codemem 0.39.0 → 0.40.0-alpha.1

package/dist/index.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { DEDUP_KEY_BACKFILL_JOB, DEFAULT_COORDINATOR_DB_PATH, DedupKeyBackfillRunner, MUTATING_TOOL_NAMES, MemoryStore, ObserverClient, REF_BACKFILL_JOB, RawEventSweeper, RefBackfillRunner, SCOPE_BACKFILL_JOB, SESSION_CONTEXT_BACKFILL_JOB, SUMMARY_DEDUP_BACKFILL_JOB, ScopeBackfillRunner, SessionContextBackfillRunner, SummaryDedupBackfillRunner, SyncRetentionRunner, VERSION, VectorModelMigrationRunner, aiBackfillStructuredContent, applyBootstrapSnapshot, applyDistillRule, backfillMemoryDedupKeys, backfillNarrativeFromBody, backfillTagsText, backfillVectors, buildAuthHeaders, buildBaseUrl, buildDistillReport, buildRawEventEnvelopeFromCodexHook, buildRawEventEnvelopeFromHook, compareMemoryRoleReports, connect, coordinatorCreateGroupAction, coordinatorCreateInviteAction, coordinatorCreateScopeAction, coordinatorDisableDeviceAction, coordinatorEnrollDeviceAction, coordinatorGrantScopeMembershipAction, coordinatorImportInviteAction, coordinatorListBootstrapGrantsAction, coordinatorListDevicesAction, coordinatorListGroupsAction, coordinatorListJoinRequestsAction, coordinatorListScopeMembershipsAction, coordinatorListScopesAction, coordinatorRemoveDeviceAction, coordinatorRenameDeviceAction, coordinatorReviewJoinRequestAction, coordinatorRevokeBootstrapGrantAction, coordinatorRevokeScopeMembershipAction, coordinatorUpdateScopeAction, createBetterSqliteCoordinatorApp, deactivateLowSignalMemories, deactivateLowSignalObservations, dedupNearDuplicateMemories, draftDistillRule, ensureDeviceIdentity, ensureSchemaBootstrapped, exportMemories, extractApplyPatchPaths, fetchAllSnapshotPages, fingerprintPublicKey, flushRawEvents, formatHostPort, getExtractionBenchmarkProfile, getInjectionEvalScenarioPack, getInjectionEvalScenarioPrompts, getMaintenanceJob, getMemoryArtifactReport, getMemoryRoleReport, getRawEventRelinkPlan, getRawEventRelinkReport, getRawEventStatus, getSemanticIndexDiagnostics, getSessionExtractionEval, getSessionExtractionEvalScenario, getWorkspaceCodememConfigPath, hasPendingDedupKeyBackfill, hasPendingRefBackfill, hasPendingScopeBackfill, hasPendingSessionContextBackfill, hasPendingSummaryDedupBackfill, hasUnsyncedSharedMemoryChanges, importMemories, initDatabase, isEmbeddingDisabled, judgeDistillReport, listMaintenanceJobs, listPerPeerScopeSyncState, listRetentionScopeIds, loadObserverConfig, loadPublicKey, loadSqliteVec, mdnsEnabled, planReplicationOpsAgePrune, projectMatchesFilter, pruneReplicationOpsUntilCaughtUp, rawEventsGate, readCodememConfigFile, readCodememConfigFileAtPath, readCoordinatorSyncConfig, readImportPayload, renderUnifiedDiff, replayBatchExtraction, replayBatchExtractionWithTierRouting, requestJson, resolveCodememConfigPath, resolveDbPath, resolveHookProject, resolveProject, resolveProjectRoot, retryRawEventFailures, runSyncDaemon, runSyncPass, scanSecretsRetroactive, schema, setPeerProjectFilter, stripJsonComments, stripPrivateObj, stripTrailingCommas, syncPassPreflight, updatePeerAddresses, vacuumDatabase, writeCodememConfigFile } from "@codemem/core";
+import { DEDUP_KEY_BACKFILL_JOB, DEFAULT_COORDINATOR_DB_PATH, DedupKeyBackfillRunner, MUTATING_TOOL_NAMES, MemoryStore, ObserverClient, REF_BACKFILL_JOB, RawEventSweeper, RefBackfillRunner, SCOPE_BACKFILL_JOB, SESSION_CONTEXT_BACKFILL_JOB, SUMMARY_DEDUP_BACKFILL_JOB, ScopeBackfillRunner, SessionContextBackfillRunner, SummaryDedupBackfillRunner, SyncRetentionRunner, VERSION, VectorModelMigrationRunner, aiBackfillStructuredContent, applyBootstrapSnapshot, applyDistillRule, backfillMemoryDedupKeys, backfillNarrativeFromBody, backfillTagsText, backfillVectors, buildAuthHeaders, buildBaseUrl, buildDistillReport, buildRawEventEnvelopeFromCodexHook, buildRawEventEnvelopeFromHook, compareMemoryRoleReports, connect, coordinatorCreateGroupAction, coordinatorCreateInviteAction, coordinatorCreateScopeAction, coordinatorDisableDeviceAction, coordinatorEnrollDeviceAction, coordinatorGrantScopeMembershipAction, coordinatorImportInviteAction, coordinatorListBootstrapGrantsAction, coordinatorListDevicesAction, coordinatorListGroupsAction, coordinatorListJoinRequestsAction, coordinatorListScopeMembershipsAction, coordinatorListScopesAction, coordinatorRemoveDeviceAction, coordinatorRenameDeviceAction, coordinatorReviewJoinRequestAction, coordinatorRevokeBootstrapGrantAction, coordinatorRevokeScopeMembershipAction, coordinatorUpdateScopeAction, createBetterSqliteCoordinatorApp, deactivateLowSignalMemories, deactivateLowSignalObservations, dedupNearDuplicateMemories, draftDistillRule, ensureDeviceIdentity, ensureSchemaBootstrapped, estimateExtractionModelCost, exportMemories, extractApplyPatchPaths, fetchAllSnapshotPages, fingerprintPublicKey, flushRawEvents, formatHostPort, getExtractionBenchmarkProfile, getExtractionModelPricing, getInjectionEvalScenarioPack, getInjectionEvalScenarioPrompts, getMaintenanceJob, getMemoryArtifactReport, getMemoryRoleReport, getRawEventRelinkPlan, getRawEventRelinkReport, getRawEventStatus, getSemanticIndexDiagnostics, getSessionExtractionEval, getSessionExtractionEvalScenario, getWorkspaceCodememConfigPath, hasPendingDedupKeyBackfill, hasPendingRefBackfill, hasPendingScopeBackfill, hasPendingSessionContextBackfill, hasPendingSummaryDedupBackfill, hasUnsyncedSharedMemoryChanges, importMemories, initDatabase, isEmbeddingDisabled, judgeDistillReport, listMaintenanceJobs, listPerPeerScopeSyncState, listRetentionScopeIds, loadObserverConfig, loadPublicKey, loadSqliteVec, mdnsEnabled, planReplicationOpsAgePrune, projectMatchesFilter, pruneReplicationOpsUntilCaughtUp, rawEventsGate, readCodememConfigFile, readCodememConfigFileAtPath, readCoordinatorSyncConfig, readImportPayload, renderUnifiedDiff, replayBatchExtraction, replayBatchExtractionWithTierRouting, requestJson, resolveCodememConfigPath, resolveDbPath, resolveHookProject, resolveProject, resolveProjectRoot, retryRawEventFailures, runSyncDaemon, runSyncPass, scanSecretsRetroactive, schema, scoreExtractionBenchmarkOutput, setPeerProjectFilter, stripJsonComments, stripPrivateObj, stripTrailingCommas, syncPassPreflight, updatePeerAddresses, vacuumDatabase, writeCodememConfigFile } from "@codemem/core";
 import { Command, Option } from "commander";
 import omelette from "omelette";
 import { appendFileSync, copyFileSync, existsSync, mkdirSync, readFileSync, readdirSync, renameSync, rmSync, rmdirSync, statSync, unlinkSync, writeFileSync } from "node:fs";
@@ -2666,12 +2666,13 @@ function buildCoordinatorCommand() {
 		}
 	});
 	cmd.addCommand(listScopeMembersCmd);
-	const grantScopeMemberCmd = new Command("grant-scope-member").configureHelp(helpStyle).description("Grant a device explicit access to a Sharing domain").argument("<group>", "group id").argument("<scope-id>", "Sharing domain scope_id").argument("<device-id>", "device id").option("--role <role>", "membership role").option("--membership-epoch <epoch>", "membership epoch").option("--manifest-hash <hash>", "membership manifest hash").option("--remote-url <url>", "remote coordinator URL override").option("--admin-secret <secret>", "remote coordinator admin secret override");
+	const grantScopeMemberCmd = new Command("grant-scope-member").configureHelp(helpStyle).description("Grant a device explicit access to a Sharing domain").argument("<group>", "group id").argument("<scope-id>", "Sharing domain scope_id").argument("<device-id>", "device id").requiredOption("--effect-id <id>", "deterministic mutation effect id").option("--role <role>", "membership role").option("--membership-epoch <epoch>", "membership epoch").option("--manifest-hash <hash>", "membership manifest hash").option("--remote-url <url>", "remote coordinator URL override").option("--admin-secret <secret>", "remote coordinator admin secret override");
 	addDbOption(grantScopeMemberCmd);
