4BAA4B;AACpD,gBAAI,KAAK,yBAAyB,OAAO;AACvC,mBAAK,eAAe;AAAA,gBAClB,QAAQ;AAAA,gBACR,QAAQ,GAAG,KAAK,qBAAqB,mCAAmC,OAAO;AAAA,cACjF;AAAA,YACF;AAAA,UACF;AAAA,QACF;AAAA,QACA,KAAK,KAAK,eAAe;AAAA,QACzB,KAAK,KAAK;AAAA,MACZ;AAAA,IACF;AAEA,SAAK,KAAK,aAAa,EAAE,MAAM,SAAS,UAAU,KAAK,UAAU,QAAQ,KAAK,KAAK,OAAO,CAAC;AAC3F,WAAO,EAAE,MAAM,cAAc,UAAU,YAAY;AAAA,EACrD;AAAA;AAAA;AAAA,EAIA,MAAM,MAA+B;AACnC,WACE,KAAK,WAAW,KAAK,KAAK,UAC1B,KAAK,iBAAiB,UACtB,CAAC,KAAK,KAAK,QAAQ,SACnB;AACA,YAAM,EAAE,KAAK,IAAI,MAAM,KAAK,KAAK;AACjC,UAAI,SAAS,KAAK,KAAK,iBAAiB,OAAW;AAAA,IACrD;AACA,WAAO,KAAK,QAAQ;AAAA,EACtB;AAAA,EAEA,WAA2B;AACzB,WAAO,KAAK,QAAQ,EAAE;AAAA,EACxB;AAAA,EAEA,WAAyB;AACvB,WAAO,CAAC,GAAG,KAAK,SAAS,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,WAAW,EAAE,QAAQ;AAAA,EACnE;AAAA,EAEA,UAA0B;AACxB,UAAM,OAAO,KAAK,YAAY,QAAQ;AAAA,MACpC,OAAO,KAAK;AAAA,MACZ,MAAM,EAAE,QAAQ,KAAK,KAAK,OAAO;AAAA,IACnC,CAAC;AACD,WAAO,aAAa;AAAA,MAClB,QAAQ,KAAK,KAAK;AAAA,MAClB,WAAW,KAAK,UAAU;AAAA,MAC1B,OAAO,KAAK;AAAA,MACZ,KAAK,KAAK;AAAA,MACV,WAAW,KAAK;AAAA,MAChB,SAAS,CAAC,GAAG,KAAK,aAAa,OAAO,CAAC;AAAA,MACvC,UAAU,KAAK;AAAA,MACf,mBAAmB,KAAK;AAAA,MACxB,UAAU,KAAK;AAAA,MACf,MAAM,OACF;AAAA,QACE,SAAS,KAAK;AAAA,QACd,iBAAiB,KAAK,UAAU,OAAO,CAAC,KAAK,QAAQ,MAAM,IAAI,eAAe,CAAC;AAAA,MACjF,IACA;AAAA,MACJ,YAAY,KAAK;AAAA,MACjB,cAAc,KAAK;AAAA,IACrB,CAAC;AAAA,EACH;AACF;;;ACrYA,eAAsB,UACpB,MACsC;AACtC,QAAM,EAAE,kBAAkB,GAAG,KAAK,IAAI;AACtC,QAAM,WAAW,IAAI,iBAAoB;AAAA,IACvC,GAAG;AAAA,IACH,WAAW,qBAAqB,oBAAoB,GAAG;AAAA,EACzD,CAAC;AACD,SAAO,EAAE,SAAS,MAAM,SAAS,IAAI,EAAE;AACzC;;;ACfO,SAAS,gBAAmB,MAA6D;AAC9F,QAAM,UAAU,KAAK,OAAO,CAAC,MAA6B,KAAK,IAAI;AACnE,SAAO;AAAA,IACL,SAAS,OAAO,UAAU,IAAI,SAAS;AACrC,iBAAW,KAAK,SAAS;AACvB,YAAI,EAAE,WAAW,CAAE,MAAM,EAAE,QAAQ,UAAU,IAAI,IAAI,EAAI,QAAO;AAAA,MAClE;AACA,aAAO;AAAA,IACT;AAAA,IACA,kBAAkB,OAAO,UAAU,IAAI,SAAS;AAC9C,iBAAW,KAAK,SAAS;AACvB,YAAI,EAAE,oBAAoB,CAAE,MAAM,EAAE,iBAAiB,UAAU,IAAI,IAAI,EAAI,QAAO;AAAA,MACpF;AACA,aAAO;AAAA,IACT;AAAA,EACF;AACF;AASO,SAAS,0BAA6B,MAMxB;AACnB,QAAM,YAAY,KAAK,oBAAoB;AAC3C,SAAO;AAAA,IACL,kBAAkB,OAAO,UAAa,KAAiB,SAAe;AACpE,YAAM,YAAY,KAAK,QAAQ,QAAQ;AACvC,UAAI,aAAa,KAAM,QAAO;AAC9B,YAAM,KAAK,MAAM,KAAK,SAAS,WAAW,IAAI;AAC9C,aAAO,GAAG,QAAQ;AAAA,IACpB;AAAA,EACF;AACF;AAMO,SAAS,kBAAqB,MAGhB;AACnB,QAAM,YAAY,KAAK,oBAAoB;AAC3C,QAAM,SAAS,KAAK,UAAU;AAC9B,SAAO;AAAA,IACL,SAAS,CAAC,WAAW,OAAmB,GAAG,SAAS,YAAY;AAAA,EAClE;AACF;;;AC/CA,IAAM,UAAU,EAAE,MAAM,UAAU,YAAY,CAAC,GAAG,sBAAsB,MAAM;AAEvE,SAAS,iBAAoB,UAAiD;AACnF,SAAO;AAAA,IACL;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,MACZ,SAAS,YAAY;AACnB,cAAM,EAAE,MAAM,SAAS,IAAI,MAAM,SAAS,KAAK;AAC/C,eAAO,EAAE,MAAM,aAAa,SAAS,QAAQ,SAAS;AAAA,MACxD;AAAA,IACF;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,MACZ,SAAS,YAAY,SAAS,SAAS;AAAA,IACzC;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aAAa;AAAA,MACb,YAAY;AAAA,QACV,MAAM;AAAA,QACN,YAAY,EAAE,OAAO,EAAE,MAAM,UAAU,aAAa,yBAAyB,EAAE;AAAA,QAC/E,sBAAsB;AAAA,MACxB;AAAA,MACA,SAAS,OAAO,SAAS;AACvB,cAAM,QAAS,MAAyC;AACxD,cAAM,MAAM,SAAS,SAAS;AAC9B,eAAO,OAAO,UAAU,WAAW,IAAI,MAAM,GAAG,KAAK,IAAI;AAAA,MAC3D;AAAA,IACF;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,aACE;AAAA,MACF,YAAY;AAAA,QACV,MAAM;AAAA,QACN,YAAY;AAAA,UACV,QAAQ,EAAE,MAAM,UAAU,MAAM,CAAC,QAAQ,MAAM,EAAE;AAAA,UACjD,aAAa,EAAE,MAAM,UAAU,aAAa,uCAAuC;AAAA,QACrF;AAAA,QACA,sBAAsB;AAAA,MACxB;AAAA,MACA,SAAS,OAAO,SAAS;AACvB,cAAM,IAAK,QAAQ,CAAC;AACpB,cAAM,UAAU,SAAS,QAAQ;AACjC,YAAI,EAAE,WAAW,OAAQ,QAAO,kBAAkB,SAAS,EAAE,aAAa,EAAE,YAAY,CAAC;AACzF,eAAO;AAAA,MACT;AAAA,IACF;AAAA,EACF;AACF;","names":["cellId"]}
|
package/dist/hosted/index.d.ts
CHANGED
|
@@ -347,9 +347,15 @@ interface InterRaterInsight {
|
|
|
347
347
|
raters: number;
|
|
348
348
|
/** Number of runs every rater scored. */
|
|
349
349
|
jointlyRated: number;
|
|
350
|
-
/**
|
|
350
|
+
/** Multi-rater weighted kappa over the jointly rated runs. */
|
|
351
351
|
kappa: number;
|
|
352
|
-
/**
|
|
352
|
+
/** Absolute agreement across raters, using ICC(2,1). */
|
|
353
|
+
icc: number;
|
|
354
|
+
/** Mean pairwise Pearson correlation. Correlation is not agreement. */
|
|
355
|
+
pearson: number;
|
|
356
|
+
/** Mean pairwise Spearman rank correlation. */
|
|
357
|
+
spearman: number;
|
|
358
|
+
/** Pairwise weighted kappa per rater pair (key = `"raterA::raterB"`). */
|
|
353
359
|
perPair: Record<string, number>;
|
|
354
360
|
/** Run ids where raters disagree the most — the high-value triage list. */
|
|
355
361
|
disagreementCases: Array<{
|
package/dist/index.d.ts
CHANGED
|
@@ -990,9 +990,10 @@ declare function evaluateActionPolicy(action: ProposedSideEffect, policy?: Actio
|
|
|
990
990
|
}): ActionPolicyDecision;
|
|
991
991
|
|
|
992
992
|
type AgentProfileCellSchemaVersion = 'agent-profile-cell/v1';
|
|
993
|
-
type
|
|
993
|
+
type AgentProfileJsonObject = {
|
|
994
994
|
[key: string]: AgentProfileJson;
|
|
995
995
|
};
|
|
996
|
+
type AgentProfileJson = string | number | boolean | null | AgentProfileJson[] | AgentProfileJsonObject;
|
|
996
997
|
type AgentProfileDimensionValue = string | number | boolean | null;
|
|
997
998
|
interface AgentProfileSource {
|
|
998
999
|
/** Runtime/profile contract being fingerprinted, e.g. `agent-interface-profile`. */
|
|
@@ -1614,6 +1615,13 @@ interface LlmClientOptions {
|
|
|
1614
1615
|
deadlineMs?: number;
|
|
1615
1616
|
/** Total provider attempts. Legacy option name; default 3 (1 initial + 2 retries). */
|
|
1616
1617
|
maxRetries?: number;
|
|
1618
|
+
/**
|
|
1619
|
+
* Transport for requests that declare `jsonSchema`. `native` sends
|
|
1620
|
+
* `response_format: json_schema`; `json-object` sends the broadly supported
|
|
1621
|
+
* JSON mode and relies on the caller to include the schema in model-visible
|
|
1622
|
+
* instructions. Default: `native`.
