AIFACTORY-FEAT-139: Demonstrate I1: Generate a microservice¶
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 Outcome verification · Increment I1 Generate a microservice · State elaborated · Route qa-engineer · Build order 91 of 185 (Factory code)
The outcome of increment I1 is demonstrated end to end and the evidence is recorded: A vision submitted through the API or the command-line client ends as an approved change proposal that contains a generated, built and tested microservice (decision D65); the run is audited from its first command and stops at its budget cap (decision D80). Demonstrations: A person submits a vision; the chain of six workflows runs without a manual start between them. A person approves the recipe and, later, the exact change set. The change proposal is open in the system of record with provenance and validation results. Every command, decision, approval and tool effect of the run is in the audit trail and the hash chain verifies. A run with a budget below its need stops at the cap, and the usage of a finished run reconciles with its receipts.
| Delivers | — |
| Stubs | — |
| Realizes | — |
| Requirements | — |
| Depends on | — |
| Sources | model/backlog/increments.yaml |
AIFACTORY-US-1339: First-outcome proof: the decided blockers are aligned before the run, and a replayable demonstration script and fixture vision brief exist¶
enabling · persona requester
As a requester, I want the first-outcome demonstration to start only when every decided blocker of the chain is aligned, from a script and a brief anyone can replay, so that a failed demonstration names a defect of the Factory and never a gap the owner already decided to close.
The first outcome rests on four links that the owner decided and the model now carries, so this PBI names them as preconditions of the proof and checks before the run that each is still in place. First, the change set reaches the system of record through the tool step propose-change-set of generate-product, which needs a short-lived credential issued by SecretBrokerService through SecretBroker.Issue after a decision of PolicyEngineService through Policy.Decide; the request policy-decision-request-v1 carries the action kind credential-issuance with its platform rule (D127). Second, the repository connection of the project is used under a grant of IntegrationService, set through SetGrant, under the action kind connection-grant and its platform rule (D129). Third, the step record-outputs of generate-product, of the class shared-write, records every generated output through the tool record-generated-output that calls Recipes.RecordOutput of GenerationRecipeService (D133, model/tools/record-generated-output.yaml). Fourth, AuditService consumes RunAccepted of RunControlService and appends it through RecordEvidence as the evidence of the submission, so the first command of the run is audited (D126, D135, model/services/governance/AuditService.yaml). The proof does not borrow another action kind and does not record outputs or submissions by another path: a precondition found missing from the model at the check is recorded as the blocker and the run is not started. In the same PBI the demonstration is made replayable: a written script of the command-line client commands and a fixture vision brief for a small product with one backend microservice. Because the change set reaches Azure DevOps through the repository connection, and D129 moved the integration gateway to I1, the script starts with a Workspace admin of the fixture project who registers, verifies and grants that connection through IntegrationHost; a grant attempted by a member without that role is refused. The register holds a second kind of precondition, the inputs of the owner (D140): every value an I1 service refuses to start without, or without which it discovers nothing, is an entry that the owner supplies before the I1 run and that is checked before the run, namely the values of Usage:Budgets:DefaultLimitUsd, Usage:Budgets:DefaultPeriod and Usage:Budgets:PlatformLimitUsd, the section Usage:Estimates and the pricing revisions of the meters of CostUsageService, the values of the Sandbox pool of ExecutionCapacityService (its security boundary, its internet egress, its region, its size and the medium of the run workspace), the template-feed project of the platform tenant that TemplateRegistryService discovers under, and the source of the tenant region that ProjectService reads. Each entry names the service, the key or the value it needs and the PBI that states it as an input of the owner, and records whether the deployment of the acceptance environment carries a value; the register never states the value itself. An entry without a value at the check is recorded as the blocker and the run is not started.