 	addJsonOption(grantScopeMemberCmd);
 	grantScopeMemberCmd.action(async (groupId, scopeId, deviceId, opts) => {
 		try {
 			const membership = await coordinatorGrantScopeMembershipAction({
+				effectId: opts.effectId,
 				groupId,
 				scopeId,
 				deviceId,
@@ -2699,12 +2700,13 @@ function buildCoordinatorCommand() {
 		}
 	});
 	cmd.addCommand(grantScopeMemberCmd);
-	const revokeScopeMemberCmd = new Command("revoke-scope-member").configureHelp(helpStyle).description("Revoke a device from a Sharing domain").argument("<group>", "group id").argument("<scope-id>", "Sharing domain scope_id").argument("<device-id>", "device id").option("--membership-epoch <epoch>", "membership epoch").option("--manifest-hash <hash>", "membership manifest hash").option("--remote-url <url>", "remote coordinator URL override").option("--admin-secret <secret>", "remote coordinator admin secret override");
+	const revokeScopeMemberCmd = new Command("revoke-scope-member").configureHelp(helpStyle).description("Revoke a device from a Sharing domain").argument("<group>", "group id").argument("<scope-id>", "Sharing domain scope_id").argument("<device-id>", "device id").requiredOption("--effect-id <id>", "deterministic mutation effect id").option("--membership-epoch <epoch>", "membership epoch").option("--manifest-hash <hash>", "membership manifest hash").option("--remote-url <url>", "remote coordinator URL override").option("--admin-secret <secret>", "remote coordinator admin secret override");
 	addDbOption(revokeScopeMemberCmd);
 	addJsonOption(revokeScopeMemberCmd);
 	revokeScopeMemberCmd.action(async (groupId, scopeId, deviceId, opts) => {
 		try {
 			if (!await coordinatorRevokeScopeMembershipAction({
+				effectId: opts.effectId,
 				groupId,
 				scopeId,
 				deviceId,
@@ -5117,8 +5119,28 @@ function createMemoryExtractionReplayCommand() {
 	});
 	return cmd;
 }
+function reconcileExtractionBenchmarkStatus(input) {
+	const quality = input.finalQuality;
+	let status = input.classification.status;
+	let reason = input.classification.reason;
+	if (input.purpose === "shape_quality" && quality && status !== "observer_no_output") {
+		if (quality.summaryDisposition.score === 0) {
+			status = "shape_fail";
+			reason = `summary disposition ${quality.summaryDisposition.actual} does not satisfy expected ${quality.summaryDisposition.expected}`;
+		} else if (status === "shape_fail" && quality.summaryDisposition.actual === "skip" && input.finalFailureReasons.length > 0 && input.finalFailureReasons.every((failure) => failure.startsWith("summary count "))) {
+			status = "pass";
+			reason = "valid low-signal skip satisfies benchmark disposition";
+		}
+	}
+	return {
+		status,
+		reason,
+		quality,
+		initialQuality: input.initialQuality
+	};
+}
 function createMemoryExtractionBenchmarkCommand() {
-	const cmd = new Command("extraction-benchmark").configureHelp(helpStyle).description("Run the formal extraction replay benchmark set and print a cost/quality scoreboard").requiredOption("--benchmark <id>", "benchmark profile id").option("--observer-provider <provider>", "override observer provider for this benchmark run").option("--observer-model <model>", "override observer model for this benchmark run").option("--observer-tier-routing", "use replay-only benchmark-backed observer tier routing").option("--openai-responses", "use OpenAI Responses API for this benchmark run").option("--reasoning-effort <level>", "set OpenAI reasoning.effort for this benchmark run (responses path)").option("--reasoning-summary <mode>", "set OpenAI reasoning.summary for this benchmark run (responses path)").option("--max-output-tokens <n>", "override OpenAI max_output_tokens for this benchmark run (responses path)").option("--observer-temperature <value>", "override observer temperature for this benchmark run").option("--transcript-budget <chars>", "override replay transcript budget in characters for this benchmark run");