|
|
1623
|
+
*/
|
|
1624
|
+
jsonSchemaTransport?: 'native' | 'json-object';
|
|
1617
1625
|
/** Fetch implementation — defaults to global `fetch`. Override for custom transport (e.g. tests). */
|
|
1618
1626
|
fetch?: typeof fetch;
|
|
1619
1627
|
/**
|
|
@@ -16788,4 +16796,4 @@ type CachedJudge<TArtifact, TScenario extends Scenario = Scenario> = JudgeConfig
|
|
|
16788
16796
|
*/
|
|
16789
16797
|
declare function cachedJudge<TArtifact, TScenario extends Scenario = Scenario>(judge: JudgeConfig<TArtifact, TScenario>, store: VerdictCacheStore, options: CachedJudgeOptions): CachedJudge<TArtifact, TScenario>;
|
|
16790
16798
|
|
|
16791
|
-
export { AGENT_PROFILE_KINDS, ATTESTATION_ALGORITHM, type ActionExecutionPolicy, type ActionPolicyDecision, type ActionableSideInfo, type ActiveLearningOptions, type AdapterRun, AgentDriver, type AgentDriverConfig, AgentEvalError, type AgentEvalErrorCode, type AgentInterfaceProfileLike, type AgentProfileCell, type AgentProfileCellInput, type AgentProfileCellSchemaVersion, AgentProfileCellValidationError, type AgentProfileDimensionValue, type AgentProfileHarness, type AgentProfileJson, type AgentProfileKind, type AgentProfileRuntimeReceipt, type AgentProfileSource, type AgentProfileSourceInput, type AgreementResult, type AlignmentOp, type Analyst, type AnalystContext, type AnalystCost, type AnalystFinding, type AnalystHooks, type AnalystInputKind, AnalystRegistry, type AnalystRegistryOptions, type AnalystRequirements, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystSeverity, type AnalystUsageReceipt, type AnalyzeTracesInput, type AnalyzeTracesOptions, type AnalyzeTracesResult, type AnalyzeTracesTurnSnapshot, type AntiSlopConfig, type AntiSlopIssue, type AntiSlopReport, type Artifact$1 as Artifact, type ArtifactCheck, type Artifact as ArtifactCheckArtifact, type ArtifactEventLike, type ArtifactResult, type ArtifactValidator, type AsiSeverity, type AssertCapabilityHeadroomOptions, type AssertCrossFamilyOptions, type AssertSingleBackendOptions, type AttestationProvenance, type AttestationVerification, type AttestedReport, type AutoPrClient, AxGepaSteeringOptimizer, type AxSteeringOptimizerConfig, BENCHMARK_SPLIT_SEED, type BackendDescriptor, BackendIntegrityError, type BackendIntegrityReport, type BaselineOptions, type BaselineReport, BehaviorAssertion, type BehavioralMetrics, type BehavioralTokenSequence, type BenchmarkAdapter, type BenchmarkDatasetItem, type BenchmarkEvaluation, type BenchmarkFamily, type BenchmarkReport$1 as BenchmarkReport, type BenchmarkResponder, BenchmarkRunner, type BenchmarkRunnerConfig, type BenchmarkScenario, type BenchmarkSource, type BenchmarkTaskKind, type BisectOptions, type BisectResult, type BisectStep, type BlendWeights, type BootstrapOptions, type BootstrapResult, BudgetBreachError, BudgetGuard, type BudgetLedgerEntry, type BudgetPolicy, type BudgetSpec, type BuildAgreementJudgeOptions, CODING_HARNESSES, type CachedJudge, type CachedJudgeOptions, type CalibrationResult, CallExpectation, CallbackResearcher, type CallbackResearcherOptions, type CampaignFactoryParams, type CampaignIntegrityPolicy, type CampaignRunContext, type CampaignRunOutcome, type CampaignRunner, type CampaignScenario, type CampaignVariant, type CanaryAlert, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CanarySeverity, type CandidateComparison, type CandidateScenario, type CandidateScore, type CanonicalRawAnalystFinding, type CapabilityHeadroomOptions, type CapabilityHeadroomResult, type CaptureFetchContext, type CaptureFetchOptions, CaptureIntegrityError, type CausalAttributionReport, type CellVerdict, type ChannelRollup, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTransport, type CheckResult, type CliBridgeTransportOpts, type CliffsMagnitude, type ClusterBootstrapInterval, type ClusterSignFlipAlternative, type ClusterSignFlipResult, type ClusteredBinaryCluster, type ClusteredMatchedPair, type ClusteredPairedBinaryOptions, type ClusteredPairedBinaryResult, type ClusteredPairedBinaryStatistics, type CollectedArtifacts, type CommandRunner, type CompareLabels, type ComparePairedArmsOptions, type CompletionCriterion, type CompletionRequirement, type CompletionVerdict, type ConceptComplexity, type ConceptFinding, type ConceptSpec, type ConceptWeightStrategy, ConfigError, type ContinuityCheck, type ContinuityCheckResult, type ContinuityReport, type ContinuitySnapshotPair, type ContinuousAgreement, type ContinuousAgreementOptions, type ContinuousCalibrationResult, type ContractCheckResult, type ContractJudgeOptions, type ContractMetric, type ContractReport, type ContractRule, type ContractRuleKind, type ContractSpan, type ContractVerdict, type ContractViolation, type ControlActionFailureMode, type ControlActionOutcome, type ControlBudget, type ControlContext, type ControlDecision, type ControlEvalResult, type ControlRunResult, type ControlRunToRunRecordOptions, type ControlRuntimeConfig, type ControlRuntimeError, type ControlSeverity, type ControlStep, type ControlStopPolicies, ConvergenceTracker, type CorpusAgreementOptions, type CorpusAgreementPerDimension, type CorpusAgreementReport, type CorpusScoreRecord, type CorrectnessChecker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, type CostChannel, type CostEntry, CostLedger, type CostLedgerEntry, type CostLedgerFilter, type CostLedgerHandle, type CostLedgerOptions, type CostLedgerPersistence, CostLedgerPersistenceError, type CostLedgerSummary, type CostReceipt, CostReceiptCaptureError, type CostReceiptInput, type CostReport, CostReservationExceededError, type CostResult, type CostSummary, CostTracker, type CostUsage, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateAnalystAiConfig, type CreateChatClientOpts, type CreateDefaultReviewerOptions, type CreateExperimentInput, type CreateSandboxPoolOpts, type CreateTraceAnalystKindOpts, CrossFamilyError, type CrossTraceDiff, type CrossTraceDiffOptions, DEFAULT_AGENT_SLOS, DEFAULT_COMPLEXITY_WEIGHTS, DEFAULT_RULES as DEFAULT_FAILURE_RULES, DEFAULT_FINDERS, DEFAULT_HARNESS_OBJECTIVES, DEFAULT_MUTATION_PRIMITIVES, DEFAULT_MUTATORS, DEFAULT_PR_REVIEW_SCORE_WEIGHTS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, type DataAcquisitionPlan, Dataset, type DatasetDifficulty, type DatasetManifest, type DatasetOverview, type DatasetProvenance, type DatasetScenario, type DatasetSplit, type DecideNextUserTurnOpts, type DefaultAnalystRegistryOptions, type DefaultVerdict, type DeployFamily, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DescriptionLengthCandidate, type DescriptionLengthConfig, type DescriptionLengthDecision, type DescriptionLengthEvidence, DescriptionLengthGate, type DescriptionLengthRejectionCode, type DetectorEvent, type DetectorSeverity, type DetectorSignal, type DiffPolicy, type DiffScorecardOptions, type DirEntry, type DirectProviderTransportOpts, type Direction, type DiscoverPersonasOptions, type DiscoveredPersona, DockerSandboxDriver, type DriverResult, type DriverState, DualAgentBench, type DualAgentBenchConfig, type DualAgentReport, type DualAgentRound, type DualAgentScenario, type DualAgentScenarioResult, type EProcess, type EProcessOptions, type EProcessState, type EProcessStep, ERROR_COUNT_PATTERNS, type EnsembleAggregate, type EnsembleJudgeOptions, type ErrorCluster, type ErrorCountPattern, type ErrorStreakOptions, type EvalCampaignOptions, type EvalCampaignResult, type EvalResult, type EvalToolDef, EvalTraceStore, type EventFilter, type EventKind, type EvidenceRef, type EvolutionRound, type ExecutorConfig, type Expectation, type Experiment, type ExperimentPlan, type ExperimentProvenance, type ExperimentRep, type ExperimentResult, type ExperimentStats, type ExperimentStore, ExperimentTracker, type ExperimentTrackerOptions, type ExperimentVerdict, type ExportableSpan, type ExportedRewardModel, type ExtractOptions, type ExtractResult, type ExtractUsageFromSseOptions, type ExtractedUsage, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, type FactorContribution, type FactorialCell, type FailedRun, type FailureClass, type FailureClassification, type FailureContext, type FailureMode, type FailureRule, type FeedbackArtifactType, type FeedbackAttempt, type FeedbackLabel, type FeedbackLabelKind, type FeedbackLabelSource, type FeedbackOptimizerRow, type FeedbackOutcome, type FeedbackPattern, type FeedbackReplayAdapter, type FeedbackReplayResult, type FeedbackSeverity, type FeedbackSplitPolicy, type FeedbackTask, type FeedbackTrajectory, type FeedbackTrajectoryFilter, type FeedbackTrajectoryStore, type FieldAgreementSpec, type FieldDestination, type FileChange, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, type FileSystemRawProviderSinkOptions, FileSystemTraceStore, type FileSystemTraceStoreOptions, type Finding, type FindingSubject, type FindingSubjectKind, type FindingToPolicyEditOptions, type FindingsDiff, FindingsStore, type FlattenOtlpOptions, type FlowAction, type FlowLayerEnv, type FlowLayerFactoryInput, type FlowRunner, type FlowRunnerStepResult, type FlowSpec, type FlowStep, type GainDistributionBin, type GainDistributionFigureSpec, type GainDistributionOptions, type GateDecision$1 as GateDecision, type