Capstone system acceptance and open-source delivery contract¶

This template turns “the project can be demonstrated” into “the system can be accepted, its artifacts can be verified, its repository can be reproduced, and its responsibilities can be handed over.”

It should be completed jointly by the Product Owner, Domain Expert, Agent Architect, Engineering, Quality, Security, SRE, Open-source Maintainer, and an Independent Reviewer who was not involved in the implementation.

Do not fill in every section at the outset. Start with one high-value Vertical Slice, establish a complete evidence chain, and then expand the system’s capabilities.

1. Acceptance charter¶

acceptance_charter:
  project_id: ""
  project_name: ""
  version: ""
  status: draft | review | accepted | conditional | rejected
  problem: ""
  users: []
  primary_goal: ""
  value_hypotheses: []
  in_scope: []
  out_of_scope: []
  anti_goals: []
  supported_intents: []
  supported_risks: []
  owners:
    product: ""
    domain: ""
    architecture: ""
    engineering: ""
    quality: ""
    security: ""
    sre: ""
    open_source: ""
  repositories: []
  target_release: ""

Checklist:

The Problem describes user harm, not the name of a solution.
The Primary Goal can be evaluated through an Outcome.
Both In-scope and Out-of-scope are defined.
Agent count, Token count, and generated word count are not used as value metrics.
Every high-risk capability has a dedicated Owner.

2. Value hypothesis¶

Hypothesis ID	User / Business Outcome	Baseline	Metric	Decision Rule	Owner

value_hypothesis:
  hypothesis_id: ""
  statement: ""
  population: ""
  baseline: ""
  metric: ""
  target: ""
  minimum_samples: 0
  measurement_window: ""
  confounders: []
  if_supported: ""
  if_rejected: ""
  owner: ""

3. Risk register¶

Risk ID	Scenario	User Impact	Likelihood	Severity	Control	Evidence	Owner	Review By

risk:
  risk_id: ""
  category: product | quality | security | privacy | reliability | operations | legal
  scenario: ""
  affected_population: ""
  impact: ""
  likelihood: ""
  severity: low | medium | high | critical
  controls: []
  residual_risk: ""
  accepted_by: ""
  review_by: ""

4. Success and safety contract¶

Metric / Invariant	Target / Rule	Gate
Goal Success
Evidence Coverage
Safe Path
Goal Latency
Cost per Success
Unauthorized Side Effect	equals 0	hard
Cross-tenant Leakage	equals 0	hard

5. Requirement record¶

requirement:
  requirement_id: ""
  version: 1
  category: business | functional | quality | security | reliability | performance | operations
  statement: ""
  rationale: ""
  user_or_system: ""
  risk: low | medium | high | critical
  preconditions: []
  acceptance:
    required_facts: []
    required_outcomes: []
    required_path: []
    thresholds: {}
  forbidden: []
  verification:
    tests: []
    evaluations: []
    drills: []
    runtime_signals: []
  owner: ""
  approved_by: []
  valid_from: ""
  review_by: ""

Checklist:

The Statement can be objectively falsified.
Acceptance distinguishes among Fact, Path, Outcome, and Threshold.
Forbidden behaviors are expressed as structured conditions.
Verification is bound to specific IDs.
The Requirement has an Owner and a validity period.

6. Traceability matrix¶

Requirement	ADR	Contract	Implementation	Test / Eval	Runtime Evidence	Sign-off

traceability_record:
  requirement_ref: ""
  decision_refs: []
  contract_refs: []
  implementation_refs: []
  verification_refs: []
  runtime_evidence_refs: []
  owner: ""
  coverage_status: complete | partial | missing | obsolete
  release: ""

Change checklist:

A Requirement change triggers a review of all affected items.
A major Contract version change lists every Consumer.
Obsolete Evidence cannot support the current Release.
Sign-off is bound to the same RC.