Acceptance criteria
| AC1 | Given docs/reference/schemas/policy-decision-request.schema.json and the platform rules of PolicyEngineService, when the preconditions are checked before the run, then the action kinds credential-issuance of SecretBrokerService (D127) and connection-grant of IntegrationService (D129) are present with their platform rules, and one found missing is recorded as the blocker and the run is not started; no other action kind is borrowed (descriptors policy-decision-request-v1, policy-decision-v1). |
| AC2 | Given model/workflows/generate-product.yaml and model/services/governance/AuditService.yaml, when the preconditions are checked, then the step record-outputs, of the class shared-write, records the generated outputs through the tool record-generated-output that calls Recipes.RecordOutput of GenerationRecipeService (D133), and AuditService records the submission of a run from RunAccepted through RecordEvidence (D126, D135); one found missing from the model at the check stops the run and is recorded as the blocker (model/tools/record-generated-output.yaml). |
| AC3 | Given the written demonstration script and the fixture vision brief, when a person who did not write them replays them on the acceptance environment with the command-line client, then a Workspace admin of the fixture project registers, verifies and grants the repository connection through IntegrationHost, the run is submitted for that project without any other instruction, and the six workflows of model/backlog/increments.yaml for I1 are the ones it exercises (IntegrationService/RegisterConnection, IntegrationService/VerifyConnection, IntegrationService/SetGrant, ADR-0027). |
| AC4 | Given the registered connection of the fixture project, when a member of the project without the role Workspace admin attempts the grant through IntegrationHost, then the grant is answered 403, no grant is recorded and the refusal is in the evidence (IntegrationService/SetGrant, D129). |
| AC5 | Given the inputs of the owner that D140 enters in the register, Usage:Budgets:DefaultLimitUsd, Usage:Budgets:DefaultPeriod, Usage:Budgets:PlatformLimitUsd, the section Usage:Estimates and the pricing revisions of CostUsageService, the Sandbox pool values of ExecutionCapacityService (boundary, egress, region, size, workspace medium), the template-feed project of the platform tenant of TemplateRegistryService and the source of the tenant region of ProjectService, when the preconditions are checked before the run, then each entry names its service, its key and the PBI that states it, the deployment of the acceptance environment carries a value for each, no value is written into the register, and an entry without a value is recorded as the blocker and the run is not started (model/services/control-plane/CostUsageService.yaml, model/services/runtime-cloud/ExecutionCapacityService.yaml, model/services/template-library/TemplateRegistryService.yaml, model/services/control-plane/ProjectService.yaml). |
| Traces: services | PolicyEngineService, SecretBrokerService, IntegrationService, IntegrationHost, GenerationRecipeService, AuditService, CostUsageService, ExecutionCapacityService, TemplateRegistryService, ProjectService |
| Traces: workflows | generate-product |
| Traces: steps | generate-product/record-outputs, generate-product/propose-change-set |
| Traces: tools | propose-change-set, record-generated-output |
| Traces: apis | IntegrationService/RegisterConnection, IntegrationService/VerifyConnection, IntegrationService/SetGrant |
| Traces: descriptors | policy-decision-request-v1, policy-decision-v1 |
| Traces: adrs | ADR-0027 |
| Depends on | feature:policy-engine-service, feature:secret-broker-service, feature:integration-service-thin, feature:integration-host, feature:governance-host, feature:audit-service-complete, feature:generation-recipe-service, feature:wf-generate-product, feature:command-line-client-commands, feature:cost-usage-service-complete, feature:execution-capacity-service-thin, feature:template-registry-service, feature:project-service |
| Sources | model/backlog/increments.yaml, model/workflows/generate-product.yaml, model/services/governance/AuditService.yaml, model/services/integration/IntegrationService.yaml, model/services/control-plane/CostUsageService.yaml, model/services/runtime-cloud/ExecutionCapacityService.yaml, model/services/template-library/TemplateRegistryService.yaml, model/services/control-plane/ProjectService.yaml, model/tools/record-generated-output.yaml, docs/reference/schemas/policy-decision.schema.json, docs/reference/schemas/policy-decision-request.schema.json, docs/architecture/adr/0027-generation-input-contract.md |
Tasks
| Task | Kind | Expected output | Repository |
|---|---|---|---|
AIFACTORY-TASK-229 Demonstration script and fixture vision brief |
documentation | A written script and a fixture vision brief for a small product with one backend microservice that anyone can replay in the acceptance environment; the script names each command of the command-line client, the person who runs it and the answer it expects, starting with the registration, verification and grant of the repository connection of the fixture project by a Workspace admin and the refused grant of a member without that role. | ConnectSoft.AI.SoftwareFactory.Documentation |