GateEvidence, type GenericSpan, type GhCliClientOptions, type GoldScenario, type GoldSplit, type GoldenItem, type GoldenSeverity, type GoldenSpec, HARNESS_NATIVE_MODEL, type HarnessAdapter, type HarnessConfig, type HarnessExperimentConfig, type HarnessExperimentResult, type HarnessIntervention, type HarnessRunRequest, type HarnessRunResult, type HarnessScenario, type HarnessSelection, type HarnessVariant, type HarnessVariantReport, type HeadroomClass, type HeadroomInput, HeldOutGate, type HeldOutGateConfig, type HeldOutGateRejectionCode, type HeldOutPartition, type HiddenCriteriaGrader, type HiddenGradeResult, type HiddenLeak, HoldoutAuditor, HoldoutLockedError, type HttpGithubClientOptions, type HypothesisManifest, type HypothesisResult, IMPROVEMENT_KIND_SPEC, INPUT_VALUE, INTENT_MATCH_JUDGE_VERSION, type ImageData, type ImprovementThresholds, type ImprovementVerdictResult, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, type InMemoryRawProviderSinkOptions, InMemoryTraceStore, InMemoryWorkspaceInspector, type InferenceScorer, type InspectorContext, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InteractionContribution, type InterimReleaseConfidence, type InterimReleaseConfidenceInput, type JudgeConfig$1 as JudgeConfig, JudgeError, type JudgeFamily, type JudgeFleetOptions, type JudgeFn, type JudgeInput, JudgeParseError, type JudgeReplayGateArgs, type JudgeReplayResult, type JudgeRetryOutcome, type JudgeRetryPolicy, type JudgeRubric, JudgeRunner, type JudgeScore$1 as JudgeScore, type JudgeScoreInput, type JudgeScoresRecord, type JudgeSpan, type JudgeVerdict, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type KnowledgeAcquisitionMode, type KnowledgeBundle, type KnowledgeFallbackPolicy, type KnowledgeFreshness, type KnowledgeImportance, type KnowledgeReadinessReport, type KnowledgeRecommendedAction, type KnowledgeRequirement, type KnowledgeRequirementCategory, type KnowledgeResponsibleSurface, type KnowledgeSensitivity, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, type LangfuseEnvelope, type LangfuseGeneration, type LangfuseScore, type Layer, type LayerResult, type LayerStatus, type LeaderboardOptions, type LeaderboardRow, type LiveProofArtifact, type LiveProofConfig, type LiveProofContext, type LiveProofResult, LlmCallError, type LlmCallMetadata, type LlmCallRequest, type LlmCallResult, LlmClient, type LlmClientOptions, type LlmCorrectnessCheckerOpts, type LlmJsonCall, type LlmJudgeDimension, type LlmJudgeOptions, type LlmMessage, LlmResponseError, type LlmReviewerConfig, LlmRouteAssertionError, type LlmRouteRequirements, type LlmSpan, type LlmSpanOtlpInput, type LlmUsage, LockedJsonlAppender, MODEL_PRICING, type MakeEvalToolsConfig, type MatchResult, type MatchedPair, type MatcherResult, type MaximumCharge, type McNemarResult, type MeasurementPolicy, type MergeOptions, type Message, type MetricSamples, type MetricVerdict, MetricsCollector, type MockTransportOpts, type ModelCostRollup, type ModelPreflight, type ModelSeats, ModelsUnreachableError, type MuffledFinder, type MuffledFinding, MultiLayerVerifier, type MultiToolchainLayerConfig, type Mutator, Mutex, type NoLeakOptions, type NoProgressOptions, NoopRawProviderSink, NoopResearcher, NotFoundError, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OUTPUT_VALUE, type Objective, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrthogonalityInput, type OrthogonalityResult, type OtelExportConfig, type OtelExporter, type OtelPipelineHandle, type OtelPipelineOptions, type OtlpExport, OtlpFileTraceStore, type OtlpFileTraceStoreOptions, type OtlpFlatLine, type OtlpResourceSpans, type OtlpSpan, type OtlpToRunRecordsOptions, type OtlpTraceRunRecord, POLICY_EDIT_AXES, POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, POLICY_EDIT_TARGET_SURFACES, type PaidCallResult, type PairArmsOptions, type PairArmsResult, type PairedArmRow, type PairedArmsComparison, type PairedBootstrapOptions, type PairedBootstrapResult, type PairedCorrectness, type PairedEvalueOptions, type PairedEvalueSequence, type PairedEvalueStep, type PairedMetricDelta, type PairedSignTestResult, PairwiseSteeringOptimizer, type ParaphraseRobustnessScenarioInput, type ParaphraseRobustnessScenarioResult, type ParetoFigureSpec, type ParetoPoint, type ParetoResult, type ParseStudentLabel, type PartitionHeldOutOptions, type PendingCostCall, type PendingCostCallView, type PersistedFinding, type PersonaConfig, type PersonaRigor, type Playbook, type PlaybookEntry, type PolicyEdit, type PolicyEditAdmission, type PolicyEditAdmissionOptions, type PolicyEditAxis, type PolicyEditCandidateRecord, type PolicyEditChange, type PolicyEditExpectedGain, type PolicyEditGainDirection, type PolicyEditGainUnit, type PolicyEditInit, type PolicyEditRisk, type PolicyEditSchemaVersion, type PolicyEditSource, type PolicyEditTarget, type PolicyEditTargetSurface, PolicyEditValidationError, type PoolSlot, type PositionalBiasResult, type PrReviewAuditCase, type PrReviewBenchmarkSummary, type PrReviewComment, type PrReviewMatchedFinding, type PrReviewOutcome, type PrReviewReferenceFinding, type PrReviewScore, type PrReviewScoreWeights, type PrReviewSeverity, type PrReviewSource, type PreferenceMemoryEntry, type PreflightModelsOptions, type PreflightOutcome, type ProducedProposal, type ProducedState, type ProductBenchmarkArm, type ProductBenchmarkArtifactPaths, type ProductBenchmarkBudgets, type ProductBenchmarkExportOptions, type ProductBenchmarkExportResult, type ProductBenchmarkManifest, type ProductBenchmarkProfileRef, type ProductBenchmarkRecord, type ProductBenchmarkRepoRef, type ProductBenchmarkRunInput, type ProductBenchmarkScenario, type ProductBenchmarkSingleRunExportOptions, type ProductBenchmarkSplit, type ProductBenchmarkSubstrateVersions, type ProductBenchmarkValidationReport, ProductClient, type ProductClientConfig, type ProfileAxisSpec, type ProjectRuntimeTrajectoryEvidenceOptions, type ProjectedOtlpSpan, type PromptHandle, PromptRegistry, type ProportionInterval, type ProposalEventLike, type ProposeAutomatedPullRequestInput, type ProposeAutomatedPullRequestResult, type ProposeFn, type ProposeInput, type ProposeOutput, type ProposeReviewConfig, type ProposeReviewControlAction, type ProposeReviewControlConfig, type ProposeReviewControlResult, type ProposeReviewControlState, type ProposeReviewReport, type ProposeReviewShot, type ProposedSideEffect, type ProvenanceReader, type ProviderRedactor, type QueryTracesPage, REDACTION_VERSION, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, RESEARCH_REPORT_HARD_PAIR_FLOOR, RUN_COST_ATTR_KEYS, type RawAnalystEvidence, type RawAnalystFinding, type RawProviderDirection, type RawProviderEvent, type RawProviderSink, type RawProviderSinkFilter, type RecordRunsOptions, type RedTeamCase, type RedTeamCategory, type RedTeamFinding, type RedTeamPayload, type RedTeamReport, type RedactionReport, type RedactionRule, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReferenceMatchResult, type ReferenceReplayAdapter, type ReferenceReplayAdapterFn, type ReferenceReplayAdapterLike, type ReferenceReplayAggregate, type ReferenceReplayCandidate, type ReferenceReplayCase, type ReferenceReplayCaseRun, type ReferenceReplayExecutionScenario, type ReferenceReplayItem, type ReferenceReplayMatch, type ReferenceReplayMatchStrategy, type ReferenceReplayMatcher, type ReferenceReplayPromotionDecision, type ReferenceReplayPromotionPolicy, type ReferenceReplayRun, type ReferenceReplayRunContext, type ReferenceReplayRunOptions, type ReferenceReplayRunStore, type ReferenceReplayScenario, type ReferenceReplayScenarioScore, type ReferenceReplayScore, type ReferenceReplayScoreOptions, type ReferenceReplaySplit, type ReferenceReplaySplitComparison, type ReferenceReplaySteeringRowsOptions, type ReflectionContext, type ReflectionProposal, type RegistryRunOpts, type ReleaseConfidenceAxis, type ReleaseConfidenceAxisName, type ReleaseConfidenceInput, type ReleaseConfidenceIssue, type ReleaseConfidenceMetrics, type ReleaseConfidenceScorecard, type ReleaseConfidenceStatus, type ReleaseConfidenceThresholds, type ReleaseTraceEvidence, type RenderReleaseReportOptions, type RenderStudentPrompt, type RepeatedActionOptions, ReplayCache, type ReplayCacheEntry, ReplayCacheMissError, type ReplayCacheStats, ReplayError, type ReplayFetchOptions, type RepoRef, type RequirementCheck, type ResearchReport, type ResearchReportCandidate, type ResearchReportDecision, type ResearchReportMethodology, type ResearchReportOptions, type ResearchReportRecommendation, type Researcher, type RetrievalSpan, type Review, type ReviewFn, type ReviewInput, type ReviewMemoryEntry, type ReviewMemoryStore, type ReviewerMemoryEntry, type ReviewerOutput, type ReviewerPromptInput, type ReviewerSoftFailDefaults, type ReviewerVerificationSummary, type RiskDifferenceResult, type RobustnessResult, type RouteMap, type RoutedField, type RouterTransportOpts, type RubricDimension, type Run, type RunCommandInput, type RunCommandResult, type RunCompleteHook, type RunCompleteHookContext, type RunCostProvenance, RunCritic, type RunCriticOptions, type RunDistillationOptions, type RunDistillationResult, type