7. Architecture blueprint¶

architecture_blueprint:
  blueprint_id: ""
  version: ""
  layers:
    experience: []
    control: []
    agent: []
    capability: []
    platform: []
  trust_boundaries: []
  state_owners: []
  data_flows: []
  side_effect_boundaries: []
  deployment_units: []
  failure_domains: []
  diagrams: []
  approved_by: []

7.1 Boundary record¶

Boundary	Caller Identity	Callee	Contract	Authorization	State Owner	Failure Semantics	Evidence

8. ADR index¶

ADR ID	Decision	Status	Alternatives	Main Consequence	Evidence	Supersedes	Owner
		proposed / accepted / superseded / rejected

adr:
  adr_id: ""
  title: ""
  status: proposed | accepted | superseded | rejected
  context: ""
  constraints: []
  options:
    - name: ""
      benefits: []
      costs: []
      risks: []
  decision: ""
  consequences:
    positive: []
    negative: []
  validation_evidence: []
  rollback_or_revisit: ""
  supersedes: []
  owner: ""

9. Contract registry¶

Contract ID	Version	Producer	Consumers	Compatibility	Schema	Tests	Deprecation	Owner

Required coverage:

9.1 Compatibility policy¶

compatibility_policy:
  public_contracts: []
  versioning: semantic
  reject_unknown_major: true
  additive_minor_fields: allowed
  required_field_removal: breaking
  enum_extension: consumer_review
  consumer_contract_tests: required
  deprecation:
    announce_at: ""
    supported_until: ""
    migration_guide: ""
  registry: ""
  owner: ""

10. System invariants¶

Invariant ID	Statement	Enforcement Point	Test	Runtime Signal	Hard Gate?

Minimum invariants:

completed has a result that satisfies the contract.
no_data and tool_error are distinguished from a business fact evaluating to false.
Every Claim is bound to the correct Tenant, Subject, Version, and Evidence.
Every side effect has authorization, approval, idempotency, and an Expected Version.
A Late Result cannot overwrite a new Plan.
Partial / Degraded results list what is missing and the resulting impact.

11. Vertical slice specification¶

vertical_slice:
  slice_id: ""
  title: ""
  user_goal: ""
  risk: ""
  entrypoint: ""
  boundaries_crossed: []
  steps:
    - step_id: ""
      component: ""
      contract: ""
      state_change: ""
      evidence: ""
  success_outcome: ""
  allowed_partial_outcomes: []
  forbidden_outcomes: []
  faults: []
  required_observability: []
  clean_room_command: ""
  owner: ""

11.1 Slice gate¶

Dimension	Acceptance	Evidence	Result	Owner
Correctness
State
Security
Reliability
Observability
Reproducibility

Expansion policy:

expansion_policy:
  require_slice_gate_pass: true
  next_capability: ""
  new_risks: []
  new_contracts: []
  new_evidence: []
  rollback_to_slice: ""

12. Milestone definition of done¶

milestone_dod:
  milestone_id: ""
  capability: ""
  contract_changes: []
  code_refs: []
  tests:
    unit: []
    contract: []
    integration: []
    e2e: []
    adversarial: []
    failure: []
  traces:
    success: []
    failure: []
  threat_updates: []
  golden_updates: []
  docs_updates: []
  migration: ""
  rollback: ""
  metric_delta: {}
  cost_delta: {}
  approved_by: []

13. Repository architecture¶

repository_architecture:
  repository: ""
  boundaries:
    docs: ""
    contracts: ""
    services: ""
    platform: ""
    evals: ""
    tests: ""
    deploy: ""
    examples: ""
  dependency_rules: []
  forbidden_imports: []
  architecture_tests: []
  code_owners: []

Checklist:

A Contract does not depend on a Service implementation.
A Team does not import another Team’s internal code.
A Worker depends on a Tool Interface.
An Eval uses the public entry point.
Docs link to versioned ADRs, Contracts, and Runbooks.

14. Reproducible environment¶

environment_contract:
  supported_os: []
  architectures: []
  runtimes: []
  package_managers: []
  lockfiles: []
  container_runtime: ""
  required_ports: []
  minimum_resources: {}
  external_dependencies: []
  environment_variables:
    - name: ""
      required: false
      secret: false
      safe_default: ""
  doctor_command: ""
  bootstrap_command: ""
  start_command: ""
  seed_command: ""
  smoke_command: ""
  eval_smoke_command: ""
  down_command: ""

