Skip to content

Risk-based governance tiers

Status
Approved
Version
2.0.0
Authority
UASK standard
Applies to
Core
Markdown source

Summary

Process costs time, and the same process does not fit a one-file script and an operating-system distribution. A project chooses one of three tiers — lightweight, standard or high-assurance — by the harm its failure can do, and the tier says how much ceremony the foundation, review and release work needs. A tier takes ceremony away; it never takes away a decision that belongs to the owner, and it never lowers a security or legal obligation.

Rules

  • R1. A project MUST choose exactly one tier. It MUST record the value in the front matter of its Blueprint (tier:) and the reason in section 4 of the Blueprint (arch.architecture-description#R12), or both in its AGENTS.md when it has no Blueprint. A Blueprint still on the earlier ten-section template records both under “Cross-cutting concerns”.
  • R2. The owner MUST decide the tier; an agent MAY propose one with the criteria it applied. When two tiers both seem to fit, the project MUST take the higher one.
  • R3. A tier MUST NOT remove an owner decision. At every tier the owner approval gate of mgmt.project-foundation (mgmt.project-foundation#R9, mgmt.project-foundation#R10) and the release approval of quality.quality-release (quality.quality-release#R14) stay.
  • R4. A tier MUST NOT lower a security, privacy or legal obligation, or the backup before a risky change (ops.backups#R4). Such obligations rank above the tier when instructions meet (the instruction precedence of a consumer project, tools/AGENTS.hub.md).
  • R5. A project MUST review its tier when its risk changes — for example when it gets paying users, holds personal data at scale, or its failure can harm people — and at each re-foundation and each MAJOR release. A change of tier MUST be recorded as an ADR (arch.adr).
  • R6. At the lightweight tier the foundation MAY be one short file that holds the purpose, the scope, the decisions made and the next steps. It MUST still name the owner, MUST list what is out of scope, and MUST record a hard-to-reverse choice as a line with its reason. Every change MUST still pass the automated gate (quality.quality-release#R5, quality.quality-release#R6), and going live (G3) MUST rest on a rollback that was tested or rehearsed (R10).
  • R7. At the standard tier the foundation MUST consist of the Vision, PRD and Blueprint, with an ADR for each hard-to-reverse choice, tailored as the playbook mgmt.project-foundation allows. Every change MUST be reviewed by someone other than its author — a person or a separate agent session — and the project MUST keep a test strategy (quality.quality-release#R3). An independent architecture review of the Blueprint (arch.architecture-review) MUST precede G1, and its verdict MUST be shown to the owner (process.project-lifecycle#R5). It comes in addition to the review before a MAJOR release or a hard-to-reverse change that the playbook asks for.
  • R8. At the high-assurance tier the project MUST meet the standard tier and, in addition, MUST: keep a risk register and a threat model next to the Blueprint; trace each requirement to the test or check that proves it; have the Blueprint and each release reviewed by a second reviewer who did not write it; rehearse the rollback before each release (quality.quality-release#R13); and keep the acceptance evidence of every release (req.acceptance-criteria#R10).
  • R9. At the high-assurance tier the independent check of Critical and High findings (arch.architecture-review#R13) MUST be done for every review, whoever designed the system.
  • R10. At every tier, going live with production of a live or client system (G3, process.project-lifecycle#R8) MUST rest on a rollback that was tested or rehearsed. What a rollback plan contains is set by quality.quality-release#R13. The high-assurance tier asks for more, a rehearsal before each release (R8).

Guidance

Choosing a tier

Ask what the worst realistic failure costs, who bears it and whether it can be undone. The questions below place a project; the owner decides.

Lightweight Standard High-assurance
Typical case a script, a personal tool, a throw-away prototype nobody depends on a product with users and a small team a system whose failure harms people, money or data at scale; an operating-system distribution
Worst realistic failure an hour lost, easy to redo lost orders, data or reputation, recoverable harm to people, large financial or legal loss, data that cannot be recreated
Who is affected the author customers and the business many users, or people who did not choose the product

What a tier changes

Lightweight Standard High-assurance
Foundation artefacts one short file Vision, PRD, Blueprint, ADRs, roadmap the same, plus a risk register, a threat model and traceability to tests
Architecture review none; a code review or an ADR discussion is enough independent review before G1; before a MAJOR release or a hard-to-reverse change as standard, with the independent check of Critical and High findings every time
Review of changes the author, against the Definition of Done someone other than the author, every change a second reviewer for the design and for each release
Release the gate passes; a tested or rehearsed rollback before go-live; a note in the changelog the whole quality.quality-release playbook, with the rollback tested before go-live the playbook, a rehearsed rollback and kept acceptance evidence
Evidence the tests that ran a test strategy and a CI run per change acceptance evidence kept for every release

Using the tier in a playbook

Each playbook has a Tailoring section that names the tier instead of repeating this table. The tier is one input to tailoring; the owner still confirms the tailoring at intake (mgmt.project-foundation#R1). The two tables above are the only home of the tier definitions: a project document, the Hub’s own Blueprint included, links to this item and does not copy them.

Common mistakes

  • Choosing the lightest tier to start faster and never revisiting it after the first paying customer (R5).
  • Treating the tier as permission to skip a backup, a licence check or a security fix (R4).
  • An agent choosing the tier alone because it “obviously” is a small project (R2).

Explanation

Tailoring a process to the project is the usual answer in process standards: the life-cycle standard ISO/IEC/IEEE 12207 includes a tailoring process for exactly this. NIST SP 800-30 supplies the vocabulary for the input to it: risk is the likelihood of harm combined with its impact, judged on a small qualitative scale. Three tiers are enough to be usable and few enough to be applied consistently; a project in between takes the higher one, because the cost of too little process is paid later and by someone else.

The tier scales process and evidence, not authority. Who approves, who accepts and who decides scope does not depend on how much paperwork the project has, which is why R3 and R4 fix what no tier can remove.

Examples

Good — a personal command-line tool with one user chooses the lightweight tier: a README that holds purpose, scope and decisions, a test script run by CI, and a changelog line per release. When the tool gains outside users, the owner raises the tier by an ADR and the Vision, PRD and Blueprint are written.

Bad — a private e-commerce project takes the lightweight tier because “it is only a theme change”, although it replaces a live shop that takes payments. Nothing records the external interfaces, no restore was ever tried, and the first cutover breaks the payment notifications. The project belonged to the standard tier (R2: when in doubt, the higher one).

Sources