{ "schemaVersion": 1, "initiative": "EVE_SOTA_GAP_CLOSURE", "task": "0.2", "capturedAt": "2026-09-01T15:09:53Z", "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "scope": { "sourceFacts": "Repository sources and committed audit evidence reachable at sourceCommit.", "runtimeFacts": "One local developer-host observation; not production, staging, or a promise of future availability.", "numbersRule": "Every numeric object names an evidenceId whose command and source commit are recorded below." }, "evidenceCommands": { "C0": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "command": "git rev-parse HEAD" }, "C1": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "command": "jq '{promptBytesHash,promptBytesComponents}' docs/audits/eve-smx-ratchet.json && jq -r '.entries[] | select(.kind==\"tool-definition\") | .source.file' docs/audits/EVE_BUILDER_PROMPT_INVENTORY_2026-08.json | sort | uniq -c | sort -nr" }, "C2": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "command": "jq '{familyFloors,familyFloorsSource,builderWilsonFloors,builderWilsonFloorsComment,telemetryFloors}' docs/audits/eve-smx-ratchet.json" }, "C3": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "historicalEvidenceCommit": "2352ee93c5391e7d8c18ecaf87fd3225380228d0", "weeklyEvidenceCommit": "cb5d630822f85933ab63b228d5336be3b8b0a6b2", "command": "jq '{generatedAt,baseCommit,liveModelVerification,weeklyEvidenceLifecycle}' docs/audits/eve-phase-11/2026-08-30-run.json && jq '{capturedDate,sourceCommit,costReport,sameDayRouteCalibration,remedy,decision}' docs/audits/eve-smx-weekly/2026-08-23-follow-up.json" }, "C4": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "historicalEvidenceCommit": "44733109326b09bddc7a1113cfb1b3dd1d877ce1", "command": "jq '{generatedAt,baseCommit,registry,phase7StrictBrowserBattery,phase8}' docs/audits/eve-phase-7-8/2026-08-30-run.json" }, "C5": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "command": "jq '{schemaVersion,contentHash,siteCount,pointCount:(.points|length),bySurface:(.points|group_by(.id|split(\".\")[0])|map({surface:.[0].id|split(\".\")[0],points:length,sites:map(.files|length)|add}))}' apps/oshun/bff/src/assistant/generated/assistant-invocation-inventory.json" }, "C6": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "command": "pnpm --dir libs/oshun/shell-assistant exec tsx -e \"import {CURATED_ASSISTANT_TOURS as t} from './src/tour-plans.ts'; console.log(JSON.stringify({total:t.length,byAudience:Object.fromEntries(['customer','studio','admin'].map(a=>[a,t.filter(x=>x.audience===a).length])),steps:t.reduce((n,x)=>n+x.steps.length,0)}))\"" }, "C7": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "command": "pnpm --dir apps/oshun/bff exec tsx -e \"import {ASSISTANT_MODEL_REGISTRY as r} from './src/assistant/model-registry.ts'; console.log(JSON.stringify({total:r.length,bound:r.filter(x=>x.status==='bound').length,unbound:r.filter(x=>x.status==='unbound').length,entries:r}))\"" }, "C8": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "command": "rg -n 'OPERATOR_MEMORY_CATEGORIES|SLOT_CEILINGS|resolveOperatorMemoryEnabled|operatorMemoryEnabled|assistant_operator_memory' apps/oshun/bff/src/assistant/operator-memory.ts" }, "C9": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "command": "rg -l 'data-admin-selection-scope=' apps/oshun/admin/src/components --glob '*.tsx' && rg -n 'data-admin-selection-scope=|Start voice input|Enable voice output|Dictate a message|Play this reply aloud' apps/oshun/web/src/components/assistant/AssistantPanel.tsx apps/oshun/admin/src/components" }, "L1": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "observedAt": "2026-09-01T15:09:53Z", "command": "docker