[Bug]: update_frontier raises IndexError when frontier has no scored programs

[blrain3]

Summary

ProgramManager.update_frontier() raises IndexError: list index out of range when get_frontier_with_scores() returns an empty list (all frontier programs have score=None).

Reproduction

from src.registry import ProgramManager

manager = ProgramManager("/path/to/repo")

# Manually mark a program as frontier without a score
manager.create_program("base", config_without_score)
manager.mark_frontier("base")

# Later, try to update frontier with a scored program
manager.update_frontier("iter-1", 0.85)

Expected: The unscored program is either included with a default score or the comparison is skipped.

Actual: get_frontier_with_scores() filters out all programs with score=None, returning an empty list. Then scored[-1] raises IndexError: list index out of range.

Root Cause

In src/registry/manager.py:

• get_frontier_with_scores() (line 325-344): if score is not None filters out unscored programs
• update_frontier() (line 423): worst_name, worst_score = scored[-1] — if all frontier programs are unscored, scored is empty and scored[-1] raises IndexError

Trigger Conditions

This happens when:

1. A program is added to frontier via mark_frontier() without being evaluated, OR
2. Config read fails for all frontier programs (the except Exception: continue on line 341 silently skips them), OR
3. ProgramConfig.get_score() returns None for all frontier members

Impact

• The self-improving loop crashes mid-run, losing all progress
• The error is silent until it happens; no validation prevents unscored programs from entering the frontier

Suggested Fix

Handle empty scored list in update_frontier:

scored = self.get_frontier_with_scores()
if not scored:
    # No scored programs yet; add unconditionally
    self.mark_frontier(name)
    return True

Or default score to 0.0 in ProgramConfig.get_score() instead of returning None.

Environment

• EvoSkill version: latest (2026-09-16)
• Python: 3.11