@agentdeck/bridge 0.2.2 → 1.0.0

package/dist/apme/index.js

@@ -13,43 +13,175 @@
 import { writeFileSync } from 'fs';
 import { join } from 'path';
 import { homedir } from 'os';
-import { debug } from '../logger.js';
+import { logError } from '../logger.js';
 import { ApmeStore } from './store.js';
 import { apmeDashboardHtml } from './dashboard-html.js';
 import { ApmeCollector } from './collector.js';
 import { ApmeRunner } from './runner.js';
-import { ApmeTuner } from './tuner.js';
 import { ApmeHwSampler } from './hw-sampler.js';
 import { ApmeRecommender } from './recommend.js';
+import { sampleEventToTimeline } from './sample-to-timeline.js';
+import { formatDurationSec } from '@agentdeck/shared';
+/** Task boundary rows read "<signal> · <N> turns · <duration>" — the turn
+ *  count says what the boundary covered, the human duration replaces the
+ *  raw "3672s" second counts. */
+function taskEndRowText(signalLabel, turns, durationSec) {
+    const parts = [signalLabel];
+    if (turns && turns > 0)
+        parts.push(turns === 1 ? '1 turn' : `${turns} turns`);
+    parts.push(formatDurationSec(durationSec));
+    return parts.join(' · ');
+}
 let singleton = null;
-/** Initialize the APME subsystem. Returns null if the SQLite store can't open. */
-export async function initApme(dbPath) {
+/** Initialize the APME subsystem. Returns null if the SQLite store can't open.
+ *  When null, the failure reason is logged at ERROR level — silent disable was
+ *  the dominant cause of multi-day data outages observed in user telemetry. */
+export async function initApme(dbPath, opts = {}) {
     if (singleton)
         return singleton;
     const store = new ApmeStore(dbPath);
     const ok = await store.init();
     if (!ok) {
-        debug('APME', 'initApme skipped (store disabled)');
+        logError(`APME disabled — ${store.lastInitError ?? 'unknown init failure'}. Agent runs will not be measured.`);
         return null;
     }
     const hwSampler = new ApmeHwSampler();
     const collector = new ApmeCollector(store, hwSampler);
     const runner = new ApmeRunner(store);
-    const tuner = new ApmeTuner(store);
     const recommender = new ApmeRecommender(store);
-    singleton = { store, collector, runner, tuner, hwSampler, recommender };
+    singleton = { store, collector, runner, hwSampler, recommender };
+    const emitTimeline = opts.emitTimeline;
+    // Task headers are DEFERRED (collector emits `task_start` only once a task
+    // proves itself multi-turn). Track which task ids actually reached the
+    // timeline so the async judge re-emit (`onTaskEvaluated`, below) can skip
+    // unpromoted single-turn tasks — otherwise it would insert an orphan
+    // `task_end` with a score badge under no header. `onTaskOpened` fires at the
+    // deferred promotion moment, so membership here == "header is on the timeline".
+    const promotedTaskIds = new Set();
     // Wire task-level judge: whenever the collector closes a task (TodoWrite
     // all-completed / /clear / session_end), enqueue a task_rollup judge call.
     // Kept here — not in the collector constructor — so the collector has no
     // hard dependency on the runner.
-    collector.onTaskClosed = ({ taskId, runId, boundarySignal, taskCategory }) => {
+    collector.onTaskClosed = ({ taskId, runId, sessionId, agentType, projectName, startedAt, endedAt, boundarySignal, taskCategory, turns, timelineEmitted, }) => {
         runner.enqueueTask({
             runId,
             taskId,
             category: taskCategory ?? undefined,
             boundarySignal,
         });
+        // Gate the `task_end` row on the header having been emitted — an
+        // unpromoted single-turn task shows only its chat rows, never a bare
+        // "Task end" marker. The eval enqueue above runs regardless.
+        if (emitTimeline && timelineEmitted) {
+            const durationSec = Math.max(0, Math.round((endedAt - startedAt) / 1000));
+            const signalLabel = boundarySignal === 'todo_complete' ? 'TODO done'
+                : boundarySignal === 'clear' ? '/clear'
+                    : boundarySignal === 'session_end' ? 'Session end'
+                        : 'Task end';
+            emitTimeline({
+                ts: endedAt,
+                type: 'task_end',
+                raw: taskEndRowText(signalLabel, turns, durationSec),
+                agentType: agentType ?? undefined,
+                projectName: projectName ?? undefined,
+                sessionId,
+                runId,
+                taskId,
+                boundarySignal,
+                startedAt,
+                endedAt,
+            });
+        }
     };
+    // Wire the SessionSample → Timeline projection. When enabled, the timeline
+    // is a projection of the canonical sample (one row per typed trajectory
+    // event), replacing display-time grouping + the race-sensitive dedup window.
+    if (emitTimeline && opts.projectSampleTimeline) {
+        const emitProjected = opts.emitProjectedTimeline ?? emitTimeline;
+        collector.onSampleEvent = ({ event, sessionId, runId, taskId, agentType, projectName }) => {
+            const entry = sampleEventToTimeline(event, { sessionId, runId, taskId, agentType, projectName });
+            if (entry)
+                emitProjected(entry);
+        };
+    }
+    // Wire task-start emission. The runner has no opinion on opens — only the
+    // dashboard does — so we keep this strictly local to the timeline path.
+    if (emitTimeline) {
+        collector.onTaskOpened = ({ taskId, runId, sessionId, agentType, projectName, taskIndex, startedAt, }) => {
+            promotedTaskIds.add(taskId);
+            emitTimeline({
+                ts: startedAt,
+                type: 'task_start',
+                raw: `Task ${taskIndex + 1}`,
+                agentType: agentType ?? undefined,
+                projectName: projectName ?? undefined,
+                sessionId,
+                runId,
+                taskId,
+                startedAt,
+            });
+        };
+    }
+    // Wire the mid-task completion milestone. TodoWrite-all-completed is a
+    // non-segmenting soft hint (fires ~18% of the time on Claude Code v2.1), so
+    // it must not split the task — but when it DOES fire it is a strong "the
+    // agent declared this chunk done" signal, and long sessions otherwise show
+    // no completion marks until /clear or session_end.
+    if (emitTimeline) {
+        collector.onTaskMilestone = ({ taskId, runId, sessionId, agentType, projectName, todoCount, at, }) => {
+            emitTimeline({
+                ts: at,
+                type: 'task_milestone',
+                raw: todoCount ? `Todos done (${todoCount})` : 'Todos done',
+                agentType: agentType ?? undefined,
+                projectName: projectName ?? undefined,
+                sessionId,
+                runId,
+                taskId,
+            });
+        };
+    }
+    // Wire task-eval completion → re-emit `task_end` with score + outcome.
+    // The original `task_end` (above) fires synchronously on closeTask so the
+    // boundary is visible immediately; the judge runs async and may take
+    // 5-30 s. When it resolves, this listener emits a second `task_end` with
+    // the same `taskId`. Timeline stores upsert by (type='task_end', taskId)
+    // and merge the new score fields onto the existing row, so dashboard task
+    // headers get the score badge as soon as the judge returns.
+    if (emitTimeline) {
+        runner.onTaskEvaluated((e) => {
+            // Skip the score re-emit for tasks whose header never reached the
+            // timeline (single-turn Q&A). Without this the merge-or-insert
+            // timeline store would insert a fresh orphan `task_end` here. Consume
+            // the id so the set doesn't grow unbounded across a long-lived daemon.
+            if (!promotedTaskIds.delete(e.taskId))
+                return;
+            const durationSec = Math.max(0, Math.round((e.endedAt - e.startedAt) / 1000));
+            const signalLabel = e.boundarySignal === 'todo_complete' ? 'TODO done'
+                : e.boundarySignal === 'clear' ? '/clear'
+                    : e.boundarySignal === 'session_end' ? 'Session end'
+                        : e.boundarySignal === 'manual' ? 'Manual'
+                            : e.boundarySignal === 'idle_gap' ? 'Idle gap'
+                                : 'Task end';
+            emitTimeline({
+                ts: e.endedAt,
+                type: 'task_end',
+                raw: taskEndRowText(signalLabel, e.turns, durationSec),
+                agentType: e.agentType,
+                projectName: e.projectName,
+                sessionId: e.sessionId,
+                runId: e.runId,
+                taskId: e.taskId,
+                boundarySignal: e.boundarySignal,
+                startedAt: e.startedAt,
+                endedAt: e.endedAt,
+                taskScore: e.compositeScore ?? undefined,
+                taskOutcome: e.outcome,
+                taskCategory: e.taskCategory,
+                taskSummary: e.summary,
+            });
+        });
+    }
     // Write dashboard HTML for Swift daemon to pick up.
     try {
         const dataDir = process.env.AGENTDECK_DATA_DIR || join(homedir(), '.agentdeck');
@@ -63,11 +195,16 @@ export async function initApme(dbPath) {
 export function getApme() {
     return singleton;
 }
+/** Phase 6 cutover switch (default OFF). Set `AGENTDECK_TIMELINE_PROJECTION=1`
+ *  to make the timeline a projection of the canonical SessionSample. Read at
+ *  daemon/session-bridge startup so it can be flipped without a recompile. */
+export function isTimelineProjectionEnabled() {
+    return process.env.AGENTDECK_TIMELINE_PROJECTION === '1';
+}
 export { loadApmeConfig, shouldJudge, DEFAULT_APME_CONFIG } from './settings.js';
 export { ApmeStore } from './store.js';
 export { ApmeCollector } from './collector.js';
 export { ApmeRunner } from './runner.js';
-export { ApmeTuner } from './tuner.js';
 export { ApmeHwSampler } from './hw-sampler.js';
 export { ApmeRecommender } from './recommend.js';
 export { classifyRun, classifyRunSmart, classifyWithLlm, classify, computeSignals, TASK_CATEGORIES } from './classifier.js';