+	const cmd = new Command("extraction-benchmark").configureHelp(helpStyle).description("Run the formal extraction replay benchmark set and print a cost/quality scoreboard").requiredOption("--benchmark <id>", "benchmark profile id").option("--observer-provider <provider>", "override observer provider for this benchmark run").option("--observer-model <model>", "override observer model for this benchmark run").option("--observer-tier-routing", "use replay-only benchmark-backed observer tier routing").option("--openai-responses", "use OpenAI Responses API for this benchmark run").option("--reasoning-effort <level>", "set OpenAI reasoning.effort for this benchmark run (responses path)").option("--reasoning-summary <mode>", "set OpenAI reasoning.summary for this benchmark run (responses path)").option("--max-output-tokens <n>", "override OpenAI max_output_tokens for this benchmark run (responses path)").option("--observer-temperature <value>", "override observer temperature for this benchmark run").option("--transcript-budget <chars>", "override replay transcript budget in characters for this benchmark run").option("--repetitions <n>", "run every benchmark batch 1-10 times to measure model stability", "1");
 	addDbOption(cmd);
 	addJsonOption(cmd);
 	cmd.action(async (opts) => {
@@ -5139,6 +5161,9 @@ function createMemoryExtractionBenchmarkCommand() {
 			const maxOutputTokensInput = opts.maxOutputTokens?.trim() ?? "";
 			const maxOutputTokens = maxOutputTokensInput.length > 0 ? parseStrictPositiveId(maxOutputTokensInput) : null;
 			if (maxOutputTokensInput.length > 0 && maxOutputTokens === null) throw new Error(`Invalid max output tokens: ${maxOutputTokensInput || opts.maxOutputTokens}`);
+			const repetitionsInput = opts.repetitions?.trim() ?? "1";
+			const repetitions = parseStrictPositiveId(repetitionsInput);
+			if (repetitions === null || repetitions > 10) throw new Error(`Invalid repetitions: ${repetitionsInput || opts.repetitions}`);
 			const observerConfig = loadObserverConfig();
 			const observerConfigWithOverrides = {
 				...observerConfig,
@@ -5152,7 +5177,7 @@ function createMemoryExtractionBenchmarkCommand() {
 			};
 			const observer = new ObserverClient(observerConfigWithOverrides);
 			const runs = [];
-			for (const batch of benchmark.batches) {
+			for (let iteration = 1; iteration <= repetitions; iteration += 1) for (const batch of benchmark.batches) {
 				const scenarioId = batch.scenarioId ?? benchmark.scenarioId;
 				const result = opts.observerTierRouting === true ? await replayBatchExtractionWithTierRouting(resolveDbOpt(opts), observerConfigWithOverrides, {
 					batchId: batch.batchId,
@@ -5163,7 +5188,51 @@ function createMemoryExtractionBenchmarkCommand() {
 					scenarioId,
 					transcriptBudget: transcriptBudget ?? void 0
 				});
+				const costModel = result.observer.modelFallbackApplied ? result.observer.resolvedModel : result.observer.resolvedModel ?? result.observer.model;
+				const initialCost = costModel ? estimateExtractionModelCost(costModel, result.observer.initialUsage) : null;
+				const repairCost = costModel ? estimateExtractionModelCost(costModel, result.observer.repairedUsage) : null;
+				const totalCost = costModel ? estimateExtractionModelCost(costModel, result.observer.totalUsage) : null;
+				const pricing = costModel ? getExtractionModelPricing(costModel) : null;
+				const costUnavailableReason = totalCost ? null : result.observer.modelFallbackApplied && !result.observer.resolvedModel ? "model_fallback_unresolved" : result.observer.totalUsage == null ? "missing_usage" : "unknown_model_pricing";
+				const initialQuality = result.observer.initialDiagnostics ? scoreExtractionBenchmarkOutput({
+					parsed: result.observer.initialParsed,
+					diagnostics: result.observer.initialDiagnostics,
+					review: batch.review ?? {
+						status: "unreviewed",
+						reviewerNotes: "No durable-fact review has been recorded for this batch."
