@workos/quickstudy 0.4.0 → 0.5.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@workos/quickstudy",
-  "version": "0.4.0",
+  "version": "0.5.0",
   "description": "An open-source, company-agnostic eval harness for measuring whether coding agents can complete real integration tasks, across an agent x feature x framework x surface matrix.",
   "license": "MIT",
   "type": "module",

package/src/adapters/echo.ts

@@ -13,6 +13,8 @@ import type { AgentAdapter, AttemptContext, AttemptOutcome, TranscriptEvent } fr
 
 export class EchoAdapter implements AgentAdapter {
   readonly name: string;
+  /** Does nothing, so every eval must fail it: a scorer it passes rewards doing nothing. */
+  readonly role = "negative-control" as const;
   readonly identity = {
     provider: "local", model: "echo-v1", reasoning: "none", cliVersion: "built-in",
     surfacePolicies: { docs: "no-network", mcp: "no-network", cli: "no-network" },

package/src/adapters/oracle.ts

@@ -27,6 +27,7 @@ function oracleMetrics(): AttemptMetrics {
 /** Host variant: the prepared fixture already carries the solution overlay. */
 export class OracleAdapter implements AgentAdapter {
   readonly name = "oracle";
+  readonly role = "positive-control" as const;
   readonly identity = ORACLE_IDENTITY;
   readonly startingState = "solution" as const;
 
@@ -50,6 +51,7 @@ export class OracleContainerAdapter implements ContainerAgentAdapter {
   readonly name = "oracle";
   readonly identity = ORACLE_IDENTITY;
   readonly startingState = "solution" as const;
+  readonly role = "positive-control" as const;
   readonly requiredHostEnv: readonly string[] = [];
 
   setupFiles(): Record<string, string> {

package/src/adapters/types.ts

@@ -82,12 +82,24 @@ export interface AttemptOutcome {
  */
 export type StartingState = "fixture" | "solution";
 
+/**
+ * What an adapter's attempts are for. "agent" is an arm under measurement; a
+ * control checks the instrument instead: "positive-control" must pass (the
+ * built-in oracle applies each eval's reference solution) and
+ * "negative-control" must fail (the built-in echo adapter does nothing). The
+ * runner records the role of the adapter that actually runs each experiment
+ * (RunConfig.roles); the report never guesses it from a name.
+ */
+export type AdapterRole = "agent" | "positive-control" | "negative-control";
+
 export interface AgentAdapter {
   /** The adapter name experiments select via `agent.adapter` (e.g. "echo"). */
   name: string;
   identity?: AdapterIdentityConfig;
   /** Starting workspace the runner prepares for this adapter (default "fixture"). */
   startingState?: StartingState;
+  /** What the adapter's attempts are for (default "agent"). */
+  role?: AdapterRole;
   runAttempt(ctx: AttemptContext): Promise<AttemptOutcome>;
 }
 
@@ -128,6 +140,8 @@ export interface ContainerAgentAdapter {
   identity?: AdapterIdentityConfig;
   /** Starting workspace the runner prepares for this adapter (default "fixture"). */
   startingState?: StartingState;
+  /** What the adapter's attempts are for (default "agent"). */
+  role?: AdapterRole;
   /** Command used during preflight to record the executable's exact version. */
   versionCommand?: readonly string[];
   /**

package/src/build-info.generated.ts

@@ -9,7 +9,7 @@ export const EMBEDDED_BUILD_INFO: {
   // (the `x-release-please-version` marker anchors the replacement). It is
   // restored as-is after a binary build (scripts/build-binary.ts), so the
   // committed source `--version` stays in lockstep with package.json.
-  version: "0.4.0", // x-release-please-version
+  version: "0.5.0", // x-release-please-version
   revision: null,
   dirty: null,
   sourceHash: null,

package/src/report/aggregate.ts

@@ -119,13 +119,16 @@ export function collectExperimentAggregate(
 }
 
 /**
- * Leaderboard order: discriminating pass rate desc (nulls last), then the
+ * Leaderboard order: agent arms before controls (the oracle would otherwise
+ * lead every board), then discriminating pass rate desc (nulls last), then the
  * all-eval pass rate desc (nulls last), then experiment id — a total order,
  * so every consumer renders the same rows in the same sequence.
  */
 export function compareLeaderboard(a: ReportExperiment, b: ReportExperiment): number {
   const rate = (value: number | null | undefined): number => (value === null || value === undefined ? -1 : value);
+  const control = (experiment: ReportExperiment): number => ((experiment.role ?? "agent") === "agent" ? 0 : 1);
   return (
+    control(a) - control(b) ||
     rate(b.aggregate?.pass_rate_discriminating) - rate(a.aggregate?.pass_rate_discriminating) ||
     rate(b.aggregate?.pass_rate_all) - rate(a.aggregate?.pass_rate_all) ||
     a.experiment_id.localeCompare(b.experiment_id)

package/src/report/report.ts

@@ -32,10 +32,16 @@ import { activeAttempts, type AttemptRecord, type ResultsStore, type RunRecord }
 import type { SurfaceUsage } from "../surface-usage.ts";
 import { collectPassAtK, displayPassAtK, type PassAtKEntry } from "./pass-at-k.ts";
 import { estimateRunCosts, type Pricing, type RunCostEstimate } from "./costs.ts";
-import { collectBands, collectScorerHealth, type EvalBand, type ScorerHealth } from "./scorer-health.ts";
+import {
+  collectBands,
+  collectScorerHealth,
+  type EvalBand,
+  type ExperimentRole,
+  type ScorerHealth,
+} from "./scorer-health.ts";
 import { collectExperimentAggregate, compareLeaderboard, type ExperimentAggregate } from "./aggregate.ts";
 
-export type { EvalBand, ScorerHealth, ScorerHealthSuspect } from "./scorer-health.ts";
+export type { EvalBand, ExperimentRole, ScorerHealth, ScorerHealthSuspect } from "./scorer-health.ts";
 export type { ExperimentAggregate } from "./aggregate.ts";
 export { compareLeaderboard } from "./aggregate.ts";
 
@@ -47,7 +53,7 @@ export class ReportError extends Error {
   }
 }
 
-export const REPORT_SCHEMA_VERSION = 9;
+export const REPORT_SCHEMA_VERSION = 10;
 
 /**
  * Substrings that must never appear in generated report prose: at n≤5 trials
@@ -79,7 +85,10 @@ export function methodologyNote(trials: number): string {
     `recorded at least one ungraded check and recalled are those whose ungraded checks all passed. Ungraded ` +
     `checks are recorded and tallied but never enter pass/fail. Scorer health lists checks that failed in every ` +
     `scored attempt of an eval under every arm while another check in the same attempts passed: a statement ` +
-    `about the instrument, not about any treatment.`
+    `about the instrument, not about any treatment. Control experiments — an oracle that applies each eval's ` +
+    `reference solution, a no-op that changes nothing — are reported alongside the agent experiments and ` +
+    `tagged as controls wherever they appear; bands and scorer health are decided over the agent experiments ` +
+    `only, and controls are ranked after every agent experiment.`
   );
 }
 
@@ -112,6 +121,13 @@ export interface ReportExperiment {
   experiment_id: string;
   treatment: string | null;
   agent: ReportAgentIdentity;
+  /**
+   * "agent" for an arm; "positive-control" (the oracle) and "negative-control"
+   * (the echo no-op) never decide a band or a scorer-health suspect. Optional on
+   * the wire: absent in schema_version < 10 artifacts; readers treat absence as
+   * "agent".
+   */
+  role?: ExperimentRole;
   /**
    * Optional on the wire: absent in schema_version < 7 artifacts, which the
    * export may still carry — readers must tolerate absence.
@@ -161,8 +177,9 @@ export interface ReportPair {
   flaky: boolean;
   checks: ReportCheck[];
   /**
-   * The eval's band, denormalised onto each of its pairs: decided over the
-   * eval's pairs with scored > 0; null when fewer than 2 arms scored. Absent
+   * The eval's band, denormalised onto each of its pairs (control pairs too):
+   * decided over the eval's agent pairs with scored > 0; null when fewer than 2
+   * agent arms scored (before schema 10, controls counted as arms). Absent
    * in schema_version < 8 artifacts; readers treat absence as unbanded.
    */
   band: EvalBand | null;
@@ -412,6 +429,21 @@ export function collectRecall(attempts: AttemptRecord[], experimentId: string):
   return { attempts: total, recalled, rate: total > 0 ? recalled / total : null };
 }
 
+/**
+ * An experiment's role as the run recorded it (RunConfig.roles, from the
+ * adapter that actually ran). A run from before roles were recorded keeps no
+ * record of which adapter ran, so every experiment in it reads as an agent —
+ * exactly as before roles existed; nothing is guessed from a name or identity.
+ */
+function roleOf(run: RunRecord, experimentId: string): ExperimentRole {
+  return run.config.roles?.[experimentId] ?? "agent";
+}
+
+/** The run's control experiments, from the roles it recorded. */
+function controlExperiments(run: RunRecord): Set<string> {
+  return new Set(run.config.experimentIds.filter((id) => roleOf(run, id) !== "agent"));
+}
+
 function collectExperiments(
   run: RunRecord,
   attempts: AttemptRecord[],
@@ -425,6 +457,7 @@ function collectExperiments(
       return {
         experiment_id: experimentId,
         treatment: treatmentOf(run, experimentId),
+        role: roleOf(run, experimentId),
         agent: {
           adapter: agent?.adapter ?? "unspecified",
           provider: agent?.provider ?? "unspecified",
@@ -599,9 +632,10 @@ export function buildReport(options: BuildReportOptions): ReportJson {
   if (publication && publication.eligible.length === 0) throw new ReportError(`no eligible complete samples for run ${runId}: ${publication.omitted.map((p) => `${p.eval_id} × ${p.experiment_id}: ${p.reasons.join("; ")}`).join(" | ")}`);
   const attempts = publication ? current.filter((a) => publication.eligible.some((p) => p.eval_id === a.evalId && p.experiment_id === a.experimentId)) : current;
   const pairs = collectPairs(attempts);
-  const bands = collectBands(pairs);
+  const controls = controlExperiments(run);
+  const bands = collectBands(pairs, controls);
   for (const pair of pairs) pair.band = bands.get(pair.eval_id) ?? null;
-  const scorerHealth = collectScorerHealth(attempts, pairs);
+  const scorerHealth = collectScorerHealth(attempts, pairs, controls);
   const costEstimate = options.pricing === undefined ? undefined : estimateRunCosts(run, attempts, options.resultsDir, options.pricing);
 
   const methodology = methodologyNote(run.config.trials) + (costEstimate === undefined ? "" :
@@ -674,11 +708,12 @@ function renderLeaderboard(experiments: ReportExperiment[]): string[] {
       (typeof agg.estimated_cost_usd === "number" ? ` +$${agg.estimated_cost_usd.toFixed(4)} est.` : "");
     const recall = experiment.sdk_recall;
     const recallText = recall && recall.rate !== null ? `${recall.recalled}/${recall.attempts}` : "n/a";
-    return [experiment.experiment_id, `${model}${reasoning}`, `${all} · ${disc}`, time, tokens, cost, recallText];
+    const id = (experiment.role ?? "agent") === "agent" ? experiment.experiment_id : `${experiment.experiment_id} (control)`;
+    return [id, `${model}${reasoning}`, `${all} · ${disc}`, time, tokens, cost, recallText];
   });
   const widths = header.map((title, i) => Math.max(title.length, ...rows.map((row) => row[i]!.length)));
   const format = (cells: string[]): string => `  ${cells.map((cell, i) => cell.padEnd(widths[i]!)).join("  ").trimEnd()}`;
-  return ["leaderboard (benchmark suite, sorted by discriminating pass rate):", format(header), ...rows.map(format)];
+  return ["leaderboard (benchmark suite, sorted by discriminating pass rate; controls last):", format(header), ...rows.map(format)];
 }
 
 /** Render the report as plain text. Renders FROM the JSON, never the store. */
@@ -794,7 +829,9 @@ export function renderReportText(report: ReportJson): string {
       lines.push(`  ${group.group}: compared (raw counts per arm)`);
       for (const arm of group.arms) {
         const cells = arm.evals.map((entry) => `${entry.eval_id} ${entry.display}`).join(", ");
-        lines.push(`    ${arm.experiment_id}${arm.treatment !== null ? ` [${arm.treatment}]` : ""}: ${cells || "no scored attempts"}`);
+        const role = experiments.find((experiment) => experiment.experiment_id === arm.experiment_id)?.role ?? "agent";
+        const control = role === "agent" ? "" : " (control)";
+        lines.push(`    ${arm.experiment_id}${control}${arm.treatment !== null ? ` [${arm.treatment}]` : ""}: ${cells || "no scored attempts"}`);
       }
     }
   }

package/src/report/scorer-health.ts

@@ -13,6 +13,15 @@ import type { ReportPair } from "./report.ts";
 
 export type EvalBand = "discriminating" | "solved-by-all" | "failed-by-all";
 
+/**
+ * What an experiment is for (the run's recorded RunConfig.roles). A control
+ * says whether the instrument works — the oracle must pass, the echo no-op must
+ * fail — so bands and scorer health are decided over agent experiments only:
+ * with an oracle and a no-op in the run, no eval could otherwise be solved or
+ * failed by all.
+ */
+export type ExperimentRole = import("../adapters/types.ts").AdapterRole;
+
 export interface ScorerHealthSuspect {
   eval_id: string;
   check: string;
@@ -30,10 +39,19 @@ export interface ScorerHealth {
   summary: string;
 }
 
-/** Per eval: over pairs with scored > 0. null when fewer than 2 arms scored. */
-export function collectBands(pairs: ReportPair[]): Map<string, EvalBand | null> {
+/**
+ * Per eval: over agent pairs with scored > 0 (`controls` names the control
+ * experiments, which never decide a band). null when fewer than 2 agent arms
+ * scored. Every eval of the run gets an entry, control-only ones as null.
+ */
+export function collectBands(
+  pairs: ReportPair[],
+  controls: ReadonlySet<string> = new Set(),
+): Map<string, EvalBand | null> {
   const byEval = new Map<string, ReportPair[]>();
   for (const pair of pairs) {
+    if (!byEval.has(pair.eval_id)) byEval.set(pair.eval_id, []);
+    if (controls.has(pair.experiment_id)) continue;
     const list = byEval.get(pair.eval_id) ?? [];
     list.push(pair);
     byEval.set(pair.eval_id, list);
@@ -62,18 +80,25 @@ export function scorerHealthSummary(suspects: ScorerHealthSuspect[]): string {
 }
 
 /**
- * Suspects come only from `failed-by-all` evals (≥2 scored arms). A check is a
- * suspect when it is present and failing in every scored (completed) attempt
- * of those arms, and at least one other graded check is present and passing
- * in every one of them. A name flagged `graded: false` in any attempt (the
- * pair tally's rule) is excluded from both sides.
+ * Suspects come only from `failed-by-all` evals (≥2 scored agent arms). A
+ * check is a suspect when it is present and failing in every scored
+ * (completed) attempt of those arms, and at least one other graded check is
+ * present and passing in every one of them. A name flagged `graded: false` in
+ * any attempt (the pair tally's rule) is excluded from both sides. Control
+ * experiments (`controls`) are never arms.
  */
-export function collectScorerHealth(attempts: AttemptRecord[], pairs: ReportPair[]): ScorerHealth {
-  const bands = collectBands(pairs);
+export function collectScorerHealth(
+  attempts: AttemptRecord[],
+  pairs: ReportPair[],
+  controls: ReadonlySet<string> = new Set(),
+): ScorerHealth {
+  const bands = collectBands(pairs, controls);
   const suspects: ScorerHealthSuspect[] = [];
   for (const [evalId, band] of bands) {
     if (band !== "failed-by-all") continue;
-    const arms = pairs.filter((pair) => pair.eval_id === evalId && pair.scored > 0);
+    const arms = pairs.filter(
+      (pair) => pair.eval_id === evalId && pair.scored > 0 && !controls.has(pair.experiment_id),
+    );
     const experiments = arms.map((pair) => pair.experiment_id).sort((a, b) => a.localeCompare(b));
     const ungraded = new Set(arms.flatMap((pair) => pair.checks.filter((check) => check.graded === false).map((check) => check.name)));
     const scored = attempts.filter(

package/src/runner/execute.ts

@@ -506,6 +506,18 @@ export async function executeRunV3(options: ExecuteRunV3Options): Promise<RunSum
       // ON THE RUN: the manifest holds runtime config only as an opaque hash,
       // so the report could never recompute this after the fact.
       groups: compareGroups(options.experiments.map((entry) => entry.experiment), manifest),
+      // Roles come from the resolved adapter, never its name: a scripted
+      // container command is an agent whatever `agent.adapter` says.
+      roles: Object.fromEntries(
+        [...experimentsById.values()].map((experiment) => [
+          experiment.id,
+          experiment.runtime.kind === "host"
+            ? (hostAdapters[experiment.agent.adapter]?.role ?? "agent")
+            : (experiment.runtime.command ?? []).length > 0
+              ? "agent"
+              : (containerAgents.get(experiment.id)?.role ?? "agent"),
+        ]),
+      ),
       ...(offeredSurfaces !== undefined ? { surfaces: offeredSurfaces } : {}),
     };
 

package/src/store/db.ts

@@ -34,6 +34,13 @@ function toLimitKind(value: string | null): LimitKind | null {
 