exec oshun-dev-postgres psql -U oshun -d oshun_dev -X -v ON_ERROR_STOP=1 -Atc ''" }, "L2": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "observedAt": "2026-09-01T15:07:00Z", "command": "docker ps --format '{{.Names}}|{{.Image}}|{{.Status}}|{{.Ports}}'; for eve_probe_port in 6379 9000 9001 6333; do timeout 1 bash -c \"'" }, "L4": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "observedAt": "2026-09-01T15:07:15Z", "command": "test -x /root/workspace/UnrealEngine-5.5/Engine/Binaries/Linux/UnrealEditor-Cmd && runuser -u ueagent -- /root/workspace/UnrealEngine-5.5/Engine/Binaries/Linux/UnrealEditor-Cmd -Version" }, "L5": { "sourceCommit": "778b9c874163fe9f958cfb44b54b23ab2b444eff", "observedAt": "2026-09-01T15:06:41Z", "command": "stat -c '%a %n' /root/Desktop/workspace/drawer/env-master.env; source that file in a subshell and print only set/unset for OPENROUTER_API_KEY, DEEPGRAM_API_KEY, OPENAI_API_KEY, ELEVENLABS_API_KEY, and CARTESIA_API_KEY" } }, "planesAndSeams": { "promptToolUnion": { "total": 89, "memberPlane": { "total": 43, "baseMember": 39, "audit": 3, "deferredRouter": 1 }, "builderPlane": { "total": 46, "adminOperations": 15, "intentWorkbench": 23, "hathorContentBridge": 2, "workbenchKit": 6 }, "qualityPlaneSeparateModelTools": 0, "evidenceId": "C1" }, "agentPlaneMcp": { "tools": 6, "evidenceId": "C4" }, "workbenchKitReadSeam": { "tools": 1, "workbenches": 4, "views": 31, "viewCounts": { "tara-workbench": 10, "launch-readiness-governance": 2, "hathor": 4, "isis": 15 }, "cardGatedMutationTools": 5, "evidenceId": "C4" } }, "prompt": { "hash": "cad7c1cb4098c319112dae202dce495dbc2c27b1143957105a3b1e9b94939b8b", "coreConductBytes": 845, "toollessConductBytes": 335, "toolCount": 89, "toolDescriptionBytes": 36462, "toolDefinitionBytes": 61551, "skillCount": 7, "skillBytes": 3945, "evidenceId": "C1" }, "evaluation": { "baselineRunsPerCase": 3, "baselineCases": 121, "baselineRuns": 363, "baselineBilledCostUsd": 0.1109, "families": [ { "family": "audit", "cases": 7, "runs": 21, "passAllFloor": 0.7142 }, { "family": "capability-smalltalk", "cases": 19, "runs": 57, "passAllFloor": 0.8421 }, { "family": "docs", "cases": 7, "runs": 21, "passAllFloor": 0.2857 }, { "family": "general", "cases": 9, "runs": 27, "passAllFloor": 0.8888 }, { "family": "member-data", "cases": 48, "runs": 144, "passAllFloor": 0.75 }, { "family": "navigate", "cases": 10, "runs": 30, "passAllFloor": 0.9 }, { "family": "safety", "cases": 3, "runs": 9, "passAllFloor": 1 }, { "family": "tour", "cases": 7, "runs": 21, "passAllFloor": 1 }, { "family": "workbench-read", "cases": 9, "runs": 27, "passAllFloor": 0.2222 }, { "family": "workbench-write", "cases": 2, "runs": 6, "passAllFloor": 0.5 } ], "builderWilson95Lower": [ { "family": "docs", "cases": 5, "pooledRuns": 50, "passedRuns": 49, "floor": 0.895 }, { "family": "workbench-read", "cases": 24, "pooledRuns": 240, "passedRuns": 220, "floor": 0.8747 }, { "family": "workbench-write", "cases": 25, "pooledRuns": 250, "passedRuns": 229, "floor": 0.875 } ], "evidenceId": "C2" }, "costCacheLatency": { "baselineDeck": { "runs": 363, "billedCostUsd": 0.1109, "cacheReadRate": null, "medianLatencySeconds": null, "evidenceId": "C2" }, "phase11CuratedTour": { "cases": 1, "runs": 10, "passed": 10, "billedCostUsd": 0.0063, "medianLatencySeconds": 3.7, "cacheReadRate": 0.472, "cacheReportingRuns": 8, "evidenceId": "C3" }, "phase11ReleaseBlocked": { "cases": 7, "runs": 70, "classifiedPassed": 59, "billedCostUsd": 0.0295, "medianLatencySeconds": 8.1, "cacheReadRate": 0.836, "cacheReportingRuns": 63, "evidenceId": "C3" }, "telemetryFloor": { "cacheReadRateFloor": 0.7, "evidenceId": "C2" }, "weeklyDriftObservation": { "totalTurns": 