Skip to content

AIFACTORY-FEAT-065: QualityScoreService: thin delivery for the first outcome

Generated from the architecture model

This page is rendered from model/**/*.yaml by tools/docs_model during mkdocs build. Edit the YAML, not this page; hand-written design lives on the linked service, platform and plane pages.

A design artefact, not a tracker

This backlog is generated from model/backlog/ and validated against the architecture model. No work item exists in any tracker, and no item carries a date, an estimate or an assignee (ADR-0029).

Epic Observability (PF-13) · Increment I1 Generate a microservice · State elaborated · Route backend-developer · Build order 79 of 185 (Factory code)

QualityScoreService is the service of bounded context BC-22 (Evaluation & Learning). Evaluation as a core capability: suites per target (models, agents, skills, tools, workflows, context and memory strategies, policies) with the quality-gate metrics, evaluation runs over datasets of historical tasks, verdicts with baseline comparison, waivers, drift monitoring and the evidence records publication reads (BC-22, D32, original design §47–49). In this increment it is delivered thin: Accepts evaluation and outcome events and stores them; no gate and no scoring. Its own repository is ConnectSoft.AI.SoftwareFactory.QualityScoreService.

Delivers QualityScoreService (thin): Accepts evaluation and outcome events and stores them; no gate and no scoring.
Stubs FeedbackService as contract-stub, completed in I4
Realizes adrs: ADR-0021, ADR-0014
Requirements FR-02, FR-04, FR-11, FUX-02, FUX-10, NFR-11, NFR-13, XL-08, XL-15
Depends on —
Sources model/services/observability/QualityScoreService.yaml

Scenarios to trace (12, catalog): SCN-QualityScoreService-001 ✓, SCN-QualityScoreService-002 ✓, SCN-QualityScoreService-003 ✓, SCN-QualityScoreService-004 ✓, SCN-QualityScoreService-005 ✓, SCN-QualityScoreService-006 ✓, SCN-QualityScoreService-007 ✓, SCN-QualityScoreService-008 ✓, SCN-QualityScoreService-009 ✓, SCN-QualityScoreService-010 ✓, SCN-QualityScoreService-011 ✓, SCN-QualityScoreService-012 ✓

AIFACTORY-US-071: QualityScoreService accepts the evaluation and outcome events of the first outcome and stores them, with no gate and no scoring

behavior · persona asset-maintainer As an asset maintainer, I want the outcomes of the first runs and the reliability signals of their model, tool, context and validation calls kept from the first outcome on, so that the evaluation gate and the drift monitor of the fourth increment start from real history instead of an empty store.

This is the integration PBI of the thin delivery of QualityScoreService, the service of bounded context BC-22 (Evaluation and Learning). The scope of the thin delivery is stated in model/backlog/features/observability.yaml and is exactly this: the service accepts evaluation and outcome events and stores them; it applies no gate and computes no score. Ten consumed events of model/services/observability/QualityScoreService.yaml are in that scope, all published by services delivered in I1: RunCompleted and AttemptOutcomeRecorded of RunControlService, AttemptStarted and AttemptOutcomeReported of AgentRuntimeService (AIFACTORY-US-1087), and ModelInvoked of ModelRouterService, ToolInvoked of ToolAdapterService, ContextAssembled of ContextBuilderService, RetrievalExecuted of VectorMemoryService, ArtifactValidated of ArtifactService and PipelineRunCompleted of PipelineExecutionService (AIFACTORY-US-1088). Each is handled once through the inbox of the quality-score messaging profile, idempotent on the event id, and stored under the tenant of its envelope in quality-score-sql, the primary store of the service (AIFACTORY-US-1089). Nothing else of the model is delivered here. The service hosts none of its five REST operations and none of its two gRPC methods, writes no EvaluationSuite and no EvaluationRun and publishes none of its five events; the handlers of AssetReleaseSubmitted (the release gate), ApprovalDecided (waivers), DatasetPublished of FeedbackService and the three retention events of RetentionResidencyService are not registered, and Evaluation.GetVerdict is not called in this increment, nor later by the Published gate of AgentRegistryService, which from I4 checks the evidence it recorded itself from EvaluationCompleted (D141); under the ADR-0021 amendment (D83) a Draft workflow or agent release runs only in the acceptance environment and the internal projects the platform rule names, and publication itself is never waived. Every one of the twelve scenarios of the service is therefore delivered by feature quality-score-service-complete in I4, against the same model, so the complete delivery adds operations and events without changing what the thin delivery stores. The feature declares a contract stub of FeedbackService that completes in I4; the thin scope calls no operation of FeedbackService, so the stub is a compiled client and a stub server of its internal dataset method kept in the test suite and nothing in the running service. The thin delivery applies no retention period of its own: the handlers of the three retention events are not registered, and the rules of ADR-0019 are applied to the records it stores by AIFACTORY-US-1097 of feature quality-score-service-complete in I4. Repository: ConnectSoft.AI.SoftwareFactory.QualityScoreService, created and scaffolded in I0 by AIFACTORY-TASK-115 of feature repository-and-pipeline-conventions (D140), by the steps of its repository checklist; this feature starts from that scaffold.