RunEvidenceMetadata, type RunFilter, RunIntegrityError, type RunIntegrityExpectations, type RunIntegrityIssue, type RunIntegrityIssueCode, type RunIntegrityReport, type RunJudgeMetadata, type RunLayer, type RunOutcome, type RunPaidCallInput, type RunRecord, type RunRecordBackend, type RunRecordFilter, RunRecordValidationError, type RunScore, type RunScoreWeights, type RunSplitTag, type RunStatus, type RunTokenUsage, type RunTrace, type RuntimeEventLike, type RuntimeResolution, type RuntimeTrajectoryEvidenceProjection, type RuntimeTrajectoryEvidenceSummary, type RuntimeTrajectoryHookEvent, type RuntimeTrajectoryRecord, type RuntimeTrajectoryRunRecord, SEMANTIC_CONCEPT_JUDGE_VERSION, SKILL_USAGE_ANALYST, SPAN_KIND_ATTR_KEYS, type SandboxDriver, SandboxHarness, type SandboxHarnessResult, type SandboxJudgeKind, type SandboxJudgeResult, type SandboxJudgeSpec, type SandboxPool, type SandboxResult, type SandboxSdkTransportOpts, type SandboxSpan, type SatisfiedBy, type ScanOptions, type Scenario$1 as Scenario, type ScenarioCost, type ScenarioFile, ScenarioRegistry, type ScenarioResult, type ScoreKnowledgeReadinessOptions, type Scorecard, type ScorecardCell, type ScorecardCellDiff, type ScorecardDiff, type ScorecardEntry, type ScorecardLogLine, type ScoredTarget, type SearchSpanResult, type SearchTraceResult, type SeatName, type SeatPresetName, SeatUnsetError, type SelfPlayOptions, type SelfPlayProposer, type SelfPlayScorer, type SelfPreferenceResult, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SequentialDecision, type SerializedRegex, type SeriesConvergenceOptions, type SeriesConvergenceResult, type Severity, type SignTestAlternative, type SignedManifest, type SignedManifestAlgo, type SingleBackendDivergence, SingleBackendError, type SingleBackendField, type SingleBackendReport, SkillUsageAnalyst, type SliceOptions, type Slo, type SloCheckResult, type SloComparator, type SloReport, type SloSeverity, type SlopCategory, type SlotFactory, type Span, type SpanBase, type SpanFilter, type SpanHandle, type SpanKind, type SpanMatchRecord, SpanNotFoundError, type SpanPredicate, type SpanStatus, type SplitGoldOptions, type SseUsageMode, type SteeringBundle, type SteeringChange, type SteeringDelta, type SteeringOptimizationResult, type SteeringOptimizationRow, type SteeringOptimizationSelector, type SteeringOptimizerBackend, type SteeringOptimizerConfig, type SteeringRolePrompt, type StepAttribution, type StopDecision, type StreamingDetector, type SuboptimalCode, type SuboptimalSignal, SubprocessSandboxDriver, type SubprocessSandboxDriverOptions, type SummaryTable, type SummaryTableOptions, type SummaryTableRow, type SynthesisReason, type SynthesisTarget, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, type TaskGold, type TaskHeadroom, type TestGradedRunOptions, type TestGradedRunResult, type TestGradedScenario, type TestOutputParser, type TestResult, type TextMatcher, type ThresholdContract, TokenCounter, type TokenSpec, type ToolCallEventLike, type ToolMatcher, type ToolSpan, type ToolSpanOtlpInput, type ToolStats, type ToolUseMetrics, type ToolUseOptions, type TraceAggregate, type TraceAnalysisStore, type TraceAnalystByteBudgets, type TraceAnalystFilters, type TraceAnalystGolden, type TraceAnalystHookOptions, type TraceAnalystKindSpec, type TraceAnalystSpan, type TraceAnalystSpanKind, type TraceAnalystSpanStatus, type TraceAnalystTraceSummary, type TraceContract, TraceContractBuilder, TraceEmitter, type TraceEmitterOptions, type TraceEvent, TraceFileMissingError, type TraceInsightContext, type TraceInsightFinding, type TraceInsightPanelRole, type TraceInsightPromptInput, type TraceInsightQualityGate, type TraceInsightQuestion, type TraceInsightReadiness, type TraceInsightSuite, type TraceInsightTask, TraceNotFoundError, type TraceStore, type TraceStoreSource, type TraceStoreToOtlpOptions, type TracedAnalystOptions, type TracedJudgeOptions, type TracesToOtlpResult, type Trajectory, type TrajectoryStep, type TreatmentClass, type TreatmentGate, type TreatmentGateInput, type TreatmentGateOptions, type TrialTrace, type Turn, type TurnMetrics, type TurnResult, UI_FINDING_SEVERITIES, UI_LENSES, UNIVERSAL_FINDERS, type UiFinding, type UiFindingScreenshot, type UiFindingSeverity, type UiLens, type UserQuestion, type ValidationContext, ValidationError, type ValidationIssue, type ValidationResult, type VerbosityBiasResult, type Verdict, type VerdictCacheStats, type VerdictCacheStore, type Verification, VerificationError, type VerificationReport, type VerifyContext, type VerifyFn, type VerifyOptions, type ViewSpansResult, type ViewTraceOversized, type ViewTraceResult, type VisualDiffOptions, type VisualDiffResult, type ViteDeployRunnerInput, type WeightedCompositeInput, type WeightedCompositeResult, type WorkerDriverContext, type WorkflowTopology, type WorkspaceAssertion, type WorkspaceAssertionResult, type WorkspaceInspector, type WorkspaceSnapshot, type WranglerDeployRunnerInput, acquisitionPlansForKnowledgeGaps, admitPolicyEdit, adversarialJudge, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, agentProfileModelId, agentVisibleFields, aggregateJudgeVerdicts, aggregateLlm, aggregatePrReviewScore, aggregateRunScore, allCriticalPassed, analyzeAntiSlop, analyzeSeries, analyzeTraces, appendScorecard, applyLlmSpanOtlpAttributes, applyPolicyEditToSurface, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertCapabilityHeadroom, assertCrossFamily, assertLlmRoute, assertModelsServed, assertNoHiddenLeak, assertProductBenchmarkRun, assertRealAgentReceipts, assertRealBackend, assertReleaseConfidence, assertRunAgentProfileCell, assertRunCaptured, assertSingleBackend, assignFeedbackSplit, assignHeldOutTag, attachCostToReport, attest, attributeCounterfactuals, backoffMs, deterministicSplit as benchmarkDeterministicSplit, index$1 as benchmarks, benjaminiHochberg, bisect, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, buildAgentInterfaceProfileCell, buildAgentProfileCell, buildAgreementJudge, buildDefaultAnalystRegistry, buildDriverSystemPrompt, buildProductBenchmarkManifest, buildReflectionPrompt, buildReviewerPrompt, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, buildWorkerDriverSystemPrompt, byteLengthRange, cachedJudge, calibrateJudge, calibrateJudgeContinuous, callLlm, callLlmJson, canaryLeakView, canonicalJson, canonicalize, capabilityHeadroom, captureFetchToRawSink, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, checkTraceContracts, clamp01, classifyFailure, classifyTreatment, cliffsDelta, clusteredPairedBinary, codeExecutionJudge, cohensD, coherenceJudge, collectionPreserved, commentsForSource, commitBisect, comparePairedArms, compareReferenceReplay, compareToBaseline, compilerJudge, completionVerdict, composeParsers, composeValidators, computeExperimentStats, computeFindingId, computePolicyEditId, computeToolUseMetrics, computeTraceMetrics, confidenceInterval, containsAll, contentHash, contextInputTokens, continuousAgreement, contractJudge, controlFailureClassFromVerification, controlRunToFeedbackTrajectory, controlRunToRunRecord, convertTraceStoresToOtlp, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, costReport, createAnalystAi, createAntiSlopJudge, createChatClient, createCustomJudge, createDefaultReviewer, createDomainExpertJudge, createFeedbackTrajectory, createIntentMatchJudge, createLlmCorrectnessChecker, createLlmReviewer, createOtelExporter, createOtelTracingStore, createReferenceEquivalenceJudge, createReplayFetch, createSandboxPool, createSemanticConceptJudge, createTokenRecallChecker, createTraceAnalystKind, crossTraceDiff, crowdingDistance, dataDescriptionBits, decideNextUserTurn, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultBlendWeights, defaultIsMaterial, defaultJudges, defaultParseStudentLabel, defaultProviderRedactor, defaultReferenceReplayMatcher, defaultRenderStudentPrompt, defaultTraceInsightPanel, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, distillPlaybook, domainEvidencePattern, dominates, eProcess, ensembleJudge, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateContract, evaluateHypothesis, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, evaluateTraceContract, executeScenario, expandProfileAxes, expectAgent, exportProductBenchmark, exportProductBenchmarkRuns, exportRewardModel, exportRunAsOtlp, extractAssetUrls, extractErrorCount, extractOtlpAttributes, extractProducedState, extractUsage, extractUsageFromResponse, extractUsageFromSse, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToDatasetScenario, feedbackTrajectoryToOptimizerRow, fieldAgreement, fileContains, fileExists, fileExperimentStore, fileVerdictCache, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findProductBenchmarkArtifacts, findSkipCountsAsPass, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, flowLayer, fnv1a32, formatBenchmarkReport, formatDriverReport, formatFindings, formatScorecardDiff, gainHistogram, gateTreatmentApplied, gateTreatmentFromMetrics, gateTreatmentFromSpans, gateTreatmentFromToolSpans, ghCliClient, gitProvenanceReader, precision as