 /** The run-level configuration recorded in `runs.config_json`. */
 export interface RunConfig {
+  /**
+   * Each experiment's role, from the adapter that actually ran it (a scripted
+   * container command is always "agent"). Recorded here — NOT in the manifest —
+   * like `groups`, so it never perturbs run identity. Absent on older runs: every
+   * experiment then reads as "agent".
+   */
+  roles?: Record<string, import("../adapters/types.ts").AdapterRole>;
   scheduler?: { concurrency: number; seed: string; policy: "seeded-hash-v1"; order: import("../plan.ts").AttemptPlan[] };
   trials: number;
   evalIds: string[];

package/ui/dist/index.html

@@ -24,7 +24,7 @@ Error generating stack: `+e.message+`
 `);return t.at(-1)===``&&t.pop(),t.map((e,t)=>Ee(e,t+1))}function A({turn:e,highlighted:t,defaultExpanded:n}){let[r,i]=(0,l.useState)(n);return(0,S.jsxs)(`div`,{className:t?`turn highlight`:`turn`,id:`L${e.line}`,"data-line":e.line,children:[(0,S.jsxs)(`button`,{type:`button`,className:`turn-head`,onClick:()=>i(!r),"aria-expanded":r,children:[(0,S.jsxs)(`span`,{className:`line-no`,children:[`L`,e.line]}),(0,S.jsx)(`span`,{className:`role role-${e.role}`,children:e.role}),(0,S.jsxs)(`span`,{className:`summary`,children:[e.isError&&`⚠ `,e.summary]}),(0,S.jsx)(`span`,{"aria-hidden":`true`,children:r?`▾`:`▸`})]}),r&&(0,S.jsx)(`div`,{className:`turn-body`,children:(0,S.jsx)(`pre`,{children:e.detail})})]})}function Oe({text:e,lines:t}){let n=(0,l.useMemo)(()=>De(e),[e]),[r,i]=(0,l.useState)(!1),a=(0,l.useRef)(null),o=!r&&n.length>k,s=n,c=null;if(o){if(t){let e=Math.min(Math.max(t[0]-1,0),n.length-1),r=Math.max(0,e-Ce),i=Math.min(n.length,Math.max(e+Ce,t[1])+1);s=n.slice(r,i),c=`showing turns ${r+1}–${i} of ${n.length} (window around L${t[0]})`}else s=n.slice(0,k),c=`showing the first ${k} of ${n.length} turns`}(0,l.useEffect)(()=>{t&&(a.current?.querySelector(`#L${t[0]}`))?.scrollIntoView({block:`center`})},[t,r,e]);let u=e=>t!==void 0&&e>=t[0]&&e<=t[1];return(0,S.jsxs)(`div`,{className:`quickstudy-viewer viewer`,ref:a,"data-testid":`transcript-viewer`,children:[c!==null&&(0,S.jsxs)(`div`,{className:`viewer-note`,children:[(0,S.jsx)(`span`,{children:c}),(0,S.jsxs)(`button`,{type:`button`,onClick:()=>i(!0),"data-testid":`load-all`,children:[`Load all `,n.length,` turns`]})]}),s.map(e=>(0,S.jsx)(A,{turn:e,highlighted:u(e.line),defaultExpanded:u(e.line)},e.line))]})}function ke({runId:e,attemptId:t,artifact:n,lines:r}){let i=C(async()=>{let r=(await _(e)).attempts.find(e=>e.id===t);if(!r)throw new d(`missing`,`data/runs/${e}/run.json`,`attempt "${t}" is not in this run's export`);let i=n===`transcript`?r.transcript:r.diff;if(i===null)throw new d(`missing`,`data/runs/${e}/attempts/${t}`,`this attempt has no ${n} artifact in the export`);return{attempt:r,text:await b(e,i)}},[e,t,n]);if(i.status===`error`)return(0,S.jsx)(ee,{error:i.error});if(i.status===`loading`)return(0,S.jsxs)(`div`,{className:`loading`,children:[`Loading `,n,`…`]});let{attempt:a,text:o}=i.value,s=n===`transcript`?`diff`:`transcript`,c=(n===`transcript`?a.diff:a.transcript)!==null,l=pe(a);return(0,S.jsxs)(`div`,{"data-testid":`attempt-view`,children:[(0,S.jsxs)(`div`,{className:`viewer-head`,children:[(0,S.jsxs)(`div`,{children:[(0,S.jsxs)(`h1`,{style:{fontSize:18,margin:0},children:[a.eval_id,` × `,a.experiment_id,a.treatment===null?``:` · ${a.treatment}`,` · trial `,a.trial_index+1]}),(0,S.jsxs)(`p`,{className:`sub`,style:{margin:`2px 0 0`},children:[`attempt `,(0,S.jsx)(`code`,{children:a.id}),` · `,n,r?` · lines ${r[0]}–${r[1]}`:``]})]}),(0,S.jsxs)(`div`,{className:`controls`,children:[(0,S.jsx)(D,{tone:me(l),children:l}),(0,S.jsx)(ge,{attempt:a}),c&&(0,S.jsxs)(`a`,{href:T.attempt(e,a.id,s),children:[`view `,s]}),(0,S.jsx)(`a`,{href:T.cell(e,{evalId:a.eval_id,experimentId:a.experiment_id}),children:`open cell`})]})]}),a.error!==null&&(0,S.jsx)(`div`,{className:`notice`,children:(0,S.jsx)(`code`,{children:a.error})}),a.checks!==null&&a.checks.length>0&&(0,S.jsx)(`div`,{className:`attempt-line`,style:{margin:`8px 0`},children:a.checks.map(e=>(0,S.jsxs)(`span`,{className:e.passed?`crit pass`:`crit fail`,title:e.notes??e.name,children:[e.passed?`✓`:`✗`,` `,e.name,!E(e)&&(0,S.jsx)(`span`,{className:`tag`,"data-testid":`ungraded-tag`,title:`Recorded, not graded: never enters pass/fail`,children:`recorded`})]},e.name))}),(0,S.jsx)(Se,{treatment:a.treatment,usage:a.surface_usage}),n===`transcript`?(0,S.jsx)(Oe,{text:o,lines:r}):(0,S.jsx)(fe,{text:o,lines:r})]})}var Ae=`unspecified`;function je(e,t){return e===Ae?null:t===Ae?e:`${e} (${t})`}function Me(e){return e.model===Ae?`model unpinned`:e.reasoning===Ae?e.model:`${e.model} · ${e.reasoning}`}function Ne(e,t){let n=e.manifest.experiments[t]?.components.agent;return n===void 0?null:je(n.model,n.reasoning)}function Pe(e,t){let n=e.manifest.experiments[t]?.components.agent;if(n===void 0||n.model===Ae)return null;let r=[`adapter ${n.adapter}`,`provider ${n.provider}`,`model ${n.model}`];return n.reasoning!==Ae&&r.push(`reasoning ${n.reasoning}`),n.cliVersion!==Ae&&r.push(`cli ${n.cliVersion}`),r.join(` · `)}function Fe(e,t){let n=new Set(t.map(t=>Ne(e,t)));return n.size===1?[...n][0]??null:null}function Ie(e){let t=new Map;for(let n of e){if(n.status!==`completed`)continue;let e=t.get(n.eval_id);e||(e=new Map,t.set(n.eval_id,e));let r=e.get(n.experiment_id)??{passed:0,scored:0};r.scored+=1,n.passed===!0&&(r.passed+=1),e.set(n.experiment_id,r)}return t}function Le(e,t,n){let r=[...t.keys()].filter(n=>e.experimentIds.some(e=>t.get(n)?.has(e))),i=[...n.keys()].filter(t=>e.experimentIds.every(e=>n.get(t)?.has(e)===!0));return[...new Set([...r,...i])].sort((e,t)=>e.localeCompare(t))}var Re=e=>e.comparison===`allowed`&&e.experimentIds.length>=2;function ze(e){let t=new Map;for(let n of e){let e=t.get(n.eval_id)??new Set;e.add(n.experiment_id),t.set(n.eval_id,e)}return t}function Be(e,t,n,r){let i=t.filter(Re);if(i.length===0)return[];let a=[];return e<3&&a.push(`With fewer than 3 trials per pair (this run: ${e}), per-arm counts are listed but not comparable.`),i.some(e=>[...r.keys()].some(t=>e.experimentIds.every(e=>r.get(t)?.has(e)===!0)&&e.experimentIds.some(e=>(n.get(t)?.get(e)?.scored??0)===0)))&&a.push(`Some shared evals have an arm with nothing scored; those rows are listed but not comparable.`),a}function Ve(e,t){return e.manifest.experiments[t]?.components.treatment??null}function He({run:e,group:t}){let n=Fe(e,t.experimentIds);return n===null?(0,S.jsx)(`h3`,{children:t.group}):(0,S.jsxs)(`h3`,{children:[n,` `,(0,S.jsx)(`span`,{className:`group-sub`,children:t.group})]})}function Ue({run:e,group:t,arms:n,attempted:r}){if(t.comparison===`withheld`)return(0,S.jsxs)(`div`,{className:`card`,"data-testid":`group-${t.group}`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(He,{run:e,group:t}),(0,S.jsx)(D,{tone:`amber`,children:`comparison withheld`})]}),(0,S.jsxs)(`div`,{className:`body`,children:[(0,S.jsx)(`p`,{className:`viewer-note`,"data-testid":`group-withheld-${t.group}`,children:`Members differ beyond their declared treatment, so no automatic comparison is rendered. Raw per-experiment results remain visible in the matrix. The differing components:`}),(0,S.jsx)(`ul`,{className:`identity-diff`,"data-testid":`group-diff-${t.group}`,children:t.details.map(e=>(0,S.jsx)(`li`,{children:(0,S.jsx)(`code`,{children:e})},e))})]})]});if(t.experimentIds.length<2)return(0,S.jsxs)(`div`,{className:`card`,"data-testid":`group-${t.group}`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(He,{run:e,group:t}),(0,S.jsx)(D,{tone:`gray`,children:`single experiment`})]}),(0,S.jsx)(`div`,{className:`body`,children:(0,S.jsx)(`p`,{className:`viewer-note`,children:`This group declares one experiment — nothing to compare. Its raw results render in the matrix.`})})]});let i=Le(t,n,r);return(0,S.jsxs)(`div`,{className:`card`,"data-testid":`group-${t.group}`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(He,{run:e,group:t}),(0,S.jsx)(D,{tone:`green`,children:`compared — raw counts`})]}),(0,S.jsxs)(`div`,{className:`body`,children:[(0,S.jsxs)(`table`,{className:`data`,"data-testid":`group-table-${t.group}`,children:[(0,S.jsx)(`thead`,{children:(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`eval`}),t.experimentIds.map(t=>(0,S.jsxs)(`th`,{children:[(0,S.jsx)(`code`,{children:t}),Ve(e,t)!==null&&(0,S.jsx)(`div`,{className:`attribution-reason`,children:Ve(e,t)})]},t))]})}),(0,S.jsx)(`tbody`,{children:i.map(e=>(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`td`,{children:(0,S.jsx)(`code`,{children:e})}),t.experimentIds.map(t=>{let r=n.get(e)?.get(t);return(0,S.jsx)(`td`,{"data-testid":`arm-${e}-${t}`,children:r===void 0?`—`:`${r.passed}/${r.scored}`},t)})]},e))})]}),(0,S.jsx)(`p`,{className:`viewer-note`,children:`Raw passed/scored counts per arm. Incomplete and errored attempts are excluded from these counts and reported in the matrix drill-down.`})]})]})}function We({meta:e,runId:t}){let n=t??``,r=C(()=>_(n),[n]),i=(0,l.useMemo)(()=>r.status===`ready`?Ie(r.value.attempts):null,[r]);if(n===``)return(0,S.jsx)(`div`,{"data-testid":`compare-view`,children:(0,S.jsxs)(`div`,{className:`empty-state`,children:[(0,S.jsx)(`h2`,{children:`No run selected`}),(0,S.jsx)(`p`,{children:`Pick a run to see its comparison groups.`})]})});if(r.status===`error`)return(0,S.jsx)(ee,{error:r.error});if(r.status===`loading`||i===null)return(0,S.jsx)(`div`,{className:`loading`,children:`Loading run…`});let a=r.value.groups,o=ze(r.value.attempts),s=Be(r.value.run.trials,a,i,o);return(0,S.jsxs)(`div`,{"data-testid":`compare-view`,children:[(0,S.jsxs)(`div`,{className:`page-head`,children:[(0,S.jsxs)(`div`,{children:[(0,S.jsx)(`h1`,{children:`Comparison groups`}),(0,S.jsx)(`p`,{className:`sub`,children:`Raw per-arm counts for experiments declared as treatments of one configuration; mismatched groups are withheld with their component diff.`})]}),(0,S.jsxs)(`div`,{className:`control`,children:[(0,S.jsx)(`label`,{htmlFor:`compare-run`,children:`Run`}),(0,S.jsx)(`select`,{id:`compare-run`,"data-testid":`compare-run-select`,value:n,onChange:e=>se(T.compare(e.currentTarget.value)),children:e.runs.map(e=>(0,S.jsxs)(`option`,{value:e.id,children:[e.id.slice(0,10),`… · `,e.started_at.slice(0,10),` · `,e.attempts,` attempts`]},e.id))})]})]}),r.value.run.finished_at===null&&(0,S.jsx)(`div`,{className:`notice`,"data-testid":`incomplete-banner`,children:`Incomplete run — this run never finished. Counts are partial.`}),s.length>0&&(0,S.jsx)(`div`,{className:`notice`,"data-testid":`compare-n-notice`,children:s.join(` `)}),a.length===0?(0,S.jsxs)(`div`,{className:`empty-state`,"data-testid":`no-groups`,children:[(0,S.jsx)(`h2`,{children:`No comparison groups in this run`}),(0,S.jsxs)(`p`,{children:[`Declare `,(0,S.jsx)(`code`,{children:`comparisonGroup`}),` on experiments meant to be treatments of one configuration, then re-run.`]})]}):a.map(e=>(0,S.jsx)(Ue,{run:r.value,group:e,arms:i,attempted:o},e.group))]})}var Ge=[`docs`,`mcp`,`cli`,`sdk`,`environment`,`eval-defect`];async function Ke(e){if(typeof navigator>`u`||!navigator.clipboard)throw Error(`the clipboard API needs a secure context (HTTPS, localhost or file://)`);await navigator.clipboard.writeText(e)}function qe({text:e,label:t=`Copy`,testId:n}){let[r,i]=(0,l.useState)({kind:`idle`});return(0,l.useEffect)(()=>{if(r.kind===`idle`)return;let e=setTimeout(()=>i({kind:`idle`}),r.kind===`failed`?4e3:1500);return()=>clearTimeout(e)},[r]),(0,S.jsx)(`button`,{type:`button`,className:`copy-button`,"data-testid":n,title:r.kind===`failed`?r.reason:void 0,onClick:()=>Ke(e()).then(()=>i({kind:`copied`}),e=>i({kind:`failed`,reason:e instanceof Error?e.message:String(e)})),children:r.kind===`idle`?t:r.kind===`copied`?`Copied`:`Copy failed`})}var Je=.25;function Ye(e,t){return`${e}\u0000${t}`}function Xe(e){let t=new Map;for(let n of e){let e=(n.checks??[]).filter(e=>E(e)&&e.passed<e.total).map(e=>e.name).sort();t.set(Ye(n.eval_id,n.experiment_id),e)}return t}function Ze(e,t){let n=t===null?new Map:Xe(t),r=new Map;for(let t of e){let e=n.get(Ye(t.eval_id,t.experiment_id))??[],i=`${t.eval_id}\u0000${e.join(`,`)}`,a=r.get(i);a===void 0&&(a={evalId:t.eval_id,failingChecks:e,findings:[],experiments:[]},r.set(i,a)),a.findings.push(t),a.experiments.includes(t.experiment_id)||a.experiments.push(t.experiment_id)}return[...r.values()].sort((e,t)=>t.findings.length-e.findings.length||(e.evalId<t.evalId?-1:+(e.evalId>t.evalId))||(e.failingChecks.join(`,`)<t.failingChecks.join(`,`)?-1:1))}function Qe(e){let t=e.filter(e=>e.target===`eval-defect`).length;return e.length===0||t/e.length<Je?[...Ge]:[`eval-defect`,...Ge.filter(e=>e!==`eval-defect`)]}function $e(e,t){let n=e.evidence.map(e=>`[${e.artifact} L${e.lines[0]}–${e.lines[1]}](${t(e)})`).join(`, `);return[`**${e.claim}** — ${e.confidence} confidence`,`\`${e.eval_id}\` × \`${e.experiment_id}\` · ${e.target}`,`Suggested change: ${e.suggested_change}`,n?`Evidence: ${n}`:`Evidence: none`].join(`
 `)}function et(e,t){let n=Ge.flatMap(n=>{let r=e.findings.filter(e=>e.target===n);return r.length===0?[]:[`## ${n} findings (${r.length})`,...r.map(e=>$e(e,t))]}),r=e.errors.length?[`## Pairs not analyzed`,...e.errors.map(e=>`- \`${e.eval_id}\` × \`${e.experiment_id}\`: ${e.error}`)]:[];return[`# Run diagnosis ${e.run_id}`,`_${e.disclaimer}_`,`Model ${e.model.requested} (resolved ${e.model.resolved??`unavailable`}) · prompt v${e.prompt_version} · generated ${e.generated_at} · ${e.pairs_diagnosed} pairs diagnosed, ${e.pairs_skipped} skipped, ${e.findings.length} findings`,...n,...r].join(`
 