package/dist/apme/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/apme/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AACnC,OAAO,EAAE,IAAI,EAAE,MAAM,MAAM,CAAC;AAC5B,OAAO,EAAE,OAAO,EAAE,MAAM,IAAI,CAAC;AAC7B,OAAO,EAAE,KAAK,EAAE,MAAM,cAAc,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,EAAE,iBAAiB,EAAE,MAAM,qBAAqB,CAAC;AACxD,OAAO,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AAWjD,IAAI,SAAS,GAAsB,IAAI,CAAC;AAExC,kFAAkF;AAClF,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,MAAe;IAC5C,IAAI,SAAS;QAAE,OAAO,SAAS,CAAC;IAChC,MAAM,KAAK,GAAG,IAAI,SAAS,CAAC,MAAM,CAAC,CAAC;IACpC,MAAM,EAAE,GAAG,MAAM,KAAK,CAAC,IAAI,EAAE,CAAC;IAC9B,IAAI,CAAC,EAAE,EAAE,CAAC;QACR,KAAK,CAAC,MAAM,EAAE,mCAAmC,CAAC,CAAC;QACnD,OAAO,IAAI,CAAC;IACd,CAAC;IACD,MAAM,SAAS,GAAG,IAAI,aAAa,EAAE,CAAC;IACtC,MAAM,SAAS,GAAG,IAAI,aAAa,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IACtD,MAAM,MAAM,GAAG,IAAI,UAAU,CAAC,KAAK,CAAC,CAAC;IACrC,MAAM,KAAK,GAAG,IAAI,SAAS,CAAC,KAAK,CAAC,CAAC;IACnC,MAAM,WAAW,GAAG,IAAI,eAAe,CAAC,KAAK,CAAC,CAAC;IAC/C,SAAS,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,CAAC;IAExE,yEAAyE;IACzE,2EAA2E;IAC3E,yEAAyE;IACzE,iCAAiC;IACjC,SAAS,CAAC,YAAY,GAAG,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,cAAc,EAAE,YAAY,EAAE,EAAE,EAAE;QAC3E,MAAM,CAAC,WAAW,CAAC;YACjB,KAAK;YACL,MAAM;YACN,QAAQ,EAAE,YAAY,IAAI,SAAS;YACnC,cAAc;SACf,CAAC,CAAC;IACL,CAAC,CAAC;IAEF,oDAAoD;IACpD,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,kBAAkB,IAAI,IAAI,CAAC,OAAO,EAAE,EAAE,YAAY,CAAC,CAAC;QAChF,aAAa,CAAC,IAAI,CAAC,OAAO,EAAE,qBAAqB,CAAC,EAAE,iBAAiB,EAAE,EAAE,OAAO,CAAC,CAAC;IACpF,CAAC;IAAC,MAAM,CAAC,CAAC,iBAAiB,CAAC,CAAC;IAE7B,sEAAsE;IACtE,8DAA8D;IAE9D,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,UAAU,OAAO;IACrB,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,OAAO,EAAE,cAAc,EAAE,WAAW,EAAE,mBAAmB,EAAE,MAAM,eAAe,CAAC;AAEjF,OAAO,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EAAE,WAAW,EAAE,gBAAgB,EAAE,eAAe,EAAE,QAAQ,EAAE,cAAc,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAC5H,OAAO,EAAE,eAAe,EAAE,aAAa,EAAE,iBAAiB,EAAE,gBAAgB,EAAE,MAAM,cAAc,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/apme/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AACnC,OAAO,EAAE,IAAI,EAAE,MAAM,MAAM,CAAC;AAC5B,OAAO,EAAE,OAAO,EAAE,MAAM,IAAI,CAAC;AAC7B,OAAO,EAAE,QAAQ,EAAE,MAAM,cAAc,CAAC;AACxC,OAAO,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,EAAE,iBAAiB,EAAE,MAAM,qBAAqB,CAAC;AACxD,OAAO,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EAAE,qBAAqB,EAAE,MAAM,yBAAyB,CAAC;AAChE,OAAO,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AAGtD;;iCAEiC;AACjC,SAAS,cAAc,CAAC,WAAmB,EAAE,KAAyB,EAAE,WAAmB;IACzF,MAAM,KAAK,GAAG,CAAC,WAAW,CAAC,CAAC;IAC5B,IAAI,KAAK,IAAI,KAAK,GAAG,CAAC;QAAE,KAAK,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,KAAK,QAAQ,CAAC,CAAC;IAC9E,KAAK,CAAC,IAAI,CAAC,iBAAiB,CAAC,WAAW,CAAC,CAAC,CAAC;IAC3C,OAAO,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;AAC3B,CAAC;AA4BD,IAAI,SAAS,GAAsB,IAAI,CAAC;AAExC;;+EAE+E;AAC/E,MAAM,CAAC,KAAK,UAAU,QAAQ,CAC5B,MAAe,EACf,OAAwB,EAAE;IAE1B,IAAI,SAAS;QAAE,OAAO,SAAS,CAAC;IAChC,MAAM,KAAK,GAAG,IAAI,SAAS,CAAC,MAAM,CAAC,CAAC;IACpC,MAAM,EAAE,GAAG,MAAM,KAAK,CAAC,IAAI,EAAE,CAAC;IAC9B,IAAI,CAAC,EAAE,EAAE,CAAC;QACR,QAAQ,CAAC,mBAAmB,KAAK,CAAC,aAAa,IAAI,sBAAsB,oCAAoC,CAAC,CAAC;QAC/G,OAAO,IAAI,CAAC;IACd,CAAC;IACD,MAAM,SAAS,GAAG,IAAI,aAAa,EAAE,CAAC;IACtC,MAAM,SAAS,GAAG,IAAI,aAAa,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IACtD,MAAM,MAAM,GAAG,IAAI,UAAU,CAAC,KAAK,CAAC,CAAC;IACrC,MAAM,WAAW,GAAG,IAAI,eAAe,CAAC,KAAK,CAAC,CAAC;IAC/C,SAAS,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,CAAC;IAEjE,MAAM,YAAY,GAAG,IAAI,CAAC,YAAY,CAAC;IAEvC,2EAA2E;IAC3E,uEAAuE;IACvE,0EAA0E;IAC1E,qEAAqE;IACrE,6EAA6E;IAC7E,gFAAgF;IAChF,MAAM,eAAe,GAAG,IAAI,GAAG,EAAU,CAAC;IAE1C,yEAAyE;IACzE,2EAA2E;IAC3E,yEAAyE;IACzE,iCAAiC;IACjC,SAAS,CAAC,YAAY,GAAG,CAAC,EACxB,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,WAAW,EAChD,SAAS,EAAE,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,KAAK,EAAE,eAAe,GACzE,EAAE,EAAE;QACH,MAAM,CAAC,WAAW,CAAC;YACjB,KAAK;YACL,MAAM;YACN,QAAQ,EAAE,YAAY,IAAI,SAAS;YACnC,cAAc;SACf,CAAC,CAAC;QAEH,iEAAiE;QACjE,qEAAqE;QACrE,6DAA6D;QAC7D,IAAI,YAAY,IAAI