goldenPrecision, gradeOnHidden, gradeSemanticStatus, groupBy, groupRunsByAgentProfileCell, harnessAxisOf, hasCapturedToolArgs, hashContent, hashJson, hashScenarios, hashToUnit, hiddenGrade, holm, htmlContainsElement, httpGithubClient, improvementVerdict, inMemoryExperimentStore, inMemoryReferenceReplayStore, inMemoryReviewStore, inMemoryRunRecordBackend, inMemoryVerdictCache, inferDomainKeywords, inferOtlpKind, interRaterReliability, interpretCliffs, iqr, isHiddenDestination, isJudgeSpan, isLlmSpan, isModelPriced, isOtelConfigured, isPolicyEdit, isRetrievalSpan, isRunRecord, isSandboxSpan, isToolSpan, isTransientLlmError, iterateRawCalls, jestTestParser, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, jsonlReviewStore, jsonlRunRecordBackend, judgeFamily, judgeReplayGate, judgeSpans, keyPreserved, knowledgeReadinessTracePayload, leaderboard, linterJudge, llmJudge, llmSpanFromProvider, llmSpans, loadGoldScenarios, loadScorecard, loadScorerFromGrader, localCommandRunner, lowercaseMutator, makeEvalTools, makeFinding, makePolicyEdit, makePolicyEditCandidateRecord, mannWhitneyU, mapConcurrent, matchGoldens, matchSpan, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, mergeLayerResults, mergeSteeringBundle, modelDescriptionBits, modelHasSnapshot, modelPriceKey, mulberry32, multiToolchainLayer, noProgressDetector, normalizeScores, notBlocked, objectiveEval, observeAll, otelRunCompleteHook, otlpRowsToRunRecords, otlpRowsToTraceRunRecords, otlpToRunRecords, otlpToTraceRunRecords, pairArms, pairedBootstrap, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedSignTest, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, paretoChart, paretoFrontier, paretoFrontierWithCrowding, parseCorrectnessResponse, parseFeedbackTrajectoriesJsonl, parseGoldJsonl, parseReflectionResponse, parseRunRecordSafe, parseRuntimeTrajectoryHookEvent, partialCredit, partitionHeldOut, passAtK, passOrthogonality, pearsonR, pixelDeltaRatio, planTraceInsightQuestions, policyEditFromFinding, policyEditsFromFindings, politenessPrefixMutator, positionalBias, preflightModels, printDriverSummary, probeLlm, productBenchmarkIntegrityFailures, productBenchmarkMutableSurfaces, productBenchmarkRepoIdentity, productBenchmarkSplits, index as profile, projectOtlpFlatLine, projectRuntimeTrajectoryEvidence, promptBisect, proposeSynthesisTargets, providerFromBaseUrl, pytestTestParser, ranks, readOtlpStatus, readProductBenchmarkManifest, readProductBenchmarkRecords, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, redactValue, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, renderMarkdownReport, renderPlaybookMarkdown, renderPreferenceMemoryMarkdown, renderPriorFindings, renderReleaseReport, renderSteeringText, renderUpstreamFindings, repeatedActionDetector, replayFeedbackTrajectories, replayFeedbackTrajectory, replayScorerOverCorpus, replayTraceThroughJudge, requireAgentProfileCell, requiredSampleSize, researchReport, resolveModelPricing, resolveRunCostProvenance, resolveSeat, roundTripRunRecord, routeFields, rowCount, rowWhere, runAgentControlLoop, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runDistillation, runE2EWorkflow, runEvalCampaign, runExpectations, runFailureClass, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runProposeReview, runProposeReviewAsControlLoop, runRecordToProductBenchmarkRecord, runReferenceEquivalenceJudge, runReferenceReplay, runScore, runSelfPlay, runSemanticConceptJudge, runTestGradedScenario, runsForScenario, scalarScore, scanForMuffledGates, scoreContinuity, scoreFromEvals, scoreKnowledgeReadiness, scorePolicyEditReadiness, scorePrReviewComments, scorePrReviewSource, scoreRedTeamOutput, scoreReferenceReplay, scoreTraceInsightReadiness, seatPresets, securityJudge, selectHarnessVariant, selfPreference, sentenceReorderMutator, serializeFeedbackTrajectoriesJsonl, signManifest, spearmanR, splitGold, statusAdvanced, stopOnNoProgress, stopOnRepeatedAction, stringField, stripFencedJson, subjectiveEval, summarizeAgentReceiptIntegrity, summarizeBackendIntegrity, summarizeHarnessResults, summarizePrReviewBenchmark, summarizePreferenceMemory, summaryTable, testJudge, textInSnapshot, throwIfRunIncomplete, toAgentProfileJson, toLangfuseEnvelope, toOpenAiTool, toPrometheusText, tokenizeDomainWords, toolNamesForRun, toolSpans, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceContract, traceJudge, traceJudgeEnsemble, traceSpanKindToOpenInferenceKind, tracedAnalyzeTraces, typoMutator, urlContains, userQuestionsForKnowledgeGaps, validateAgentProfileCell, validatePolicyEdit, validatePolicyEditCandidateRecord, validateProductBenchmarkManifest, validateProductBenchmarkRecord, validateProductBenchmarkRun, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, verifyManifest, visualDiff, viteDeployRunner, vitestTestParser, weightedComposite, weightedMean, weightedRecall, welchsTTest, whitespaceCollapseMutator, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, withOtelPipeline, wranglerDeployRunner };
|
|
16799
|
+
export { AGENT_PROFILE_KINDS, ATTESTATION_ALGORITHM, type ActionExecutionPolicy, type ActionPolicyDecision, type ActionableSideInfo, type ActiveLearningOptions, type AdapterRun, AgentDriver, type AgentDriverConfig, AgentEvalError, type AgentEvalErrorCode, type AgentInterfaceProfileLike, type AgentProfileCell, type AgentProfileCellInput, type AgentProfileCellSchemaVersion, AgentProfileCellValidationError, type AgentProfileDimensionValue, type AgentProfileHarness, type AgentProfileJson, type AgentProfileJsonObject, type AgentProfileKind, type AgentProfileRuntimeReceipt, type AgentProfileSource, type AgentProfileSourceInput, type AgreementResult, type AlignmentOp, type Analyst, type AnalystContext, type AnalystCost, type AnalystFinding, type AnalystHooks, type AnalystInputKind, AnalystRegistry, type AnalystRegistryOptions, type AnalystRequirements, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystSeverity, type AnalystUsageReceipt, type AnalyzeTracesInput, type AnalyzeTracesOptions, type AnalyzeTracesResult, type AnalyzeTracesTurnSnapshot, type AntiSlopConfig, type AntiSlopIssue, type AntiSlopReport, type Artifact$1 as Artifact, type ArtifactCheck, type Artifact as ArtifactCheckArtifact, type ArtifactEventLike, type ArtifactResult, type ArtifactValidator, type AsiSeverity, type AssertCapabilityHeadroomOptions, type AssertCrossFamilyOptions, type AssertSingleBackendOptions, type AttestationProvenance, type AttestationVerification, type AttestedReport, type AutoPrClient, AxGepaSteeringOptimizer, type AxSteeringOptimizerConfig, BENCHMARK_SPLIT_SEED, type BackendDescriptor, BackendIntegrityError, type BackendIntegrityReport, type BaselineOptions, type BaselineReport, BehaviorAssertion, type BehavioralMetrics, type BehavioralTokenSequence, type BenchmarkAdapter, type BenchmarkDatasetItem, type BenchmarkEvaluation, type BenchmarkFamily, type BenchmarkReport$1 as BenchmarkReport, type BenchmarkResponder, BenchmarkRunner, type BenchmarkRunnerConfig, type BenchmarkScenario, type BenchmarkSource, type BenchmarkTaskKind, type BisectOptions, type BisectResult, type BisectStep, type BlendWeights, type BootstrapOptions, type BootstrapResult, BudgetBreachError, BudgetGuard, type BudgetLedgerEntry, type BudgetPolicy, type BudgetSpec, type BuildAgreementJudgeOptions, CODING_HARNESSES, type CachedJudge, type CachedJudgeOptions, type CalibrationResult, CallExpectation, CallbackResearcher, type CallbackResearcherOptions, type CampaignFactoryParams, type CampaignIntegrityPolicy, type CampaignRunContext, type CampaignRunOutcome, type CampaignRunner, type CampaignScenario, type CampaignVariant, type CanaryAlert, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CanarySeverity, type CandidateComparison, type CandidateScenario, type CandidateScore, type CanonicalRawAnalystFinding, type CapabilityHeadroomOptions, type CapabilityHeadroomResult, type CaptureFetchContext, type CaptureFetchOptions, CaptureIntegrityError, type CausalAttributionReport, type CellVerdict, type ChannelRollup, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTransport, type CheckResult, type CliBridgeTransportOpts, type CliffsMagnitude, type ClusterBootstrapInterval, type ClusterSignFlipAlternative, type ClusterSignFlipResult, type ClusteredBinaryCluster, type ClusteredMatchedPair, type ClusteredPairedBinaryOptions, type ClusteredPairedBinaryResult, type ClusteredPairedBinaryStatistics, type CollectedArtifacts, type CommandRunner, type CompareLabels, type ComparePairedArmsOptions, type CompletionCriterion, type CompletionRequirement, type CompletionVerdict, type ConceptComplexity, type ConceptFinding, type ConceptSpec, type ConceptWeightStrategy, ConfigError, type ContinuityCheck, type ContinuityCheckResult, type ContinuityReport, type ContinuitySnapshotPair, type ContinuousAgreement, type ContinuousAgreementOptions, type ContinuousCalibrationResult, type ContractCheckResult, type ContractJudgeOptions, type ContractMetric, type ContractReport, type ContractRule, type ContractRuleKind, type ContractSpan, type ContractVerdict, type ContractViolation, type ControlActionFailureMode, type ControlActionOutcome, type ControlBudget, type ControlContext, type ControlDecision, type ControlEvalResult, type ControlRunResult, type