-`)}function tt(e){return e===`high`?`green`:e===`medium`?`amber`:`gray`}function nt(e,t){return T.attempt(e,t.attempt_id,t.artifact,t.lines)}function rt(e){return t=>new URL(nt(e,t),window.location.href).href}function it({runId:e,evidence:t}){return(0,S.jsxs)(`a`,{className:`evidence-chip`,"data-testid":`evidence-link`,href:nt(e,t),children:[t.artifact,` L`,t.lines[0],`–`,t.lines[1]]})}function at({runId:e,finding:t}){return(0,S.jsxs)(`div`,{className:`card`,"data-testid":`diagnosis-finding`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:t.claim}),(0,S.jsxs)(`span`,{className:`head-actions`,children:[(0,S.jsxs)(D,{tone:tt(t.confidence),children:[t.confidence,` confidence`]}),(0,S.jsx)(qe,{testId:`copy-finding`,text:()=>$e(t,rt(e))})]})]}),(0,S.jsxs)(`div`,{className:`body`,children:[(0,S.jsxs)(`div`,{className:`attribution-reason`,children:[(0,S.jsx)(`code`,{children:t.eval_id}),` × `,(0,S.jsx)(`code`,{children:t.experiment_id})]}),(0,S.jsx)(`p`,{children:t.suggested_change}),(0,S.jsx)(`div`,{className:`evidence-row`,children:t.evidence.map((t,n)=>(0,S.jsx)(it,{runId:e,evidence:t},`${t.attempt_id}-${t.artifact}-${n}`))})]})]})}var ot=(e,t)=>`${e} ${t}${e===1?``:`s`}`;function st(e){let t=ot(e.findings.length,`finding`);return e.failingChecks.length===0?`${e.evalId} · ${t}`:`${e.evalId} · ${e.failingChecks.join(`, `)} failed under ${ot(e.experiments.length,`arm`)} · ${t}`}function ct({runId:e,diagnosis:t,pairs:n}){let[r,i]=(0,l.useState)(null);if(t.schema_version!==1)return(0,S.jsxs)(`div`,{className:`error-banner`,"data-testid":`diagnosis-schema-banner`,role:`alert`,children:[(0,S.jsx)(`strong`,{children:`Diagnosis schema mismatch.`}),` This diagnosis.json is schema_version`,` `,String(t.schema_version),` but this UI build expects `,1,` — regenerate it with the matching quickstudy version (`,(0,S.jsxs)(`code`,{children:[`quickstudy diagnose `,e]}),`).`]});let a=new Map;for(let e of t.findings){let t=a.get(e.target);t?t.push(e):a.set(e.target,[e])}let o=Qe(t.findings),s=Ze(t.findings,n);return(0,S.jsxs)(S.Fragment,{children:[(0,S.jsx)(`div`,{className:`notice`,"data-testid":`diagnosis-disclaimer`,children:t.disclaimer}),(0,S.jsxs)(`div`,{className:`kpi-row`,"data-testid":`diagnosis-provenance`,children:[(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Model`}),(0,S.jsx)(`div`,{className:`kpi-val`,style:{fontSize:15},children:t.model.requested}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`resolved: `,t.model.resolved??`unavailable`]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Pairs diagnosed`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:t.pairs_diagnosed}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[t.pairs_skipped,` skipped`]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Findings`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:t.findings.length}),(0,S.jsx)(`div`,{className:`kpi-meta`,children:t.errors.length>0?`${t.errors.length} pair(s) not analyzed`:`all selected pairs analyzed`})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Provenance`}),(0,S.jsx)(`div`,{className:`kpi-val`,style:{fontSize:15},children:t.generated_at.slice(0,10)}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`run `,(0,S.jsxs)(`code`,{children:[t.run_id.slice(0,10),`…`]}),` · prompt v`,t.prompt_version]})]})]}),t.findings.length===0&&(0,S.jsxs)(`div`,{className:`empty-state`,"data-testid":`diagnosis-no-findings`,children:[(0,S.jsx)(`h2`,{children:`Diagnosis ran; no findings.`}),(0,S.jsx)(`p`,{children:`The analyst drafted no hypotheses for this run's diagnosable pairs.`})]}),t.findings.length>0&&(0,S.jsx)(`div`,{className:`toolbar`,style:{justifyContent:`flex-start`},children:(0,S.jsxs)(`div`,{className:`seg`,role:`group`,"aria-label":`Finding target`,"data-testid":`diagnosis-target-filter`,children:[(0,S.jsxs)(`button`,{type:`button`,className:r===null?`active`:``,"aria-pressed":r===null,onClick:()=>i(null),children:[`all (`,t.findings.length,`)`]}),o.filter(e=>a.has(e)).map(e=>(0,S.jsxs)(`button`,{type:`button`,className:r===e?`active`:``,"aria-pressed":r===e,onClick:()=>i(e),children:[e,` (`,a.get(e)?.length,`)`]},e))]})}),s.map(t=>{let n=o.filter(e=>(r===null||r===e)&&t.findings.some(t=>t.target===e));return n.length===0?null:(0,S.jsxs)(`section`,{className:`cluster`,"data-testid":`diagnosis-cluster`,children:[(0,S.jsx)(`h2`,{children:st(t)}),n.map(n=>(0,S.jsxs)(`section`,{"data-testid":`diagnosis-target-${n}`,children:[(0,S.jsxs)(`h3`,{children:[(0,S.jsx)(`code`,{children:n}),` findings`]}),t.findings.filter(e=>e.target===n).map((t,n)=>(0,S.jsx)(at,{runId:e,finding:t},`${t.eval_id}|${t.experiment_id}|${n}`))]},n))]},`${t.evalId}|${t.failingChecks.join(`,`)}`)}),t.errors.length>0&&(0,S.jsxs)(`div`,{className:`card`,"data-testid":`diagnosis-errors`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:`Pairs that could not be analyzed`}),(0,S.jsx)(`span`,{className:`kpi-meta`,children:`model failures are recorded, never silently dropped`})]}),(0,S.jsx)(`div`,{className:`body`,children:(0,S.jsxs)(`table`,{className:`data`,children:[(0,S.jsx)(`thead`,{children:(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`eval`}),(0,S.jsx)(`th`,{children:`experiment`}),(0,S.jsx)(`th`,{children:`error`})]})}),(0,S.jsx)(`tbody`,{children:t.errors.map(e=>(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`td`,{children:(0,S.jsx)(`code`,{children:e.eval_id})}),(0,S.jsx)(`td`,{children:(0,S.jsx)(`code`,{children:e.experiment_id})}),(0,S.jsx)(`td`,{children:e.error})]},`${e.eval_id}|${e.experiment_id}`))})]})})]})]})}function lt({meta:e,runId:t}){let n=C(()=>y(t),[t]),r=C(()=>v(t).then(e=>e.pairs??null,()=>null),[t]);if(n.status===`error`)return(0,S.jsx)(ee,{error:n.error});if(n.status===`loading`||r.status===`loading`)return(0,S.jsx)(`div`,{className:`loading`,children:`Loading diagnosis…`});let i=r.status===`ready`?r.value:null,a=n.value;return(0,S.jsxs)(`div`,{"data-testid":`diagnosis-view`,children:[(0,S.jsxs)(`div`,{className:`page-head`,children:[(0,S.jsxs)(`div`,{children:[(0,S.jsx)(`h1`,{children:`Run diagnosis`}),(0,S.jsx)(`p`,{className:`sub`,children:`Model-drafted hypotheses with harness-selected evidence — verify cited lines before acting.`})]}),(0,S.jsxs)(`div`,{className:`control`,children:[(0,S.jsx)(`label`,{htmlFor:`diagnosis-run`,children:`Run`}),(0,S.jsx)(`select`,{id:`diagnosis-run`,"data-testid":`diagnosis-run-select`,value:t,onChange:e=>se(T.diagnosis(e.currentTarget.value)),children:e.runs.map(e=>(0,S.jsxs)(`option`,{value:e.id,children:[e.id.slice(0,10),`… · `,e.started_at.slice(0,10)]},e.id))})]})]}),a!==null&&a.schema_version===1&&(0,S.jsx)(`div`,{className:`toolbar`,"data-testid":`diagnosis-toolbar`,children:(0,S.jsx)(qe,{testId:`copy-diagnosis`,label:`Copy all findings as Markdown`,text:()=>et(a,rt(t))})}),a===null?(0,S.jsxs)(`div`,{className:`empty-state`,"data-testid":`diagnosis-empty`,children:[(0,S.jsx)(`h2`,{children:`No diagnosis for this run`}),(0,S.jsx)(`p`,{children:`Generate one (requires a model API key — never done at export time), then re-export:`}),(0,S.jsx)(`pre`,{children:(0,S.jsxs)(`code`,{children:[`quickstudy diagnose `,t]})})]}):(0,S.jsx)(ct,{runId:t,diagnosis:a,pairs:i})]})}function ut({coverage:e}){return e?(0,S.jsxs)(`details`,{"data-testid":`publication-omissions`,open:e.omitted.length>0,children:[(0,S.jsxs)(`summary`,{children:[`Complete-sample publication: `,e.eligible.length,` pairs included, `,e.omitted.length,` omitted;`,` `,e.expected_trials,` trials required.`]}),(0,S.jsx)(`p`,{children:`These counts cover the included pairs. Source attempts and retry history remain available in the diagnostic artifacts.`}),e.omitted.map(e=>(0,S.jsxs)(`p`,{children:[e.eval_id,` × `,e.experiment_id,`: `,e.reasons.join(`; `)]},`${e.eval_id}:${e.experiment_id}`))]}):null}function dt(e,t){let n=e=>e??-1;return n(t.aggregate?.pass_rate_discriminating)-n(e.aggregate?.pass_rate_discriminating)||n(t.aggregate?.pass_rate_all)-n(e.aggregate?.pass_rate_all)||e.experiment_id.localeCompare(t.experiment_id)}function ft(e){if(e===null)return`—`;let t=e=>e.toFixed(1).replace(/\.0$/,``);return e>=1e6?`${t(e/1e6)}M`:e>=1e3?`${t(e/1e3)}k`:String(Math.round(e))}function pt(e){if(e===null)return`—`;let t=Math.round(e/1e3);return t<60?`${t}s`:`${Math.floor(t/60)}m ${t%60}s`}function mt(e){return e===null?`—`:e>0&&e<.005?`<$0.01`:`$${e.toFixed(2)}`}function ht(e){let t=e.timedOut??0,n=e.incomplete-t,r=(e,t)=>`${e} ${t}${e===1?``:`s`}`,i=[e.unfinished?`the run never finished`:``,t>0?`${r(t,`attempt`)} timed out`:``,n>0?`${r(n,`attempt`)} incomplete`:``,e.errors>0?`${r(e.errors,`attempt`)} errored`:``].filter(Boolean);return i.length>0?i.join(`; `):`missing trial coverage`}function gt({experiment:e}){let t=e.aggregate;if(t===void 0||t.pass_rate_all===null)return`—`;let n=`${t.passed}/${t.passed+t.failed}`;return t.discriminating_evals===0?(0,S.jsxs)(S.Fragment,{children:[n,` · `,(0,S.jsx)(`span`,{className:`muted`,title:`No discriminating evals scored for this experiment`,children:`—`})]}):`${n} · ${t.discriminating_passed}/${t.discriminating_scored}`}function _t({experiment:e}){let t=e.aggregate;return t===void 0||t.attempts===0?`—`:`${t.cost_attempts>0&&t.cost_usd!==null?`${mt(t.cost_usd/t.cost_attempts)} · ${t.cost_attempts}/${t.attempts} reported`:`not reported (0/${t.attempts})`}${typeof t.estimated_cost_usd==`number`&&t.estimated_attempts?` · ${mt(t.estimated_cost_usd/t.estimated_attempts)} est.`:``}`}function vt({report:e,runId:t}){let n=Array.isArray(e.experiments)?e.experiments:[];if(n.length===0||n.some(e=>e.aggregate===void 0))return(0,S.jsxs)(`div`,{className:`notice`,"data-testid":`leaderboard-stale`,children:[`This report predates the leaderboard. Regenerate it with `,(0,S.jsxs)(`code`,{children:[`quickstudy report `,t]}),`.`]});let r=[...n].sort(dt);return(0,S.jsxs)(`div`,{className:`card`,"data-testid":`leaderboard`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:`Which experiment is best?`}),(0,S.jsx)(`span`,{className:`muted`,children:`sorted by pass rate on discriminating evals`})]}),(0,S.jsx)(`div`,{className:`body`,children:(0,S.jsxs)(`table`,{className:`data`,children:[(0,S.jsx)(`thead`,{children:(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`Experiment`}),(0,S.jsx)(`th`,{children:`Pass rate (all · discriminating)`}),(0,S.jsx)(`th`,{children:`Agent time (median)`}),(0,S.jsx)(`th`,{children:`Tokens/run (in)`}),(0,S.jsx)(`th`,{children:`Cost/run`}),(0,S.jsx)(`th`,{children:`SDK recall`})]})}),(0,S.jsx)(`tbody`,{children:r.map(e=>{let t=e.aggregate,n=e.sdk_recall;return(0,S.jsxs)(`tr`,{"data-testid":`leaderboard-row-${e.experiment_id}`,children:[(0,S.jsxs)(`td`,{children:[Me(e.agent),(0,S.jsxs)(`span`,{className:`group-sub`,children:[(0,S.jsx)(`code`,{children:e.experiment_id}),e.treatment===null?``:` · ${e.treatment}`]})]}),(0,S.jsx)(`td`,{"data-testid":`leaderboard-pass`,children:(0,S.jsx)(gt,{experiment:e})}),(0,S.jsx)(`td`,{title:t?`median over ${t.agent_time_attempts} of ${t.completed} scored attempt(s)`:void 0,children:pt(t?.median_agent_ms??null)}),(0,S.jsx)(`td`,{title:t?`mean over ${t.token_attempts} attempt(s) that reported tokens`:void 0,children:ft(t?.mean_tokens_in??null)}),(0,S.jsx)(`td`,{"data-testid":`leaderboard-cost`,children:(0,S.jsx)(_t,{experiment:e})}),(0,S.jsx)(`td`,{children:n&&n.rate!==null?`${n.recalled}/${n.attempts}`:`n/a`})]},e.experiment_id)})})]})})]})}function yt(e){return e===null?{bg:`var(--gray-bg)`,fg:`var(--muted)`}:e>=.95?{bg:`#1a7f37`,fg:`#ffffff`}:e>=.9?{bg:`#7ecb8f`,fg:`#0f3d1e`}:e>=.8?{bg:`#ffdf80`,fg:`#5c3d00`}:e>=.5?{bg:`#ffb066`,fg:`#5c2800`}:e>0?{bg:`#f28b82`,fg:`#520f08`}:{bg:`#d93025`,fg:`#ffffff`}}function bt(e){return e===null?`—`:`${Math.round(e*100)}%`}function xt(){return(0,S.jsxs)(`div`,{className:`legend`,"aria-label":`Pass-rate color scale (fixed 0-100%)`,children:[(0,S.jsx)(`span`,{children:`0%`}),[0,.5,.8,.9,.95].map(e=>{let t=yt(e);return(0,S.jsx)(`span`,{className:`legend-sw`,style:{background:t.bg},title:`≥ ${Math.round(e*100)}%`},e)}),(0,S.jsx)(`span`,{children:`100%`}),(0,S.jsx)(`span`,{style:{marginLeft:8},children:`ticks: 80 / 90 / 95`})]})}function St(e){let t=[];for(let n of e){let e=n.group===null?`agent:${n.agentLabel??``}`:`group:${n.group}`,r=t.at(-1);r&&r.key===e?r.columns.push(n):t.push({key:e,group:n.group,columns:[n]})}return t.map(({group:e,columns:t})=>{let n=new Set(t.map(e=>e.agentLabel)),r=n.size===1?[...n][0]??null:null;return{label:r??e??``,sub:r!==null&&e!==null?e:null,title:r===null?null:t[0]?.agentTitle??null,span:t.length}})}function Ct({cells:e,rows:t,columns:n,onCellClick:r,band:i}){let a=new Map(e.map(e=>[`${e.evalId}|${e.experimentId}`,e])),o=n.some(e=>e.group!==null||e.agentLabel!==null);return(0,S.jsxs)(`div`,{className:`matrix-region`,children:[(0,S.jsx)(`div`,{className:`scroll-hint`,"aria-hidden":`true`,children:`Scroll horizontally to see every experiment →`}),(0,S.jsx)(`div`,{className:`matrix-scroll`,role:`region`,"aria-label":i===void 0?`Pass-rate matrix`:`Pass-rate matrix — ${i}`,tabIndex:0,children:(0,S.jsxs)(`table`,{className:`matrix`,children:[(0,S.jsxs)(`thead`,{children:[o&&(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{"aria-hidden":`true`}),St(n).map((e,t)=>(0,S.jsxs)(`th`,{colSpan:e.span,className:`matrix-group-head`,title:e.title??void 0,children:[e.label,e.sub!==null&&(0,S.jsx)(`span`,{className:`matrix-group-sub`,children:e.sub})]},`${e.label}-${t}`))]}),(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`eval`}),n.map(e=>(0,S.jsxs)(`th`,{children:[(0,S.jsx)(`code`,{children:e.experimentId}),e.treatment!==null&&(0,S.jsx)(`span`,{className:`matrix-row-feature`,children:e.treatment})]},e.experimentId))]})]}),(0,S.jsx)(`tbody`,{children:t.map(e=>(0,S.jsxs)(`tr`,{children:[(0,S.jsxs)(`th`,{children:[(0,S.jsx)(`code`,{children:e.evalId}),e.framework!==void 0&&(0,S.jsx)(`span`,{className:`matrix-row-feature`,children:e.framework})]}),n.map(t=>{let n=a.get(`${e.evalId}|${t.experimentId}`);if(!n)return(0,S.jsx)(`td`,{className:`cell`,style:{background:`var(--gray-bg)`,color:`var(--muted)`},children:(0,S.jsx)(`div`,{className:`cell-rate`,children:`—`})},t.experimentId);let i=yt(n.rate),o=`${n.evalId}, ${n.experimentId}: ${bt(n.rate)}, ${n.passed} of ${n.scored} scored passed${n.checks?`; ${n.checks.passed} of ${n.checks.total} checks passed`:``}${n.errors>0?`; ${n.errors} errored`:``}${n.flaky?`; flaky trials`:``}${n.incomplete?`; incomplete`:``}`;return(0,S.jsx)(`td`,{className:`cell`,style:{background:i.bg,color:i.fg},children:(0,S.jsxs)(`button`,{type:`button`,className:`cell-button`,"aria-label":o,"data-testid":`cell-${n.evalId}-${n.experimentId}`,"data-rate":n.rate===null?``:String(Math.round(n.rate*100)),"data-flaky":n.flaky?`true`:`false`,"data-incomplete":n.incomplete?`true`:`false`,onClick:()=>r(n),children:[n.flaky&&(0,S.jsx)(`span`,{className:`flaky-marker`,title:`flaky: trials disagreed`,"aria-hidden":`true`}),(0,S.jsx)(`span`,{className:`cell-rate`,children:bt(n.rate)}),(0,S.jsxs)(`span`,{className:`cell-sub`,children:[n.passed,`/`,n.scored,n.checks?` · ${n.checks.passed}/${n.checks.total} checks`:``,n.errors>0?` · ${n.errors} err`:``]}),n.incomplete&&(0,S.jsx)(`span`,{className:`incomplete-marker`,title:`incomplete: fewer attempts than planned`,"aria-hidden":`true`,children:`◌`})]})},t.experimentId)})]},e.evalId))})]})})]})}var wt=`all`,Tt=[`benchmark`,`regression`,`other`];function Et(e,t){return e.manifest.evals[t]?.metadata??{}}function Dt(e,t){for(let n of e.groups)if(n.experimentIds.includes(t))return n.group;return null}function Ot(e,t){return[...new Set(t.map(e=>e.experiment_id))].map(t=>({experimentId:t,treatment:e.manifest.experiments[t]?.components.treatment??null,group:Dt(e,t),agentLabel:Ne(e,t),agentTitle:Pe(e,t)})).sort((e,t)=>(e.group??`￿`).localeCompare(t.group??`￿`)||(e.treatment??``).localeCompare(t.treatment??