Fragile keyword matching in _normalize_focus_mode() for agent focus detection

[DeveshParagiri]

Problem

_normalize_focus_mode() in extropy/population/sampler/core.py:58-72 uses hardcoded keyword substring matching to determine which household members become agents:

focus_lower = agent_focus.lower()

if any(kw in focus_lower for kw in ("famil", "household", "everyone")):
    return "all"

if any(kw in focus_lower for kw in ("couple", "pair", "partners", "spouses")):
    return "couples"

return "primary_only"

Silent Failure Cases

agent_focus Value	Expected Mode	Actual	Why
families	all	all	✅ "famil" matches
all_members	all	primary_only	❌ No keywords match
whole_household	all	all	✅ "household" matches
complete_units	all	primary_only	❌ No keywords match
couples	couples	couples	✅ "couple" matches
married	couples	primary_only	❌ No keywords match
dual_income	couples	primary_only	❌ No keywords match
both_adults	couples	primary_only	❌ No keywords match

When the spec builder or user uses different phrasing for agent_focus, the wrong focus mode is silently applied.

Suggested Fix

Use an enum or explicit mapping in the spec:

# population.v1.yaml
agent_focus: all  # enum: primary_only | couples | all

Or if keeping natural language, use an LLM call during spec compilation to normalize to the enum, not runtime heuristics.

Impact

Medium-High - determines whether children/partners become reasoning agents. Wrong mode means either:

• Missing agents (primary_only when should be all)
• Wasted compute (all when should be primary_only)

References

• extropy/population/sampler/core.py:58-72
• Called by sample_population()

[DeveshParagiri]

Fixed in PR #107 - agent_focus_mode is now an enum set by LLM during sufficiency check