AIFACTORY-TASK-5389 Command sequence of the command-line client for the first-outcome demonstration |
fixture | A replayable command sequence in the command-line client repository that signs in, creates the project, registers, verifies and grants the repository connection of the project through IntegrationHost as a Workspace admin, attempts the same grant as a member without that role and expects 403, submits the fixture vision brief, follows the run, approves the recipe and approves the change proposal, each step printing the identifier or the answer it produced. | ConnectSoft.AI.SoftwareFactory.Cli |
AIFACTORY-TASK-5390 Precondition register of the first-outcome proof |
documentation | A precondition register on the evidence page that lists the four decided links with the decision that settles each (D127 for the action kind credential-issuance, D129 for the action kind connection-grant, D133 for the tool record-generated-output, D126 and D135 for the audit of RunAccepted), the model file that holds it (docs/reference/schemas/policy-decision-request.schema.json, model/tools/record-generated-output.yaml, model/workflows/generate-product.yaml, model/services/governance/AuditService.yaml) and its state at the check, aligned or blocking; and, as its second part, one entry per input of the owner that D140 enters: Usage:Budgets:DefaultLimitUsd, Usage:Budgets:DefaultPeriod, Usage:Budgets:PlatformLimitUsd, the section Usage:Estimates and the pricing revisions of the meters of CostUsageService, the Sandbox pool boundary, egress, region, size and workspace medium of ExecutionCapacityService, the template-feed project of the platform tenant of TemplateRegistryService and the source of the tenant region of ProjectService, each with its service, its key, the PBI that states it as an input of the owner and its state at the check, supplied or blocking, and none with its value. |
ConnectSoft.AI.SoftwareFactory.Documentation |
AIFACTORY-TASK-5391 Verify the preconditions and the replay of the script |
verification | The register is compared with docs/reference/schemas/policy-decision-request.schema.json, model/tools/record-generated-output.yaml, model/workflows/generate-product.yaml and model/services/governance/AuditService.yaml and proves AC1 and AC2 of AIFACTORY-US-1339; a replay of the script by a second person on a fresh acceptance environment grants the connection, records the refused grant (IntegrationService/SetGrant) and submits the run, which proves AC3 and AC4; the owner-input entries of the register are compared with model/services/control-plane/CostUsageService.yaml, model/services/runtime-cloud/ExecutionCapacityService.yaml, model/services/template-library/TemplateRegistryService.yaml and model/services/control-plane/ProjectService.yaml and with the configuration of the acceptance environment, every entry is supplied and none carries its value, which proves AC5; python -I tools/docs_model validate passes. |
ConnectSoft.AI.SoftwareFactory.Documentation |
AIFACTORY-US-1340: First-outcome proof: one submitted vision runs the six workflows without a manual start and ends as an approved, open change proposal whose service builds outside the Factory¶
evidence · persona requester
As a requester, I want one vision submitted from the command-line client to end as an open change proposal for a generated, built and tested microservice, so that the Factory has proven its first outcome on a real product idea.
This PBI proves the first three demonstrations of increment I1 (decision D65). A requester submits the fixture vision brief from the command-line client; the call passes the control plane gateway with the user token and reaches SubmitRun of RunControlService. That submission is the run of vision-to-blueprint, the only workflow of the chain a person submits. From there no workflow is started by hand: each of architecture-decomposition, module-blueprints, application-design, database-blueprint and generate-product is then submitted by WorkflowEngineService when its readiness bundle is satisfied, which ReadinessRunSubmitted records. Two people decide. An asset maintainer approves the validated recipe through ApproveRecipe of GenerationRecipeService, which publishes RecipeApproved and is the single gate of the generation run (ADR-0027 decision 2). Later a reviewer, never the requester of the run, approves the exact change set at the step approve-proposal through DecideApproval, bound to the hash of the proposal (FR-06, ADR-0015). The generation run records each output with its provenance, which GeneratedOutputRecorded carries, and the tool of the step propose-change-set opens the pull request in the system of record through the connection and records it through Pipelines.RecordProposal of PipelineExecutionService with the run, the attempt, the agent release, the pull-request reference and the tool receipt (D135), which ChangeProposalOpened records, with the provenance and the validation results attached. The proof ends outside the Factory: the branch of the proposal is cloned and built by a person, and the service builds and its tests pass.
