ACTIVE vs DROP Node Logic Explained | Generated by AI

Home 2026.05 English 中文

---

Here’s the full picture:

How ACTIVE vs DROP works

The two key enums (models.py:12-54)

class NodeStatus(str, Enum):
    ACTIVE = "ACTIVE"           # viable, can be expanded further
    PRUNED_BY_RULE = "PRUNED_BY_RULE"  # killed by hard physics rule
    PRUNED_BY_SLM = "PRUNED_BY_SLM"   # killed by soft scoring (reserved)

class NodeResultState(str, Enum):
    PASS = "PASS"    # survived this round
    DROP = "DROP"    # dead — won't be expanded

A node only gets DROP when both are set together:

self.node.status = NodeStatus.PRUNED_BY_RULE
self.node.result_state = NodeResultState.DROP
self.node.fsm_state = FSMState.FINALIZED   # no further transitions

---

Where DROP is triggered

1. During calculate stage (builder.py:295-310) — tot_hard_rule_check from skills.py returns physical_violations:

if not hard_rule_result["passed"]:
    if physical_violations:
        → status = PRUNED_BY_RULE, result_state = DROP

2. During evaluate stage (builder.py:396-416) — the evaluation breakdown itself contains hard_rule_violations:

if breakdown.hard_rule_violations:
    → status = PRUNED_BY_RULE, result_state = DROP

---

What does NOT cause DROP

• Low score — the docstring at builder.py:76 is explicit: “Low scores never prune the node; they only reduce downstream expansion priority.” The node stays ACTIVE, just ranked lower.
• Recoverable violations — sent to REFLECT state for a retry instead of being dropped (builder.py:317).
• Root node violations — special-cased via _bounce_root_rule_violations() — gets a reflection attempt rather than immediate DROP.

---

One-liner: Only a hard physics rule violation (detected by skills.py) flips a node to DROP. Soft scores just affect expansion priority, leaving the node ACTIVE.

---

Back Donate