ControlRunToRunRecordOptions, type ControlRuntimeConfig, type ControlRuntimeError, type ControlSeverity, type ControlStep, type ControlStopPolicies, ConvergenceTracker, type CorpusAgreementOptions, type CorpusAgreementPerDimension, type CorpusAgreementReport, type CorpusScoreRecord, type CorrectnessChecker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, type CostChannel, type CostEntry, CostLedger, type CostLedgerEntry, type CostLedgerFilter, type CostLedgerHandle, type CostLedgerOptions, type CostLedgerPersistence, CostLedgerPersistenceError, type CostLedgerSummary, type CostReceipt, CostReceiptCaptureError, type CostReceiptInput, type CostReport, CostReservationExceededError, type CostResult, type CostSummary, CostTracker, type CostUsage, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateAnalystAiConfig, type CreateChatClientOpts, type CreateDefaultReviewerOptions, type CreateExperimentInput, type CreateSandboxPoolOpts, type CreateTraceAnalystKindOpts, CrossFamilyError, type CrossTraceDiff, type CrossTraceDiffOptions, DEFAULT_AGENT_SLOS, DEFAULT_COMPLEXITY_WEIGHTS, DEFAULT_RULES as DEFAULT_FAILURE_RULES, DEFAULT_FINDERS, DEFAULT_HARNESS_OBJECTIVES, DEFAULT_MUTATION_PRIMITIVES, DEFAULT_MUTATORS, DEFAULT_PR_REVIEW_SCORE_WEIGHTS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, type DataAcquisitionPlan, Dataset, type DatasetDifficulty, type DatasetManifest, type DatasetOverview, type DatasetProvenance, type DatasetScenario, type DatasetSplit, type DecideNextUserTurnOpts, type DefaultAnalystRegistryOptions, type DefaultVerdict, type DeployFamily, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DescriptionLengthCandidate, type DescriptionLengthConfig, type DescriptionLengthDecision, type DescriptionLengthEvidence, DescriptionLengthGate, type DescriptionLengthRejectionCode, type DetectorEvent, type DetectorSeverity, type DetectorSignal, type DiffPolicy, type DiffScorecardOptions, type DirEntry, type DirectProviderTransportOpts, type Direction, type DiscoverPersonasOptions, type DiscoveredPersona, DockerSandboxDriver, type DriverResult, type DriverState, DualAgentBench, type DualAgentBenchConfig, type DualAgentReport, type DualAgentRound, type DualAgentScenario, type DualAgentScenarioResult, type EProcess, type EProcessOptions, type EProcessState, type EProcessStep, ERROR_COUNT_PATTERNS, type EnsembleAggregate, type EnsembleJudgeOptions, type ErrorCluster, type ErrorCountPattern, type ErrorStreakOptions, type EvalCampaignOptions, type EvalCampaignResult, type EvalResult, type EvalToolDef, EvalTraceStore, type EventFilter, type EventKind, type EvidenceRef, type EvolutionRound, type ExecutorConfig, type Expectation, type Experiment, type ExperimentPlan, type ExperimentProvenance, type ExperimentRep, type ExperimentResult, type ExperimentStats, type ExperimentStore, ExperimentTracker, type ExperimentTrackerOptions, type ExperimentVerdict, type ExportableSpan, type ExportedRewardModel, type ExtractOptions, type ExtractResult, type ExtractUsageFromSseOptions, type ExtractedUsage, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, type FactorContribution, type FactorialCell, type FailedRun, type FailureClass, type FailureClassification, type FailureContext, type FailureMode, type FailureRule, type FeedbackArtifactType, type FeedbackAttempt, type FeedbackLabel, type FeedbackLabelKind, type FeedbackLabelSource, type FeedbackOptimizerRow, type FeedbackOutcome, type FeedbackPattern, type FeedbackReplayAdapter, type FeedbackReplayResult, type FeedbackSeverity, type FeedbackSplitPolicy, type FeedbackTask, type FeedbackTrajectory, type FeedbackTrajectoryFilter, type FeedbackTrajectoryStore, type FieldAgreementSpec, type FieldDestination, type FileChange, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, type FileSystemRawProviderSinkOptions, FileSystemTraceStore, type FileSystemTraceStoreOptions, type Finding, type FindingSubject, type FindingSubjectKind, type FindingToPolicyEditOptions, type FindingsDiff, FindingsStore, type FlattenOtlpOptions, type FlowAction, type FlowLayerEnv, type FlowLayerFactoryInput, type FlowRunner, type FlowRunnerStepResult, type FlowSpec, type FlowStep, type GainDistributionBin, type GainDistributionFigureSpec, type GainDistributionOptions, type GateDecision$1 as GateDecision, type GateEvidence, type GenericSpan, type GhCliClientOptions, type GoldScenario, type GoldSplit, type GoldenItem, type GoldenSeverity, type GoldenSpec, HARNESS_NATIVE_MODEL, type HarnessAdapter, type HarnessConfig, type HarnessExperimentConfig, type HarnessExperimentResult, type HarnessIntervention, type HarnessRunRequest, type HarnessRunResult, type HarnessScenario, type HarnessSelection, type HarnessVariant, type HarnessVariantReport, type HeadroomClass, type HeadroomInput, HeldOutGate, type HeldOutGateConfig, type HeldOutGateRejectionCode, type HeldOutPartition, type HiddenCriteriaGrader, type HiddenGradeResult, type HiddenLeak, HoldoutAuditor, HoldoutLockedError, type HttpGithubClientOptions, type HypothesisManifest, type HypothesisResult, IMPROVEMENT_KIND_SPEC, INPUT_VALUE, INTENT_MATCH_JUDGE_VERSION, type ImageData, type ImprovementThresholds, type ImprovementVerdictResult, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, type InMemoryRawProviderSinkOptions, InMemoryTraceStore, InMemoryWorkspaceInspector, type InferenceScorer, type InspectorContext, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InteractionContribution, type InterimReleaseConfidence, type InterimReleaseConfidenceInput, type JudgeConfig$1 as JudgeConfig, JudgeError, type JudgeFamily, type JudgeFleetOptions, type JudgeFn, type JudgeInput, JudgeParseError, type JudgeReplayGateArgs, type JudgeReplayResult, type JudgeRetryOutcome, type JudgeRetryPolicy, type JudgeRubric, JudgeRunner, type JudgeScore$1 as JudgeScore, type JudgeScoreInput, type JudgeScoresRecord, type JudgeSpan, type JudgeVerdict, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type KnowledgeAcquisitionMode, type KnowledgeBundle, type KnowledgeFallbackPolicy, type KnowledgeFreshness, type KnowledgeImportance, type KnowledgeReadinessReport, type KnowledgeRecommendedAction, type KnowledgeRequirement, type KnowledgeRequirementCategory, type KnowledgeResponsibleSurface, type KnowledgeSensitivity, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, type LangfuseEnvelope, type LangfuseGeneration, type LangfuseScore, type Layer, type LayerResult, type LayerStatus, type LeaderboardOptions, type LeaderboardRow, type LiveProofArtifact, type LiveProofConfig, type LiveProofContext, type LiveProofResult, LlmCallError, type LlmCallMetadata, type LlmCallRequest, type LlmCallResult, LlmClient, type LlmClientOptions, type LlmCorrectnessCheckerOpts, type LlmJsonCall, type LlmJudgeDimension, type LlmJudgeOptions, type LlmMessage, LlmResponseError, type LlmReviewerConfig, LlmRouteAssertionError, type LlmRouteRequirements, type LlmSpan, type LlmSpanOtlpInput, type LlmUsage, LockedJsonlAppender, MODEL_PRICING, type MakeEvalToolsConfig, type MatchResult, type MatchedPair, type MatcherResult, type MaximumCharge, type McNemarResult, type MeasurementPolicy, type MergeOptions, type Message, type MetricSamples, type MetricVerdict, MetricsCollector, type MockTransportOpts, type ModelCostRollup, type ModelPreflight, type ModelSeats, ModelsUnreachableError, type MuffledFinder, type MuffledFinding, MultiLayerVerifier, type MultiToolchainLayerConfig, type Mutator, Mutex, type NoLeakOptions, type NoProgressOptions, NoopRawProviderSink, NoopResearcher, NotFoundError, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OUTPUT_VALUE, type Objective, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrthogonalityInput, type OrthogonalityResult, type OtelExportConfig, type OtelExporter, type OtelPipelineHandle, type OtelPipelineOptions, type OtlpExport, OtlpFileTraceStore, type OtlpFileTraceStoreOptions, type OtlpFlatLine, type OtlpResourceSpans, type OtlpSpan, type OtlpToRunRecordsOptions, type OtlpTraceRunRecord, POLICY_EDIT_AXES, POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, POLICY_EDIT_TARGET_SURFACES, type PaidCallResult, type PairArmsOptions, type PairArmsResult, type PairedArmRow, type PairedArmsComparison, type PairedBootstrapOptions, type PairedBootstrapResult, type PairedCorrectness, type PairedEvalueOptions, type PairedEvalueSequence, type PairedEvalueStep, type PairedMetricDelta, type PairedSignTestResult, PairwiseSteeringOptimizer, type ParaphraseRobustnessScenarioInput, type ParaphraseRobustnessScenarioResult, type ParetoFigureSpec, type ParetoPoint, type ParetoResult, type ParseStudentLabel, type PartitionHeldOutOptions, type PendingCostCall, type PendingCostCallView, type PersistedFinding, type PersonaConfig, type PersonaRigor, type Playbook, type PlaybookEntry, type PolicyEdit, type PolicyEditAdmission, type PolicyEditAdmissionOptions, type PolicyEditAxis, type PolicyEditCandidateRecord, type PolicyEditChange, type PolicyEditExpectedGain, type PolicyEditGainDirection, type PolicyEditGainUnit, type PolicyEditInit, type PolicyEditRisk, type PolicyEditSchemaVersion, type PolicyEditSource, type PolicyEditTarget, type PolicyEditTargetSurface, PolicyEditValidationError, type