Acceptance criteria
| AC1 | Given a new project and the fixture vision brief, when the requester submits it from the command-line client through SubmitRun of RunControlService, then the six workflows run in the order of model/backlog/increments.yaml and each later run is submitted by WorkflowEngineService, as ReadinessRunSubmitted records, with no workflow started by hand (RunControlService/SubmitRun, WorkflowEngineService/ReadinessRunSubmitted). |
| AC2 | Given the validated recipe, when the asset maintainer approves it, then RecipeApproved names the approved version and the generation run reads its pins from it; and given the change set, when a reviewer who is not the requester approves it at the step generate-product/approve-proposal, then the decision is bound to the proposal hash (GenerationRecipeService/RecipeApproved, RunControlService/ApprovalDecided, FR-06, ADR-0027). |
| AC3 | Given the approved change set, when the step generate-product/propose-change-set runs, then the change proposal is open in the system of record with the provenance of every generated output and the validation results attached, recorded through Pipelines.RecordProposal with the run, the attempt, the agent release, the pull-request reference and the tool receipt, as GeneratedOutputRecorded and ChangeProposalOpened record them (GenerationRecipeService/GeneratedOutputRecorded, PipelineExecutionService/Pipelines.RecordProposal, PipelineExecutionService/ChangeProposalOpened, D135). |
| AC4 | Given the open proposal, when its branch is cloned and built by a person outside the Factory, then the service builds in Release and its generated tests pass, and the build log is recorded (ADR-0027, docs/architecture/adr/0027-generation-input-contract.md). |
| Traces: services | ControlPlaneApi, RunControlService, WorkflowEngineService, GenerationRecipeService, PipelineExecutionService |
| Traces: workflows | vision-to-blueprint, architecture-decomposition, module-blueprints, application-design, database-blueprint, generate-product |
| Traces: steps | generate-product/approve-proposal, generate-product/propose-change-set |
| Traces: apis | RunControlService/SubmitRun, RunControlService/DecideApproval, GenerationRecipeService/ApproveRecipe, PipelineExecutionService/Pipelines.RecordProposal |
| Traces: events | WorkflowEngineService/ReadinessRunSubmitted, GenerationRecipeService/RecipeApproved, GenerationRecipeService/GeneratedOutputRecorded, RunControlService/ApprovalDecided, PipelineExecutionService/ChangeProposalOpened |
| Traces: adrs | ADR-0015, ADR-0027 |
| Traces: descriptors | change-proposal-record-v1 |
| Traces: requirements | FR-06 |
| Depends on | AIFACTORY-US-1339, feature:command-line-client-commands, feature:control-plane-api, feature:run-control-service, feature:workflow-engine-service, feature:generation-recipe-service, feature:pipeline-execution-service-thin, feature:wf-vision-to-blueprint, feature:wf-architecture-decomposition, feature:wf-module-blueprints, feature:wf-application-design, feature:wf-database-blueprint, feature:wf-generate-product |
| Sources | model/backlog/increments.yaml, model/workflows/generate-product.yaml, docs/architecture/adr/0027-generation-input-contract.md |
Tasks
| Task | Kind | Expected output | Repository |
|---|---|---|---|
AIFACTORY-TASK-230 Recorded evidence of one full run |
documentation | Run identifiers of the six workflow runs, the approved recipe version, the approval decision of the change set with its proposal hash and the reference of the change proposal on the evidence page, all from one replay of the script of AIFACTORY-TASK-229. | ConnectSoft.AI.SoftwareFactory.Documentation |
AIFACTORY-TASK-5392 Out-of-Factory build of the proposal branch |
operations | A clone of the branch of the change proposal built in Release and tested by a person on a machine with only the ConnectSoft feed as package source, with the build and test log recorded beside the reference of the proposal. | ConnectSoft.AI.SoftwareFactory.Documentation |
AIFACTORY-TASK-5393 Verify the chain run against its criteria |
verification | The recorded run identifiers, approvals, proposal and build log are compared with AC1 to AC4 of AIFACTORY-US-1340: six runs with no manual start (WorkflowEngineService/ReadinessRunSubmitted), the recipe approval and the change-set approval by two people bound to the proposal hash (RunControlService/ApprovalDecided), an open proposal (PipelineExecutionService/ChangeProposalOpened) and a green build outside the Factory. | ConnectSoft.AI.SoftwareFactory.Documentation |
AIFACTORY-US-1341: First-outcome proof: every act of the run is in the audit trail with a verified hash chain, a run under budget stops at its cap and usage reconciles with receipts¶
evidence · persona observer
As an observer, I want the run of the first outcome audited from its first command and bounded by its budget, so that the first outcome is accountable and cannot spend beyond what the project allowed.