15. Clean-room reproduction¶

clean_room_reproduction:
  run_id: ""
  reviewer: ""
  reviewer_role: ""
  prior_project_knowledge: none | limited | substantial
  environment:
    os: ""
    architecture: ""
    runtime_versions: {}
  source_commit: ""
  release_artifact_digest: ""
  cache_state: empty
  steps:
    - command: ""
      expected: ""
      observed: ""
      duration: ""
      result: pass | fail
  successful_scenarios: []
  failure_scenarios: []
  trace_refs: []
  problems_found: []
  undocumented_interventions: []
  total_duration: ""
  result: pass | conditional | fail
  signed_at: ""

Hard requirements:

Do not use caches from the author’s machine.
Do not use undocumented commands.
Do not obtain required steps through private messages.
Run both the success and failure paths.
Trace / Evidence can be inspected.
The environment can be cleaned up completely.

16. Seed manifest¶

seed_manifest:
  dataset_id: ""
  version: ""
  synthetic: true
  generator_ref: ""
  generator_version: ""
  random_seed: 0
  entities: {}
  edge_cases: []
  expected_hashes: {}
  pii_status: none | deidentified | restricted
  redistribution_rights: ""
  license: ""
  valid_from: ""
  owner: ""

17. Public data and secret review¶

Surface	Check	Tool / Reviewer	Findings	Remediation	Result
Git history	Secrets
Source	Internal URLs / Tokens
Dataset	PII / Reidentification
Prompt / Policy	Confidential rules
Notebook	Output / Credentials
Trace / Dashboard	User data
Media	Faces / Logos / Copyright

18. Evidence package manifest¶

evidence_package:
  package_id: ""
  release_candidate: ""
  source_commit: ""
  generated_at: ""
  product:
    charter: ""
    journeys: []
    acceptance: ""
  architecture:
    blueprint: ""
    adrs: []
    contracts: []
  implementation:
    build_run: ""
    deployment_manifest: ""
    artifact_digests: []
  quality:
    golden_dataset: ""
    eval_run: ""
    n_run_report: ""
    known_failures: []
  security:
    threat_model: ""
    red_team_run: ""
    sbom: ""
    exceptions: []
  reliability:
    failure_matrix: ""
    recovery_report: ""
    gameday: ""
  operations:
    slos: []
    dashboards: []
    alerts: []
    runbooks: []
    orr: ""
  traceability_index: ""
  content_digest: ""

19. Evidence record¶

evidence_record:
  evidence_id: ""
  claim: ""
  requirement_refs: []
  decision_refs: []
  contract_refs: []
  implementation_refs: []
  test_refs: []
  runtime_refs:
    traces: []
    context_graphs: []
    metrics: []
  artifact_digests: []
  limitations: []
  signed_by: []
  valid_for_release: ""
  expires_at: ""

20. Quality acceptance¶

Layer	Quality Claim	Dataset / Fixture	Evaluator	N-run	Threshold / Rule	Result
Planner
Router / Team
Worker / Tool
Consolidator
Decision
System
Safety

Checklist:

Golden and Fixture are bound to the current version.
N-run preserves the path from every run and the worst-performing slice.
Safety conditions are excluded from any compensable aggregate score.
The Judge is calibrated and allowed to abstain.
Baseline and Candidate use the same conditions.

21. Security acceptance¶

Threat Surface	Scenario	Control	Red-team Case	Hard Gate	Result	Exception
Prompt Injection
Tool / MCP
A2A Delegation
Memory / RAG
Side Effect
Output / DLP
Supply Chain
Operations

security_acceptance:
  run_id: ""
  release_candidate: ""
  threat_model_version: ""
  red_team_cases: 0
  high_risk_n_run: 0
  unauthorized_side_effects: 0
  cross_tenant_leakage: 0
  replay_block_rate: ""
  pii_block_rate: ""
  secret_scan: pass | fail
  sbom_ref: ""
  dependency_critical_vulnerabilities: 0
  open_exceptions: []
  approved_by: []

22. Failure matrix¶

Failure	Injection	Expected Control	Allowed Outcome	State / Side-effect Check	Evidence	Result
Provider unavailable
Tool timeout
Schema drift
Worker crash
Knowledge stale
Graph / Trace unavailable
Cache confusion