PoolSlot, type PositionalBiasResult, type PrReviewAuditCase, type PrReviewBenchmarkSummary, type PrReviewComment, type PrReviewMatchedFinding, type PrReviewOutcome, type PrReviewReferenceFinding, type PrReviewScore, type PrReviewScoreWeights, type PrReviewSeverity, type PrReviewSource, type PreferenceMemoryEntry, type PreflightModelsOptions, type PreflightOutcome, type ProducedProposal, type ProducedState, type ProductBenchmarkArm, type ProductBenchmarkArtifactPaths, type ProductBenchmarkBudgets, type ProductBenchmarkExportOptions, type ProductBenchmarkExportResult, type ProductBenchmarkManifest, type ProductBenchmarkProfileRef, type ProductBenchmarkRecord, type ProductBenchmarkRepoRef, type ProductBenchmarkRunInput, type ProductBenchmarkScenario, type ProductBenchmarkSingleRunExportOptions, type ProductBenchmarkSplit, type ProductBenchmarkSubstrateVersions, type ProductBenchmarkValidationReport, ProductClient, type ProductClientConfig, type ProfileAxisSpec, type ProjectRuntimeTrajectoryEvidenceOptions, type ProjectedOtlpSpan, type PromptHandle, PromptRegistry, type ProportionInterval, type ProposalEventLike, type ProposeAutomatedPullRequestInput, type ProposeAutomatedPullRequestResult, type ProposeFn, type ProposeInput, type ProposeOutput, type ProposeReviewConfig, type ProposeReviewControlAction, type ProposeReviewControlConfig, type ProposeReviewControlResult, type ProposeReviewControlState, type ProposeReviewReport, type ProposeReviewShot, type ProposedSideEffect, type ProvenanceReader, type ProviderRedactor, type QueryTracesPage, REDACTION_VERSION, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, RESEARCH_REPORT_HARD_PAIR_FLOOR, RUN_COST_ATTR_KEYS, type RawAnalystEvidence, type RawAnalystFinding, type RawProviderDirection, type RawProviderEvent, type RawProviderSink, type RawProviderSinkFilter, type RecordRunsOptions, type RedTeamCase, type RedTeamCategory, type RedTeamFinding, type RedTeamPayload, type RedTeamReport, type RedactionReport, type RedactionRule, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReferenceMatchResult, type ReferenceReplayAdapter, type ReferenceReplayAdapterFn, type ReferenceReplayAdapterLike, type ReferenceReplayAggregate, type ReferenceReplayCandidate, type ReferenceReplayCase, type ReferenceReplayCaseRun, type ReferenceReplayExecutionScenario, type ReferenceReplayItem, type ReferenceReplayMatch, type ReferenceReplayMatchStrategy, type ReferenceReplayMatcher, type ReferenceReplayPromotionDecision, type ReferenceReplayPromotionPolicy, type ReferenceReplayRun, type ReferenceReplayRunContext, type ReferenceReplayRunOptions, type ReferenceReplayRunStore, type ReferenceReplayScenario, type ReferenceReplayScenarioScore, type ReferenceReplayScore, type ReferenceReplayScoreOptions, type ReferenceReplaySplit, type ReferenceReplaySplitComparison, type ReferenceReplaySteeringRowsOptions, type ReflectionContext, type ReflectionProposal, type RegistryRunOpts, type ReleaseConfidenceAxis, type ReleaseConfidenceAxisName, type ReleaseConfidenceInput, type ReleaseConfidenceIssue, type ReleaseConfidenceMetrics, type ReleaseConfidenceScorecard, type ReleaseConfidenceStatus, type ReleaseConfidenceThresholds, type ReleaseTraceEvidence, type RenderReleaseReportOptions, type RenderStudentPrompt, type RepeatedActionOptions, ReplayCache, type ReplayCacheEntry, ReplayCacheMissError, type ReplayCacheStats, ReplayError, type ReplayFetchOptions, type RepoRef, type RequirementCheck, type ResearchReport, type ResearchReportCandidate, type ResearchReportDecision, type ResearchReportMethodology, type ResearchReportOptions, type ResearchReportRecommendation, type Researcher, type RetrievalSpan, type Review, type ReviewFn, type ReviewInput, type ReviewMemoryEntry, type ReviewMemoryStore, type ReviewerMemoryEntry, type ReviewerOutput, type ReviewerPromptInput, type ReviewerSoftFailDefaults, type ReviewerVerificationSummary, type RiskDifferenceResult, type RobustnessResult, type RouteMap, type RoutedField, type RouterTransportOpts, type RubricDimension, type Run, type RunCommandInput, type RunCommandResult, type RunCompleteHook, type RunCompleteHookContext, type RunCostProvenance, RunCritic, type RunCriticOptions, type RunDistillationOptions, type RunDistillationResult, type RunEvidenceMetadata, type RunFilter, RunIntegrityError, type RunIntegrityExpectations, type RunIntegrityIssue, type RunIntegrityIssueCode, type RunIntegrityReport, type RunJudgeMetadata, type RunLayer, type RunOutcome, type RunPaidCallInput, type RunRecord, type RunRecordBackend, type RunRecordFilter, RunRecordValidationError, type RunScore, type RunScoreWeights, type RunSplitTag, type RunStatus, type RunTokenUsage, type RunTrace, type RuntimeEventLike, type RuntimeResolution, type RuntimeTrajectoryEvidenceProjection, type RuntimeTrajectoryEvidenceSummary, type RuntimeTrajectoryHookEvent, type RuntimeTrajectoryRecord, type RuntimeTrajectoryRunRecord, SEMANTIC_CONCEPT_JUDGE_VERSION, SKILL_USAGE_ANALYST, SPAN_KIND_ATTR_KEYS, type SandboxDriver, SandboxHarness, type SandboxHarnessResult, type SandboxJudgeKind, type SandboxJudgeResult, type SandboxJudgeSpec, type SandboxPool, type SandboxResult, type SandboxSdkTransportOpts, type SandboxSpan, type SatisfiedBy, type ScanOptions, type Scenario$1 as Scenario, type ScenarioCost, type ScenarioFile, ScenarioRegistry, type ScenarioResult, type ScoreKnowledgeReadinessOptions, type Scorecard, type ScorecardCell, type ScorecardCellDiff, type ScorecardDiff, type ScorecardEntry, type ScorecardLogLine, type ScoredTarget, type SearchSpanResult, type SearchTraceResult, type SeatName, type SeatPresetName, SeatUnsetError, type SelfPlayOptions, type SelfPlayProposer, type SelfPlayScorer, type SelfPreferenceResult, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SequentialDecision, type SerializedRegex, type SeriesConvergenceOptions, type SeriesConvergenceResult, type Severity, type SignTestAlternative, type SignedManifest, type SignedManifestAlgo, type SingleBackendDivergence, SingleBackendError, type SingleBackendField, type SingleBackendReport, SkillUsageAnalyst, type SliceOptions, type Slo, type SloCheckResult, type SloComparator, type SloReport, type SloSeverity, type SlopCategory, type SlotFactory, type Span, type SpanBase, type SpanFilter, type SpanHandle, type SpanKind, type SpanMatchRecord, SpanNotFoundError, type SpanPredicate, type SpanStatus, type SplitGoldOptions, type SseUsageMode, type SteeringBundle, type SteeringChange, type SteeringDelta, type SteeringOptimizationResult, type SteeringOptimizationRow, type SteeringOptimizationSelector, type SteeringOptimizerBackend, type SteeringOptimizerConfig, type SteeringRolePrompt, type StepAttribution, type StopDecision, type StreamingDetector, type SuboptimalCode, type SuboptimalSignal, SubprocessSandboxDriver, type SubprocessSandboxDriverOptions, type SummaryTable, type SummaryTableOptions, type SummaryTableRow, type SynthesisReason, type SynthesisTarget, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, type TaskGold, type TaskHeadroom, type TestGradedRunOptions, type TestGradedRunResult, type TestGradedScenario, type TestOutputParser, type TestResult, type TextMatcher, type ThresholdContract, TokenCounter, type TokenSpec, type ToolCallEventLike, type ToolMatcher, type ToolSpan, type ToolSpanOtlpInput, type ToolStats, type ToolUseMetrics, type ToolUseOptions, type TraceAggregate, type TraceAnalysisStore, type TraceAnalystByteBudgets, type TraceAnalystFilters, type TraceAnalystGolden, type TraceAnalystHookOptions, type TraceAnalystKindSpec, type TraceAnalystSpan, type TraceAnalystSpanKind, type TraceAnalystSpanStatus, type TraceAnalystTraceSummary, type TraceContract, TraceContractBuilder, TraceEmitter, type TraceEmitterOptions, type TraceEvent, TraceFileMissingError, type TraceInsightContext, type TraceInsightFinding, type TraceInsightPanelRole, type TraceInsightPromptInput, type TraceInsightQualityGate, type TraceInsightQuestion, type TraceInsightReadiness, type TraceInsightSuite, type TraceInsightTask, TraceNotFoundError, type TraceStore, type TraceStoreSource, type TraceStoreToOtlpOptions, type TracedAnalystOptions, type TracedJudgeOptions, type TracesToOtlpResult, type Trajectory, type TrajectoryStep, type TreatmentClass, type TreatmentGate, type TreatmentGateInput, type TreatmentGateOptions, type TrialTrace, type Turn, type TurnMetrics, type TurnResult, UI_FINDING_SEVERITIES, UI_LENSES, UNIVERSAL_FINDERS, type UiFinding, type UiFindingScreenshot, type UiFindingSeverity, type UiLens, type UserQuestion, type ValidationContext, ValidationError, type ValidationIssue, type ValidationResult, type VerbosityBiasResult, type Verdict, type VerdictCacheStats, type VerdictCacheStore, type Verification, VerificationError, type VerificationReport, type VerifyContext, type VerifyFn, type VerifyOptions, type ViewSpansResult, type ViewTraceOversized, type ViewTraceResult, type VisualDiffOptions, type VisualDiffResult, type ViteDeployRunnerInput, type WeightedCompositeInput, type WeightedCompositeResult, type WorkerDriverContext, type WorkflowTopology, type WorkspaceAssertion, type WorkspaceAssertionResult, type