Replace story points with Context, Iteration, Review, and Risk.
Task: fix bug in payment calculation
C: 3 | I: 2 | R: 3 | K: 3
→ requires human review + controlled deployment
AI writes the code. CIRK governs the execution.
The shift
But AI changed the bottleneck. What matters now is not just how long work takes. It is:
CIRK models that reality.
The model
Each task is scored from 1 to 3 across four dimensions. The vector defines how it runs.
How much system understanding is required for correct execution?
| C1 | Isolated logic, local change, minimal context |
| C2 | Multiple components, shared patterns, moderate system understanding |
| C3 | Cross-system behavior, architectural reasoning, or domain-critical context |
How many cycles are likely required before the solution converges?
| I1 | Deterministic or near one-pass execution |
| I2 | Some iteration expected, usually 2–3 cycles |
| I3 | Ambiguous, exploratory, or high-complexity convergence |
How much human validation is required before the work is trusted?
| R1 | Quick validation or spot-check |
| R2 | Rule-level or functional review required |
| R3 | Deep architectural, security, or mission-critical review |
How risky is integration, release, or rollout?
| K1 | Safe, additive, low-risk rollout |
| K2 | Existing behavior changes, moderate release caution |
| K3 | Coordinated rollout, migration, or cross-team dependency |
Composite score = C + I + R + K
"Score the execution reality, not the political preference."
CIRK Scoring Guidance
Example — Login API
What CIRK is not
CIRK is an execution standard for AI-native development.
CIRK in 60 seconds
Any unit of work: a feature, a fix, a refactor, a migration.
Assign C, I, R, K values from 1 to 3. Ask: how much context? how many cycles? how much review? how risky is rollout?
The vector maps to an execution mode — autonomous, guided, draft-first, supervised, or blocked. Agents and humans follow the same rules.
Teams gain a shared language: "This is high R." "Low K, we can ship." "I3 — let the agent draft first."
Rule of thumb
Execution mapping
The vector defines what happens — not just how big the task is.
Agent executes without intervention. Auto-approval allowed. No checkpoints.
Agent executes. Human review required before merge.
Agent produces a draft. Human validates before any commit.
Step-by-step execution. Approval per step. Deploy runbook required.
Task must be decomposed. Execution not allowed.
Policy rules
Examples
Each vector maps to concrete execution behavior, with reasoning for each dimension.
Backend
Frontend
Agent
Add read-only status endpoint
Update button color token
Rename internal variable and update references
Login API
Dashboard layout revision
Refactor repository layer for shared patterns
Billing webhook change
Auth UI redesign (login + recovery flows)
Change identity propagation across services
Open standard
It is an open standard for execution governance in AI-native software development.
It can be adopted in issue trackers, coding agents, pull request workflows, internal platforms, or governance layers such as Orbit618.
FAQ
Yes, in AI-assisted development contexts where execution governance matters more than effort estimation.
Story points estimate human effort. CIRK estimates execution conditions for AI-native workflows.
Not directly. CIRK is about execution complexity, review intensity, and rollout risk.
Teams may later derive time insights from calibration data, but duration is not the primary output.
Yes. Even when humans perform the implementation, CIRK still helps classify review burden, context depth, and deployment sensitivity.
No. Orbit618 is one possible implementation environment for CIRK, but CIRK is designed as a standalone open standard.
Because effort is no longer the most important variable for AI-assisted execution.
CIRK makes the real constraints explicit instead of treating them as side rules layered on top of an effort model.
The vector, not just the score.
C3 I1 R1 K3 and C1 I3 R3 K1 may have the same sum but require very different execution policies. One is deploy-sensitive. The other is review-sensitive.
No. CIRK is a standard and a shared execution language. Products and platforms may implement it, but the model itself is implementation-agnostic and MIT licensed.
MIT licensed. No dependencies. Works with anything.