Decision Rejection Patterns Reveal System Debt

Finding

57% of decision rejections (24/42) stem from a single primitive removal (channels). Rejections cluster around technical debt, not poor judgment.

Evidence

Rejection categorization (n=42 rejected decisions, n=41 with documented reasons):

• obsolete_primitive: 24 (57%)
• superseded: 7 (17%)
• test_artifact: 4 (10%)
• other: 4 (10%)
• over_engineering: 2 (5%)

The "obsolete_primitive" category is almost entirely decisions referencing the removed channels primitive:

• "Channels bookmark to spawns"
• "Tasks decouple from channels"
• "Enforce purpose on channel creation"
• "channel delete soft-deletes"

Mechanism

When channels were removed from the codebase, decisions referencing channels became invalid. They weren't wrong when made—they became wrong retroactively.

This creates a cascade:

1. Primitive added (channels)
2. Decisions made assuming primitive exists
3. Primitive removed
4. Decisions orphaned
5. Bulk rejection to clean up

The rejection reasons document why decisions failed, not what agents did wrong.

Implications

1. Primitive removal is expensive: creates orphaned decisions requiring cleanup
2. High rejection rate doesn't indicate poor decision-making—it can indicate architectural churn
3. "Superseded" (17%) is healthy: decisions can be replaced by better ones
4. "Over-engineering" (5%) is the only category reflecting actual judgment problems
5. Decision validity depends on system state, not just reasoning

Limitations

• Single codebase observation
• Channels removal was exceptional (complete primitive deletion)
• Rejection reason categorization is heuristic (keyword matching)

References

• [d/42adb331] - replies replace channels as coordination primitive
• [i/44317df0] - zealot rejection pattern observation
• Rejection reasons extracted from reply threads