23. Performance, capacity and cost¶

performance_budget:
  population: ""
  load_model:
    arrival_pattern: ""
    concurrency: 0
    intent_mix: {}
    risk_mix: {}
    conversation_length: {}
  latency:
    p50_ms: 0
    p95_ms: 0
    p99_ms: 0
  throughput: ""
  queue_age: ""
  model_tokens: ""
  tool_calls: ""
  cost_per_success: ""
  retry_amplification: ""
  test_environment: ""
  known_differences_from_production: []

24. Sequential release gates¶

Gate	Entry Condition	Required Evidence	Hard Conditions	Decision	Owner	Signed At
G0 Problem
G1 Design
G2 Slice
G3 Quality
G4 Security
G5 Operations

gate_decision:
  gate: ""
  release_candidate: ""
  decision: pass | conditional | fail
  evidence_refs: []
  hard_gate_results: {}
  conditions:
    - condition: ""
      scope: ""
      owner: ""
      due_at: ""
      verifier: ""
  signed_by: []
  signed_at: ""

25. Release candidate manifest¶

release_manifest:
  release: ""
  source_commit: ""
  contracts: {}
  agents: {}
  prompts: {}
  models: {}
  tools: {}
  knowledge_snapshot: ""
  capability_snapshot: ""
  policy_version: ""
  checkpoint_schema: ""
  container_digests: []
  datasets: []
  evaluation_run: ""
  security_run: ""
  sbom_ref: ""
  provenance_ref: ""
  evidence_index: ""
  rollback_to: ""
  created_at: ""

Consistency checklist:

All evidence points to the same Source Commit and RC.
Artifacts use Digests, not Tags alone.
Prompt, Policy, Knowledge, and Contract versions are all pinned.
Every critical change generates a new RC.

26. SBOM and third-party inventory¶

third_party_asset:
  asset_id: ""
  name: ""
  version: ""
  asset_type: package | container | model | dataset | tool | media | document
  source: ""
  package_identifier: ""
  license_id: ""
  copyright: ""
  transitive: false
  modified: false
  redistribution_allowed: false
  notice_required: false
  vulnerabilities: []
  end_of_life: ""
  owner: ""

SBOM checklist:

Includes Direct and Transitive Dependencies.
Records the Package Identifier, Version, and License.
Bound to the current Release Artifact.
Critical vulnerabilities are blocked or covered by an exception.
NOTICE / Attribution has been generated.

27. Build provenance and attestation¶

build_provenance:
  artifact_name: ""
  artifact_digest: ""
  source_repository: ""
  source_commit: ""
  builder_identity: ""
  workflow_ref: ""
  build_type: ""
  external_parameters: {}
  resolved_dependencies: []
  started_at: ""
  finished_at: ""
  attestation_ref: ""
  signature_verified: false
  verification_policy: ""
  verification_result: pass | fail

Verification:

The Artifact Digest matches the Attestation Subject.
The Builder Identity is on the allowlist.
The Workflow and Build Type match expectations.
External Parameters contain no unapproved fields.
The Source Commit is a reviewed revision on a protected branch.
The presence of an Attestation is not misrepresented as “no vulnerabilities.”

28. Deployment, canary and rollback¶

release_strategy:
  shadow:
    population: ""
    exit: []
  canary:
    population: ""
    slices: []
    duration: ""
    exit: []
  dual_run:
    enabled: false
    comparison: ""
  promote:
    conditions: []
  rollback:
    target_release: ""
    command_or_control: ""
    state_migration: ""
    cache: ""
    knowledge: ""
    verification: []
  retire:
    drain: ""
    data_retention: ""
    support_end: ""

29. Kill / pause criteria¶

kill_or_pause:
  conditions:
    - signal: ""
      operator: gt | gte | lt | lte | eq
      threshold: ""
      window: ""
      scope: ""
      action: pause | rollback | degrade | block
  manual_triggers:
    - business_owner_request
    - security_incident
    - legal_request
  authority:
    requestors: []
    approvers: []
  audit_ref: ""
  recovery_conditions: []

30. Operational readiness¶

Area	Evidence	Result	Owner
SLI / SLO
Error Budget
Dashboard
Alert
Control Plane
Runbook
On-call
GameDay
Incident Bundle
Cost / Budget