WorkspaceInspector, type WorkspaceSnapshot, type WranglerDeployRunnerInput, acquisitionPlansForKnowledgeGaps, admitPolicyEdit, adversarialJudge, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, agentProfileModelId, agentVisibleFields, aggregateJudgeVerdicts, aggregateLlm, aggregatePrReviewScore, aggregateRunScore, allCriticalPassed, analyzeAntiSlop, analyzeSeries, analyzeTraces, appendScorecard, applyLlmSpanOtlpAttributes, applyPolicyEditToSurface, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertCapabilityHeadroom, assertCrossFamily, assertLlmRoute, assertModelsServed, assertNoHiddenLeak, assertProductBenchmarkRun, assertRealAgentReceipts, assertRealBackend, assertReleaseConfidence, assertRunAgentProfileCell, assertRunCaptured, assertSingleBackend, assignFeedbackSplit, assignHeldOutTag, attachCostToReport, attest, attributeCounterfactuals, backoffMs, deterministicSplit as benchmarkDeterministicSplit, index$1 as benchmarks, benjaminiHochberg, bisect, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, buildAgentInterfaceProfileCell, buildAgentProfileCell, buildAgreementJudge, buildDefaultAnalystRegistry, buildDriverSystemPrompt, buildProductBenchmarkManifest, buildReflectionPrompt, buildReviewerPrompt, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, buildWorkerDriverSystemPrompt, byteLengthRange, cachedJudge, calibrateJudge, calibrateJudgeContinuous, callLlm, callLlmJson, canaryLeakView, canonicalJson, canonicalize, capabilityHeadroom, captureFetchToRawSink, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, checkTraceContracts, clamp01, classifyFailure, classifyTreatment, cliffsDelta, clusteredPairedBinary, codeExecutionJudge, cohensD, coherenceJudge, collectionPreserved, commentsForSource, commitBisect, comparePairedArms, compareReferenceReplay, compareToBaseline, compilerJudge, completionVerdict, composeParsers, composeValidators, computeExperimentStats, computeFindingId, computePolicyEditId, computeToolUseMetrics, computeTraceMetrics, confidenceInterval, containsAll, contentHash, contextInputTokens, continuousAgreement, contractJudge, controlFailureClassFromVerification, controlRunToFeedbackTrajectory, controlRunToRunRecord, convertTraceStoresToOtlp, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, costReport, createAnalystAi, createAntiSlopJudge, createChatClient, createCustomJudge, createDefaultReviewer, createDomainExpertJudge, createFeedbackTrajectory, createIntentMatchJudge, createLlmCorrectnessChecker, createLlmReviewer, createOtelExporter, createOtelTracingStore, createReferenceEquivalenceJudge, createReplayFetch, createSandboxPool, createSemanticConceptJudge, createTokenRecallChecker, createTraceAnalystKind, crossTraceDiff, crowdingDistance, dataDescriptionBits, decideNextUserTurn, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultBlendWeights, defaultIsMaterial, defaultJudges, defaultParseStudentLabel, defaultProviderRedactor, defaultReferenceReplayMatcher, defaultRenderStudentPrompt, defaultTraceInsightPanel, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, distillPlaybook, domainEvidencePattern, dominates, eProcess, ensembleJudge, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateContract, evaluateHypothesis, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, evaluateTraceContract, executeScenario, expandProfileAxes, expectAgent, exportProductBenchmark, exportProductBenchmarkRuns, exportRewardModel, exportRunAsOtlp, extractAssetUrls, extractErrorCount, extractOtlpAttributes, extractProducedState, extractUsage, extractUsageFromResponse, extractUsageFromSse, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToDatasetScenario, feedbackTrajectoryToOptimizerRow, fieldAgreement, fileContains, fileExists, fileExperimentStore, fileVerdictCache, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findProductBenchmarkArtifacts, findSkipCountsAsPass, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, flowLayer, fnv1a32, formatBenchmarkReport, formatDriverReport, formatFindings, formatScorecardDiff, gainHistogram, gateTreatmentApplied, gateTreatmentFromMetrics, gateTreatmentFromSpans, gateTreatmentFromToolSpans, ghCliClient, gitProvenanceReader, precision as goldenPrecision, gradeOnHidden, gradeSemanticStatus, groupBy, groupRunsByAgentProfileCell, harnessAxisOf, hasCapturedToolArgs, hashContent, hashJson, hashScenarios, hashToUnit, hiddenGrade, holm, htmlContainsElement, httpGithubClient, improvementVerdict, inMemoryExperimentStore, inMemoryReferenceReplayStore, inMemoryReviewStore, inMemoryRunRecordBackend, inMemoryVerdictCache, inferDomainKeywords, inferOtlpKind, interRaterReliability, interpretCliffs, iqr, isHiddenDestination, isJudgeSpan, isLlmSpan, isModelPriced, isOtelConfigured, isPolicyEdit, isRetrievalSpan, isRunRecord, isSandboxSpan, isToolSpan, isTransientLlmError, iterateRawCalls, jestTestParser, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, jsonlReviewStore, jsonlRunRecordBackend, judgeFamily, judgeReplayGate, judgeSpans, keyPreserved, knowledgeReadinessTracePayload, leaderboard, linterJudge, llmJudge, llmSpanFromProvider, llmSpans, loadGoldScenarios, loadScorecard, loadScorerFromGrader, localCommandRunner, lowercaseMutator, makeEvalTools, makeFinding, makePolicyEdit, makePolicyEditCandidateRecord, mannWhitneyU, mapConcurrent, matchGoldens, matchSpan, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, mergeLayerResults, mergeSteeringBundle, modelDescriptionBits, modelHasSnapshot, modelPriceKey, mulberry32, multiToolchainLayer, noProgressDetector, normalizeScores, notBlocked, objectiveEval, observeAll, otelRunCompleteHook, otlpRowsToRunRecords, otlpRowsToTraceRunRecords, otlpToRunRecords, otlpToTraceRunRecords, pairArms, pairedBootstrap, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedSignTest, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, paretoChart, paretoFrontier, paretoFrontierWithCrowding, parseCorrectnessResponse, parseFeedbackTrajectoriesJsonl, parseGoldJsonl, parseReflectionResponse, parseRunRecordSafe, parseRuntimeTrajectoryHookEvent, partialCredit, partitionHeldOut, passAtK, passOrthogonality, pearsonR, pixelDeltaRatio, planTraceInsightQuestions, policyEditFromFinding, policyEditsFromFindings, politenessPrefixMutator, positionalBias, preflightModels, printDriverSummary, probeLlm, productBenchmarkIntegrityFailures, productBenchmarkMutableSurfaces, productBenchmarkRepoIdentity, productBenchmarkSplits, index as profile, projectOtlpFlatLine, projectRuntimeTrajectoryEvidence, promptBisect, proposeSynthesisTargets, providerFromBaseUrl, pytestTestParser, ranks, readOtlpStatus, readProductBenchmarkManifest, readProductBenchmarkRecords, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, redactValue, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, renderMarkdownReport, renderPlaybookMarkdown, renderPreferenceMemoryMarkdown, renderPriorFindings, renderReleaseReport, renderSteeringText, renderUpstreamFindings, repeatedActionDetector, replayFeedbackTrajectories, replayFeedbackTrajectory, replayScorerOverCorpus, replayTraceThroughJudge, requireAgentProfileCell, requiredSampleSize, researchReport, resolveModelPricing, resolveRunCostProvenance, resolveSeat, roundTripRunRecord, routeFields, rowCount, rowWhere, runAgentControlLoop, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runDistillation, runE2EWorkflow, runEvalCampaign, runExpectations, runFailureClass, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runProposeReview, runProposeReviewAsControlLoop, runRecordToProductBenchmarkRecord, runReferenceEquivalenceJudge, runReferenceReplay, runScore, runSelfPlay, runSemanticConceptJudge, runTestGradedScenario, runsForScenario, scalarScore, scanForMuffledGates, scoreContinuity, scoreFromEvals, scoreKnowledgeReadiness, scorePolicyEditReadiness, scorePrReviewComments, scorePrReviewSource, scoreRedTeamOutput, scoreReferenceReplay, scoreTraceInsightReadiness, seatPresets, securityJudge, selectHarnessVariant, selfPreference, sentenceReorderMutator, serializeFeedbackTrajectoriesJsonl, signManifest, spearmanR, splitGold, statusAdvanced, stopOnNoProgress, stopOnRepeatedAction, stringField, stripFencedJson, subjectiveEval, summarizeAgentReceiptIntegrity, summarizeBackendIntegrity, summarizeHarnessResults, summarizePrReviewBenchmark, summarizePreferenceMemory, summaryTable, testJudge, textInSnapshot, throwIfRunIncomplete, toAgentProfileJson, toLangfuseEnvelope, toOpenAiTool, toPrometheusText, tokenizeDomainWords, toolNamesForRun, toolSpans, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceContract, traceJudge, traceJudgeEnsemble, traceSpanKindToOpenInferenceKind, tracedAnalyzeTraces, typoMutator, urlContains, userQuestionsForKnowledgeGaps, validateAgentProfileCell, validatePolicyEdit, validatePolicyEditCandidateRecord, validateProductBenchmarkManifest, validateProductBenchmarkRecord, validateProductBenchmarkRun, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, verifyManifest, visualDiff, viteDeployRunner, vitestTestParser, weightedComposite, weightedMean, weightedRecall, welchsTTest, whitespaceCollapseMutator, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, withOtelPipeline, wranglerDeployRunner };
|