This PBI proves the last two demonstrations of increment I1, which decision D80 makes part of the first outcome by delivering AuditService and CostUsageService complete in I1, and the completeness of the run record. For the run of AIFACTORY-US-1340 an observer exports the audit trail of the project through ExportAudit and checks that each family of acts of the run has its entries, each one recorded by AuditEntryRecorded: the submission of the run, the policy decisions, the approval requests and decisions, the tool invocations and, for a cancelled run, the cancellation; and that the chain verifies, which AuditChainVerified reports with valid true. The submission entry comes from RunAccepted, which AuditService appends through RecordEvidence (D126, D135), a precondition checked by AIFACTORY-US-1339; a missing submission entry fails the proof and is recorded with the run identifier. The observer then reads the run record: GetRun of RunControlService, the model receipts of ListRunReceipts of ModelRouterService and the tool receipts of ListRunToolReceipts of ToolAdapterService, and checks that every step has an attempt with one outcome, every model call a receipt and every tool call an invocation (ADR-0014). A second run of the same brief is submitted under a project budget set through SetProjectBudget whose limit and approval threshold are both above the estimate CostUsageService reserves for the run, so the reservation succeeds and no spend approval opens, while the limit is below the usage the run observes. When the observed usage takes the committed amount above the limit, CostUsageService publishes BudgetExceeded for that run only; RunControlService records the cancellation with the reason budget-exceeded and the run ends Cancelled, which RunCancelled records, never Failed. The refusal case is a third submission under a budget whose limit is below the estimate: the reservation is denied, which BudgetReservationDenied records, and the run ends Failed with the failure class policy-denied, which RunFailed records (model/services/control-plane/CostUsageService.yaml). For the finished first run the usage that GetRunUsage reports reconciles with its invocation receipts, and the reservation is settled, which ReservationSettled records. A missing entry, a broken chain, an incomplete run record or a usage that does not match its receipts fails the proof and is recorded with the run identifier.