Acceptance criteria

AC1 Given the repository ConnectSoft.AI.SoftwareFactory.QualityScoreService created and scaffolded in I0 by AIFACTORY-TASK-115 of feature repository-and-pipeline-conventions, whose first commit is the unchanged output of dotnet new connectsoft-base -n ConnectSoft.AI.SoftwareFactory.QualityScoreService --UseFactoryProfile true from the pinned template package, when its pipeline runs with the work of this feature, then the solution builds, the unit and acceptance tests pass with complete reports and a container image is published (ADR-0028, model/services/observability/QualityScoreService.yaml).
AC2 Given an I1 run of the chain that ends, when its events reach the service, then RunControlService/RunCompleted, RunControlService/AttemptOutcomeRecorded, AgentRuntimeService/AttemptStarted, AgentRuntimeService/AttemptOutcomeReported, ModelRouterService/ModelInvoked and ToolAdapterService/ToolInvoked are each stored once under the tenant of their envelope, as AIFACTORY-US-1087 and AIFACTORY-US-1088 prove event by event, and no score, verdict or evaluation run exists afterwards (ADR-0014, ADR-0021).
AC3 Given the thin service deployed in the chain, when the descriptor test of AIFACTORY-US-1089 runs in its pipeline, then the hosted surface is the thin scope of model/backlog/features/observability.yaml and nothing more, as AIFACTORY-US-1089 proves (ADR-0028).
AC4 Given the service deployed in the chain, when it starts, then readiness waits for quality-score-sql and the transport and the records of the chain run are read only under the tenant row filter, as AIFACTORY-US-1089 proves (ADR-0015).
Traces: services QualityScoreService
Traces: events RunControlService/RunCompleted, RunControlService/AttemptOutcomeRecorded, AgentRuntimeService/AttemptStarted, AgentRuntimeService/AttemptOutcomeReported, ModelRouterService/ModelInvoked, ToolAdapterService/ToolInvoked
Traces: adrs ADR-0014, ADR-0015, ADR-0021, ADR-0028
Traces: requirements NFR-11, XL-08
Depends on AIFACTORY-US-1087, AIFACTORY-US-1088, AIFACTORY-US-1089, feature:repository-and-pipeline-conventions, feature:component-base-template-factory-profile, feature:run-control-service, feature:agent-runtime-service, feature:model-router-service, feature:tool-adapter-service
Sources model/services/observability/QualityScoreService.yaml, model/backlog/features/observability.yaml, docs/architecture/adr/0021-quality-gates.md, docs/planes/trust-operations/observability/services/quality-score-service/index.md#purpose-and-responsibilities

Tasks

Task Kind Expected output Repository
AIFACTORY-TASK-116 Implement the thin scope of QualityScoreService for the first outcome implementation The service accepts the ten evaluation and outcome events of the thin scope and stores them, with no gate and no scoring; it hosts no operation and publishes no event, so the complete delivery of I4 adds the five REST operations, the two gRPC methods, the five events and the remaining handlers of model/services/observability/QualityScoreService.yaml without changing what is stored here. ConnectSoft.AI.SoftwareFactory.QualityScoreService
AIFACTORY-TASK-773 Verify quality-score-service-thin-contract by running the declared command verification Run dotnet build -c Release and dotnet test -c Release --no-build in the pipeline of the repository; a run that discovers no test counts as a failure; they pass and prove AC1 to AC4 of AIFACTORY-US-071: the scaffold with the Factory profile, one chain run whose events are stored once each with no score or verdict, and the thin surface, readiness and tenant row filter proven by AIFACTORY-US-1089 holding in the deployed chain. ConnectSoft.AI.SoftwareFactory.QualityScoreService

AIFACTORY-US-1087: Run and attempt outcome events are accepted once and stored per attempt and release, without scoring

behavior · persona asset-maintainer As an asset maintainer, I want every run end and every attempt outcome of the first outcome kept with the release that produced it, so that the drift monitor of the complete delivery can compare production outcomes per release with evaluated scores from the first run on.