,eAAe,EAAE,CAAC;YACpC,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,OAAO,GAAG,SAAS,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC;YAC1E,MAAM,WAAW,GAAG,cAAc,KAAK,eAAe,CAAC,CAAC,CAAC,WAAW;gBAClE,CAAC,CAAC,cAAc,KAAK,OAAO,CAAC,CAAC,CAAC,QAAQ;oBACvC,CAAC,CAAC,cAAc,KAAK,aAAa,CAAC,CAAC,CAAC,aAAa;wBAClD,CAAC,CAAC,UAAU,CAAC;YACf,YAAY,CAAC;gBACX,EAAE,EAAE,OAAO;gBACX,IAAI,EAAE,UAAU;gBAChB,GAAG,EAAE,cAAc,CAAC,WAAW,EAAE,KAAK,EAAE,WAAW,CAAC;gBACpD,SAAS,EAAE,SAAS,IAAI,SAAS;gBACjC,WAAW,EAAE,WAAW,IAAI,SAAS;gBACrC,SAAS;gBACT,KAAK;gBACL,MAAM;gBACN,cAAc;gBACd,SAAS;gBACT,OAAO;aACR,CAAC,CAAC;QACL,CAAC;IACH,CAAC,CAAC;IAEF,2EAA2E;IAC3E,wEAAwE;IACxE,6EAA6E;IAC7E,IAAI,YAAY,IAAI,IAAI,CAAC,qBAAqB,EAAE,CAAC;QAC/C,MAAM,aAAa,GAAG,IAAI,CAAC,qBAAqB,IAAI,YAAY,CAAC;QACjE,SAAS,CAAC,aAAa,GAAG,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,EAAE,EAAE;YACxF,MAAM,KAAK,GAAG,qBAAqB,CAAC,KAAK,EAAE,EAAE,SAAS,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,CAAC,CAAC;YACjG,IAAI,KAAK;gBAAE,aAAa,CAAC,KAAK,CAAC,CAAC;QAClC,CAAC,CAAC;IACJ,CAAC;IAED,0EAA0E;IAC1E,wEAAwE;IACxE,IAAI,YAAY,EAAE,CAAC;QACjB,SAAS,CAAC,YAAY,GAAG,CAAC,EACxB,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,WAAW,EAAE,SAAS,EAAE,SAAS,GACvE,EAAE,EAAE;YACH,eAAe,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;YAC5B,YAAY,CAAC;gBACX,EAAE,EAAE,SAAS;gBACb,IAAI,EAAE,YAAY;gBAClB,GAAG,EAAE,QAAQ,SAAS,GAAG,CAAC,EAAE;gBAC5B,SAAS,EAAE,SAAS,IAAI,SAAS;gBACjC,WAAW,EAAE,WAAW,IAAI,SAAS;gBACrC,SAAS;gBACT,KAAK;gBACL,MAAM;gBACN,SAAS;aACV,CAAC,CAAC;QACL,CAAC,CAAC;IACJ,CAAC;IAED,uEAAuE;IACvE,4EAA4E;IAC5E,yEAAyE;IACzE,2EAA2E;IAC3E,mDAAmD;IACnD,IAAI,YAAY,EAAE,CAAC;QACjB,SAAS,CAAC,eAAe,GAAG,CAAC,EAC3B,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,WAAW,EAAE,SAAS,EAAE,EAAE,GAChE,EAAE,EAAE;YACH,YAAY,CAAC;gBACX,EAAE,EAAE,EAAE;gBACN,IAAI,EAAE,gBAAgB;gBACtB,GAAG,EAAE,SAAS,CAAC,CAAC,CAAC,eAAe,SAAS,GAAG,CAAC,CAAC,CAAC,YAAY;gBAC3D,SAAS,EAAE,SAAS,IAAI,SAAS;gBACjC,WAAW,EAAE,WAAW,IAAI,SAAS;gBACrC,SAAS;gBACT,KAAK;gBACL,MAAM;aACP,CAAC,CAAC;QACL,CAAC,CAAC;IACJ,CAAC;IAED,uEAAuE;IACvE,0EAA0E;IAC1E,qEAAqE;IACrE,yEAAyE;IACzE,yEAAyE;IACzE,0EAA0E;IAC1E,4DAA4D;IAC5D,IAAI,YAAY,EAAE,CAAC;QACjB,MAAM,CAAC,eAAe,CAAC,CAAC,CAAC,EAAE,EAAE;YAC3B,kEAAkE;YAClE,+DAA+D;YAC/D,sEAAsE;YACtE,uEAAuE;YACvE,IAAI,CAAC,eAAe,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;gBAAE,OAAO;YAC9C,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,SAAS,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC;YAC9E,MAAM,WAAW,GAAG,CAAC,CAAC,cAAc,KAAK,eAAe,CAAC,CAAC,CAAC,WAAW;gBACpE,CAAC,CAAC,CAAC,CAAC,cAAc,KAAK,OAAO,CAAC,CAAC,CAAC,QAAQ;oBACzC,CAAC,CAAC,CAAC,CAAC,cAAc,KAAK,aAAa,CAAC,CAAC,CAAC,aAAa;wBACpD,CAAC,CAAC,CAAC,CAAC,cAAc,KAAK,QAAQ,CAAC,CAAC,CAAC,QAAQ;4BAC1C,CAAC,CAAC,CAAC,CAAC,cAAc,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU;gCAC9C,CAAC,CAAC,UAAU,CAAC;YACf,YAAY,CAAC;gBACX,EAAE,EAAE,CAAC,CAAC,OAAO;gBACb,IAAI,EAAE,UAAU;gBAChB,GAAG,EAAE,cAAc,CAAC,WAAW,EAAE,CAAC,CAAC,KAAK,EAAE,WAAW,CAAC;gBACtD,SAAS,EAAE,CAAC,CAAC,SAAS;gBACtB,WAAW,EAAE,CAAC,CAAC,WAAW;gBAC1B,SAAS,EAAE,CAAC,CAAC,SAAS;gBACtB,KAAK,EAAE,CAAC,CAAC,KAAK;gBACd,MAAM,EAAE,CAAC,CAAC,MAAM;gBAChB,cAAc,EAAE,CAAC,CAAC,cAAc;gBAChC,SAAS,EAAE,CAAC,CAAC,SAAS;gBACtB,OAAO,EAAE,CAAC,CAAC,OAAO;gBAClB,SAAS,EAAE,CAAC,CAAC,cAAc,IAAI,SAAS;gBACxC,WAAW,EAAE,CAAC,CAAC,OAAO;gBACtB,YAAY,EAAE,CAAC,CAAC,YAAY;gBAC5B,WAAW,EAAE,CAAC,CAAC,OAAO;aACvB,CAAC,CAAC;QACL,CAAC,CAAC,CAAC;IACL,CAAC;IAED,oDAAoD;IACpD,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,kBAAkB,IAAI,IAAI,CAAC,OAAO,EAAE,EAAE,YAAY,CAAC,CAAC;QAChF,aAAa,CAAC,IAAI,CAAC,OAAO,EAAE,qBAAqB,CAAC,EAAE,iBAAiB,EAAE,EAAE,OAAO,CAAC,CAAC;IACpF,CAAC;IAAC,MAAM,CAAC,CAAC,iBAAiB,CAAC,CAAC;IAE7B,sEAAsE;IACtE,8DAA8D;IAE9D,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,UAAU,OAAO;IACrB,OAAO,SAAS,CAAC;AACnB,CAAC;AAED;;8EAE8E;AAC9E,MAAM,UAAU,2BAA2B;IACzC,OAAO,OAAO,CAAC,GAAG,CAAC,6BAA6B,KAAK,GAAG,CAAC;AAC3D,CAAC;AAED,OAAO,EAAE,cAAc,EAAE,WAAW,EAAE,mBAAmB,EAAE,MAAM,eAAe,CAAC;AAEjF,OAAO,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EAAE,WAAW,EAAE,gBAAgB,EAAE,eAAe,EAAE,QAAQ,EAAE,cAAc,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAC5H,OAAO,EAAE,eAAe,EAAE,aAAa,EAAE,iBAAiB,EAAE,gBAAgB,EAAE,MAAM,cAAc,CAAC"}