Acceptance criteria
| AC1 | Given the finished run of AIFACTORY-US-1340, when the observer exports the audit trail of its project, then the submission of the run, every policy decision, every approval request and decision, every tool invocation and, for the capped run, the cancellation have an entry, and the chain verification reports valid for the head sequence of the tenant; the submission entry is the one appended from RunAccepted (D126, D135), and a missing one fails the proof and is recorded with the run identifier (AuditService/ExportAudit, AuditService/AuditEntryRecorded, AuditService/AuditChainVerified, RunControlService/RunAccepted). |
| AC2 | Given the finished run, when GetRun, the model receipts and the tool receipts of the run are read, then every step has an attempt with exactly one outcome, every model call a receipt and every tool call an invocation, and a step, call or invocation without its record is named (RunControlService/GetRun, ModelRouterService/ListRunReceipts, ToolAdapterService/ListRunToolReceipts, ADR-0014). |
| AC3 | Given a project budget whose limit and approval threshold are above the reserved estimate of the run and whose limit is below the usage the run observes, when the observed usage crosses the limit, then the reservation was granted with no spend approval, CostUsageService publishes BudgetExceeded for that run only, the run ends Cancelled with the reason budget-exceeded and not Failed, and the next reservation of the budget is denied (CostUsageService/SetProjectBudget, CostUsageService/BudgetExceeded, RunControlService/RunCancelled, FR-09). |
| AC4 | Given a project budget whose limit is below the estimate of the run, when the run is accepted, then the reservation is denied and the run ends Failed with the failure class policy-denied, and nothing is reserved (CostUsageService/BudgetReservationDenied, RunControlService/RunFailed, model/services/control-plane/CostUsageService.yaml). |
| AC5 | Given the finished first run, when its usage is read and compared with its model invocation receipts, then the input and output tokens and the cost reported for the run equal the sum of its receipts, and the reservation of the run is settled (CostUsageService/GetRunUsage, ModelRouterService/ListRunReceipts, CostUsageService/ReservationSettled, FR-09). |
| Traces: services | AuditService, CostUsageService, ModelRouterService, ToolAdapterService, RunControlService |
| Traces: apis | AuditService/ExportAudit, CostUsageService/SetProjectBudget, CostUsageService/GetRunUsage, ModelRouterService/ListRunReceipts, ToolAdapterService/ListRunToolReceipts, RunControlService/GetRun |
| Traces: events | AuditService/AuditEntryRecorded, AuditService/AuditChainVerified, CostUsageService/BudgetExceeded, CostUsageService/BudgetReservationDenied, CostUsageService/ReservationSettled, RunControlService/RunAccepted, RunControlService/RunCancelled, RunControlService/RunFailed |
| Traces: adrs | ADR-0014 |
| Traces: requirements | FR-09 |
| Depends on | AIFACTORY-US-1339, AIFACTORY-US-1340, feature:audit-service-complete, feature:cost-usage-service-complete, feature:model-router-service, feature:tool-adapter-service, feature:run-control-service |
| Sources | model/backlog/increments.yaml, model/services/governance/AuditService.yaml, model/services/control-plane/CostUsageService.yaml, model/services/control-plane/RunControlService.yaml |
Tasks
| Task | Kind | Expected output | Repository |
|---|---|---|---|
AIFACTORY-TASK-5394 Audit export, chain verification and run record of the demonstrated run |
operations | The exported audit trail of the project of the demonstrated run, a list that maps the submission, every policy decision, approval request and decision, tool invocation and cancellation of the run to its audit entry, the chain verification result with its head sequence, and the run record read through GetRun, ListRunReceipts and ListRunToolReceipts with every step, attempt, model receipt and tool invocation matched, recorded on the evidence page. |
ConnectSoft.AI.SoftwareFactory.Documentation |
AIFACTORY-TASK-5395 Budget-capped run, denied reservation and usage reconciliation |
operations | A second run of the fixture brief under a budget whose limit and threshold are above its estimate and below its observed usage, recorded with its granted reservation, its BudgetExceeded event and its Cancelled state with the reason budget-exceeded; a third run under a limit below its estimate, recorded with its denied reservation and its Failed state with the failure class policy-denied; and for the first run the usage per meter beside the sum of its model invocation receipts and its settled reservation. |
ConnectSoft.AI.SoftwareFactory.Documentation |
AIFACTORY-TASK-5396 Verify the audit, run record and budget proof against its criteria |
verification | The recorded audit export, chain result, run record, capped run, denied run and reconciliation are compared with AC1 to AC5 of AIFACTORY-US-1341 (AuditService/AuditChainVerified, RunControlService/GetRun, ToolAdapterService/ListRunToolReceipts, CostUsageService/BudgetExceeded, CostUsageService/BudgetReservationDenied, RunControlService/RunCancelled, ModelRouterService/ListRunReceipts); any unmatched entry, attempt or receipt fails the check and is named. | ConnectSoft.AI.SoftwareFactory.Documentation |
AIFACTORY-US-018: The first outcome is demonstrated end to end¶
evidence · persona requester
As a requester, I want to submit one vision and end with an open change proposal for a generated, built and tested microservice, so that the Factory has proven its first outcome on a real product idea.