``)||e.experimentId.localeCompare(t.experimentId))}function kt(e,t){let n=new Map;for(let t of e){let e=`${t.eval_id}|${t.experiment_id}`,r=n.get(e);r?r.push(t):n.set(e,[t])}let r=[];for(let e of n.values()){let n=e[0],i=e.filter(e=>e.status===`completed`),a=i.filter(e=>e.passed===!0).length,o=i.length,s=new Set(i.map(e=>e.passed===!0)),c=le(i.flatMap(e=>e.checks??[]));r.push({evalId:n.eval_id,experimentId:n.experiment_id,rate:o>0?a/o:null,passed:a,scored:o,flaky:s.size>1,incomplete:e.length<t||e.some(e=>e.status===`incomplete`),errors:e.filter(e=>e.status===`error`).length,checks:c.total>0?c:null})}return r}function At(e){if(e.started_at===null||e.finished_at===null)return`—`;let t=Math.round((Date.parse(e.finished_at)-Date.parse(e.started_at))/1e3);return`${Math.floor(t/60)}m${String(t%60).padStart(2,`0`)}s`}var jt={codex:{label:`thread turns`,title:`codex exec reports one thread turn per prompt, not per agent step — not comparable to Claude's turn count`}};function Mt(e){return e.status===`incomplete`&&e.error?.startsWith(`attempt timed out`)===!0}function Nt({attempt:e,runId:t,adapter:n}){let r=pe(e),i=n!==null&&jt[n]||{label:`turns`,title:`agent turns, as counted by the adapter`};return(0,S.jsxs)(`div`,{className:`attempt-row`,"data-testid":`drill-attempt`,children:[(0,S.jsxs)(`div`,{className:`attempt-line`,children:[(0,S.jsxs)(`span`,{className:`kpi-meta`,children:[`trial `,e.trial_index+1]}),(0,S.jsx)(D,{tone:me(r),children:r}),(0,S.jsx)(ge,{attempt:e}),(0,S.jsx)(`span`,{className:`spacer`}),e.transcript!==null&&(0,S.jsx)(`a`,{href:T.attempt(t,e.id,`transcript`),children:`transcript`}),e.diff!==null&&(0,S.jsx)(`a`,{href:T.attempt(t,e.id,`diff`),children:`diff`})]}),e.checks!==null&&e.checks.length>0&&(0,S.jsx)(`div`,{className:`attempt-line`,children:e.checks.map(e=>(0,S.jsxs)(`span`,{className:e.passed?`crit pass`:`crit fail`,title:e.notes??e.name,children:[e.passed?`✓`:`✗`,` `,e.name,!E(e)&&(0,S.jsx)(`span`,{className:`tag`,"data-testid":`ungraded-tag`,title:`Recorded, not graded: never enters pass/fail`,children:`recorded`})]},e.name))}),e.error!==null&&(0,S.jsx)(`div`,{className:`attempt-line`,children:(0,S.jsx)(`code`,{children:e.error})}),(0,S.jsxs)(`div`,{className:`telemetry`,children:[(0,S.jsxs)(`span`,{children:[`wall `,At(e)]}),(0,S.jsxs)(`span`,{children:[`tokens `,ft(e.tokens_in),` in / `,ft(e.tokens_out),` out`]}),(0,S.jsxs)(`span`,{children:[`cost `,e.cost_usd===null?`—`:`$${e.cost_usd.toFixed(2)}`]}),(0,S.jsxs)(`span`,{title:i.title,children:[i.label,` `,e.turns??`—`]})]}),(0,S.jsx)(Se,{treatment:e.treatment,usage:e.surface_usage})]})}function Pt({runId:e,run:t,cell:n,attempts:r}){let i=(0,l.useRef)(null),a=(0,l.useRef)(!1),o=(0,l.useId)(),s=r.filter(e=>e.eval_id===n.evalId&&e.experiment_id===n.experimentId).sort((e,t)=>e.trial_index-t.trial_index),c=t.manifest.experiments[n.experimentId]?.components.treatment??null,u=s.filter(e=>e.status===`completed`),d=u.filter(e=>e.passed===!0).length,f=`cell-${n.evalId}-${n.experimentId}`,p=(0,l.useCallback)(()=>{a.current=!0,se(T.matrix(e))},[e]);return(0,l.useEffect)(()=>{let e=i.current;if(!e)return;let t=document.body.style.overflow;document.body.style.overflow=`hidden`;let n=()=>[...e.querySelectorAll(`a[href], button:not([disabled]), input:not([disabled]), select:not([disabled]), textarea:not([disabled]), [tabindex]:not([tabindex="-1"])`)].filter(e=>!e.hasAttribute(`hidden`));n()[0]?.focus();let r=t=>{if(t.key===`Escape`){t.preventDefault(),p();return}if(t.key!==`Tab`)return;let r=n();if(r.length===0){t.preventDefault(),e.focus();return}let i=r[0],a=r.at(-1);t.shiftKey&&document.activeElement===i?(t.preventDefault(),a.focus()):!t.shiftKey&&document.activeElement===a?(t.preventDefault(),i.focus()):e.contains(document.activeElement)||(t.preventDefault(),i.focus())};return document.addEventListener(`keydown`,r),()=>{if(document.removeEventListener(`keydown`,r),document.body.style.overflow=t,!a.current)return;let e=()=>{let e=[...document.querySelectorAll(`[data-testid]`)].find(e=>e.dataset.testid===f);return e?.focus(),e!==void 0};requestAnimationFrame(()=>{e()||window.setTimeout(e,0)})}},[p,f]),(0,S.jsxs)(S.Fragment,{children:[(0,S.jsx)(`div`,{className:`drill-overlay`,"aria-hidden":`true`,onClick:p,onKeyDown:e=>e.preventDefault()}),(0,S.jsxs)(`aside`,{ref:i,className:`drill`,"data-testid":`drill-down`,role:`dialog`,"aria-modal":`true`,"aria-labelledby":o,tabIndex:-1,children:[(0,S.jsx)(`button`,{type:`button`,className:`drill-close`,onClick:p,"data-testid":`drill-close`,children:`Close ✕`}),(0,S.jsxs)(`h2`,{id:o,children:[n.evalId,` · `,n.experimentId,c===null?``:` · ${c}`]}),(0,S.jsxs)(`p`,{className:`sub`,children:[s.length,` attempt(s) · `,d,`/`,u.length,` scored passed · `,bt(u.length>0?d/u.length:null)]}),s.length===0&&(0,S.jsx)(`div`,{className:`viewer-note`,children:`No attempts in this cell.`}),s.map(n=>(0,S.jsx)(Nt,{attempt:n,runId:e,adapter:t.manifest.experiments[n.experiment_id]?.components.agent?.adapter??null},n.id))]})]})}function Ft({id:e,label:t,values:n,value:r,onChange:i}){return n.length===0?null:(0,S.jsxs)(`div`,{className:`control`,children:[(0,S.jsx)(`label`,{htmlFor:e,children:t}),(0,S.jsxs)(`select`,{id:e,"data-testid":e,value:r,onChange:e=>i(e.currentTarget.value),children:[(0,S.jsx)(`option`,{value:wt,children:`all`}),n.map(e=>(0,S.jsx)(`option`,{value:e,children:e},e))]})]})}var It=e=>e===1?``:`s`;function Lt({meta:e,runId:t,cell:n}){let r=C(()=>_(t),[t]),i=C(()=>v(t).then(e=>e,()=>null),[t]),[a,o]=(0,l.useState)(`benchmark`),s=r.status===`ready`?Tt.filter(e=>r.value.attempts.some(t=>t.suite===e)):[],c=s.includes(a)?a:s[0]??`benchmark`,[u,d]=(0,l.useState)({product:wt,framework:wt,topic:wt});(0,l.useEffect)(()=>{d({product:wt,framework:wt,topic:wt})},[t]);let f=(0,l.useMemo)(()=>{if(r.status!==`ready`)return null;let e=r.value,t=e.attempts.filter(e=>e.suite===c).filter(t=>{let n=Et(e,t.eval_id);return!(u.product!==wt&&n.product!==u.product||u.framework!==wt&&n.framework!==u.framework||u.topic!==wt&&!(n.topics??[]).includes(u.topic))});return{rows:[...new Set(t.map(e=>e.eval_id))].sort((e,t)=>e.localeCompare(t)).map(t=>({evalId:t,...Et(e,t).framework===void 0?{}:{framework:Et(e,t).framework}})),columns:Ot(e,t),cells:kt(t,e.run.trials),attempts:t}},[r,c,u]);if(r.status===`error`)return(0,S.jsx)(ee,{error:r.error});if(r.status===`loading`||i.status===`loading`||f===null)return(0,S.jsx)(`div`,{className:`loading`,children:`Loading run…`});let p=r.value,m=p.run,h=Tt.filter(e=>p.attempts.some(t=>t.suite===e)),g=[...new Set(p.attempts.map(e=>Et(p,e.eval_id).product).filter(e=>e!==void 0))].sort(),y=[...new Set(p.attempts.map(e=>Et(p,e.eval_id).framework).filter(e=>e!==void 0))].sort(),b=[...new Set(p.attempts.flatMap(e=>Et(p,e.eval_id).topics??[]))].sort(),x={attempts:p.attempts.length,passed:p.attempts.filter(e=>e.status===`completed`&&e.passed===!0).length,failed:p.attempts.filter(e=>e.status===`completed`&&e.passed!==!0).length,incomplete:p.attempts.filter(e=>e.status===`incomplete`).length,timedOut:p.attempts.filter(e=>e.status===`completed`&&e.limit===`timeout`||Mt(e)).length,unscoredTimedOut:p.attempts.filter(Mt).length,errors:p.attempts.filter(e=>e.status===`error`).length},te=x.passed+x.failed,ne=i.status===`ready`?i.value:null,w=Array.isArray(ne?.pairs)?ne.pairs:[],re=w.some(e=>e.band!==void 0),ie=new Map(w.map(e=>[e.eval_id,e.band??null])),ae=e=>f.rows.filter(t=>(ie.get(t.evalId)??null)===e),oe=f.rows.filter(e=>{let t=ie.get(e.evalId)??null;return t===null||t===`discriminating`}),ce=ae(null).length,E=ae(`solved-by-all`),le=ae(`failed-by-all`),ue=ne?.scorer_health?.suspects??[],de=e=>se(T.cell(t,{evalId:e.evalId,experimentId:e.experimentId})),fe=e=>{o(e),n&&se(T.matrix(t))},pe=e=>r=>{d(t=>({...t,[e]:r})),n&&se(T.matrix(t))};return(0,S.jsxs)(`div`,{"data-testid":`matrix-view`,children:[(0,S.jsx)(ut,{coverage:p.publication}),(0,S.jsxs)(`div`,{className:`page-head`,children:[(0,S.jsxs)(`div`,{children:[(0,S.jsx)(`h1`,{children:`Eval × experiment matrix`}),(0,S.jsx)(`p`,{className:`sub`,children:`Pass rate over scored attempts per cell; experiment columns are the treatment arms, grouped by comparison group. Select a cell for attempt evidence.`})]}),(0,S.jsxs)(`div`,{className:`controls`,children:[(0,S.jsxs)(`div`,{className:`control`,children:[(0,S.jsx)(`label`,{htmlFor:`run-select`,children:`Run`}),(0,S.jsx)(`select`,{id:`run-select`,"data-testid":`run-select`,value:t,onChange:e=>se(T.matrix(e.currentTarget.value)),children:e.runs.map(e=>(0,S.jsxs)(`option`,{value:e.id,children:[e.id.slice(0,10),`… · `,e.started_at.slice(0,10),` · `,e.attempts,` attempts`]},e.id))})]}),(0,S.jsx)(Ft,{id:`filter-product`,label:`Product`,values:g,value:u.product,onChange:pe(`product`)}),(0,S.jsx)(Ft,{id:`filter-framework`,label:`Framework`,values:y,value:u.framework,onChange:pe(`framework`)}),(0,S.jsx)(Ft,{id:`filter-topic`,label:`Topic`,values:b,value:u.topic,onChange:pe(`topic`)}),h.length>1&&(0,S.jsxs)(`div`,{className:`control`,children:[(0,S.jsx)(`label`,{children:`Suite`}),(0,S.jsx)(`div`,{className:`seg`,role:`group`,"aria-label":`Suite`,children:h.map(e=>(0,S.jsx)(`button`,{type:`button`,className:c===e?`active`:``,"aria-pressed":c===e,"data-testid":`suite-${e}`,onClick:()=>fe(e),children:e},e))})]})]})]}),(m.finished_at===null||m.incomplete)&&(0,S.jsxs)(`div`,{className:`notice`,"data-testid":`incomplete-banner`,children:[`Incomplete run — `,ht({unfinished:m.finished_at===null,incomplete:x.incomplete,errors:x.errors,timedOut:x.unscoredTimedOut}),`. Numbers are partial.`]}),(0,S.jsxs)(`div`,{className:`kpi-row`,children:[(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsxs)(`div`,{className:`kpi-label`,children:[`Overall pass rate`,h.length>1?` (all suites)`:``]}),(0,S.jsx)(`div`,{className:`kpi-val`,children:te===0?`—`:`${Math.round(x.passed/te*100)}%`}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[(0,S.jsxs)(D,{tone:`green`,children:[x.passed,` passed`]}),` `,(0,S.jsxs)(D,{tone:`red`,children:[x.failed,` failed`]}),` `,(0,S.jsxs)(D,{tone:`gray`,children:[x.incomplete,` incomplete`]}),` `,(0,S.jsxs)(D,{tone:`amber`,children:[x.errors,` errored`]}),x.timedOut>0&&(0,S.jsxs)(S.Fragment,{children:[` `,(0,S.jsxs)(D,{tone:`gray`,children:[x.timedOut,` timed out`]})]})]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Evals × experiments`}),(0,S.jsxs)(`div`,{className:`kpi-val`,children:[m.eval_ids.length,` × `,m.experiment_ids.length]}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[m.attempts,` attempts`]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Trials per pair`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:m.trials}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`manifest `,m.manifest_hash.slice(0,10),`…`]})]})]}),ne!==null&&(0,S.jsx)(vt,{report:ne,runId:t}),(0,S.jsxs)(`div`,{className:`card`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsxs)(`h3`,{children:[`Pass rate per eval × experiment`,c===`regression`&&(0,S.jsxs)(S.Fragment,{children:[` `,(0,S.jsx)(D,{tone:`amber`,title:`Known failure modes — tracked separately from benchmark results.`,children:`regression suite`})]})]}),(0,S.jsx)(xt,{})]}),ue.length>0&&ne?.scorer_health&&(0,S.jsxs)(`div`,{className:`notice band-notice`,"data-testid":`scorer-health-banner`,children:[ne.scorer_health.summary,(0,S.jsx)(`ul`,{children:ue.map(e=>(0,S.jsxs)(`li`,{children:[(0,S.jsx)(`code`,{children:e.eval_id}),` · `,(0,S.jsx)(`code`,{children:e.check}),` failed in `,e.attempts,` attempts under`,` `,e.experiments.length,` arms · passing: `,e.passing_checks.join(`, `)]},`${e.eval_id}|${e.check}`))})]}),re?(0,S.jsxs)(S.Fragment,{children:[oe.length>0&&(0,S.jsxs)(`section`,{className:`band`,"data-testid":`matrix-band-discriminating`,children:[(0,S.jsxs)(`h4`,{children:[`Discriminating (`,oe.length,` eval`,It(oe.length),`) — arms disagree`,ce>0&&` (${ce} eval${It(ce)} with one scored arm)`]}),(0,S.jsx)(Ct,{cells:f.cells,rows:oe,columns:f.columns,onCellClick:de,band:`discriminating`})]}),E.length>0&&(0,S.jsxs)(`details`,{className:`band`,"data-testid":`matrix-band-solved`,children:[(0,S.jsxs)(`summary`,{children:[`Solved by every arm (`,E.length,` eval`,It(E.length),`)`]}),(0,S.jsx)(Ct,{cells:f.cells,rows:E,columns:f.columns,onCellClick:de,band:`solved by every arm`})]}),le.length>0&&(0,S.jsxs)(`details`,{className:`band`,"data-testid":`matrix-band-failed`,children:[(0,S.jsxs)(`summary`,{children:[`Failed by every arm (`,le.length,` eval`,It(le.length),`) — suspect the eval`]}),(0,S.jsx)(Ct,{cells:f.cells,rows:le,columns:f.columns,onCellClick:de,band:`failed by every arm`})]})]}):(0,S.jsx)(Ct,{cells:f.cells,rows:f.rows,columns:f.columns,onCellClick:de})]}),n&&(0,S.jsx)(Pt,{runId:t,run:p,cell:n,attempts:p.attempts})]})}var Rt=[2,3,5,10];function zt(e){return(e??[]).filter(e=>Rt.includes(e.k))}function Bt({estimate:e}){return e===void 0?null:(0,S.jsxs)(`div`,{className:`kpi`,"data-testid":`cost-estimate`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Estimated token cost`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:e.cost_usd===null?`—`:e.cost_usd>0&&e.cost_usd<.01?`<$0.01`:`$${e.cost_usd.toFixed(2)}`}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`In addition to reported cost · `,e.cost_attempts,` estimated, `,e.unpriced_attempts,` unpriced attempt(s)`]}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`Pricing: `,e.pricing.version,`. Excludes tool fees, diagnosis, infrastructure and superseded retries.`]})]})}function Vt(e){return e===null?`—`:`${Math.round(e*100)}%`}function Ht({pair:e}){let t=zt(e.pass_at_k);return t.length===0?(0,S.jsx)(S.Fragment,{children:`—`}):(0,S.jsx)(`div`,{children:t.map(e=>(0,S.jsxs)(`div`,{children:[`k=`,e.k,`: `,Vt(e.pass_at_k),` / `,Vt(e.pass_to_the_k)]},e.k))})}function Ut({pair:e}){return e.checks.length===0?(0,S.jsx)(`span`,{className:`attribution-reason`,children:`no named checks recorded`}):(0,S.jsx)(`div`,{className:`attribution-effects`,children:e.checks.map(e=>(0,S.jsxs)(`span`,{className:e.passed===e.total?`crit pass`:`crit fail`,title:e.name,children:[e.passed===e.total?`✓`:`✗`,` `,e.name,` (`,e.passed,`/`,e.total,`)`,!E(e)&&(0,S.jsx)(`span`,{className:`tag`,"data-testid":`ungraded-tag`,title:`Recorded, not graded: never enters pass/fail`,children:`recorded`})]},e.name))})}function Wt({group:e,labelOf:t}){let n=new Set(e.experiment_ids.map(t)),r=n.size===1?[...n][0]??null:null;return r===null?(0,S.jsx)(`h3`,{children:e.group}):(0,S.jsxs)(`h3`,{children:[r,` `,(0,S.jsx)(`span`,{className:`group-sub`,children:e.group})]})}function Gt({group:e,labelOf:t}){if(e.comparison===`withheld`)return(0,S.jsxs)(`div`,{className:`card`,"data-testid":`report-group-${e.group}`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(Wt,{group:e,labelOf:t}),(0,S.jsx)(D,{tone:`amber`,children:`comparison withheld`})]}),(0,S.jsxs)(`div`,{className:`body`,children:[(0,S.jsx)(`p`,{className:`viewer-note`,children:`Members differ beyond their declared treatment — the differing components:`}),(0,S.jsx)(`ul`,{children:e.details.map(e=>(0,S.jsx)(`li`,{children:(0,S.jsx)(`code`,{children:e})},e))})]})]});let n=[...new Set(e.arms.flatMap(e=>e.evals.map(e=>e.eval_id)))].sort();return(0,S.jsxs)(`div`,{className:`card`,"data-testid":`report-group-${e.group}`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(Wt,{group:e,labelOf:t}),(0,S.jsx)(D,{tone:`green`,children:`compared — raw counts`})]}),(0,S.jsx)(`div`,{className:`body`,children:(0,S.jsxs)(`table`,{className:`data`,children:[(0,S.jsx)(`thead`,{children:(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`eval`}),e.arms.map(e=>(0,S.jsxs)(`th`,{children:[(0,S.jsx)(`code`,{children:e.experiment_id}),e.treatment!==null&&(0,S.jsx)(`div`,{className:`attribution-reason`,children:e.treatment})]},e.experiment_id))]})}),(0,S.jsx)(`tbody`,{children:n.map(t=>(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`td`,{children:(0,S.jsx)(`code`,{children:t})}),e.arms.map(e=>{let n=e.evals.find(e=>e.eval_id===t);return(0,S.jsx)(`td`,{children:n===void 0?