+					},
+					estimatedCostUsd: initialCost?.totalCostUsd ?? null,
+					expectedSummaryDisposition: batch.expectedSummaryDisposition
+				}) : null;
+				const repairQuality = result.observer.repairedParsed && result.observer.repairedDiagnostics ? scoreExtractionBenchmarkOutput({
+					parsed: result.observer.repairedParsed,
+					diagnostics: result.observer.repairedDiagnostics,
+					review: batch.review ?? {
+						status: "unreviewed",
+						reviewerNotes: "No durable-fact review has been recorded for this batch."
+					},
+					estimatedCostUsd: repairCost?.totalCostUsd ?? null,
+					expectedSummaryDisposition: batch.expectedSummaryDisposition
+				}) : null;
+				const finalQuality = result.observer.diagnostics ? scoreExtractionBenchmarkOutput({
+					parsed: result.observer.parsed,
+					diagnostics: result.observer.diagnostics,
+					review: batch.review ?? {
+						status: "unreviewed",
+						reviewerNotes: "No durable-fact review has been recorded for this batch."
+					},
+					estimatedCostUsd: totalCost?.totalCostUsd ?? null,
+					expectedSummaryDisposition: batch.expectedSummaryDisposition
+				}) : null;
+				const reconciled = reconcileExtractionBenchmarkStatus({
+					purpose: batch.purpose,
+					classification: result.classification,
+					finalFailureReasons: result.evaluation.failureReasons,
+					initialQuality,
+					finalQuality
+				});
 				runs.push({
+					iteration,
 					batchId: batch.batchId,
 					sessionId: batch.sessionId,
 					label: batch.label,
@@ -5171,17 +5240,23 @@ function createMemoryExtractionBenchmarkCommand() {
 					complexity: batch.complexity,
 					scenarioId,
 					expectedTier: batch.expectedTier ?? null,
+					expectedSummaryDisposition: batch.expectedSummaryDisposition,
 					analysis: {
 						eventSpan: result.analysis.eventSpan,
 						promptCount: result.analysis.promptCount,
 						toolCount: result.analysis.toolCount,
 						transcriptLength: result.analysis.transcriptLength
 					},
-					status: result.classification.status,
-					reason: result.classification.reason,
+					status: reconciled.status,
+					reason: reconciled.reason,
 					tier: result.observer.tier ?? "manual",
 					provider: result.observer.provider,
 					model: result.observer.model,
+					transport: result.observer.transport,
+					requestedModel: result.observer.requestedModel,
+					resolvedModel: result.observer.resolvedModel,
+					modelFallbackApplied: result.observer.modelFallbackApplied,
+					modelFallbackReason: result.observer.modelFallbackReason,
 					openaiUseResponses: result.observer.openaiUseResponses,
 					reasoningEffort: result.observer.reasoningEffort,
 					reasoningSummary: result.observer.reasoningSummary,
@@ -5189,22 +5264,89 @@ function createMemoryExtractionBenchmarkCommand() {
 					temperature: result.observer.temperature,
 					summaries: result.evaluation.counts.summaries,
 					observations: result.evaluation.counts.observations,
-					repairApplied: result.observer.repairApplied
+					repairApplied: result.observer.repairApplied,
+					initial: {
+						raw: result.observer.initialRaw,
+						status: result.initialClassification.status,
+						reason: result.initialClassification.reason,
+						pass: result.initialEvaluation.pass,
+						failureReasons: result.initialEvaluation.failureReasons,
+						summaries: result.initialEvaluation.counts.summaries,
+						observations: result.initialEvaluation.counts.observations,
+						diagnostics: result.observer.initialDiagnostics,
+						elapsedMs: result.observer.initialElapsedMs,
+						usage: result.observer.initialUsage,
+						quality: reconciled.initialQuality