3233, "costReportedTurns": 246, "costPerReportedTurnUsd": 0.000727, "cacheReadRate": 0.265, "fixedBatteryRuns": 150, "fixedBatteryCacheReadRate": 0.92, "evidenceId": "C3" } }, "localQueueAndLeases": { "scope": "developer Postgres oshun_dev, not production or staging", "observedAt": "2026-09-01T15:09:53Z", "workItems": 4, "byStatus": { "draft": 2, "ready": 1, "verified": 1 }, "queueEligible": 1, "leasedOrInProgress": 0, "leaseRows": 0, "liveLeases": 0, "expiredLeases": 0, "distinctHolders": 0, "evidenceId": "L1" }, "memory": { "sourcePosture": { "defaultOnWhenDatabaseBound": true, "databaseRequiredForDurability": true, "sessionOnlyWithoutDatabase": true, "categories": 3, "slotCeilings": { "preferences": 12, "standing-initiative": 1, "pinned-contexts": 8 }, "evidenceId": "C8" }, "localObservation": { "databaseUrlBoundInCurrentShell": false, "killSwitchExplicitlySetInCurrentShell": false, "tablePresent": true, "rows": 0, "operators": 0, "populatedCategories": 0, "evidenceId": "L1" } }, "interactionSurfaces": { "invocations": { "points": 12, "sites": 25, "customerWeb": { "points": 9, "sites": 22 }, "adminWeb": { "points": 3, "sites": 3 }, "evidenceId": "C5" }, "tours": { "reviewedPlans": 7, "steps": 17, "customerPlans": 3, "studioPlans": 1, "adminPlans": 3, "evidenceId": "C6" }, "voice": { "webShells": 2, "controls": 4, "providerBackedVoiceCredentialsConfigured": 0, "providerBackedVoiceCredentialNamesChecked": 4, "browserFallbackImplemented": true, "evidenceId": "C9", "credentialEvidenceId": "L5" }, "selection": { "registeredInvocationPoints": 2, "adminScopeLabels": 3, "adminSourceComponents": 4, "evidenceId": "C5", "scopeEvidenceId": "C9" } }, "modelRegistry": { "legs": 4, "bound": 3, "unbound": 1, "entries": [ { "leg": "turn", "status": "bound", "slug": "deepseek/deepseek-v4-flash-0731", "inputPerMillionUsd": 0.04494, "outputPerMillionUsd": 0.08988, "priceSnapshotDate": "2026-08-30", "measuredEndpoints": 3 }, { "leg": "escalation", "status": "bound", "slug": "deepseek/deepseek-v4-pro-0813", "inputPerMillionUsd": 1.218, "outputPerMillionUsd": 2.436, "priceSnapshotDate": "2026-08-17" }, { "leg": "judge", "status": "bound", "slug": "deepseek/deepseek-v4-pro-0813", "inputPerMillionUsd": 1.218, "outputPerMillionUsd": 2.436, "priceSnapshotDate": "2026-08-17" }, { "leg": "embedding", "status": "unbound" } ], "evidenceId": "C7" }, "localDependencies": { "scope": "developer host observation", "node": { "available": true, "version": "24.15.0", "evidenceId": "L3" }, "pnpm": { "available": true, "version": "10.25.0", "evidenceId": "L3" }, "postgres": { "available": true, "version": "16.14", "evidenceId": "L1" }, "pgvector": { "available": true, "version": "0.8.2", "evidenceId": "L1" }, "redis": { "available": true, "image": "redis:7-alpine", "ping": "PONG", "evidenceId": "L2" }, "minio": { "available": true, "healthy": true, "imageUsesMutableLatestTag": true, "evidenceId": "L2" }, "qdrant": { "available": false, "port": 6333, "evidenceId": "L2" }, "playwright": { "available": true, "version": "1.60.0", "installedEngines": 3, "chromium": true, "firefox": true, "webkit": true, "evidenceId": "L3" }, "mobile": { "adbAvailable": true, "adbVersion": "1.0.41", "platformToolsVersion": "34.0.4", "maestroAvailable": true, "maestroVersion": "2.5.1", "attachedDevices": 0, "evidenceId": "L3" }, "blender": { "available": true, "version": "4.0.2", "evidenceId": "L3" }, "unreal": { "requiredPathExecutable": true, "engineVersion": "5.5.4", "headlessVersionProbeExitCode": 1, "limitation": "Binary initializes as ueagent and reports its version, then exits because this host has no usable Vulkan/display