`—`:n.display},e.experiment_id)})]},t))})]})})]})}function Kt({meta:e,runId:t}){let n=C(()=>v(t),[t]);if(n.status===`error`)return(0,S.jsx)(ee,{error:n.error});if(n.status===`loading`)return(0,S.jsx)(`div`,{className:`loading`,children:`Loading report…`});let r=n.value,i=r.totals.passed+r.totals.failed,a=new Map((r.experiments??[]).map(e=>[e.experiment_id,je(e.agent.model,e.agent.reasoning)])),o=e=>a.get(e)??null,s=r.run.trials>1;return(0,S.jsxs)(`div`,{"data-testid":`report-view`,children:[(0,S.jsx)(ut,{coverage:n.value.publication}),(0,S.jsxs)(`div`,{className:`page-head`,children:[(0,S.jsxs)(`div`,{children:[(0,S.jsx)(`h1`,{children:`Run report`}),(0,S.jsxs)(`p`,{className:`sub`,children:[r.run.eval_ids.length,` evals × `,r.run.experiment_ids.length,` experiments · `,r.run.trials,` `,`trials/pair · started `,r.run.started_at.slice(0,10)]})]}),(0,S.jsxs)(`div`,{className:`control`,children:[(0,S.jsx)(`label`,{htmlFor:`report-run`,children:`Run`}),(0,S.jsx)(`select`,{id:`report-run`,"data-testid":`report-run-select`,value:t,onChange:e=>se(T.report(e.currentTarget.value)),children:e.runs.map(e=>(0,S.jsxs)(`option`,{value:e.id,children:[e.id.slice(0,10),`… · `,e.started_at.slice(0,10)]},e.id))})]})]}),r.incomplete&&(0,S.jsxs)(`div`,{className:`notice`,"data-testid":`incomplete-banner`,children:[`⚠️ INCOMPLETE RUN — `,ht({unfinished:r.run.finished_at===null,...r.totals}),`. Numbers below are partial and NOT comparable to a full sweep.`]}),(0,S.jsxs)(`div`,{className:`kpi-row`,children:[(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Attempts`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:r.totals.attempts}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[(0,S.jsxs)(D,{tone:`green`,children:[r.totals.passed,` passed`]}),` `,(0,S.jsxs)(D,{tone:`red`,children:[r.totals.failed,` failed`]}),` `,(0,S.jsxs)(D,{tone:`gray`,children:[r.totals.incomplete,` incomplete`]}),` `,(0,S.jsxs)(D,{tone:`amber`,children:[r.totals.errors,` errored`]})]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Overall pass rate`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:i===0?`—`:`${Math.round(r.totals.passed/i*100)}%`}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[i,` scored attempt(s)`]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Reported cost`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:r.telemetry.cost_usd===null?`—`:`$${r.telemetry.cost_usd.toFixed(2)}`}),(0,S.jsx)(`div`,{className:`kpi-meta`,children:r.telemetry.cost_usd===null?`unavailable`:`${r.telemetry.cost_attempts} of ${r.totals.attempts} attempts reported a cost`+(r.totals.attempts>r.telemetry.cost_attempts?` · ${r.totals.attempts-r.telemetry.cost_attempts} unpriced, so this understates the spend`:``)})]}),(0,S.jsx)(Bt,{estimate:r.cost_estimate}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Tokens`}),(0,S.jsx)(`div`,{className:`kpi-val`,style:{fontSize:15},children:r.telemetry.tokens_in===null&&r.telemetry.tokens_out===null?`unavailable`:`${ft(r.telemetry.tokens_in)} in / ${ft(r.telemetry.tokens_out)} out`}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`from `,r.telemetry.token_attempts,` of `,r.totals.attempts,` attempts`]})]})]}),(0,S.jsxs)(`div`,{className:`card`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:`Pass rates per eval × experiment`}),(0,S.jsx)(`span`,{className:`kpi-meta`,children:`scored = passed + failed; incomplete and errored attempts never enter the denominator`})]}),(0,S.jsx)(`div`,{className:`body`,children:(0,S.jsxs)(`table`,{className:`data`,"data-testid":`pairs-table`,children:[(0,S.jsx)(`thead`,{children:(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`eval`}),(0,S.jsx)(`th`,{children:`experiment`}),(0,S.jsx)(`th`,{children:`pass rate`}),(0,S.jsx)(`th`,{children:`raw`}),s&&(0,S.jsx)(`th`,{title:`pass@k: fraction of size-k subsets of the scored trials with ≥1 pass; pass^k: (passed/scored)^k`,children:`pass@k / pass^k`}),(0,S.jsx)(`th`,{children:`notes`}),(0,S.jsx)(`th`,{children:`checks`})]})}),(0,S.jsx)(`tbody`,{children:r.pairs.map(e=>(0,S.jsxs)(`tr`,{children:[(0,S.jsxs)(`td`,{children:[(0,S.jsx)(`code`,{children:e.eval_id}),e.suite!==`benchmark`&&(0,S.jsx)(`div`,{className:`attribution-reason`,children:e.suite})]}),(0,S.jsxs)(`td`,{children:[(0,S.jsx)(`code`,{children:e.experiment_id}),o(e.experiment_id)!==null&&(0,S.jsx)(`div`,{className:`attribution-reason`,children:o(e.experiment_id)})]}),(0,S.jsx)(`td`,{children:(0,S.jsx)(`strong`,{children:Vt(e.pass_rate)})}),(0,S.jsxs)(`td`,{children:[e.passed,`/`,e.scored]}),s&&(0,S.jsx)(`td`,{children:(0,S.jsx)(Ht,{pair:e})}),(0,S.jsx)(`td`,{children:[e.incomplete>0?`${e.incomplete} incomplete`:``,e.errors>0?`${e.errors} errored`:``,e.flaky?`trials disagreed`:``].filter(Boolean).join(`; `)||`—`}),(0,S.jsx)(`td`,{children:(0,S.jsx)(Ut,{pair:e})})]},`${e.eval_id}|${e.experiment_id}`))})]})})]}),r.groups.length>0&&(0,S.jsxs)(`section`,{"data-testid":`report-groups`,children:[(0,S.jsx)(`h2`,{children:`Comparison groups`}),r.groups.map(e=>(0,S.jsx)(Gt,{group:e,labelOf:o},e.group))]}),(r.experiments??[]).length>0&&(0,S.jsxs)(`div`,{className:`card`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:`SDK recall`}),(0,S.jsx)(`span`,{className:`kpi-meta`,children:`scored attempts whose ungraded checks all passed`})]}),(0,S.jsx)(`div`,{className:`body`,"data-testid":`sdk-recall`,children:(r.experiments??[]).map(e=>{let t=e.sdk_recall;return(0,S.jsx)(`div`,{className:`attribution-arm`,children:(0,S.jsxs)(`div`,{className:`attribution-arm-head`,children:[(0,S.jsx)(`strong`,{children:(0,S.jsx)(`code`,{children:e.experiment_id})}),e.treatment!==null&&(0,S.jsx)(D,{tone:`blue`,children:e.treatment}),(0,S.jsxs)(`span`,{children:[`SDK recall `,t&&t.rate!==null?`${t.recalled}/${t.attempts}`:`n/a`]})]})},e.experiment_id)})})]}),r.usage.length>0&&(0,S.jsxs)(`div`,{className:`card`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:`Surface usage (offered vs observed)`}),(0,S.jsx)(`span`,{className:`kpi-meta`,children:`null telemetry counts as unavailable, never as zero use`})]}),(0,S.jsx)(`div`,{className:`body`,"data-testid":`usage-summary`,children:r.usage.map(e=>(0,S.jsxs)(`div`,{className:`attribution-arm`,children:[(0,S.jsxs)(`div`,{className:`attribution-arm-head`,children:[(0,S.jsx)(`strong`,{children:(0,S.jsx)(`code`,{children:e.experiment_id})}),e.treatment!==null&&(0,S.jsxs)(D,{tone:`blue`,children:[`offered: `,e.treatment]})]}),(0,S.jsx)(O,{usage:e})]},e.experiment_id))})]}),r.egress_denials.length>0&&(0,S.jsxs)(`div`,{className:`card`,children:[(0,S.jsx)(`div`,{className:`head`,children:(0,S.jsx)(`h3`,{children:`Egress denials (proxied run)`})}),(0,S.jsx)(`div`,{className:`body`,children:(0,S.jsxs)(`table`,{className:`data`,children:[(0,S.jsx)(`thead`,{children:(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`attempt`}),(0,S.jsx)(`th`,{children:`eval`}),(0,S.jsx)(`th`,{children:`experiment`}),(0,S.jsx)(`th`,{children:`denials`}),(0,S.jsx)(`th`,{children:`top hosts`})]})}),(0,S.jsx)(`tbody`,{children:r.egress_denials.map(e=>(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`td`,{children:(0,S.jsxs)(`code`,{children:[e.attempt_id.slice(0,8),`…`]})}),(0,S.jsx)(`td`,{children:e.eval_id}),(0,S.jsx)(`td`,{children:e.experiment_id}),(0,S.jsx)(`td`,{children:e.count}),(0,S.jsx)(`td`,{children:e.top_hosts.map(({host:e,count:t})=>`${e} (${t})`).join(`, `)})]},e.attempt_id))})]})})]}),(0,S.jsxs)(`div`,{className:`methodology`,"data-testid":`methodology`,children:[(0,S.jsx)(`strong`,{children:`Methodology`}),`: `,r.methodology]})]})}function qt(){return(0,S.jsxs)(`div`,{className:`empty-state`,"data-testid":`empty-state`,children:[(0,S.jsx)(`h2`,{children:`No runs in this data directory yet`}),(0,S.jsx)(`p`,{children:`Execute a sweep, then re-export (or restart) the UI:`}),(0,S.jsx)(`pre`,{children:(0,S.jsxs)(`code`,{children:[`quickstudy run --eval my-eval --experiment my-experiment`,`
+`)}function tt(e){return e===`high`?`green`:e===`medium`?`amber`:`gray`}function nt(e,t){return T.attempt(e,t.attempt_id,t.artifact,t.lines)}function rt(e){return t=>new URL(nt(e,t),window.location.href).href}function it({runId:e,evidence:t}){return(0,S.jsxs)(`a`,{className:`evidence-chip`,"data-testid":`evidence-link`,href:nt(e,t),children:[t.artifact,` L`,t.lines[0],`–`,t.lines[1]]})}function at({runId:e,finding:t}){return(0,S.jsxs)(`div`,{className:`card`,"data-testid":`diagnosis-finding`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:t.claim}),(0,S.jsxs)(`span`,{className:`head-actions`,children:[(0,S.jsxs)(D,{tone:tt(t.confidence),children:[t.confidence,` confidence`]}),(0,S.jsx)(qe,{testId:`copy-finding`,text:()=>$e(t,rt(e))})]})]}),(0,S.jsxs)(`div`,{className:`body`,children:[(0,S.jsxs)(`div`,{className:`attribution-reason`,children:[(0,S.jsx)(`code`,{children:t.eval_id}),` × `,(0,S.jsx)(`code`,{children:t.experiment_id})]}),(0,S.jsx)(`p`,{children:t.suggested_change}),(0,S.jsx)(`div`,{className:`evidence-row`,children:t.evidence.map((t,n)=>(0,S.jsx)(it,{runId:e,evidence:t},`${t.attempt_id}-${t.artifact}-${n}`))})]})]})}var ot=(e,t)=>`${e} ${t}${e===1?``:`s`}`;function st(e){let t=ot(e.findings.length,`finding`);return e.failingChecks.length===0?`${e.evalId} · ${t}`:`${e.evalId} · ${e.failingChecks.join(`, `)} failed under ${ot(e.experiments.length,`arm`)} · ${t}`}function ct({runId:e,diagnosis:t,pairs:n}){let[r,i]=(0,l.useState)(null);if(t.schema_version!==1)return(0,S.jsxs)(`div`,{className:`error-banner`,"data-testid":`diagnosis-schema-banner`,role:`alert`,children:[(0,S.jsx)(`strong`,{children:`Diagnosis schema mismatch.`}),` This diagnosis.json is schema_version`,` `,String(t.schema_version),` but this UI build expects `,1,` — regenerate it with the matching quickstudy version (`,(0,S.jsxs)(`code`,{children:[`quickstudy diagnose `,e]}),`).`]});let a=new Map;for(let e of t.findings){let t=a.get(e.target);t?t.push(e):a.set(e.target,[e])}let o=Qe(t.findings),s=Ze(t.findings,n);return(0,S.jsxs)(S.Fragment,{children:[(0,S.jsx)(`div`,{className:`notice`,"data-testid":`diagnosis-disclaimer`,children:t.disclaimer}),(0,S.jsxs)(`div`,{className:`kpi-row`,"data-testid":`diagnosis-provenance`,children:[(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Model`}),(0,S.jsx)(`div`,{className:`kpi-val`,style:{fontSize:15},children:t.model.requested}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`resolved: `,t.model.resolved??`unavailable`]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Pairs diagnosed`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:t.pairs_diagnosed}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[t.pairs_skipped,` skipped`]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Findings`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:t.findings.length}),(0,S.jsx)(`div`,{className:`kpi-meta`,children:t.errors.length>0?`${t.errors.length} pair(s) not analyzed`:`all selected pairs analyzed`})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Provenance`}),(0,S.jsx)(`div`,{className:`kpi-val`,style:{fontSize:15},children:t.generated_at.slice(0,10)}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`run `,(0,S.jsxs)(`code`,{children:[t.run_id.slice(0,10),`…`]}),` · prompt v`,t.prompt_version]})]})]}),t.findings.length===0&&(0,S.jsxs)(`div`,{className:`empty-state`,"data-testid":`diagnosis-no-findings`,children:[(0,S.jsx)(`h2`,{children:`Diagnosis ran; no findings.`}),(0,S.jsx)(`p`,{children:`The analyst drafted no hypotheses for this run's diagnosable pairs.`})]}),t.findings.length>0&&(0,S.jsx)(`div`,{className:`toolbar`,style:{justifyContent:`flex-start`},children:(0,S.jsxs)(`div`,{className:`seg`,role:`group`,"aria-label":`Finding target`,"data-testid":`diagnosis-target-filter`,children:[(0,S.jsxs)(`button`,{type:`button`,className:r===null?`active`:``,"aria-pressed":r===null,onClick:()=>i(null),children:[`all (`,t.findings.length,`)`]}),o.filter(e=>a.has(e)).map(e=>(0,S.jsxs)(`button`,{type:`button`,className:r===e?`active`:``,"aria-pressed":r===e,onClick:()=>i(e),children:[e,` (`,a.get(e)?.length,`)`]},e))]})}),s.map(t=>{let n=o.filter(e=>(r===null||r===e)&&t.findings.some(t=>t.target===e));return n.length===0?null:(0,S.jsxs)(`section`,{className:`cluster`,"data-testid":`diagnosis-cluster`,children:[(0,S.jsx)(`h2`,{children:st(t)}),n.map(n=>(0,S.jsxs)(`section`,{"data-testid":`diagnosis-target-${n}`,children:[(0,S.jsxs)(`h3`,{children:[(0,S.jsx)(`code`,{children:n}),` findings`]}),t.findings.filter(e=>e.target===n).map((t,n)=>(0,S.jsx)(at,{runId:e,finding:t},`${t.eval_id}|${t.experiment_id}|${n}`))]},n))]},`${t.evalId}|${t.failingChecks.join(`,`)}`)}),t.errors.length>0&&(0,S.jsxs)(`div`,{className:`card`,"data-testid":`diagnosis-errors`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:`Pairs that could not be analyzed`}),(0,S.jsx)(`span`,{className:`kpi-meta`,children:`model failures are recorded, never silently dropped`})]}),(0,S.jsx)(`div`,{className:`body`,children:(0,S.jsxs)(`table`,{className:`data`,children:[(0,S.jsx)(`thead`,{children:(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`eval`}),(0,S.jsx)(`th`,{children:`experiment`}),(0,S.jsx)(`th`,{children:`error`})]})}),(0,S.jsx)(`tbody`,{children:t.errors.map(e=>(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`td`,{children:(0,S.jsx)(`code`,{children:e.eval_id})}),(0,S.jsx)(`td`,{children:(0,S.jsx)(`code`,{children:e.experiment_id})}),(0,S.jsx)(`td`,{children:e.error})]},`${e.eval_id}|${e.experiment_id}`))})]})})]})]})}function lt({meta:e,runId:t}){let n=C(()=>y(t),[t]),r=C(()=>v(t).then(e=>e.pairs??null,()=>null),[t]);if(n.status===`error`)return(0,S.jsx)(ee,{error:n.error});if(n.status===`loading`||r.status===`loading`)return(0,S.jsx)(`div`,{className:`loading`,children:`Loading diagnosis…`});let i=r.status===`ready`?r.value:null,a=n.value;return(0,S.jsxs)(`div`,{"data-testid":`diagnosis-view`,children:[(0,S.jsxs)(`div`,{className:`page-head`,children:[(0,S.jsxs)(`div`,{children:[(0,S.jsx)(`h1`,{children:`Run diagnosis`}),(0,S.jsx)(`p`,{className:`sub`,children:`Model-drafted hypotheses with harness-selected evidence — verify cited lines before acting.`})]}),(0,S.jsxs)(`div`,{className:`control`,children:[(0,S.jsx)(`label`,{htmlFor:`diagnosis-run`,children:`Run`}),(0,S.jsx)(`select`,{id:`diagnosis-run`,"data-testid":`diagnosis-run-select`,value:t,onChange:e=>se(T.diagnosis(e.currentTarget.value)),children:e.runs.map(e=>(0,S.jsxs)(`option`,{value:e.id,children:[e.id.slice(0,10),`… · `,e.started_at.slice(0,10)]},e.id))})]})]}),a!==null&&a.schema_version===1&&(0,S.jsx)(`div`,{className:`toolbar`,"data-testid":`diagnosis-toolbar`,children:(0,S.jsx)(qe,{testId:`copy-diagnosis`,label:`Copy all findings as Markdown`,text:()=>et(a,rt(t))})}),a===null?(0,S.jsxs)(`div`,{className:`empty-state`,"data-testid":`diagnosis-empty`,children:[(0,S.jsx)(`h2`,{children:`No diagnosis for this run`}),(0,S.jsx)(`p`,{children:`Generate one (requires a model API key — never done at export time), then re-export:`}),(0,S.jsx)(`pre`,{children:(0,S.jsxs)(`code`,{children:[`quickstudy diagnose `,t]})})]}):(0,S.jsx)(ct,{runId:t,diagnosis:a,pairs:i})]})}function ut({coverage:e}){return e?(0,S.jsxs)(`details`,{"data-testid":`publication-omissions`,open:e.omitted.length>0,children:[(0,S.jsxs)(`summary`,{children:[`Complete-sample publication: `,e.eligible.length,` pairs included, `,e.omitted.length,` omitted;`,` `,e.expected_trials,` trials required.`]}),(0,S.jsx)(`p`,{children:`These counts cover the included pairs. Source attempts and retry history remain available in the diagnostic artifacts.`}),e.omitted.map(e=>(0,S.jsxs)(`p`,{children:[e.eval_id,` × `,e.experiment_id,`: `,e.reasons.join(`; `)]},`${e.eval_id}:${e.experiment_id}`))]}):null}function dt(e,t){let n=e=>e??-1,r=e=>(e.role??`agent`)===`agent`?0:1;return r(e)-r(t)||n(t.aggregate?.pass_rate_discriminating)-n(e.aggregate?.pass_rate_discriminating)||n(t.aggregate?.pass_rate_all)-n(e.aggregate?.pass_rate_all)||e.experiment_id.localeCompare(t.experiment_id)}function ft(e){if(e===null)return`—`;let t=e=>e.toFixed(1).replace(/\.0$/,``);return e>=1e6?`${t(e/1e6)}M`:e>=1e3?