package/dist/apme/judge-detect.d.ts

@@ -0,0 +1,26 @@
+/**
+ * HTTP-only detection of locally-running, de-facto-standard inference servers
+ * so onboarding + the REVIEW judge-setup guide can offer "use what you already
+ * have" instead of asking the user to configure an endpoint by hand.
+ *
+ * App Store safety: this is pure network I/O against loopback ports — NO
+ * subprocess, NO CLI probe (`ollama list` etc.). The macOS sandbox's
+ * `com.apple.security.network.client` covers loopback HTTP, so this runs
+ * unchanged in the signed App Store daemon. Mirror: apple ApmeJudgeDetect.swift.
+ */
+export interface DetectedProvider {
+    /** Stable key: 'ollama' | 'lmstudio' | 'mlx' | 'openai-generic'. */
+    provider: string;
+    label: string;
+    /** OpenAI-compatible base to store as apme.judge.endpoint. */
+    endpoint: string;
+    /** Chat model ids advertised by the server (first is a sensible default). */
+    models: string[];
+}
+/**
+ * Probe the standard local endpoints concurrently. Returns only servers that
+ * are reachable AND advertise at least one usable chat model. De-duplicated by
+ * (provider, endpoint) so the two MLX ports don't both show when one answers.
+ */
+export declare function detectLocalJudgeProviders(timeoutMs?: number): Promise<DetectedProvider[]>;
+//# sourceMappingURL=judge-detect.d.ts.map

package/dist/apme/judge-detect.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"judge-detect.d.ts","sourceRoot":"","sources":["../../src/apme/judge-detect.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,MAAM,WAAW,gBAAgB;IAC/B,oEAAoE;IACpE,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,8DAA8D;IAC9D,QAAQ,EAAE,MAAM,CAAC;IACjB,6EAA6E;IAC7E,MAAM,EAAE,MAAM,EAAE,CAAC;CAClB;AA2CD;;;;GAIG;AACH,wBAAsB,yBAAyB,CAAC,SAAS,SAAO,GAAG,OAAO,CAAC,gBAAgB,EAAE,CAAC,CAgB7F"}