31. Repository public contract¶

repository_public_contract:
  repository: ""
  visibility: public
  project_status: experimental | active | maintenance | archived
  readme: ""
  license: ""
  contributing: ""
  code_of_conduct: ""
  security: ""
  governance: ""
  support: ""
  changelog: ""
  citation: ""
  issue_templates: []
  pull_request_template: ""
  maintainers: []
  support_window: ""

32. README acceptance¶

The first screen explains what the project does and who it is for.
The Out-of-scope boundary and production responsibility statement are explicit.
The architecture diagram matches the repository version.
Quickstart runs in a Clean-room environment.
Both success and failure paths are demonstrated.
A Trace / Evidence entry point is provided.
Test, evaluation, and security commands are provided.
LICENSE, SECURITY, CONTRIBUTING, and SUPPORT are linked.
Maintenance status and supported versions are stated.

33. Asset license matrix¶

Path / Asset	Type	Copyright	License ID	Source	Modified	Redistribution	Notice

asset_license:
  path: ""
  asset_type: code | documentation | data | model | media
  copyright_holder: ""
  license_id: ""
  source: ""
  modified: false
  redistribution_allowed: false
  notice_required: false
  approved_by: ""

34. SECURITY.md contract¶

security_policy:
  supported_versions: []
  private_reporting_channel: ""
  do_not_report_publicly: []
  required_report_fields: []
  acknowledgement_target: ""
  triage_target: ""
  remediation_process: ""
  coordinated_disclosure: ""
  advisory_process: ""
  cve_process: ""
  security_contacts: []

35. Contribution and governance¶

contribution_contract:
  development_setup: ""
  branch_policy: ""
  commit_policy: ""
  required_checks: []
  required_evidence: []
  review_owners: []
  security_review_triggers: []
  compatibility_review_triggers: []
  dco_or_cla: ""
  release_authority: []

governance:
  maintainers: []
  decision_process: ""
  voting_or_consensus: ""
  escalation: ""
  conflict_of_interest: ""
  moderation: ""
  maintainer_addition: ""
  maintainer_removal: ""
  succession: ""
  archival_policy: ""

36. Pull request definition of done¶

37. Version and release notes¶

release_notes:
  release: ""
  date: ""
  status: rc | stable | deprecated | retired
  what_works: []
  target_users: []
  quality_evidence: []
  security_evidence: []
  known_limitations: []
  compatibility: []
  migration: []
  rollback: []
  artifact_digests: []
  provenance: ""
  support_until: ""
  maintainers: []

38. Maintenance and sustainability¶

Responsibility	Primary	Backup	Response Policy	Automation	End Condition
Issues
PR Review
Security Reports
Dependency Updates
Release
Documentation
Community Moderation

39. Known limitations¶

known_limitation:
  limitation_id: ""
  description: ""
  affected_users: []
  affected_versions: []
  impact: ""
  workaround: ""
  detection: ""
  planned_resolution: ""
  owner: ""
  review_by: ""

40. Independent acceptance¶

independent_acceptance:
  review_id: ""
  reviewer: ""
  reviewer_role: ""
  independent_from_implementation: true
  source_commit: ""
  release_artifact_digest: ""
  environment: ""
  requirements_sampled: []
  evidence_verified: []
  clean_room_run: ""
  success_scenarios: []
  failure_scenarios: []
  security_checks: []
  operational_checks: []
  limitations_confirmed: []
  findings: []
  result: accepted | conditional | rejected
  conditions:
    - condition: ""
      owner: ""
      due_at: ""
      verifier: ""
  signed_at: ""

41. Final acceptance decision¶

final_acceptance:
  project_id: ""
  release: ""
  evidence_package: ""
  gate_results:
    g0_problem: ""
    g1_design: ""
    g2_slice: ""
    g3_quality: ""
    g4_security: ""
    g5_operations: ""
  independent_review: ""
  known_risks: []
  decision: accepted | conditional | rejected
  conditions: []
  approved_by:
    product: ""
    architecture: ""
    engineering: ""
    quality: ""
    security: ""
    operations: ""
    open_source: ""
  signed_at: ""

Final principle:

Delivery is complete only when claims are traceable, paths are reproducible, risks are verifiable, artifacts can be validated, limitations are visible, and responsibilities can be handed over.