`${t(e/1e3)}k`:String(Math.round(e))}function pt(e){if(e===null)return`—`;let t=Math.round(e/1e3);return t<60?`${t}s`:`${Math.floor(t/60)}m ${t%60}s`}function mt(e){return e===null?`—`:e>0&&e<.005?`<$0.01`:`$${e.toFixed(2)}`}function ht(e){let t=e.timedOut??0,n=e.incomplete-t,r=(e,t)=>`${e} ${t}${e===1?``:`s`}`,i=[e.unfinished?`the run never finished`:``,t>0?`${r(t,`attempt`)} timed out`:``,n>0?`${r(n,`attempt`)} incomplete`:``,e.errors>0?`${r(e.errors,`attempt`)} errored`:``].filter(Boolean);return i.length>0?i.join(`; `):`missing trial coverage`}function gt({experiment:e}){let t=e.aggregate;if(t===void 0||t.pass_rate_all===null)return`—`;let n=`${t.passed}/${t.passed+t.failed}`;return t.discriminating_evals===0?(0,S.jsxs)(S.Fragment,{children:[n,` · `,(0,S.jsx)(`span`,{className:`muted`,title:`No discriminating evals scored for this experiment`,children:`—`})]}):`${n} · ${t.discriminating_passed}/${t.discriminating_scored}`}function _t({experiment:e}){let t=e.aggregate;return t===void 0||t.attempts===0?`—`:`${t.cost_attempts>0&&t.cost_usd!==null?`${mt(t.cost_usd/t.cost_attempts)} · ${t.cost_attempts}/${t.attempts} reported`:`not reported (0/${t.attempts})`}${typeof t.estimated_cost_usd==`number`&&t.estimated_attempts?` · ${mt(t.estimated_cost_usd/t.estimated_attempts)} est.`:``}`}function vt({report:e,runId:t}){let n=Array.isArray(e.experiments)?e.experiments:[];if(n.length===0||n.some(e=>e.aggregate===void 0))return(0,S.jsxs)(`div`,{className:`notice`,"data-testid":`leaderboard-stale`,children:[`This report predates the leaderboard. Regenerate it with `,(0,S.jsxs)(`code`,{children:[`quickstudy report `,t]}),`.`]});let r=[...n].sort(dt);return(0,S.jsxs)(`div`,{className:`card`,"data-testid":`leaderboard`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:`Which experiment is best?`}),(0,S.jsx)(`span`,{className:`muted`,children:`sorted by pass rate on discriminating evals; controls last`})]}),(0,S.jsx)(`div`,{className:`body`,children:(0,S.jsxs)(`table`,{className:`data`,children:[(0,S.jsx)(`thead`,{children:(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`Experiment`}),(0,S.jsx)(`th`,{children:`Pass rate (all · discriminating)`}),(0,S.jsx)(`th`,{children:`Agent time (median)`}),(0,S.jsx)(`th`,{children:`Tokens/run (in)`}),(0,S.jsx)(`th`,{children:`Cost/run`}),(0,S.jsx)(`th`,{children:`SDK recall`})]})}),(0,S.jsx)(`tbody`,{children:r.map(e=>{let t=e.aggregate,n=e.sdk_recall;return(0,S.jsxs)(`tr`,{"data-testid":`leaderboard-row-${e.experiment_id}`,children:[(0,S.jsxs)(`td`,{children:[Me(e.agent),(0,S.jsxs)(`span`,{className:`group-sub`,children:[(0,S.jsx)(`code`,{children:e.experiment_id}),e.treatment===null?``:` · ${e.treatment}`]}),(e.role??`agent`)!==`agent`&&(0,S.jsx)(`span`,{className:`tag`,"data-testid":`leaderboard-control`,title:`A control checks the instrument; it is not an arm and never decides a band.`,children:e.role===`positive-control`?`positive control`:`negative control`})]}),(0,S.jsx)(`td`,{"data-testid":`leaderboard-pass`,children:(0,S.jsx)(gt,{experiment:e})}),(0,S.jsx)(`td`,{title:t?`median over ${t.agent_time_attempts} of ${t.completed} scored attempt(s)`:void 0,children:pt(t?.median_agent_ms??null)}),(0,S.jsx)(`td`,{title:t?`mean over ${t.token_attempts} attempt(s) that reported tokens`:void 0,children:ft(t?.mean_tokens_in??null)}),(0,S.jsx)(`td`,{"data-testid":`leaderboard-cost`,children:(0,S.jsx)(_t,{experiment:e})}),(0,S.jsx)(`td`,{children:n&&n.rate!==null?`${n.recalled}/${n.attempts}`:`n/a`})]},e.experiment_id)})})]})})]})}function yt(e){return e===null?{bg:`var(--gray-bg)`,fg:`var(--muted)`}:e>=.95?{bg:`#1a7f37`,fg:`#ffffff`}:e>=.9?{bg:`#7ecb8f`,fg:`#0f3d1e`}:e>=.8?{bg:`#ffdf80`,fg:`#5c3d00`}:e>=.5?{bg:`#ffb066`,fg:`#5c2800`}:e>0?{bg:`#f28b82`,fg:`#520f08`}:{bg:`#d93025`,fg:`#ffffff`}}function bt(e){return e===null?`—`:`${Math.round(e*100)}%`}function xt(){return(0,S.jsxs)(`div`,{className:`legend`,"aria-label":`Pass-rate color scale (fixed 0-100%)`,children:[(0,S.jsx)(`span`,{children:`0%`}),[0,.5,.8,.9,.95].map(e=>{let t=yt(e);return(0,S.jsx)(`span`,{className:`legend-sw`,style:{background:t.bg},title:`≥ ${Math.round(e*100)}%`},e)}),(0,S.jsx)(`span`,{children:`100%`}),(0,S.jsx)(`span`,{style:{marginLeft:8},children:`ticks: 80 / 90 / 95`})]})}function St(e){let t=[];for(let n of e){let e=n.group===null?`agent:${n.agentLabel??``}`:`group:${n.group}`,r=t.at(-1);r&&r.key===e?r.columns.push(n):t.push({key:e,group:n.group,columns:[n]})}return t.map(({group:e,columns:t})=>{let n=new Set(t.map(e=>e.agentLabel)),r=n.size===1?[...n][0]??null:null;return{label:r??e??``,sub:r!==null&&e!==null?e:null,title:r===null?null:t[0]?.agentTitle??null,span:t.length}})}function Ct({cells:e,rows:t,columns:n,onCellClick:r,band:i}){let a=new Map(e.map(e=>[`${e.evalId}|${e.experimentId}`,e])),o=n.some(e=>e.group!==null||e.agentLabel!==null);return(0,S.jsxs)(`div`,{className:`matrix-region`,children:[(0,S.jsx)(`div`,{className:`scroll-hint`,"aria-hidden":`true`,children:`Scroll horizontally to see every experiment →`}),(0,S.jsx)(`div`,{className:`matrix-scroll`,role:`region`,"aria-label":i===void 0?`Pass-rate matrix`:`Pass-rate matrix — ${i}`,tabIndex:0,children:(0,S.jsxs)(`table`,{className:`matrix`,children:[(0,S.jsxs)(`thead`,{children:[o&&(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{"aria-hidden":`true`}),St(n).map((e,t)=>(0,S.jsxs)(`th`,{colSpan:e.span,className:`matrix-group-head`,title:e.title??void 0,children:[e.label,e.sub!==null&&(0,S.jsx)(`span`,{className:`matrix-group-sub`,children:e.sub})]},`${e.label}-${t}`))]}),(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`eval`}),n.map(e=>(0,S.jsxs)(`th`,{children:[(0,S.jsx)(`code`,{children:e.experimentId}),e.treatment!==null&&(0,S.jsx)(`span`,{className:`matrix-row-feature`,children:e.treatment})]},e.experimentId))]})]}),(0,S.jsx)(`tbody`,{children:t.map(e=>(0,S.jsxs)(`tr`,{children:[(0,S.jsxs)(`th`,{children:[(0,S.jsx)(`code`,{children:e.evalId}),e.framework!==void 0&&(0,S.jsx)(`span`,{className:`matrix-row-feature`,children:e.framework})]}),n.map(t=>{let n=a.get(`${e.evalId}|${t.experimentId}`);if(!n)return(0,S.jsx)(`td`,{className:`cell`,style:{background:`var(--gray-bg)`,color:`var(--muted)`},children:(0,S.jsx)(`div`,{className:`cell-rate`,children:`—`})},t.experimentId);let i=yt(n.rate),o=`${n.evalId}, ${n.experimentId}: ${bt(n.rate)}, ${n.passed} of ${n.scored} scored passed${n.checks?`; ${n.checks.passed} of ${n.checks.total} checks passed`:``}${n.errors>0?`; ${n.errors} errored`:``}${n.flaky?`; flaky trials`:``}${n.incomplete?`; incomplete`:``}`;return(0,S.jsx)(`td`,{className:`cell`,style:{background:i.bg,color:i.fg},children:(0,S.jsxs)(`button`,{type:`button`,className:`cell-button`,"aria-label":o,"data-testid":`cell-${n.evalId}-${n.experimentId}`,"data-rate":n.rate===null?``:String(Math.round(n.rate*100)),"data-flaky":n.flaky?`true`:`false`,"data-incomplete":n.incomplete?`true`:`false`,onClick:()=>r(n),children:[n.flaky&&(0,S.jsx)(`span`,{className:`flaky-marker`,title:`flaky: trials disagreed`,"aria-hidden":`true`}),(0,S.jsx)(`span`,{className:`cell-rate`,children:bt(n.rate)}),(0,S.jsxs)(`span`,{className:`cell-sub`,children:[n.passed,`/`,n.scored,n.checks?` · ${n.checks.passed}/${n.checks.total} checks`:``,n.errors>0?` · ${n.errors} err`:``]}),n.incomplete&&(0,S.jsx)(`span`,{className:`incomplete-marker`,title:`incomplete: fewer attempts than planned`,"aria-hidden":`true`,children:`◌`})]})},t.experimentId)})]},e.evalId))})]})})]})}var wt=`all`,Tt=[`benchmark`,`regression`,`other`];function Et(e,t){return e.manifest.evals[t]?.metadata??{}}function Dt(e,t){for(let n of e.groups)if(n.experimentIds.includes(t))return n.group;return null}function Ot(e,t){return[...new Set(t.map(e=>e.experiment_id))].map(t=>({experimentId:t,treatment:e.manifest.experiments[t]?.components.treatment??null,group:Dt(e,t),agentLabel:Ne(e,t),agentTitle:Pe(e,t)})).sort((e,t)=>(e.group??`￿`).localeCompare(t.group??`￿`)||(e.treatment??``).localeCompare(t.treatment??``)||e.experimentId.localeCompare(t.experimentId))}function kt(e,t){let n=new Map;for(let t of e){let e=`${t.eval_id}|${t.experiment_id}`,r=n.get(e);r?r.push(t):n.set(e,[t])}let r=[];for(let e of n.values()){let n=e[0],i=e.filter(e=>e.status===`completed`),a=i.filter(e=>e.passed===!0).length,o=i.length,s=new Set(i.map(e=>e.passed===!0)),c=le(i.flatMap(e=>e.checks??[]));r.push({evalId:n.eval_id,experimentId:n.experiment_id,rate:o>0?a/o:null,passed:a,scored:o,flaky:s.size>1,incomplete:e.length<t||e.some(e=>e.status===`incomplete`),errors:e.filter(e=>e.status===`error`).length,checks:c.total>0?c:null})}return r}function At(e){if(e.started_at===null||e.finished_at===null)return`—`;let t=Math.round((Date.parse(e.finished_at)-Date.parse(e.started_at))/1e3);return`${Math.floor(t/60)}m${String(t%60).padStart(2,`0`)}s`}var jt={codex:{label:`thread turns`,title:`codex exec reports one thread turn per prompt, not per agent step — not comparable to Claude's turn count`}};function Mt(e){return e.status===`incomplete`&&e.error?.startsWith(`attempt timed out`)===!0}function Nt({attempt:e,runId:t,adapter:n}){let r=pe(e),i=n!==null&&jt[n]||{label:`turns`,title:`agent turns, as counted by the adapter`};return(0,S.jsxs)(`div`,{className:`attempt-row`,"data-testid":`drill-attempt`,children:[(0,S.jsxs)(`div`,{className:`attempt-line`,children:[(0,S.jsxs)(`span`,{className:`kpi-meta`,children:[`trial `,e.trial_index+1]}),(0,S.jsx)(D,{tone:me(r),children:r}),(0,S.jsx)(ge,{attempt:e}),(0,S.jsx)(`span`,{className:`spacer`}),e.transcript!==null&&(0,S.jsx)(`a`,{href:T.attempt(t,e.id,`transcript`),children:`transcript`}),e.diff!==null&&(0,S.jsx)(`a`,{href:T.attempt(t,e.id,`diff`),children:`diff`})]}),e.checks!==null&&e.checks.length>0&&(0,S.jsx)(`div`,{className:`attempt-line`,children:e.checks.map(e=>(0,S.jsxs)(`span`,{className:e.passed?`crit pass`:`crit fail`,title:e.notes??e.name,children:[e.passed?`✓`:`✗`,` `,e.name,!E(e)&&(0,S.jsx)(`span`,{className:`tag`,"data-testid":`ungraded-tag`,title:`Recorded, not graded: never enters pass/fail`,children:`recorded`})]},e.name))}),e.error!==null&&(0,S.jsx)(`div`,{className:`attempt-line`,children:(0,S.jsx)(`code`,{children:e.error})}),(0,S.jsxs)(`div`,{className:`telemetry`,children:[(0,S.jsxs)(`span`,{children:[`wall `,At(e)]}),(0,S.jsxs)(`span`,{children:[`tokens `,ft(e.tokens_in),` in / `,ft(e.tokens_out),` out`]}),(0,S.jsxs)(`span`,{children:[`cost `,e.cost_usd===null?`—`:`$${e.cost_usd.toFixed(2)}`]}),(0,S.jsxs)(`span`,{title:i.title,children:[i.label,` `,e.turns??`—`]})]}),(0,S.jsx)(Se,{treatment:e.treatment,usage:e.surface_usage})]})}function Pt({runId:e,run:t,cell:n,attempts:r}){let i=(0,l.useRef)(null),a=(0,l.useRef)(!1),o=(0,l.useId)(),s=r.filter(e=>e.eval_id===n.evalId&&e.experiment_id===n.experimentId).sort((e,t)=>e.trial_index-t.trial_index),c=t.manifest.experiments[n.experimentId]?.components.treatment??null,u=s.filter(e=>e.status===`completed`),d=u.filter(e=>e.passed===!0).length,f=`cell-${n.evalId}-${n.experimentId}`,p=(0,l.useCallback)(()=>{a.current=!0,se(T.matrix(e))},[e]);return(0,l.useEffect)(()=>{let e=i.current;if(!e)return;let t=document.body.style.overflow;document.body.style.overflow=`hidden`;let n=()=>[...e.querySelectorAll(`a[href], button:not([disabled]), input:not([disabled]), select:not([disabled]), textarea:not([disabled]), [tabindex]:not([tabindex="-1"])`)].filter(e=>!e.hasAttribute(`hidden`));n()[0]?.focus();let r=t=>{if(t.key===`Escape`){t.preventDefault(),p();return}if(t.key!==`Tab`)return;let r=n();if(r.length===0){t.preventDefault(),e.focus();return}let i=r[0],a=r.at(-1);t.shiftKey&&document.activeElement===i?(t.preventDefault(),a.focus()):!t.shiftKey&&document.activeElement===a?(t.preventDefault(),i.focus()):e.contains(document.activeElement)||(t.preventDefault(),i.focus())};return document.addEventListener(`keydown`,r),()=>{if(document.removeEventListener(`keydown`,r),document.body.style.overflow=t,!a.current)return;let e=()=>{let e=[...document.querySelectorAll(`[data-testid]`)].find(e=>e.dataset.testid===f);return e?.focus(),e!==void 0};requestAnimationFrame(()=>{e()||window.setTimeout(e,0)})}},[p,f]),(0,S.jsxs)(S.Fragment,{children:[(0,S.jsx)(`div`,{className:`drill-overlay`,"aria-hidden":`true`,onClick:p,onKeyDown:e=>e.preventDefault()}),(0,S.jsxs)(`aside`,{ref:i,className:`drill`,"data-testid":`drill-down`,role:`dialog`,"aria-modal":`true`,"aria-labelledby":o,tabIndex:-1,children:[(0,S.jsx)(`button`,{type:`button`,className:`drill-close`,onClick:p,"data-testid":`drill-close`,children:`Close ✕`}),(0,S.jsxs)(`h2`,{id:o,children:[n.evalId,` · `,n.experimentId,c===null?``:` · ${c}`]}),(0,S.jsxs)(`p`,{className:`sub`,children:[s.length,` attempt(s) · `,d,`/`,u.length,` scored passed · `,bt(u.length>0?d/u.length:null)]}),s.length===0&&(0,S.jsx)(`div`,{className:`viewer-note`,children:`No attempts in this cell.`}),s.map(n=>(0,S.jsx)(Nt,{attempt:n,runId:e,adapter:t.manifest.experiments[n.experiment_id]?.components.agent?.adapter??null},n.id))]})]})}function Ft({id:e,label:t,values:n,value:r,onChange:i}){return n.length===0?null:(0,S.jsxs)(`div`,{className:`control`,children:[(0,S.jsx)(`label`,{htmlFor:e,children:t}),(0,S.jsxs)(`select`,{id:e,"data-testid":e,value:r,onChange:e=>i(e.currentTarget.value),children:[(0,S.jsx)(`option`,{value:wt,children:`all`}),n.map(e=>(0,S.jsx)(`option`,{value:e,children:e},e))]})]})}var It=e=>e===1?``:`s`;function Lt({meta:e,runId:t,cell:n}){let r=C(()=>_(t),[t]),i=C(()=>v(t).then(e=>e,()=>null),[t]),[a,o]=(0,l.useState)(`benchmark`),s=r.status===`ready`?Tt.filter(e=>r.value.attempts.some(t=>t.suite===e)):[],c=s.includes(a)?a:s[0]??`benchmark`,[u,d]=(0,l.useState)({product:wt,framework:wt,topic:wt});(0,l.useEffect)(()=>{d({product:wt,framework:wt,topic:wt})},[t]);let f=(0,l.useMemo)(()=>{if(r.status!==`ready`)return null;let e=r.value,t=e.attempts.filter(e=>e.suite===c).filter(t=>{let n=Et(e,t.eval_id);return!(u.product!==wt&&n.product!==u.product||u.framework!==wt&&n.framework!==u.framework||u.topic!==wt&&!(n.topics??[]).includes(u.topic))});return{rows:[...new Set(t.map(e=>e.eval_id))].sort((e,t)=>e.localeCompare(t)).map(t=>({evalId:t,...Et(e,t).framework===void 0?{}:{framework:Et(e,t).framework}})),columns:Ot(e,t),cells:kt(t,e.run.trials),attempts:t}},[r,c,u]);if(r.status===`error`)return(0,S.jsx)(ee,{error:r.error});if(r.status===`loading`||i.status===`loading`||f===null)return(0,S.jsx)(`div`,{className:`loading`,children:`Loading run…`});let p=r.value,m=p.run,h=Tt.filter(e=>p.attempts.some(t=>t.suite===e)),g=[...new Set(p.attempts.map(e=>Et(p,e.eval_id).product).filter(e=>e!==void 0))].sort(),y=[...new Set(p.attempts.map(e=>Et(p,e.eval_id).framework).filter(e=>e!==void 0))].sort(),b=[...new Set(p.attempts.flatMap(e=>Et(p,e.eval_id).topics??[]))].sort(),x={attempts:p.attempts.length,passed:p.attempts.filter(e=>e.status===`completed`&&e.passed===!0).length,failed:p.attempts.filter(e=>e.status===`completed`&&e.passed!==!0).length,incomplete:p.attempts.filter(e=>e.status===`incomplete`).length,timedOut:p.attempts.filter(e=>e.status===`completed`&&e.limit===`timeout`||Mt(e)).length,unscoredTimedOut:p.attempts.filter(Mt).length,errors:p.attempts.filter(e=>e.status===`error`).length},te=x.passed+x.failed,ne=i.status===`ready`?i.value:null,w=Array.isArray(ne?.pairs)?ne.pairs:[],re=w.some(e=>e.band!==void 0),ie=new Map(w.map(e=>[e.eval_id,e.band??null])),ae=e=>f.rows.filter(t=>(ie.get(t.evalId)??null)===e),oe=f.rows.filter(e=>{let t=ie.get(e.evalId)??null;return t===null||t===`discriminating`}),ce=ae(null).length,E=ae(`solved-by-all`),le=ae(`failed-by-all`),ue=ne?.scorer_health?.suspects??[],de=e=>se(T.cell(t,{evalId:e.evalId,experimentId:e.experimentId})),fe=e=>{o(e),n&&se(T.matrix(t))},pe=e=>r=>{d(t=>({...t,[e]:r})),n&&se(T.matrix(t))};return(0,S.jsxs)(`div`,{"data-testid":`matrix-view`,children:[(0,S.jsx)(ut,{coverage:p.publication}),(0,S.jsxs)(`div`,{className:`page-head`,children:[(0,S.jsxs)(`div`,{children:[(0,S.jsx)(`h1`,{children:`Eval × experiment matrix`}),(0,S.jsx)(`p`,{className:`sub`,children:`Pass rate over scored attempts per cell; experiment columns are the treatment arms, grouped by comparison group. Select a cell for attempt evidence.