package/dist/apme/judge-detect.js

@@ -0,0 +1,69 @@
+/**
+ * HTTP-only detection of locally-running, de-facto-standard inference servers
+ * so onboarding + the REVIEW judge-setup guide can offer "use what you already
+ * have" instead of asking the user to configure an endpoint by hand.
+ *
+ * App Store safety: this is pure network I/O against loopback ports — NO
+ * subprocess, NO CLI probe (`ollama list` etc.). The macOS sandbox's
+ * `com.apple.security.network.client` covers loopback HTTP, so this runs
+ * unchanged in the signed App Store daemon. Mirror: apple ApmeJudgeDetect.swift.
+ */
+const CANDIDATES = [
+    { provider: 'ollama', label: 'Ollama', base: 'http://127.0.0.1:11434', endpoint: 'http://127.0.0.1:11434/v1', tags: true },
+    { provider: 'lmstudio', label: 'LM Studio', base: 'http://127.0.0.1:1234', endpoint: 'http://127.0.0.1:1234/v1' },
+    { provider: 'mlx', label: 'Local MLX server', base: 'http://127.0.0.1:8800', endpoint: 'http://127.0.0.1:8800/v1' },
+    { provider: 'mlx', label: 'Local MLX server', base: 'http://127.0.0.1:8080', endpoint: 'http://127.0.0.1:8080/v1' },
+];
+const NANO = 'nanollava';
+async function modelsFor(c, timeoutMs) {
+    // Ollama's canonical list is /api/tags; every OpenAI-compatible server
+    // (LM Studio, MLX, vLLM, Ollama's compat shim) exposes /v1/models.
+    if (c.tags) {
+        try {
+            const r = await fetch(`${c.base}/api/tags`, { signal: AbortSignal.timeout(timeoutMs) }).catch(() => null);
+            if (r?.ok) {
+                const j = await r.json();
+                const names = (j.models ?? []).map((m) => m.name).filter((n) => !!n);
+                if (names.length)
+                    return names;
+            }
+        }
+        catch { /* fall through to /v1/models */ }
+    }
+    try {
+        const r = await fetch(`${c.base}/v1/models`, { signal: AbortSignal.timeout(timeoutMs) }).catch(() => null);
+        if (r?.ok) {
+            const j = await r.json();
+            const ids = (j.data ?? []).map((m) => m.id).filter((id) => !!id && !id.toLowerCase().includes(NANO));
+            return ids; // empty array = reachable but no usable model
+        }
+    }
+    catch { /* unreachable */ }
+    return null;
+}
+/**
+ * Probe the standard local endpoints concurrently. Returns only servers that
+ * are reachable AND advertise at least one usable chat model. De-duplicated by
+ * (provider, endpoint) so the two MLX ports don't both show when one answers.
+ */
+export async function detectLocalJudgeProviders(timeoutMs = 1200) {
+    const results = await Promise.all(CANDIDATES.map(async (c) => {
+        const models = await modelsFor(c, timeoutMs);
+        if (!models || models.length === 0)
+            return null;
+        return { provider: c.provider, label: c.label, endpoint: c.endpoint, models: models.slice(0, 12) };
+    }));
+    const seen = new Set();
+    const out = [];
+    for (const r of results) {
+        if (!r)
+            continue;
+        const key = `${r.provider}:${r.endpoint}`;
+        if (seen.has(key))
+            continue;
+        seen.add(key);
+        out.push(r);
+    }
+    return out;
+}
+//# sourceMappingURL=judge-detect.js.map

package/dist/apme/judge-detect.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"judge-detect.js","sourceRoot":"","sources":["../../src/apme/judge-detect.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAoBH,MAAM,UAAU,GAAgB;IAC9B,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,QAAQ,EAAE,IAAI,EAAE,wBAAwB,EAAE,QAAQ,EAAE,2BAA2B,EAAE,IAAI,EAAE,IAAI,EAAE;IAC1H,EAAE,QAAQ,EAAE,UAAU,EAAE,KAAK,EAAE,WAAW,EAAE,IAAI,EAAE,uBAAuB,EAAE,QAAQ,EAAE,0BAA0B,EAAE;IACjH,EAAE,QAAQ,EAAE,KAAK,EAAE,KAAK,EAAE,kBAAkB,EAAE,IAAI,EAAE,uBAAuB,EAAE,QAAQ,EAAE,0BAA0B,EAAE;IACnH,EAAE,QAAQ,EAAE,KAAK,EAAE,KAAK,EAAE,kBAAkB,EAAE,IAAI,EAAE,uBAAuB,EAAE,QAAQ,EAAE,0BAA0B,EAAE;CACpH,CAAC;AAEF,MAAM,IAAI,GAAG,WAAW,CAAC;AAEzB,KAAK,UAAU,SAAS,CAAC,CAAY,EAAE,SAAiB;IACtD,uEAAuE;IACvE,mEAAmE;IACnE,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC;QACX,IAAI,CAAC;YACH,MAAM,CAAC,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,WAAW,EAAE,EAAE,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC;YAC1G,IAAI,CAAC,EAAE,EAAE,EAAE,CAAC;gBACV,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,EAA2C,CAAC;gBAClE,MAAM,KAAK,GAAG,CAAC,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;gBAClF,IAAI,KAAK,CAAC,MAAM;oBAAE,OAAO,KAAK,CAAC;YACjC,CAAC;QACH,CAAC;QAAC,MAAM,CAAC,CAAC,gCAAgC,CAAC,CAAC;IAC9C,CAAC;IACD,IAAI,CAAC;QACH,MAAM,CAAC,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,YAAY,EAAE,EAAE,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC;QAC3G,IAAI,CAAC,EAAE,EAAE,EAAE,CAAC;YACV,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,EAAuC,CAAC;YAC9D,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,EAAE,EAAgB,EAAE,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC;YACnH,OAAO,GAAG,CAAC,CAAC,8CAA8C;QAC5D,CAAC;IACH,CAAC;IAAC,MAAM,CAAC,CAAC,iBAAiB,CAAC,CAAC;IAC7B,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,SAAS,GAAG,IAAI;IAC9D,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,GAAG,CAAC,KAAK,EAAE,CAAC,EAAE,EAAE;QAC3D,MAAM,MAAM,GAAG,MAAM,SAAS,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;QAC7C,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC;QAChD,OAAO,EAAE,QAAQ,EAAE,CAAC,CAAC,QAAQ,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,QAAQ,EAAE,CAAC,CAAC,QAAQ,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAsB,CAAC;IACzH,CAAC,CAAC,CAAC,CAAC;IACJ,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;IAC/B,MAAM,GAAG,GAAuB,EAAE,CAAC;IACnC,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,IAAI,CAAC,CAAC;YAAE,SAAS;QACjB,MAAM,GAAG,GAAG,GAAG,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,QAAQ,EAAE,CAAC;QAC1C,IAAI,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC;YAAE,SAAS;QAC5B,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACd,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACd,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC"}