Four consumed events of model/services/observability/QualityScoreService.yaml carry the outcomes of runs and attempts, and the thin delivery accepts and stores them. AttemptStarted of AgentRuntimeService (handler CollectRuntimeStats) opens the runtime sample of an attempt with its attemptId, runId, stepId, releaseId and snapshotId; it is the only one of the four that names the release, so the release of an attempt is taken from it. AttemptOutcomeReported of AgentRuntimeService (handler CollectRuntimeStats) adds the state, the failure class, the outcome hash and the counts of corrections, model calls and tool calls of the attempt. AttemptOutcomeRecorded of RunControlService (handler CollectOutcome) adds the authoritative outcome and failure class that Run Control recorded exactly once. RunCompleted of RunControlService (handler ScoreEvaluationTask) records the end of a run with its outcome hash and, when present, its output reference; in the complete delivery the same handler scores the tasks of evaluation runs, but no evaluation run exists in this increment, so here it only stores the run end as a production outcome. Each event arrives through the inbox of the quality-score messaging profile and is applied once, idempotent on its event id, so a redelivered event changes nothing. The tenant and the project come from the run envelope of the event, never from a payload field, and every record is written and read under the tenant row filter of the service (ADR-0015). The events of one attempt may arrive in any order: an outcome that arrives before its AttemptStarted is stored under its attemptId and joins the release when the start arrives, and no record is dropped for order. Nothing is scored, no threshold is read, no EvaluationRun is written and no event is published; the comparison of these outcomes with evaluated scores is the drift monitor of feature quality-score-service-complete in I4. When the store is unavailable the inbox does not acknowledge the event, the transport redelivers it and it is applied once when the store is back (ADR-0012). The records hold identifiers, states, classes, hashes and counts, never a prompt, a tool argument or an artifact body.

Acceptance criteria

AC1 Given an attempt of an I1 run, when AgentRuntimeService/AttemptStarted, AgentRuntimeService/AttemptOutcomeReported and RunControlService/AttemptOutcomeRecorded arrive, then one record of the attempt exists with its run, step, release and snapshot, the reported state, failure class, outcome hash and counts of corrections, model calls and tool calls, and the recorded outcome of Run Control, under the tenant and project of the run envelope (model/services/observability/QualityScoreService.yaml, ADR-0014).
AC2 Given the outcome events of an attempt arrive before its AgentRuntimeService/AttemptStarted, when the start arrives later, then the stored outcome joins the release and snapshot of the start and nothing is lost; given any of the four events delivered twice, then the second delivery changes nothing (ADR-0012).
AC3 Given a run ends, when RunControlService/RunCompleted arrives, then the run end is stored with its outcome hash and output reference as a production outcome, no evaluation task is scored, no EvaluationRun is written and no event of QualityScoreService is published (ADR-0021, model/backlog/features/observability.yaml).
AC4 Given quality-score-sql is unavailable while RunControlService/AttemptOutcomeRecorded arrives, when the store returns, then the event is redelivered and applied once; given records of two tenants, then a read under one tenant returns only its own records, and no record holds a prompt, a tool argument or an artifact body (ADR-0012, ADR-0015).
Traces: services QualityScoreService, RunControlService, AgentRuntimeService
Traces: events RunControlService/RunCompleted, RunControlService/AttemptOutcomeRecorded, AgentRuntimeService/AttemptStarted, AgentRuntimeService/AttemptOutcomeReported
Traces: adrs ADR-0012, ADR-0014, ADR-0015, ADR-0021
Traces: requirements NFR-11, XL-08
Depends on AIFACTORY-US-1089, feature:run-control-service, feature:agent-runtime-service
Sources model/services/observability/QualityScoreService.yaml, model/services/control-plane/RunControlService.yaml, model/services/agent-mesh/AgentRuntimeService.yaml, docs/planes/trust-operations/observability/services/quality-score-service/index.md#messages

