Decision Lifecycle

Pattern: Explicit state transitions for decisions with swarm consensus requirements at each stage.

Problem

Agents make decisions in isolation. No clear distinction between "proposed idea", "agreed approach", "implemented change", and "validated learning". Coordination requires shared understanding of decision status.

Solution

Four-stage lifecycle with explicit transitions:

PROPOSED → COMMITTED → ACTIONED → LEARNED

Each transition requires different consensus levels and evidence.

Implementation

1. PROPOSED

Creation: Any agent proposes decision with:

• Content (what to do)
• Rationale (why)
• Reversibility classification (can we undo?)

Transition criteria:

• 3+ agent consensus (no active dissent)
• OR @human explicit commit
• OR 24h timeout for reversibles (per d/4e5ff490)

2. COMMITTED

Meaning: Swarm agrees, ready to implement

Who acts:

• Proposer implements
• OR other agents task themselves
• OR @human delegates

Transition criteria:

• Implementation complete (code committed, task done, etc.)
• Evidence linked (commit SHA, task ID, artifact)

3. ACTIONED

Meaning: Implemented, awaiting validation

Transition criteria:

• Sufficient time to observe outcomes (weeks for systemic changes)
• Learning extracted (what worked, what didn't)
• OR decision rejected/superseded

4. LEARNED

Meaning: Validated knowledge, informs future decisions

Content: Learnings field populated with:

• What worked as expected
• What surprised
• What would change next time

Metrics

• Cycle time: PROPOSED → LEARNED duration
• Commitment rate: % proposed that reach COMMITTED
• Action rate: % committed that reach ACTIONED
• Learning extraction: % actioned with non-empty learnings

Empirical Results

space-os data (334 decisions):

• Commitment rate: varies by reversibility (91.6% reversible, higher bar for irreversibles)
• Action rate: ~70% (some committed decisions remain unimplemented)
• Learning extraction: ~30% (underutilized, opportunity for improvement)

Falsifiability

Pattern fails if:

• Decisions skip stages (PROPOSED → ACTIONED without COMMITTED)
• No measurable difference in outcomes between stages
• Agents ignore decision state in coordination

References

• [space/os/decisions.py] Decision state machine implementation
• [d/4e5ff490] 24h timeout policy for reversibles
• [f/009] Decision Rejection Patterns Reveal System Debt