package/dist/apme/pareto.d.ts

@@ -0,0 +1,39 @@
+/**
+ * Pareto frontier over (quality, cost) for model orchestration.
+ *
+ * A model is on the frontier if no other model dominates it — i.e. no other
+ * model is BOTH higher quality AND lower (or equal) cost per sample. Dominated
+ * models are strictly worse on both axes and should never be recommended; the
+ * frontier is the real menu of quality/cost tradeoffs.
+ *
+ * "Same quality, 40% cheaper" decisions fall straight out of this: walk the
+ * frontier from cheapest to best and pick the knee that fits the budget.
+ *
+ * Pure + synchronous + unit-testable. Input is the sample-granularity
+ * scorecard; output partitions candidates into frontier vs dominated.
+ */
+import type { ApmeSampleScorecardRow } from '@agentdeck/shared';
+export interface ParetoPoint {
+    agentType: string;
+    modelId: string;
+    taskCategory: string | null;
+    /** Quality axis (avg composite, [0,1]). Higher is better. */
+    quality: number;
+    /** Cost axis: average USD per sample. Lower is better. */
+    costPerSample: number;
+    avgLatencyMs: number | null;
+    samples: number;
+}
+export interface ParetoResult {
+    frontier: ParetoPoint[];
+    dominated: ParetoPoint[];
+}
+/**
+ * Partition scorecard rows into the Pareto frontier and the dominated set.
+ * @param rows         sample scorecard rows (optionally pre-filtered by category)
+ * @param minSamples   ignore models with too little data to trust
+ */
+export declare function computePareto(rows: ApmeSampleScorecardRow[], minSamples?: number): ParetoResult;
+/** Convenience: filter scorecard to a category before computing the frontier. */
+export declare function paretoForCategory(rows: ApmeSampleScorecardRow[], category: string | null | undefined, minSamples?: number): ParetoResult;
+//# sourceMappingURL=pareto.d.ts.map

package/dist/apme/pareto.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"pareto.d.ts","sourceRoot":"","sources":["../../src/apme/pareto.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,mBAAmB,CAAC;AAEhE,MAAM,WAAW,WAAW;IAC1B,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,6DAA6D;IAC7D,OAAO,EAAE,MAAM,CAAC;IAChB,0DAA0D;IAC1D,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,YAAY;IAC3B,QAAQ,EAAE,WAAW,EAAE,CAAC;IACxB,SAAS,EAAE,WAAW,EAAE,CAAC;CAC1B;AAsBD;;;;GAIG;AACH,wBAAgB,aAAa,CAC3B,IAAI,EAAE,sBAAsB,EAAE,EAC9B,UAAU,SAAI,GACb,YAAY,CAcd;AAED,iFAAiF;AACjF,wBAAgB,iBAAiB,CAC/B,IAAI,EAAE,sBAAsB,EAAE,EAC9B,QAAQ,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,EACnC,UAAU,SAAI,GACb,YAAY,CAGd"}

package/dist/apme/pareto.js

@@ -0,0 +1,59 @@
+/**
+ * Pareto frontier over (quality, cost) for model orchestration.
+ *
+ * A model is on the frontier if no other model dominates it — i.e. no other
+ * model is BOTH higher quality AND lower (or equal) cost per sample. Dominated
+ * models are strictly worse on both axes and should never be recommended; the
+ * frontier is the real menu of quality/cost tradeoffs.
+ *
+ * "Same quality, 40% cheaper" decisions fall straight out of this: walk the
+ * frontier from cheapest to best and pick the knee that fits the budget.
+ *
+ * Pure + synchronous + unit-testable. Input is the sample-granularity
+ * scorecard; output partitions candidates into frontier vs dominated.
+ */
+function toPoint(r) {
+    return {
+        agentType: r.agentType,
+        modelId: r.modelId,
+        taskCategory: r.taskCategory,
+        quality: r.avgQuality ?? 0,
+        costPerSample: r.samples > 0 ? (r.totalCost ?? 0) / r.samples : 0,
+        avgLatencyMs: r.avgLatencyMs,
+        samples: r.samples,
+    };
+}
+/** True when `b` dominates `a`: at least as good on both axes and strictly
+ *  better on at least one (higher quality, lower cost). */
+function dominates(b, a) {
+    const betterOrEqual = b.quality >= a.quality && b.costPerSample <= a.costPerSample;
+    const strictlyBetter = b.quality > a.quality || b.costPerSample < a.costPerSample;
+    return betterOrEqual && strictlyBetter;
+}
+/**
+ * Partition scorecard rows into the Pareto frontier and the dominated set.
+ * @param rows         sample scorecard rows (optionally pre-filtered by category)
+ * @param minSamples   ignore models with too little data to trust
+ */
+export function computePareto(rows, minSamples = 3) {
+    const points = rows
+        .filter((r) => r.samples >= minSamples && (r.avgQuality ?? 0) > 0)
+        .map(toPoint);
+    const frontier = [];
+    const dominated = [];
+    for (const a of points) {
+        if (points.some((b) => b !== a && dominates(b, a)))
+            dominated.push(a);
+        else
+            frontier.push(a);
+    }
+    // Frontier sorted cheapest → most expensive (the natural tradeoff curve).
+    frontier.sort((x, y) => x.costPerSample - y.costPerSample || y.quality - x.quality);
+    return { frontier, dominated };
+}
+/** Convenience: filter scorecard to a category before computing the frontier. */
+export function paretoForCategory(rows, category, minSamples = 3) {
+    const scoped = category ? rows.filter((r) => r.taskCategory === category) : rows;
+    return computePareto(scoped, minSamples);
+}
+//# sourceMappingURL=pareto.js.map