`})]}),(0,S.jsxs)(`div`,{className:`controls`,children:[(0,S.jsxs)(`div`,{className:`control`,children:[(0,S.jsx)(`label`,{htmlFor:`run-select`,children:`Run`}),(0,S.jsx)(`select`,{id:`run-select`,"data-testid":`run-select`,value:t,onChange:e=>se(T.matrix(e.currentTarget.value)),children:e.runs.map(e=>(0,S.jsxs)(`option`,{value:e.id,children:[e.id.slice(0,10),`… · `,e.started_at.slice(0,10),` · `,e.attempts,` attempts`]},e.id))})]}),(0,S.jsx)(Ft,{id:`filter-product`,label:`Product`,values:g,value:u.product,onChange:pe(`product`)}),(0,S.jsx)(Ft,{id:`filter-framework`,label:`Framework`,values:y,value:u.framework,onChange:pe(`framework`)}),(0,S.jsx)(Ft,{id:`filter-topic`,label:`Topic`,values:b,value:u.topic,onChange:pe(`topic`)}),h.length>1&&(0,S.jsxs)(`div`,{className:`control`,children:[(0,S.jsx)(`label`,{children:`Suite`}),(0,S.jsx)(`div`,{className:`seg`,role:`group`,"aria-label":`Suite`,children:h.map(e=>(0,S.jsx)(`button`,{type:`button`,className:c===e?`active`:``,"aria-pressed":c===e,"data-testid":`suite-${e}`,onClick:()=>fe(e),children:e},e))})]})]})]}),(m.finished_at===null||m.incomplete)&&(0,S.jsxs)(`div`,{className:`notice`,"data-testid":`incomplete-banner`,children:[`Incomplete run — `,ht({unfinished:m.finished_at===null,incomplete:x.incomplete,errors:x.errors,timedOut:x.unscoredTimedOut}),`. Numbers are partial.`]}),(0,S.jsxs)(`div`,{className:`kpi-row`,children:[(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsxs)(`div`,{className:`kpi-label`,children:[`Overall pass rate`,h.length>1?` (all suites)`:``]}),(0,S.jsx)(`div`,{className:`kpi-val`,children:te===0?`—`:`${Math.round(x.passed/te*100)}%`}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[(0,S.jsxs)(D,{tone:`green`,children:[x.passed,` passed`]}),` `,(0,S.jsxs)(D,{tone:`red`,children:[x.failed,` failed`]}),` `,(0,S.jsxs)(D,{tone:`gray`,children:[x.incomplete,` incomplete`]}),` `,(0,S.jsxs)(D,{tone:`amber`,children:[x.errors,` errored`]}),x.timedOut>0&&(0,S.jsxs)(S.Fragment,{children:[` `,(0,S.jsxs)(D,{tone:`gray`,children:[x.timedOut,` timed out`]})]})]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Evals × experiments`}),(0,S.jsxs)(`div`,{className:`kpi-val`,children:[m.eval_ids.length,` × `,m.experiment_ids.length]}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[m.attempts,` attempts`]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Trials per pair`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:m.trials}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`manifest `,m.manifest_hash.slice(0,10),`…`]})]})]}),ne!==null&&(0,S.jsx)(vt,{report:ne,runId:t}),(0,S.jsxs)(`div`,{className:`card`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsxs)(`h3`,{children:[`Pass rate per eval × experiment`,c===`regression`&&(0,S.jsxs)(S.Fragment,{children:[` `,(0,S.jsx)(D,{tone:`amber`,title:`Known failure modes — tracked separately from benchmark results.`,children:`regression suite`})]})]}),(0,S.jsx)(xt,{})]}),ue.length>0&&ne?.scorer_health&&(0,S.jsxs)(`div`,{className:`notice band-notice`,"data-testid":`scorer-health-banner`,children:[ne.scorer_health.summary,(0,S.jsx)(`ul`,{children:ue.map(e=>(0,S.jsxs)(`li`,{children:[(0,S.jsx)(`code`,{children:e.eval_id}),` · `,(0,S.jsx)(`code`,{children:e.check}),` failed in `,e.attempts,` attempts under`,` `,e.experiments.length,` arms · passing: `,e.passing_checks.join(`, `)]},`${e.eval_id}|${e.check}`))})]}),re?(0,S.jsxs)(S.Fragment,{children:[oe.length>0&&(0,S.jsxs)(`section`,{className:`band`,"data-testid":`matrix-band-discriminating`,children:[(0,S.jsxs)(`h4`,{children:[`Discriminating (`,oe.length,` eval`,It(oe.length),`) — arms disagree`,ce>0&&` (${ce} eval${It(ce)} with one scored arm)`]}),(0,S.jsx)(Ct,{cells:f.cells,rows:oe,columns:f.columns,onCellClick:de,band:`discriminating`})]}),E.length>0&&(0,S.jsxs)(`details`,{className:`band`,"data-testid":`matrix-band-solved`,children:[(0,S.jsxs)(`summary`,{children:[`Solved by every arm (`,E.length,` eval`,It(E.length),`)`]}),(0,S.jsx)(Ct,{cells:f.cells,rows:E,columns:f.columns,onCellClick:de,band:`solved by every arm`})]}),le.length>0&&(0,S.jsxs)(`details`,{className:`band`,"data-testid":`matrix-band-failed`,children:[(0,S.jsxs)(`summary`,{children:[`Failed by every arm (`,le.length,` eval`,It(le.length),`) — suspect the eval`]}),(0,S.jsx)(Ct,{cells:f.cells,rows:le,columns:f.columns,onCellClick:de,band:`failed by every arm`})]})]}):(0,S.jsx)(Ct,{cells:f.cells,rows:f.rows,columns:f.columns,onCellClick:de})]}),n&&(0,S.jsx)(Pt,{runId:t,run:p,cell:n,attempts:p.attempts})]})}var Rt=[2,3,5,10];function zt(e){return(e??[]).filter(e=>Rt.includes(e.k))}function Bt({estimate:e}){return e===void 0?null:(0,S.jsxs)(`div`,{className:`kpi`,"data-testid":`cost-estimate`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Estimated token cost`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:e.cost_usd===null?`—`:e.cost_usd>0&&e.cost_usd<.01?`<$0.01`:`$${e.cost_usd.toFixed(2)}`}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`In addition to reported cost · `,e.cost_attempts,` estimated, `,e.unpriced_attempts,` unpriced attempt(s)`]}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`Pricing: `,e.pricing.version,`. Excludes tool fees, diagnosis, infrastructure and superseded retries.`]})]})}function Vt(e){return e===null?`—`:`${Math.round(e*100)}%`}function Ht({pair:e}){let t=zt(e.pass_at_k);return t.length===0?(0,S.jsx)(S.Fragment,{children:`—`}):(0,S.jsx)(`div`,{children:t.map(e=>(0,S.jsxs)(`div`,{children:[`k=`,e.k,`: `,Vt(e.pass_at_k),` / `,Vt(e.pass_to_the_k)]},e.k))})}function Ut({pair:e}){return e.checks.length===0?(0,S.jsx)(`span`,{className:`attribution-reason`,children:`no named checks recorded`}):(0,S.jsx)(`div`,{className:`attribution-effects`,children:e.checks.map(e=>(0,S.jsxs)(`span`,{className:e.passed===e.total?`crit pass`:`crit fail`,title:e.name,children:[e.passed===e.total?`✓`:`✗`,` `,e.name,` (`,e.passed,`/`,e.total,`)`,!E(e)&&(0,S.jsx)(`span`,{className:`tag`,"data-testid":`ungraded-tag`,title:`Recorded, not graded: never enters pass/fail`,children:`recorded`})]},e.name))})}function Wt({group:e,labelOf:t}){let n=new Set(e.experiment_ids.map(t)),r=n.size===1?[...n][0]??null:null;return r===null?(0,S.jsx)(`h3`,{children:e.group}):(0,S.jsxs)(`h3`,{children:[r,` `,(0,S.jsx)(`span`,{className:`group-sub`,children:e.group})]})}function Gt({group:e,labelOf:t}){if(e.comparison===`withheld`)return(0,S.jsxs)(`div`,{className:`card`,"data-testid":`report-group-${e.group}`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(Wt,{group:e,labelOf:t}),(0,S.jsx)(D,{tone:`amber`,children:`comparison withheld`})]}),(0,S.jsxs)(`div`,{className:`body`,children:[(0,S.jsx)(`p`,{className:`viewer-note`,children:`Members differ beyond their declared treatment — the differing components:`}),(0,S.jsx)(`ul`,{children:e.details.map(e=>(0,S.jsx)(`li`,{children:(0,S.jsx)(`code`,{children:e})},e))})]})]});let n=[...new Set(e.arms.flatMap(e=>e.evals.map(e=>e.eval_id)))].sort();return(0,S.jsxs)(`div`,{className:`card`,"data-testid":`report-group-${e.group}`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(Wt,{group:e,labelOf:t}),(0,S.jsx)(D,{tone:`green`,children:`compared — raw counts`})]}),(0,S.jsx)(`div`,{className:`body`,children:(0,S.jsxs)(`table`,{className:`data`,children:[(0,S.jsx)(`thead`,{children:(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`eval`}),e.arms.map(e=>(0,S.jsxs)(`th`,{children:[(0,S.jsx)(`code`,{children:e.experiment_id}),e.treatment!==null&&(0,S.jsx)(`div`,{className:`attribution-reason`,children:e.treatment})]},e.experiment_id))]})}),(0,S.jsx)(`tbody`,{children:n.map(t=>(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`td`,{children:(0,S.jsx)(`code`,{children:t})}),e.arms.map(e=>{let n=e.evals.find(e=>e.eval_id===t);return(0,S.jsx)(`td`,{children:n===void 0?`—`:n.display},e.experiment_id)})]},t))})]})})]})}function Kt({meta:e,runId:t}){let n=C(()=>v(t),[t]);if(n.status===`error`)return(0,S.jsx)(ee,{error:n.error});if(n.status===`loading`)return(0,S.jsx)(`div`,{className:`loading`,children:`Loading report…`});let r=n.value,i=r.totals.passed+r.totals.failed,a=new Map((r.experiments??[]).map(e=>[e.experiment_id,je(e.agent.model,e.agent.reasoning)])),o=e=>a.get(e)??null,s=r.run.trials>1;return(0,S.jsxs)(`div`,{"data-testid":`report-view`,children:[(0,S.jsx)(ut,{coverage:n.value.publication}),(0,S.jsxs)(`div`,{className:`page-head`,children:[(0,S.jsxs)(`div`,{children:[(0,S.jsx)(`h1`,{children:`Run report`}),(0,S.jsxs)(`p`,{className:`sub`,children:[r.run.eval_ids.length,` evals × `,r.run.experiment_ids.length,` experiments · `,r.run.trials,` `,`trials/pair · started `,r.run.started_at.slice(0,10)]})]}),(0,S.jsxs)(`div`,{className:`control`,children:[(0,S.jsx)(`label`,{htmlFor:`report-run`,children:`Run`}),(0,S.jsx)(`select`,{id:`report-run`,"data-testid":`report-run-select`,value:t,onChange:e=>se(T.report(e.currentTarget.value)),children:e.runs.map(e=>(0,S.jsxs)(`option`,{value:e.id,children:[e.id.slice(0,10),`… · `,e.started_at.slice(0,10)]},e.id))})]})]}),r.incomplete&&(0,S.jsxs)(`div`,{className:`notice`,"data-testid":`incomplete-banner`,children:[`⚠️ INCOMPLETE RUN — `,ht({unfinished:r.run.finished_at===null,...r.totals}),`. Numbers below are partial and NOT comparable to a full sweep.`]}),(0,S.jsxs)(`div`,{className:`kpi-row`,children:[(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Attempts`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:r.totals.attempts}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[(0,S.jsxs)(D,{tone:`green`,children:[r.totals.passed,` passed`]}),` `,(0,S.jsxs)(D,{tone:`red`,children:[r.totals.failed,` failed`]}),` `,(0,S.jsxs)(D,{tone:`gray`,children:[r.totals.incomplete,` incomplete`]}),` `,(0,S.jsxs)(D,{tone:`amber`,children:[r.totals.errors,` errored`]})]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Overall pass rate`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:i===0?`—`:`${Math.round(r.totals.passed/i*100)}%`}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[i,` scored attempt(s)`]})]}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Reported cost`}),(0,S.jsx)(`div`,{className:`kpi-val`,children:r.telemetry.cost_usd===null?`—`:`$${r.telemetry.cost_usd.toFixed(2)}`}),(0,S.jsx)(`div`,{className:`kpi-meta`,children:r.telemetry.cost_usd===null?`unavailable`:`${r.telemetry.cost_attempts} of ${r.totals.attempts} attempts reported a cost`+(r.totals.attempts>r.telemetry.cost_attempts?` · ${r.totals.attempts-r.telemetry.cost_attempts} unpriced, so this understates the spend`:``)})]}),(0,S.jsx)(Bt,{estimate:r.cost_estimate}),(0,S.jsxs)(`div`,{className:`kpi`,children:[(0,S.jsx)(`div`,{className:`kpi-label`,children:`Tokens`}),(0,S.jsx)(`div`,{className:`kpi-val`,style:{fontSize:15},children:r.telemetry.tokens_in===null&&r.telemetry.tokens_out===null?`unavailable`:`${ft(r.telemetry.tokens_in)} in / ${ft(r.telemetry.tokens_out)} out`}),(0,S.jsxs)(`div`,{className:`kpi-meta`,children:[`from `,r.telemetry.token_attempts,` of `,r.totals.attempts,` attempts`]})]})]}),(0,S.jsxs)(`div`,{className:`card`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:`Pass rates per eval × experiment`}),(0,S.jsx)(`span`,{className:`kpi-meta`,children:`scored = passed + failed; incomplete and errored attempts never enter the denominator`})]}),(0,S.jsx)(`div`,{className:`body`,children:(0,S.jsxs)(`table`,{className:`data`,"data-testid":`pairs-table`,children:[(0,S.jsx)(`thead`,{children:(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`eval`}),(0,S.jsx)(`th`,{children:`experiment`}),(0,S.jsx)(`th`,{children:`pass rate`}),(0,S.jsx)(`th`,{children:`raw`}),s&&(0,S.jsx)(`th`,{title:`pass@k: fraction of size-k subsets of the scored trials with ≥1 pass; pass^k: (passed/scored)^k`,children:`pass@k / pass^k`}),(0,S.jsx)(`th`,{children:`notes`}),(0,S.jsx)(`th`,{children:`checks`})]})}),(0,S.jsx)(`tbody`,{children:r.pairs.map(e=>(0,S.jsxs)(`tr`,{children:[(0,S.jsxs)(`td`,{children:[(0,S.jsx)(`code`,{children:e.eval_id}),e.suite!==`benchmark`&&(0,S.jsx)(`div`,{className:`attribution-reason`,children:e.suite})]}),(0,S.jsxs)(`td`,{children:[(0,S.jsx)(`code`,{children:e.experiment_id}),o(e.experiment_id)!==null&&(0,S.jsx)(`div`,{className:`attribution-reason`,children:o(e.experiment_id)})]}),(0,S.jsx)(`td`,{children:(0,S.jsx)(`strong`,{children:Vt(e.pass_rate)})}),(0,S.jsxs)(`td`,{children:[e.passed,`/`,e.scored]}),s&&(0,S.jsx)(`td`,{children:(0,S.jsx)(Ht,{pair:e})}),(0,S.jsx)(`td`,{children:[e.incomplete>0?`${e.incomplete} incomplete`:``,e.errors>0?`${e.errors} errored`:``,e.flaky?`trials disagreed`:``].filter(Boolean).join(`; `)||`—`}),(0,S.jsx)(`td`,{children:(0,S.jsx)(Ut,{pair:e})})]},`${e.eval_id}|${e.experiment_id}`))})]})})]}),r.groups.length>0&&(0,S.jsxs)(`section`,{"data-testid":`report-groups`,children:[(0,S.jsx)(`h2`,{children:`Comparison groups`}),r.groups.map(e=>(0,S.jsx)(Gt,{group:e,labelOf:o},e.group))]}),(r.experiments??[]).length>0&&(0,S.jsxs)(`div`,{className:`card`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:`SDK recall`}),(0,S.jsx)(`span`,{className:`kpi-meta`,children:`scored attempts whose ungraded checks all passed`})]}),(0,S.jsx)(`div`,{className:`body`,"data-testid":`sdk-recall`,children:(r.experiments??[]).map(e=>{let t=e.sdk_recall;return(0,S.jsx)(`div`,{className:`attribution-arm`,children:(0,S.jsxs)(`div`,{className:`attribution-arm-head`,children:[(0,S.jsx)(`strong`,{children:(0,S.jsx)(`code`,{children:e.experiment_id})}),e.treatment!==null&&(0,S.jsx)(D,{tone:`blue`,children:e.treatment}),(0,S.jsxs)(`span`,{children:[`SDK recall `,t&&t.rate!==null?`${t.recalled}/${t.attempts}`:`n/a`]})]})},e.experiment_id)})})]}),r.usage.length>0&&(0,S.jsxs)(`div`,{className:`card`,children:[(0,S.jsxs)(`div`,{className:`head`,children:[(0,S.jsx)(`h3`,{children:`Surface usage (offered vs observed)`}),(0,S.jsx)(`span`,{className:`kpi-meta`,children:`null telemetry counts as unavailable, never as zero use`})]}),(0,S.jsx)(`div`,{className:`body`,"data-testid":`usage-summary`,children:r.usage.map(e=>(0,S.jsxs)(`div`,{className:`attribution-arm`,children:[(0,S.jsxs)(`div`,{className:`attribution-arm-head`,children:[(0,S.jsx)(`strong`,{children:(0,S.jsx)(`code`,{children:e.experiment_id})}),e.treatment!==null&&(0,S.jsxs)(D,{tone:`blue`,children:[`offered: `,e.treatment]})]}),(0,S.jsx)(O,{usage:e})]},e.experiment_id))})]}),r.egress_denials.length>0&&(0,S.jsxs)(`div`,{className:`card`,children:[(0,S.jsx)(`div`,{className:`head`,children:(0,S.jsx)(`h3`,{children:`Egress denials (proxied run)`})}),(0,S.jsx)(`div`,{className:`body`,children:(0,S.jsxs)(`table`,{className:`data`,children:[(0,S.jsx)(`thead`,{children:(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`th`,{children:`attempt`}),(0,S.jsx)(`th`,{children:`eval`}),(0,S.jsx)(`th`,{children:`experiment`}),(0,S.jsx)(`th`,{children:`denials`}),(0,S.jsx)(`th`,{children:`top hosts`})]})}),(0,S.jsx)(`tbody`,{children:r.egress_denials.map(e=>(0,S.jsxs)(`tr`,{children:[(0,S.jsx)(`td`,{children:(0,S.jsxs)(`code`,{children:[e.attempt_id.slice(0,8),`…`]})}),(0,S.jsx)(`td`,{children:e.eval_id}),(0,S.jsx)(`td`,{children:e.experiment_id}),(0,S.jsx)(`td`,{children:e.count}),(0,S.jsx)(`td`,{children:e.top_hosts.map(({host:e,count:t})=>`${e} (${t})`).join(`, `)})]},e.attempt_id))})]})})]}),(0,S.jsxs)(`div`,{className:`methodology`,"data-testid":`methodology`,children:[(0,S.jsx)(`strong`,{children:`Methodology`}),`: `,r.methodology]})]})}function qt(){return(0,S.jsxs)(`div`,{className:`empty-state`,"data-testid":`empty-state`,children:[(0,S.jsx)(`h2`,{children:`No runs in this data directory yet`}),(0,S.jsx)(`p`,{children:`Execute a sweep, then re-export (or restart) the UI:`}),(0,S.jsx)(`pre`,{children:(0,S.jsxs)(`code`,{children:[`quickstudy run --eval my-eval --experiment my-experiment`,`