Tasks

Task Kind Expected output Repository
AIFACTORY-TASK-4357 Handlers CollectRuntimeStats, CollectOutcome and the storing half of ScoreEvaluationTask implementation The handlers of AttemptStarted and AttemptOutcomeReported (CollectRuntimeStats), AttemptOutcomeRecorded (CollectOutcome) and RunCompleted (ScoreEvaluationTask, storing only) write one record per attempt and one per run end, keyed by attemptId and runId, with the tenant and project of the run envelope, join an early outcome to the release of a later start, are idempotent on the event id through the inbox, and score nothing. ConnectSoft.AI.SoftwareFactory.QualityScoreService
AIFACTORY-TASK-4358 Event fixtures of the four run and attempt events in every arrival order fixture The acceptance suite holds fixtures of AttemptStarted, AttemptOutcomeReported, AttemptOutcomeRecorded and RunCompleted built from the contracts package under the run envelope of two tenants, in order, out of order and redelivered, and a switch that makes quality-score-sql unavailable. ConnectSoft.AI.SoftwareFactory.QualityScoreService
AIFACTORY-TASK-4359 Verify the run and attempt outcome records through the acceptance suite verification Run dotnet test -c Release --filter "Category=Acceptance&Feature=RunAndAttemptOutcomes"; a zero-match filter counts as a failure; the tests prove AC1 to AC4 of AIFACTORY-US-1087 for RunControlService/RunCompleted, RunControlService/AttemptOutcomeRecorded, AgentRuntimeService/AttemptStarted and AgentRuntimeService/AttemptOutcomeReported: the joined record, arrival out of order, the redelivery without a second effect, the store outage and the tenant row filter, with no EvaluationRun written and no event published. ConnectSoft.AI.SoftwareFactory.QualityScoreService

AIFACTORY-US-1088: Model, tool, context, retrieval and validation events are accepted once and stored as reliability signals, without scoring

behavior · persona asset-maintainer As an asset maintainer, I want the reliability signals of model calls, tool calls, context assembly, retrieval and validators kept per run and attempt, so that the quality-gate metrics of the complete delivery can be computed over history the Factory already has.

Six consumed events of model/services/observability/QualityScoreService.yaml carry the signals from which the quality-gate metrics of ADR-0021 are computed later, and the thin delivery accepts and stores them. ModelInvoked of ModelRouterService (handler CollectModelStats) gives the receipt of a model call with its profile, deployment and outcome, the reliability of structured output and tool calls per deployment; a receipt of a call outside a run names no run and no attempt and is stored under its tenant and project without one. ToolInvoked of ToolAdapterService (handler CollectToolStats) gives the receipt of a tool call with the tool, its version, the effective class and the state, the signal of tool selection and argument validity. ContextAssembled of ContextBuilderService (handler CollectContextStats) gives the status of a context snapshot and the tokens used against the budget. RetrievalExecuted of VectorMemoryService (handler CollectRetrievalStats) gives the result count and the latency of a retrieval. ArtifactValidated of ArtifactService and PipelineRunCompleted of PipelineExecutionService (handler CollectValidationStats) give validator outcomes and build and test results. Each event is applied once through the inbox of the quality-score messaging profile, idempotent on its event id, and stored under the tenant of its envelope. ModelInvoked of a call in a run, ToolInvoked and ContextAssembled name their run and attempt and are stored with them, so these signals are later attributed to the release of the attempt through the records of AIFACTORY-US-1087. RetrievalExecuted names at most a run and no attempt, ArtifactValidated names an artifact and PipelineRunCompleted a project and a commit; these three are stored by tenant and project with the run, artifact or commit reference they carry where present, and are not attributed to a release. The results of a pipeline run are stored as its state and the counts of its build and test results, never as the results body. The handlers are registered for all six events from the start; the thin deliveries of ContextBuilderService and VectorMemoryService and the thin PipelineExecutionService of this increment may publish few or none of their events, and the handlers store what arrives without waiting for them. Nothing is scored and no threshold, suite or verdict is read or written. A record holds identifiers, classes, states, counts, latencies and hashes; it never holds a prompt, a completion, a tool argument, retrieved content or a validator report body, which stay with their owners and are referenced only.

