Right-Sizing Process to the Work
Four tiers match the weight of the change to the weight of the review. A copy edit and a payment system rewrite do not get the same gate treatment.
The Four Tiers
Trivial
Changes with negligible risk and no architectural implications.
Evidence required: one-line rationale in the commit message. Gates may collapse into a single approval.
Typical examples: copy and content updates, small styling tweaks, typo fixes, meta tag adjustments, adding an entry to Stack Quirks.
Standard
Scoped features with predictable structure. The default tier for most work.
Evidence required: documented responses to each agent concern, inline in the spec or checklist. Decision Rationale section required.
Typical examples: a new static page, a new interactive component, analytics integration, a campaign landing page, a contact form variant.
Complex
Multi-surface changes, new architectural patterns, or cross-cutting strategic work.
Evidence required: full written responses to all agent concerns. Anchor impact evaluated at Arch Gate; update may ship in the same PR.
Typical examples: framework migrations, new public URL trees, AI agent integration on a new surface, major navigation reorganizations, introducing a new dependency class.
Critical
Changes touching authentication, payments, PII, or core domain models.
Evidence required: full agent responses, a 24-hour waiting period, and an external second reviewer. No exceptions.
Typical examples: auth and user account systems, payment and subscription flows, any PII handling, security-boundary changes, schema changes on authoritative data.
How to Pick the Tier
Tiers are not a taste question. They are determined by escalation triggers built into the spec template. If any of the following apply, the spec is Critical:
- Authentication or authorization touched
- Payment or financial data touched
- PII or PHI handled
- New external integration
- Core domain model change
- New architectural pattern introduced
- Framework migration
- AI or ML integration where agent output drives production behavior
Absent those triggers, the tier defaults are calibrated to keep the review overhead proportional to the actual risk: Trivial for content, Standard for most scoped work, Complex for cross-cutting or strategic changes.
A tier-discipline signal to watch: if more than 80% of your specs are landing as Trivial, something is wrong — either triggers are being under-read, or the team is gaming the process to avoid review.