package/dist/apme/pareto.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"pareto.js","sourceRoot":"","sources":["../../src/apme/pareto.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAqBH,SAAS,OAAO,CAAC,CAAyB;IACxC,OAAO;QACL,SAAS,EAAE,CAAC,CAAC,SAAS;QACtB,OAAO,EAAE,CAAC,CAAC,OAAO;QAClB,YAAY,EAAE,CAAC,CAAC,YAAY;QAC5B,OAAO,EAAE,CAAC,CAAC,UAAU,IAAI,CAAC;QAC1B,aAAa,EAAE,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QACjE,YAAY,EAAE,CAAC,CAAC,YAAY;QAC5B,OAAO,EAAE,CAAC,CAAC,OAAO;KACnB,CAAC;AACJ,CAAC;AAED;2DAC2D;AAC3D,SAAS,SAAS,CAAC,CAAc,EAAE,CAAc;IAC/C,MAAM,aAAa,GAAG,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,aAAa,IAAI,CAAC,CAAC,aAAa,CAAC;IACnF,MAAM,cAAc,GAAG,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,aAAa,GAAG,CAAC,CAAC,aAAa,CAAC;IAClF,OAAO,aAAa,IAAI,cAAc,CAAC;AACzC,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,aAAa,CAC3B,IAA8B,EAC9B,UAAU,GAAG,CAAC;IAEd,MAAM,MAAM,GAAG,IAAI;SAChB,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,IAAI,UAAU,IAAI,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;SACjE,GAAG,CAAC,OAAO,CAAC,CAAC;IAEhB,MAAM,QAAQ,GAAkB,EAAE,CAAC;IACnC,MAAM,SAAS,GAAkB,EAAE,CAAC;IACpC,KAAK,MAAM,CAAC,IAAI,MAAM,EAAE,CAAC;QACvB,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,SAAS,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAAE,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;;YACjE,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACxB,CAAC;IACD,0EAA0E;IAC1E,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,GAAG,CAAC,CAAC,aAAa,IAAI,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC;IACpF,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,CAAC;AACjC,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,iBAAiB,CAC/B,IAA8B,EAC9B,QAAmC,EACnC,UAAU,GAAG,CAAC;IAEd,MAAM,MAAM,GAAG,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IACjF,OAAO,aAAa,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC;AAC3C,CAAC"}

package/dist/apme/recommend.d.ts

@@ -28,5 +28,8 @@ export declare class ApmeRecommender {
     private readonly store;
     constructor(store: ApmeStore);
     recommend(input: RecommendInput): RecommendCandidate[];
+    /** Legacy run-level recommendation (v_model_scorecard). Used when no
+     *  sample-granularity composite scores exist yet. */
+    private recommendFromRuns;
 }
 //# sourceMappingURL=recommend.d.ts.map

package/dist/apme/recommend.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"recommend.d.ts","sourceRoot":"","sources":["../../src/apme/recommend.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AACnD,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AAE5C,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,mFAAmF;IACnF,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;CAC5B;AAED,MAAM,WAAW,kBAAkB;IACjC,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,SAAS,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,eAAe,EAAE,MAAM,CAAC;IACxB,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,qBAAa,eAAe;IACd,OAAO,CAAC,QAAQ,CAAC,KAAK;gBAAL,KAAK,EAAE,SAAS;IAE7C,SAAS,CAAC,KAAK,EAAE,cAAc,GAAG,kBAAkB,EAAE;CA2BvD"}
+{"version":3,"file":"recommend.d.ts","sourceRoot":"","sources":["../../src/apme/recommend.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AACnD,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AAG5C,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,mFAAmF;IACnF,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;CAC5B;AAED,MAAM,WAAW,kBAAkB;IACjC,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,SAAS,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,eAAe,EAAE,MAAM,CAAC;IACxB,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,qBAAa,eAAe;IACd,OAAO,CAAC,QAAQ,CAAC,KAAK;gBAAL,KAAK,EAAE,SAAS;IAE7C,SAAS,CAAC,KAAK,EAAE,cAAc,GAAG,kBAAkB,EAAE;IAgDtD;yDACqD;IACrD,OAAO,CAAC,iBAAiB;CAyB1B"}

package/dist/apme/recommend.js

@@ -6,6 +6,7 @@
  * confidence. v1 uses the v_model_scorecard view; v2 (stretch) layers in
  * task-similarity via local embeddings.
  */
+import { paretoForCategory } from './pareto.js';
 export class ApmeRecommender {
     store;
     constructor(store) {
@@ -14,6 +15,48 @@ export class ApmeRecommender {
     recommend(input) {
         if (!this.store.enabled)
             return [];
+        // Work at sample granularity (the canonical eval unit) and restrict to the
+        // Pareto frontier — dominated models (strictly worse on quality AND cost)
+        // are never worth recommending. This is the model-orchestration payoff:
+        // the menu is the quality/cost tradeoff curve, not a flat ranking.
+        const scorecard = this.store.sampleScorecard();
+        let { frontier } = paretoForCategory(scorecard, input.taskKind);
+        // Fall back to the whole frontier if the category has too little data.
+        if (frontier.length === 0)
+            frontier = paretoForCategory(scorecard, undefined).frontier;
+        // Fall back to the run-level scorecard when no sample-granularity data has
+        // accumulated yet (legacy runs, or before the first task judge resolves).
+        if (frontier.length === 0)
+            return this.recommendFromRuns(input);
+        let candidates = frontier;
+        if (input.availableModels) {
+            candidates = candidates.filter((p) => input.availableModels.includes(p.modelId));
+        }
+        if (input.budgetUsd !== undefined) {
+            candidates = candidates.filter((p) => p.costPerSample <= input.budgetUsd);
+        }
+        if (input.latencyBudgetMs !== undefined) {
+            candidates = candidates.filter((p) => p.avgLatencyMs == null || p.avgLatencyMs <= input.latencyBudgetMs);
+        }
+        candidates = [...candidates].sort((a, b) => {
+            // Budget-conscious or local-preferring → cheapest first; else best quality.
+            if (input.preferLocal || (input.budgetUsd !== undefined && input.budgetUsd < 5)) {
+                return a.costPerSample - b.costPerSample || b.quality - a.quality;
+            }
+            return b.quality - a.quality || a.costPerSample - b.costPerSample;
+        });
+        return candidates.slice(0, 3).map((p) => ({
+            modelId: p.modelId,
+            agentType: p.agentType,
+            expectedScore: p.quality,
+            expectedCostUsd: p.costPerSample,
+            confidence: Math.min(1, p.samples / 20),
+            rationale: `${p.samples} samples, avg ${(p.quality * 100).toFixed(0)}%, $${p.costPerSample.toFixed(4)}/sample${p.avgLatencyMs != null ? `, ${Math.round(p.avgLatencyMs)}ms` : ''} — on the cost/quality frontier`,
+        }));
+    }
+    /** Legacy run-level recommendation (v_model_scorecard). Used when no
+     *  sample-granularity composite scores exist yet. */
+    recommendFromRuns(input) {
         const rows = this.store.scorecard();
         const filtered = input.availableModels
             ? rows.filter((r) => input.availableModels.includes(r.modelId))
@@ -21,7 +64,6 @@ export class ApmeRecommender {
         return filtered
             .filter((r) => r.runs >= 3 && (r.avgOverall ?? 0) > 0)
             .sort((a, b) => {
-            // Prefer cost-per-quality if budget is tight, otherwise raw quality.
             if (input.budgetUsd !== undefined && input.budgetUsd < 5) {
                 return (a.costPerQuality ?? Infinity) - (b.costPerQuality ?? Infinity);
             }