Acceptance criteria

AC1 Given a model call and a tool call of an attempt, when ModelRouterService/ModelInvoked and ToolAdapterService/ToolInvoked arrive, then each is stored once with its run, attempt, tenant and project and the profile, deployment, tool, version, effective class and state of its receipt; given a ModelRouterService/ModelInvoked of a call outside a run, then it is stored under its tenant and project without a run (model/services/observability/QualityScoreService.yaml).
AC2 Given ContextBuilderService/ContextAssembled, VectorMemoryService/RetrievalExecuted, ArtifactService/ArtifactValidated and PipelineExecutionService/PipelineRunCompleted, when each arrives, then it is stored once with the status and tokens of the snapshot, the result count and latency of the retrieval, the validator verdict with its artifact and validation reference, and the state, commit and counts of build and test results of the pipeline run without the results body, under the tenant and project of its envelope and without a release (model/services/observability/QualityScoreService.yaml).
AC3 Given any of the six events delivered twice, when the second delivery is handled, then nothing changes; given a stored signal, then it holds no prompt, completion, tool argument, retrieved content or report body, and no threshold, suite, verdict or score is read or written (ADR-0012, ADR-0021).
Traces: services QualityScoreService, ModelRouterService, ToolAdapterService, ContextBuilderService, VectorMemoryService, ArtifactService, PipelineExecutionService
Traces: events ModelRouterService/ModelInvoked, ToolAdapterService/ToolInvoked, ContextBuilderService/ContextAssembled, VectorMemoryService/RetrievalExecuted, ArtifactService/ArtifactValidated, PipelineExecutionService/PipelineRunCompleted
Traces: adrs ADR-0012, ADR-0019, ADR-0021
Traces: requirements FR-04, NFR-11
Depends on AIFACTORY-US-1089, feature:model-router-service, feature:tool-adapter-service, feature:context-builder-service-thin, feature:vector-memory-service-thin, feature:artifact-service, feature:pipeline-execution-service-thin
Sources model/services/observability/QualityScoreService.yaml, model/services/model-compute/ModelRouterService.yaml, model/services/tools-mcp/ToolAdapterService.yaml, docs/planes/trust-operations/observability/services/quality-score-service/index.md#messages

Tasks

Task Kind Expected output Repository
AIFACTORY-TASK-4360 Handlers CollectModelStats, CollectToolStats, CollectContextStats, CollectRetrievalStats and CollectValidationStats implementation The five handlers store ModelInvoked, ToolInvoked, ContextAssembled, RetrievalExecuted, ArtifactValidated and PipelineRunCompleted once each, idempotent on the event id, with tenant and project from the envelope, run and attempt only where the event names them, the artifact or commit reference of a validation or pipeline signal, pipeline results as state and counts, a model receipt outside a run without a run, and no payload content beyond identifiers, classes, states, counts, latencies and hashes. ConnectSoft.AI.SoftwareFactory.QualityScoreService
AIFACTORY-TASK-4361 Event fixtures of the six signal events from the contracts package fixture The acceptance suite holds fixtures of the six events built from the contracts package, a model receipt with and without a run, a redelivered event of each kind, and a scan helper that fails when a stored record holds a field outside the allowed identifiers, classes, states, counts, latencies and hashes. ConnectSoft.AI.SoftwareFactory.QualityScoreService
AIFACTORY-TASK-4362 Verify the stored reliability signals through the acceptance suite verification Run dotnet test -c Release --filter "Category=Acceptance&Feature=CapabilitySignals"; a zero-match filter counts as a failure; the tests prove AC1 to AC3 of AIFACTORY-US-1088 for ModelRouterService/ModelInvoked, ToolAdapterService/ToolInvoked, ContextBuilderService/ContextAssembled, VectorMemoryService/RetrievalExecuted, ArtifactService/ArtifactValidated and PipelineExecutionService/PipelineRunCompleted, the redelivery without effect and the scan for payload content. ConnectSoft.AI.SoftwareFactory.QualityScoreService

AIFACTORY-US-1089: Thin surface of QualityScoreService: the inbox of ten handlers, the store of accepted events, no operation and no published event

enabling · persona platform-admin As a platform admin, I want the thin QualityScoreService to host exactly its stated scope and nothing that looks like a gate, so that no caller mistakes the thin service for an evaluation gate and the complete delivery adds the gate without changing what is stored.