device.", "evidenceId": "L4" }, "codingAgents": { "codexAvailable": true, "codexVersion": "0.152.0", "claudeCodeAvailable": true, "claudeCodeVersion": "2.1.247", "evidenceId": "L3" }, "providerCredentials": { "secretFileMode": "0600", "openRouter": true, "deepgram": false, "openAi": false, "elevenLabs": false, "cartesia": false, "evidenceId": "L5" } }, "claimBoundary": { "productionStateObserved": false, "stagingStateObserved": false, "fullDeckRerunForThisBaseline": false, "externalProviderCallsMadeForThisBaseline": false, "dependencyRecordIsNonExpiring": false, "nextOwner": "Task 0.7 replaces the host observation with a stale-expiring capability record." }, "documentedDrift": [ { "section": "modelRegistry", "task": "3.2, 15.1", "reason": "Two later tasks moved this section, and the capture stays frozen. Task 3.2 bound the embedding leg: it was unbound at capture because docs retrieval is lexical BM25 and binding a slug nothing calls would have been a paper pin, and 3.2 priced the whole enumerated field from billed calls and bound perplexity/pplx-embed-v1-0.6b at $0.004/1M \u2014 a pricing and posture decision, with relevance still unmeasured and owned by task 3.5 (docs/audits/EVE_SOTA_EMBEDDING_PRICING_2026-09.md). Task 15.1 then widened the vocabulary from four legs to nine, because a leg omitted from a registry is a leg nobody examined: two operator-configured speech legs and three not-admitted legs (reranker, vision, media) joined the four bound ones, so the capture's `unbound` status no longer exists and the count is 0. The same task re-priced the chat legs from the endpoint feed under unchanged pins and recorded genuine auction drift: turn $0.04494 -> $0.04998 per million input (+11.2%) and escalation/judge $1.218/$2.436 -> $1.1088/$3.3264 (input -9.0%, output +36.6%), the floor having moved from GMICloud to Novita (docs/audits/EVE_SOTA_MODEL_LEG_CONTRACT_2026-09.md). Reproduce with the capture's own command, C7.", "capturedValue": { "legs": 4, "bound": 3, "unbound": 1, "entries": [ { "leg": "turn", "status": "bound", "slug": "deepseek/deepseek-v4-flash-0731", "inputPerMillionUsd": 0.04494, "outputPerMillionUsd": 0.08988, "priceSnapshotDate": "2026-08-30", "measuredEndpoints": 3 }, { "leg": "escalation", "status": "bound", "slug": "deepseek/deepseek-v4-pro-0813", "inputPerMillionUsd": 1.218, "outputPerMillionUsd": 2.436, "priceSnapshotDate": "2026-08-17" }, { "leg": "judge", "status": "bound", "slug": "deepseek/deepseek-v4-pro-0813", "inputPerMillionUsd": 1.218, "outputPerMillionUsd": 2.436, "priceSnapshotDate": "2026-08-17" }, { "leg": "embedding", "status": "unbound" } ], "evidenceId": "C7" }, "currentValue": { "legs": 9, "bound": 4, "unbound": 0, "entries": [ { "leg": "turn", "status": "bound", "slug": "deepseek/deepseek-v4-flash-0731", "inputPerMillionUsd": 0.04998, "outputPerMillionUsd": 0.09996, "priceSnapshotDate": "2026-09-05", "measuredEndpoints": 3 }, { "leg": "escalation", "status": "bound", "slug": "deepseek/deepseek-v4-pro-0813", "inputPerMillionUsd": 1.1088, "outputPerMillionUsd": 3.3264, "priceSnapshotDate": "2026-09-05" }, { "leg": "judge", "status": "bound", "slug": "deepseek/deepseek-v4-pro-0813", "inputPerMillionUsd": 1.1088, "outputPerMillionUsd": 3.3264, "priceSnapshotDate": "2026-09-05" }, { "leg": "embedding", "status": "bound", "slug": "perplexity/pplx-embed-v1-0.6b", "inputPerMillionUsd": 0.004, "outputPerMillionUsd": 0, "priceSnapshotDate": "2026-09-05" }, { "leg": "reranker", "status": "not-admitted" }, { "leg": "speech-to-text", "status": "operator-configured" }, { "leg": "text-to-speech", "status": "operator-configured" }, { "leg": "vision", "status": "not-admitted" }, { "leg": "media", "status": "not-admitted" } ], "evidenceId": "C7" } } ] }