+					},
+					repair: {
+						applied: result.observer.repairApplied,
+						raw: result.observer.repairedRaw,
+						status: result.repairedClassification?.status ?? null,
+						reason: result.repairedClassification?.reason ?? null,
+						pass: result.repairedEvaluation?.pass ?? null,
+						failureReasons: result.repairedEvaluation?.failureReasons ?? [],
+						summaries: result.repairedEvaluation?.counts.summaries ?? null,
+						observations: result.repairedEvaluation?.counts.observations ?? null,
+						diagnostics: result.observer.repairedDiagnostics,
+						elapsedMs: result.observer.repairedElapsedMs,
+						usage: result.observer.repairedUsage,
+						quality: repairQuality
+					},
+					telemetry: {
+						totalElapsedMs: result.observer.totalElapsedMs,
+						totalUsage: result.observer.totalUsage
+					},
+					pricing,
+					cost: {
+						initial: initialCost,
+						repair: repairCost,
+						total: totalCost,
+						unavailableReason: costUnavailableReason
+					},
+					quality: reconciled.quality
 				});
 			}
+			const reviewedQualityRuns = runs.filter((run) => run.quality?.weightedQualityScore != null);
+			const knownCostRuns = runs.filter((run) => run.cost.total != null);
+			const knownElapsedRuns = runs.filter((run) => run.telemetry.totalElapsedMs != null);
 			const summary = {
+				repetitions,
 				total: runs.length,
 				shapeQualityTotal: runs.filter((run) => run.purpose === "shape_quality").length,
 				shapeQualityPasses: runs.filter((run) => run.purpose === "shape_quality" && run.status === "pass").length,
 				shapeQualityFails: runs.filter((run) => run.purpose === "shape_quality" && run.status === "shape_fail").length,
 				expectedTierTotal: runs.filter((run) => run.expectedTier != null).length,
 				expectedTierMatches: runs.filter((run) => run.expectedTier != null && run.expectedTier === run.tier).length,
-				robustnessNoOutput: runs.filter((run) => run.status === "observer_no_output").length
+				robustnessNoOutput: runs.filter((run) => run.status === "observer_no_output").length,
+				summaryDispositionTotal: runs.filter((run) => run.quality != null).length,
+				summaryDispositionMatches: runs.filter((run) => run.quality?.summaryDisposition.score === 1).length,
+				reviewedQualityRuns: reviewedQualityRuns.length,
+				knownCostRuns: knownCostRuns.length,
+				unknownCostRuns: runs.length - knownCostRuns.length,
+				missingUsageRuns: runs.filter((run) => run.cost.unavailableReason === "missing_usage").length,
+				unknownPricingRuns: runs.filter((run) => run.cost.unavailableReason === "unknown_model_pricing").length,
+				fallbackUnresolvedRuns: runs.filter((run) => run.cost.unavailableReason === "model_fallback_unresolved").length,
+				totalKnownCostUsd: knownCostRuns.reduce((sum, run) => sum + (run.cost.total?.totalCostUsd ?? 0), 0),
+				knownElapsedRuns: knownElapsedRuns.length,
+				totalKnownElapsedMs: knownElapsedRuns.reduce((sum, run) => sum + (run.telemetry.totalElapsedMs ?? 0), 0),
+				perBatchStability: benchmark.batches.map((batch) => {
+					const batchRuns = runs.filter((run) => run.batchId === batch.batchId);
+					const passes = batchRuns.filter((run) => run.status === "pass").length;
+					return {
+						batchId: batch.batchId,
+						purpose: batch.purpose,
+						passes,
+						total: batchRuns.length,
+						passRate: batchRuns.length > 0 ? passes / batchRuns.length : null,
+						statuses: batchRuns.map((run) => run.status)
+					};
+				})
 			};
-			const uniqueObserverKeys = Array.from(new Set(runs.map((run) => `${run.provider}::${run.model}::${run.openaiUseResponses ? "responses" : "chat"}`)));