This PBI sets up what the two behaviour PBIs of the thin delivery run on. The inbox of the quality-score messaging profile (option OrleansMessaging:ServiceName, default quality-score, the inbox and the outbox belonging to this service alone under ADR-0028) registers exactly the ten handlers of the thin scope and no other; the outbox is wired in the transaction pipeline but declares no event type, because the thin delivery publishes none of the five events of the model. The REST and gRPC hosts are configured with token validation and the problem-response pipeline but map no operation, so a call to any path of the model, among them ListEvaluationSuites, StartEvaluation and Evaluation.GetVerdict, is answered as an unknown route and no handler runs. The store quality-score-sql is migrated with the tenant row filter of the service and holds the records of the accepted events. The model of QualityScoreService names three tables, EvaluationSuites, EvaluationRuns and EvaluationResults, and none of them fits a record of a run, an attempt or a call that belongs to no evaluation, so the table that holds these records is an open point of the model reported with this feature; the migration of this PBI creates it under a name the owner may change, and the complete delivery reads it for drift without a second migration of the records. A descriptor test compares the hosted surface, the registered handlers and the declared event types with the thin scope in model/backlog/features/observability.yaml and fails when anything outside it appears. The contract stub of FeedbackService that the feature declares lives in the test suite only, as a compiled client of the dataset method and a stub server answering it, since nothing in the thin service calls FeedbackService. Metrics follow the observability conventions and carry tenant and project by default (D115); the inbox lag of the service is reported as connectsoft.factory.evaluation.inbox_lag.

Acceptance criteria

AC1 Given the built service, when its descriptor test runs, then the inbox registers exactly the ten handlers of the thin scope, the outbox declares no event type, no REST operation and no gRPC method of model/services/observability/QualityScoreService.yaml is mapped, and a call to the path of any of them runs no handler and changes nothing (ADR-0028, model/backlog/features/observability.yaml).
AC2 Given a fresh database, when the service starts, then quality-score-sql is migrated with the table of accepted events under the tenant row filter, a query without a tenant in scope returns nothing, and readiness turns ready only after the migration and the transport check pass (ADR-0015, model/services/observability/QualityScoreService.yaml).
AC3 Given messages waiting in the inbox, when the metrics are read, then connectsoft.factory.evaluation.inbox_lag reports the age of the oldest unprocessed message with the dimension service, and the series carry tenant and project as the helper of the pinned observability library adds them (ADR-0018, model/services/observability/QualityScoreService.yaml).
Traces: services QualityScoreService
Traces: adrs ADR-0012, ADR-0015, ADR-0018, ADR-0028
Depends on feature:component-base-template-factory-profile, feature:component-extensions-ai, feature:component-observability-factory-conventions
Sources model/services/observability/QualityScoreService.yaml, model/backlog/features/observability.yaml, docs/architecture/adr/0028-one-repository-and-deployable-per-service.md

Tasks

Task Kind Expected output Repository
AIFACTORY-TASK-117 Descriptor test of the thin surface of QualityScoreService verification A test compares the mapped REST and gRPC operations, the registered inbox handlers and the declared outbox event types with the thin scope of model/backlog/features/observability.yaml and fails, naming the entry, when an operation, a published event or a handler outside the ten of the scope appears; the compiled client and stub server of the FeedbackService dataset method live in the test suite only. ConnectSoft.AI.SoftwareFactory.QualityScoreService
AIFACTORY-TASK-4363 Messaging profile, hosts without operations and the migration of the accepted-event store implementation The quality-score messaging profile registers the ten handlers through the inbox with idempotency on the event id and wires an outbox with no event type; the REST and gRPC hosts validate tokens and map no operation; the migration of quality-score-sql creates the table of accepted events with tenant, project, run, attempt and event id columns, the tenant row filter and the index by tenant and attempt; readiness waits for the migration and the transport; the inbox lag gauge is emitted. ConnectSoft.AI.SoftwareFactory.QualityScoreService
AIFACTORY-TASK-4364 Verify the thin surface, the store and the inbox lag verification Run dotnet test -c Release --filter "Category=Acceptance&Feature=ThinSurface"; a zero-match filter counts as a failure; the tests prove AC1 to AC3 of AIFACTORY-US-1089: the ten handlers and nothing more, unmapped operations that run no handler, the migrated store under the tenant row filter, readiness after migration and transport, and connectsoft.factory.evaluation.inbox_lag as model/services/observability/QualityScoreService.yaml names it. ConnectSoft.AI.SoftwareFactory.QualityScoreService