This is the integration PBI of the feature and the closing proof of increment I1. The outcome of I1, as model/backlog/increments.yaml states it, is that a vision submitted through the API or the command-line client ends as an approved change proposal that contains a generated, built and tested microservice (decision D65), and that the run is audited from its first command and stops at its budget cap (decision D80). The increment is finished when this proof passes, not when its services are deployed. The preconditions of AIFACTORY-US-1339 are checked first: the action kinds credential-issuance (D127) and connection-grant (D129) with their platform rules, the tool record-generated-output (D133) and the audit of RunAccepted by AuditService (D126, D135), all decided and carried by the model, and the inputs of the owner that D140 enters in the same register, which the owner supplies before the run; one found missing at the check stops the run and its blocker is the evidence. Then the five demonstrations are replayed in one recorded acceptance run from the script and the fixture brief, after a Workspace admin of the fixture project has granted its repository connection through IntegrationHost: the chain and the two approvals of AIFACTORY-US-1340, and the audit chain, the run record, the budget cap and the reconciliation of AIFACTORY-US-1341. The evidence pack holds every run identifier, approval, receipt, audit export and the reference of the change proposal.
Acceptance criteria
| AC1 | Given the preconditions of AIFACTORY-US-1339 are aligned, when the script is replayed in the recorded acceptance run, then the six workflows run in order with no workflow started by hand and the run identifiers are in the evidence pack (WorkflowEngineService, RunControlService, ADR-0027). |
| AC2 | Given the recipe and the change set of that run, when the asset maintainer and the reviewer approve them, then the change proposal is open in the system of record with provenance and validation results, and its branch builds and tests outside the Factory, as AIFACTORY-US-1340 proves it (GenerationRecipeService, PipelineExecutionService). |
| AC3 | Given the run ended, when its record and its audit trail are read, then every step has an attempt, every model call a receipt and every tool call an invocation, the audit chain verifies from the submission of the run, and a run whose observed usage crossed its budget stopped at the cap, as AIFACTORY-US-1341 proves it (RunControlService, ADR-0014). |
| AC4 | Given the evidence pack, when it is reviewed against the outcome of I1 in model/backlog/increments.yaml and the done_when of the epic outcome-verification in model/backlog/epics.yaml, then each of the five demonstrations has its evidence, and a failed demonstration or an unaligned precondition is recorded instead of being waived (ADR-0029). |
| Traces: services | RunControlService, WorkflowEngineService, GenerationRecipeService, PipelineExecutionService |
| Traces: workflows | vision-to-blueprint, architecture-decomposition, module-blueprints, application-design, database-blueprint, generate-product |
| Traces: adrs | ADR-0014, ADR-0027, ADR-0029 |
| Depends on | AIFACTORY-US-1339, AIFACTORY-US-1340, AIFACTORY-US-1341 |
| Sources | model/backlog/increments.yaml, model/backlog/epics.yaml |
Tasks
| Task | Kind | Expected output | Repository |
|---|---|---|---|
AIFACTORY-TASK-185 Evidence pack for increment I1 |
documentation | A recorded run of every demonstration below with its run identifiers, artifacts and logs: A person submits a vision; the chain of six workflows runs without a manual start between them. A person approves the recipe and, later, the exact change set. The change proposal is open in the system of record with provenance and validation results. Every command, decision, approval and tool effect of the run is in the audit trail and the hash chain verifies. A run with a budget below its need stops at the cap, and the usage of a finished run reconciles with its receipts. | ConnectSoft.AI.SoftwareFactory.Documentation |
AIFACTORY-TASK-776 Verify first-outcome-demonstrated against the recorded acceptance run |
verification | The evidence pack is compared with AC1 to AC4 of AIFACTORY-US-018 and with the outcome of I1 in model/backlog/increments.yaml; every demonstration and every precondition has its evidence, and python -I tools/docs_model validate passes with the evidence page in place. |
ConnectSoft.AI.SoftwareFactory.Documentation |