 `,`quickstudy ui`]})}),(0,S.jsx)(`p`,{children:`The matrix, compare, report, and diagnosis views light up as soon as a run exists.`})]})}function Jt({route:e,meta:t}){let n=t.runs[0]?.id,r=e.view===`matrix`||e.view===`attempt`;return(0,S.jsxs)(`nav`,{className:`tabs`,"aria-label":`Views`,children:[(0,S.jsx)(`a`,{href:T.matrix(),className:r?`tab active`:`tab`,"aria-current":r?`page`:void 0,children:`Matrix`}),(0,S.jsx)(`a`,{href:T.compare(),className:e.view===`compare`?`tab active`:`tab`,"aria-current":e.view===`compare`?`page`:void 0,children:`Compare`}),(0,S.jsx)(`a`,{href:T.report(n),className:e.view===`report`?`tab active`:`tab`,"aria-current":e.view===`report`?`page`:void 0,children:`Report`}),(0,S.jsx)(`a`,{href:T.diagnosis(n),className:e.view===`diagnosis`?`tab active`:`tab`,"aria-current":e.view===`diagnosis`?`page`:void 0,children:`Diagnosis`})]})}function Yt({route:e,meta:t}){let n=t.runs[0]?.id;switch(e.view){case`matrix`:return(0,S.jsx)(Lt,{meta:t,runId:e.runId??n??``,cell:e.cell});case`attempt`:return(0,S.jsx)(ke,{runId:e.runId,attemptId:e.attemptId,artifact:e.artifact,lines:e.lines});case`compare`:return(0,S.jsx)(We,{meta:t,runId:e.runId??n});case`report`:return(0,S.jsx)(Kt,{meta:t,runId:e.runId??n??``});case`diagnosis`:return(0,S.jsx)(lt,{meta:t,runId:e.runId??n??``})}}function Xt(){let e=oe(),t=C(()=>h(),[]),n=(e.view===`matrix`||e.view===`compare`)&&e.legacyLink===!0;return(0,S.jsxs)(`div`,{className:`app`,children:[(0,S.jsxs)(`header`,{className:`topbar`,children:[(0,S.jsxs)(`div`,{className:`brand`,children:[(0,S.jsx)(`span`,{className:`brand-mark`,"aria-hidden":`true`}),`quickstudy`]}),t.status===`ready`&&(0,S.jsx)(Jt,{route:e,meta:t.value})]}),(0,S.jsxs)(`main`,{className:`content`,children:[n&&(0,S.jsx)(`div`,{className:`notice`,"data-testid":`legacy-link-notice`,children:`This link uses a URL format from an earlier quickstudy version — showing the closest current view.`}),t.status===`loading`&&(0,S.jsx)(`div`,{className:`loading`,children:`Loading data index…`}),t.status===`error`&&(0,S.jsx)(ee,{error:t.error}),t.status===`ready`&&(t.value.runs.length===0?(0,S.jsx)(qt,{}):(0,S.jsx)(te,{children:(0,S.jsx)(Yt,{route:e,meta:t.value})},JSON.stringify(e)))]})]})}var Zt=document.getElementById(`root`);if(Zt===null)throw Error(`missing #root element`);(0,u.createRoot)(Zt).render((0,S.jsx)(l.StrictMode,{children:(0,S.jsx)(Xt,{})}));</script>
     <style>.quickstudy-viewer.viewer{background:var(--surface);border:1px solid var(--line);border-radius:10px;overflow:hidden}.quickstudy-viewer .turn{border-bottom:1px solid var(--line)}.quickstudy-viewer .turn-head{width:100%;font:inherit;text-align:left;background:var(--surface);cursor:pointer;border:0;align-items:center;gap:8px;padding:6px 12px;display:flex}.quickstudy-viewer .turn-head:hover{background:var(--bg)}.quickstudy-viewer .turn-head .line-no{font-family:var(--mono);color:var(--muted);min-width:46px;font-size:11px}.quickstudy-viewer .turn-head .role{text-transform:uppercase;letter-spacing:.04em;min-width:72px;font-size:11px;font-weight:700}.quickstudy-viewer .role-assistant{color:var(--accent)}.quickstudy-viewer .role-user{color:var(--green)}.quickstudy-viewer .role-tool{color:var(--amber)}.quickstudy-viewer .role-system{color:var(--muted)}.quickstudy-viewer .turn-head .summary{color:var(--muted);white-space:nowrap;text-overflow:ellipsis;flex:1;font-size:12px;overflow:hidden}.quickstudy-viewer .turn-body{background:var(--bg);padding:8px 12px 12px 70px}.quickstudy-viewer .turn-body pre{white-space:pre-wrap;word-break:break-word;margin:0}.quickstudy-viewer .turn.highlight,.quickstudy-viewer .turn.highlight .turn-head{background:var(--amber-bg)}.quickstudy-viewer .viewer-note{color:var(--muted);border-bottom:1px solid var(--line);justify-content:space-between;align-items:center;gap:12px;padding:10px 14px;font-size:12px;display:flex}.quickstudy-viewer .viewer-note button{font:inherit;border:1px solid var(--line);background:var(--surface);cursor:pointer;border-radius:6px;padding:3px 10px}.quickstudy-viewer.diff{font-family:var(--mono);font-size:12px;line-height:1.45;overflow-x:auto}.quickstudy-viewer .diff-line{white-space:pre;display:flex}.quickstudy-viewer .diff-line .line-no{min-width:46px;color:var(--muted);text-align:right;-webkit-user-select:none;user-select:none;flex-shrink:0;padding:0 8px}.quickstudy-viewer .diff-line .text{flex:1;padding-right:12px}.quickstudy-viewer .diff-add{background:var(--green-bg)}.quickstudy-viewer .diff-del{background:var(--red-bg)}.quickstudy-viewer .diff-hunk{background:var(--gray-bg);color:var(--accent)}.quickstudy-viewer .diff-file{background:var(--gray-bg);font-weight:700}.quickstudy-viewer .diff-line.highlight{outline:2px solid var(--amber);outline-offset:-2px}:root{--bg:#f7f7f5;--surface:#fff;--ink:#1c1e21;--muted:#6b7280;--line:#e3e4e0;--accent:#4c5fd5;--green:#1a7f37;--green-bg:#d3f2dd;--amber:#9a6700;--amber-bg:#fff1c2;--red:#b42318;--red-bg:#ffe0dc;--gray-bg:#eceef0;--accent-bg:#e2e6f9;--mono:ui-monospace, "SF Mono", SFMono-Regular, Menlo, Consolas, monospace}*{box-sizing:border-box}body{background:var(--bg);color:var(--ink);margin:0;font:14px/1.5 -apple-system,BlinkMacSystemFont,Segoe UI,Roboto,sans-serif}code,pre{font-family:var(--mono);font-size:12px}a{color:var(--accent);text-decoration:none}a:hover{text-decoration:underline}:where(a,button,select,[tabindex]):focus-visible{outline:3px solid var(--accent);outline-offset:2px}.app{flex-direction:column;min-height:100vh;display:flex}.topbar{background:var(--surface);border-bottom:1px solid var(--line);z-index:20;align-items:center;gap:24px;padding:10px 20px;display:flex;position:sticky;top:0}.brand{letter-spacing:-.01em;align-items:center;gap:8px;font-weight:700;display:flex}.brand-mark{background:linear-gradient(135deg, var(--accent), var(--green));border-radius:3px;width:12px;height:12px}.tabs{gap:4px;display:flex}.tab{color:var(--muted);border-radius:6px;padding:5px 12px;font-weight:500}.tab:hover{color:var(--ink);text-decoration:none}.tab.active{background:var(--gray-bg);color:var(--ink)}.content{flex:1;width:100%;max-width:1200px;margin:0 auto;padding:24px 20px 64px}.loading{color:var(--muted);text-align:center;padding:48px 0}.error-banner{background:var(--red-bg);border:1px solid var(--red);color:var(--red);border-radius:8px;padding:14px 16px}.error-panel{background:var(--surface);border:1px solid var(--red);border-radius:8px;gap:6px;padding:16px;display:grid}.empty-state{background:var(--surface);border:1px dashed var(--line);text-align:center;border-radius:10px;padding:40px}.empty-state pre{text-align:left;background:var(--gray-bg);border-radius:8px;padding:12px 16px;display:inline-block}.notice{background:var(--amber-bg);border:1px solid var(--amber);color:var(--amber);border-radius:8px;margin-bottom:16px;padding:10px 14px}.page-head{flex-wrap:wrap;justify-content:space-between;align-items:flex-end;gap:16px;margin-bottom:16px;display:flex}.page-head h1{letter-spacing:-.01em;margin:0 0 2px;font-size:20px}.page-head .sub{color:var(--muted);margin:0}.controls{flex-wrap:wrap;align-items:center;gap:12px;display:flex}.control label{text-transform:uppercase;letter-spacing:.04em;color:var(--muted);margin-bottom:2px;font-size:11px;display:block}select{font:inherit;border:1px solid var(--line);background:var(--surface);border-radius:6px;max-width:320px;padding:5px 8px}.seg{border:1px solid var(--line);background:var(--surface);border-radius:7px;display:inline-flex;overflow:hidden}.seg button{font:inherit;cursor:pointer;color:var(--muted);background:0 0;border:0;padding:5px 12px}.seg button+button{border-left:1px solid var(--line)}.seg button.active{background:var(--ink);color:#fff}.kpi-row{grid-template-columns:repeat(auto-fit,minmax(180px,1fr));gap:12px;margin-bottom:16px;display:grid}.kpi{background:var(--surface);border:1px solid var(--line);border-radius:10px;padding:12px 14px}.kpi-label{text-transform:uppercase;letter-spacing:.04em;color:var(--muted);font-size:11px}.kpi-val{letter-spacing:-.02em;font-size:22px;font-weight:700}.kpi-meta{color:var(--muted);font-size:12px}.badge{white-space:nowrap;border-radius:999px;align-items:center;gap:5px;padding:1px 9px;font-size:12px;font-weight:600;display:inline-flex}.badge .dot{background:currentColor;border-radius:999px;width:6px;height:6px}.badge.green{background:var(--green-bg);color:var(--green)}.badge.red{background:var(--red-bg);color:var(--red)}.badge.amber{background:var(--amber-bg);color:var(--amber)}.badge.gray{background:var(--gray-bg);color:var(--muted)}.badge.blue{background:var(--accent-bg);color:var(--accent)}.card{background:var(--surface);border:1px solid var(--line);border-radius:10px;margin-bottom:16px}.card>.head{border-bottom:1px solid var(--line);flex-wrap:wrap;justify-content:space-between;align-items:center;gap:12px;padding:12px 16px;display:flex}.card>.head h3{margin:0;font-size:14px}.card>.body{padding:16px;overflow-x:auto}.legend{color:var(--muted);align-items:center;gap:4px;font-size:11px;display:inline-flex}.legend-sw{border-radius:3px;width:22px;height:12px;display:inline-block}.matrix-scroll{padding:8px 16px 16px;overflow-x:auto}.band+.band{border-top:1px solid var(--line)}.band>h4,details.band>summary{margin:0;padding:10px 16px 0;font-size:13px}details.band>summary{cursor:pointer;padding-bottom:10px}.band-notice{margin:12px 16px 0}.band-notice ul{margin:6px 0 0;padding-left:18px}.scroll-hint{color:var(--muted);padding:8px 16px 0;font-size:11px;display:none}table.matrix{border-collapse:separate;border-spacing:4px}table.matrix th{color:var(--muted);text-align:left;white-space:nowrap;padding:6px 8px;font-size:12px;font-weight:600}.matrix-row-feature{color:var(--muted);text-transform:none;letter-spacing:0;margin-top:2px;font-size:10px;font-weight:400;display:block}table.matrix th.matrix-group-head{color:var(--ink);font-size:13px}.matrix-group-sub,.group-sub{color:var(--muted);margin-top:2px;font-size:10px;font-weight:400;display:block}.muted{color:var(--muted);font-size:12px}h3 .group-sub{margin-left:6px;font-size:12px;display:inline}table.matrix td.cell{border-radius:8px;min-width:110px;padding:0;transition:box-shadow .1s;position:relative}.cell-button{appearance:none;border-radius:inherit;width:100%;min-height:78px;color:inherit;font:inherit;text-align:left;cursor:pointer;background:0 0;border:0;flex-direction:column;align-items:flex-start;padding:8px 10px;display:flex;position:relative}table.matrix td.cell:hover,table.matrix td.cell:focus-within{outline:2px solid var(--accent);outline-offset:1px}.cell-button:focus-visible{outline:0}.cell-rate{letter-spacing:-.02em;font-size:18px;font-weight:700;display:block}.cell-sub{opacity:.75;font-size:11px;display:block}.agent-dots{gap:4px;margin-top:4px;display:flex}@media (width<=640px){.topbar{gap:8px;padding-inline:12px;overflow-x:auto}.brand,.tabs{flex:none}.content{padding:16px 12px 48px}.controls,.control,.control select{width:100%}.seg{max-width:100%;overflow-x:auto}.scroll-hint{display:block}.matrix-scroll{padding-inline:12px}table.data{min-width:640px}}.agent-dot{border:1px solid #00000040;border-radius:999px;width:9px;height:9px}.flaky-marker{border-top:12px solid var(--amber);border-left:12px solid #0000;border-top-right-radius:8px;width:0;height:0;position:absolute;top:0;right:0}.incomplete-marker{font-size:10px;font-weight:700;position:absolute;bottom:4px;right:6px}.drill-overlay{z-index:30;background:#14161a59;position:fixed;inset:0}.drill{background:var(--surface);border-left:1px solid var(--line);z-index:31;width:min(680px,92vw);padding:20px;position:fixed;top:0;bottom:0;right:0;overflow-y:auto;box-shadow:-12px 0 40px #00000026}.drill h2{margin:0 0 4px;font-size:17px}.drill .sub{color:var(--muted);margin:0 0 16px}.drill-close{border:1px solid var(--line);background:var(--surface);cursor:pointer;font:inherit;border-radius:6px;padding:4px 10px;position:absolute;top:14px;right:14px}.drill-agent{border:1px solid var(--line);border-radius:8px;margin-bottom:12px}.drill-agent>.head{border-bottom:1px solid var(--line);align-items:center;gap:10px;padding:10px 12px;font-weight:600;display:flex}.attempt-row{gap:6px;padding:10px 12px;display:grid}.attempt-row+.attempt-row{border-top:1px solid var(--line)}.attempt-line{flex-wrap:wrap;align-items:center;gap:8px;display:flex}.attempt-line .spacer{flex:1}.crit{border-radius:5px;align-items:center;gap:4px;padding:1px 7px;font-size:11px;display:inline-flex}.crit.pass{background:var(--green-bg);color:var(--green)}.crit.fail{background:var(--red-bg);color:var(--red)}.tag{background:var(--gray-bg);color:var(--muted);border-radius:4px;padding:0 5px;font-size:10px}.telemetry{color:var(--muted);flex-wrap:wrap;gap:12px;font-size:12px;display:flex}.judge-scores{color:var(--muted);font-size:12px}.surface-usage{gap:5px;font-size:12px;display:grid}.usage-badges{flex-wrap:wrap;gap:5px;display:flex}.help-tip{border:1px solid var(--muted);width:15px;height:15px;color:var(--muted);cursor:help;-webkit-user-select:none;user-select:none;border-radius:50%;justify-content:center;align-self:center;align-items:center;font-size:10px;line-height:1;display:inline-flex}.help-tip:hover,.help-tip:focus-visible{color:var(--ink);border-color:var(--ink)}.usage-values{color:var(--muted)}.usage-values summary{cursor:pointer;width:max-content}.usage-values ul{margin:5px 0 0;padding-left:22px}.usage-values code{overflow-wrap:anywhere}table.data{border-collapse:collapse;width:100%}table.data th,table.data td{text-align:left;border-bottom:1px solid var(--line);font-variant-numeric:tabular-nums;padding:7px 10px}table.data th{text-transform:uppercase;letter-spacing:.04em;color:var(--muted);font-size:11px}td.delta-pos{background:var(--green-bg);color:var(--green);font-weight:700}td.delta-neg{background:var(--red-bg);color:var(--red);font-weight:700}td.delta-zero{color:var(--muted)}.identity-controls{border:1px solid var(--line);background:var(--surface);border-radius:8px;margin:0 0 16px;padding:12px 14px}.identity-controls legend{padding:0 5px;font-weight:700}.identity-controls p{color:var(--muted);margin:0 0 10px}.identity-choices{gap:6px;display:grid}.identity-choice{cursor:pointer;align-items:center;gap:8px;display:flex}.identity-choice>span{color:var(--muted);font-size:12px}.identity-choice.disabled{cursor:not-allowed;opacity:.7}.chip-row{flex-wrap:wrap;gap:6px;margin-bottom:12px;display:flex}.chip{font:inherit;border:1px solid var(--line);background:var(--surface);cursor:pointer;color:var(--muted);border-radius:999px;padding:3px 11px;font-size:12px}.chip.active{background:var(--ink);border-color:var(--ink);color:#fff}.rec-card{border:1px solid var(--line);background:var(--surface);border-radius:10px;margin-bottom:12px;padding:14px 16px}.rec-card h3{margin:0 0 6px;font-size:14px}.rec-meta{flex-wrap:wrap;gap:6px;margin-bottom:8px;display:flex}.attribution-arms,.attribution-effects{gap:8px;min-width:270px;display:grid}.attribution-arm,.attribution-effect{border:1px solid var(--line);background:var(--gray-bg);border-radius:7px;gap:5px;padding:8px;display:grid}.attribution-arm-head,.attribution-effect-head{flex-wrap:wrap;align-items:center;gap:7px;display:flex}.attribution-counts,.attribution-interval,.attribution-reason{color:var(--muted);font-size:12px}.evidence-row{flex-wrap:wrap;gap:6px;margin-top:8px;display:flex}.evidence-chip{font-size:12px;font-family:var(--mono);border:1px solid var(--line);background:var(--gray-bg);border-radius:6px;padding:2px 8px}.methodology{color:var(--muted);border-top:1px solid var(--line);margin-top:24px;padding-top:12px;font-size:12px}.viewer-head{flex-wrap:wrap;justify-content:space-between;align-items:baseline;gap:12px;margin-bottom:12px;display:flex}.copy-button{font:inherit;border:1px solid var(--line);background:var(--surface);cursor:pointer;border-radius:6px;padding:3px 10px;font-size:12px}.head-actions{align-items:center;gap:8px;display:inline-flex}.toolbar{justify-content:flex-end;margin-bottom:12px;display:flex}.cluster{margin-bottom:24px}
 </style>