+			const uniqueObserverKeys = Array.from(new Set(runs.map((run) => `${run.provider}::${run.model}::${run.transport}`)));
 			const observerSummary = opts.observerTierRouting === true ? {
 				provider: uniqueObserverKeys.length === 1 ? runs[0]?.provider ?? observer.provider : "mixed",
 				model: uniqueObserverKeys.length === 1 ? runs[0]?.model ?? observer.model : "mixed",
+				transport: uniqueObserverKeys.length === 1 ? runs[0]?.transport ?? "unknown" : "mixed",
 				tierRouting: true,
 				openaiUseResponses: uniqueObserverKeys.length === 1 ? runs[0]?.openaiUseResponses ?? observer.openaiUseResponses : null,
 				reasoningEffort: uniqueObserverKeys.length === 1 ? runs[0]?.reasoningEffort ?? observer.reasoningEffort : "mixed",
@@ -5216,6 +5358,7 @@ function createMemoryExtractionBenchmarkCommand() {
 			} : {
 				provider: observer.provider,
 				model: observer.model,
+				transport: runs[0]?.transport ?? observer.getStatus().runtime,
 				tierRouting: false,
 				openaiUseResponses: observer.openaiUseResponses,
 				reasoningEffort: observer.reasoningEffort,
@@ -5229,7 +5372,8 @@ function createMemoryExtractionBenchmarkCommand() {
 				benchmark: {
 					id: benchmark.id,
 					title: benchmark.title,
-					scenarioId: benchmark.scenarioId
+					scenarioId: benchmark.scenarioId,
+					modelCandidates: benchmark.modelCandidates
 				},
 				observer: observerSummary,
 				summary,
@@ -5243,6 +5387,7 @@ function createMemoryExtractionBenchmarkCommand() {
 			p.log.info([
 				`Benchmark: ${benchmark.id} — ${benchmark.title}`,
 				`Observer: ${observerSummary.provider}/${observerSummary.model}`,
+				`Transport: ${observerSummary.transport}`,
 				`Tier routing: ${opts.observerTierRouting === true ? "yes" : "no"}`,
 				`OpenAI Responses: ${observerSummary.openaiUseResponses === null ? "mixed" : observerSummary.openaiUseResponses ? "yes" : "no"}`,
 				`Reasoning effort: ${observerSummary.reasoningEffort ?? "none"}`,
@@ -5250,12 +5395,23 @@ function createMemoryExtractionBenchmarkCommand() {
 				`Max output tokens: ${observerSummary.maxOutputTokens ?? "mixed"}`,
 				`Temperature: ${observerSummary.temperature ?? "mixed"}`,
 				`Transcript budget override: ${transcriptBudget ?? "default"}`,
+				`Repetitions: ${summary.repetitions}`,
 				`Shape-quality passes: ${summary.shapeQualityPasses}/${summary.shapeQualityTotal}`,
 				`Shape-quality fails: ${summary.shapeQualityFails}`,
 				`Expected-tier matches: ${summary.expectedTierMatches}/${summary.expectedTierTotal}`,
-				`Observer no-output cases: ${summary.robustnessNoOutput}`
+				`Observer no-output cases: ${summary.robustnessNoOutput}`,
+				`Summary disposition matches: ${summary.summaryDispositionMatches}/${summary.summaryDispositionTotal}`,
+				`Reviewed quality runs: ${summary.reviewedQualityRuns} (compare per-run dimensions; scores are fixture-specific)`,
+				`Known estimated cost: $${summary.totalKnownCostUsd.toFixed(6)} (${summary.knownCostRuns}/${summary.total}; missing usage=${summary.missingUsageRuns}, unknown pricing=${summary.unknownPricingRuns}, unresolved fallback=${summary.fallbackUnresolvedRuns})`,
+				`Known elapsed time: ${summary.totalKnownElapsedMs}ms (${summary.knownElapsedRuns}/${summary.total} run(s))`
 			].join("\n"));
-			for (const run of runs) p.log.message(`  [${run.batchId}] ${run.status.padEnd(18)} ${run.complexity.padEnd(10)} tier=${run.tier.padEnd(6)} expected=${(run.expectedTier ?? "n/a").padEnd(6)} span=${String(run.analysis.eventSpan).padEnd(3)} prompts=${run.analysis.promptCount} tools=${String(run.analysis.toolCount).padEnd(2)} transcript=${run.analysis.transcriptLength} ${run.provider}/${run.model}${run.openaiUseResponses ? " [responses]" : ""} summaries=${run.summaries} observations=${run.observations} repair=${run.repairApplied ? "yes" : "no"} — ${run.label}`);