package/dist/apme/recommend.js.map

@@ -1 +1 @@
-{"version":3,"file":"recommend.js","sourceRoot":"","sources":["../../src/apme/recommend.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAuBH,MAAM,OAAO,eAAe;IACG;IAA7B,YAA6B,KAAgB;QAAhB,UAAK,GAAL,KAAK,CAAW;IAAG,CAAC;IAEjD,SAAS,CAAC,KAAqB;QAC7B,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO;YAAE,OAAO,EAAE,CAAC;QACnC,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC;QACpC,MAAM,QAAQ,GAAG,KAAK,CAAC,eAAe;YACpC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,eAAgB,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;YAChE,CAAC,CAAC,IAAI,CAAC;QACT,OAAO,QAAQ;aACZ,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;aACrD,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;YACb,qEAAqE;YACrE,IAAI,KAAK,CAAC,SAAS,KAAK,SAAS,IAAI,KAAK,CAAC,SAAS,GAAG,CAAC,EAAE,CAAC;gBACzD,OAAO,CAAC,CAAC,CAAC,cAAc,IAAI,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,cAAc,IAAI,QAAQ,CAAC,CAAC;YACzE,CAAC;YACD,OAAO,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,CAAC;QACnD,CAAC,CAAC;aACD,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC;aACX,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACX,OAAO,EAAE,CAAC,CAAC,OAAO;YAClB,SAAS,EAAE,CAAC,CAAC,SAAsB;YACnC,aAAa,EAAE,CAAC,CAAC,UAAU,IAAI,CAAC;YAChC,eAAe,EAAE,CAAC,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;YACzD,UAAU,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,GAAG,EAAE,CAAC;YACpC,SAAS,EAAE,GAAG,CAAC,CAAC,IAAI,cAAc,CAAC,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IACtE,CAAC,CAAC,YAAY,IAAI,IAAI,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAC/E,EAAE;SACH,CAAC,CAAC,CAAC;IACR,CAAC;CACF"}
+{"version":3,"file":"recommend.js","sourceRoot":"","sources":["../../src/apme/recommend.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAIH,OAAO,EAAE,iBAAiB,EAAE,MAAM,aAAa,CAAC;AAoBhD,MAAM,OAAO,eAAe;IACG;IAA7B,YAA6B,KAAgB;QAAhB,UAAK,GAAL,KAAK,CAAW;IAAG,CAAC;IAEjD,SAAS,CAAC,KAAqB;QAC7B,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO;YAAE,OAAO,EAAE,CAAC;QAEnC,2EAA2E;QAC3E,0EAA0E;QAC1E,wEAAwE;QACxE,mEAAmE;QACnE,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,eAAe,EAAE,CAAC;QAC/C,IAAI,EAAE,QAAQ,EAAE,GAAG,iBAAiB,CAAC,SAAS,EAAE,KAAK,CAAC,QAAQ,CAAC,CAAC;QAEhE,uEAAuE;QACvE,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;YAAE,QAAQ,GAAG,iBAAiB,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC,QAAQ,CAAC;QAEvF,2EAA2E;QAC3E,0EAA0E;QAC1E,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC,iBAAiB,CAAC,KAAK,CAAC,CAAC;QAEhE,IAAI,UAAU,GAAG,QAAQ,CAAC;QAC1B,IAAI,KAAK,CAAC,eAAe,EAAE,CAAC;YAC1B,UAAU,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,eAAgB,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;QACpF,CAAC;QACD,IAAI,KAAK,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;YAClC,UAAU,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,IAAI,KAAK,CAAC,SAAU,CAAC,CAAC;QAC7E,CAAC;QACD,IAAI,KAAK,CAAC,eAAe,KAAK,SAAS,EAAE,CAAC;YACxC,UAAU,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,IAAI,IAAI,IAAI,CAAC,CAAC,YAAY,IAAI,KAAK,CAAC,eAAgB,CAAC,CAAC;QAC5G,CAAC;QAED,UAAU,GAAG,CAAC,GAAG,UAAU,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;YACzC,4EAA4E;YAC5E,IAAI,KAAK,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,SAAS,KAAK,SAAS,IAAI,KAAK,CAAC,SAAS,GAAG,CAAC,CAAC,EAAE,CAAC;gBAChF,OAAO,CAAC,CAAC,aAAa,GAAG,CAAC,CAAC,aAAa,IAAI,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,OAAO,CAAC;YACpE,CAAC;YACD,OAAO,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,aAAa,GAAG,CAAC,CAAC,aAAa,CAAC;QACpE,CAAC,CAAC,CAAC;QAEH,OAAO,UAAU,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACxC,OAAO,EAAE,CAAC,CAAC,OAAO;YAClB,SAAS,EAAE,CAAC,CAAC,SAAsB;YACnC,aAAa,EAAE,CAAC,CAAC,OAAO;YACxB,eAAe,EAAE,CAAC,CAAC,aAAa;YAChC,UAAU,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,OAAO,GAAG,EAAE,CAAC;YACvC,SAAS,EAAE,GAAG,CAAC,CAAC,OAAO,iBAAiB,CAAC,CAAC,CAAC,OAAO,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,UACnG,CAAC,CAAC,YAAY,IAAI,IAAI,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC,EACjE,iCAAiC;SAClC,CAAC,CAAC,CAAC;IACN,CAAC;IAED;yDACqD;IAC7C,iBAAiB,CAAC,KAAqB;QAC7C,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC;QACpC,MAAM,QAAQ,GAAG,KAAK,CAAC,eAAe;YACpC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,eAAgB,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;YAChE,CAAC,CAAC,IAAI,CAAC;QACT,OAAO,QAAQ;aACZ,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;aACrD,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;YACb,IAAI,KAAK,CAAC,SAAS,KAAK,SAAS,IAAI,KAAK,CAAC,SAAS,GAAG,CAAC,EAAE,CAAC;gBACzD,OAAO,CAAC,CAAC,CAAC,cAAc,IAAI,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,cAAc,IAAI,QAAQ,CAAC,CAAC;YACzE,CAAC;YACD,OAAO,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,CAAC;QACnD,CAAC,CAAC;aACD,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC;aACX,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACX,OAAO,EAAE,CAAC,CAAC,OAAO;YAClB,SAAS,EAAE,CAAC,CAAC,SAAsB;YACnC,aAAa,EAAE,CAAC,CAAC,UAAU,IAAI,CAAC;YAChC,eAAe,EAAE,CAAC,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;YACzD,UAAU,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,GAAG,EAAE,CAAC;YACpC,SAAS,EAAE,GAAG,CAAC,CAAC,IAAI,cAAc,CAAC,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IACtE,CAAC,CAAC,YAAY,IAAI,IAAI,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAC/E,EAAE;SACH,CAAC,CAAC,CAAC;IACR,CAAC;CACF"}