+			for (const run of runs) {
+				const qualityLabel = run.quality?.weightedQualityScore == null ? "n/a" : run.quality.weightedQualityScore.toFixed(3);
+				const costLabel = run.cost.total == null ? "n/a" : `$${run.cost.total.totalCostUsd.toFixed(6)}`;
+				const latencyLabel = run.telemetry.totalElapsedMs == null ? "n/a" : `${run.telemetry.totalElapsedMs}ms`;
+				const missingRequired = run.quality?.requiredRecall.missingLabelIds.join(",") || "none";
+				p.log.message(`  [${run.batchId}#${run.iteration}] ${run.status.padEnd(18)} ${run.complexity.padEnd(10)} tier=${run.tier.padEnd(6)} expected=${(run.expectedTier ?? "n/a").padEnd(6)} disposition=${run.quality?.summaryDisposition.actual ?? "n/a"}/${run.expectedSummaryDisposition} span=${String(run.analysis.eventSpan).padEnd(3)} prompts=${run.analysis.promptCount} tools=${String(run.analysis.toolCount).padEnd(2)} transcript=${run.analysis.transcriptLength} ${run.provider}/${run.model} [${run.transport}] initial=${run.initial.summaries}s/${run.initial.observations}o final=${run.summaries}s/${run.observations}o quality=${qualityLabel} coverage=${run.quality?.weightedQualityCoverage?.toFixed(3) ?? "n/a"} required_missing=${missingRequired} cost=${costLabel} latency=${latencyLabel} schema_loss=${run.initial.diagnostics?.dataLoss === true ? "yes" : "no"} fallback=${run.modelFallbackApplied ? "yes" : "no"} repair=${run.repairApplied ? "yes" : "no"} — ${run.label}`);
+			}
 			p.outro("done");
 		} catch (error) {
 			const message = error instanceof Error ? error.message : "Extraction benchmark failed";
@@ -5976,6 +6132,14 @@ function sqliteVecFailureDiagnostics(error, dbPath) {
 		`error=${message}`
 	];
 }
+async function runServeCoordinatorMaintenance(store, dependencies) {
+	const projectShares = await dependencies.advancePendingProjectShares(store, { limit: 3 });
+	if (projectShares.failed > 0) throw new Error(`share operation maintenance failed for ${projectShares.failed} of ${projectShares.processed} operations`);
+	return {
+		projectShares,
+		recipientPolicies: await dependencies.reconcileRecipientPolicyProjects(store, { limit: 3 })
+	};
+}
 async function startBackgroundViewer(invocation) {
 	warnIfViewerExposed(invocation.host, invocation.port);
 	if (await isPortOpen(invocation.host, invocation.port)) {
@@ -6008,7 +6172,7 @@ async function startBackgroundViewer(invocation) {
 	p.outro(`Viewer started in background (pid ${child.pid}) at http://${invocation.host}:${invocation.port}`);
 }
 async function startForegroundViewer(invocation) {
-	const { createApp, createSyncApp, closeStore, getStore } = await import("@codemem/server");
+	const { advancePendingProjectShares, createApp, createSyncApp, closeStore, getStore, reconcileRecipientPolicyProjects } = await import("@codemem/server");
 	const { serve } = await import("@hono/node-server");
 	if (invocation.dbPath) process.env.CODEMEM_DB = invocation.dbPath;
 	if (invocation.configPath) process.env.CODEMEM_CONFIG = invocation.configPath;
@@ -6099,6 +6263,12 @@ async function startForegroundViewer(invocation) {
 					port: syncConfig.syncPort,
 					signal: syncAbort.signal,
 					scanner: store.scanner,
+					onAfterCoordinatorRefresh: async () => {
+						await runServeCoordinatorMaintenance(store, {
+							advancePendingProjectShares,
+							reconcileRecipientPolicyProjects
+						});
+					},
 					onPhaseChange: (phase) => {
 						if (phase === "running") {
 							syncRuntimeStatus.phase = null;