Fragile keyword matching in _position_action_friction() for outcome friction estimation

[DeveshParagiri]

Problem

_position_action_friction() in extropy/simulation/engine.py:1350-1385 uses hardcoded keyword substring matching to estimate behavioral friction from position labels:

token = position.lower()
low = ("maintain", "continue", "keep", "stay", "deny", "remove_shared_access", "no_change")
medium = ("reduce", "abandon", "pause")
high = ("switch", "migrate", "cancel", "subscribe", "purchase", "buy", "upgrade")

if any(k in token for k in low):
    return 0.2
if any(k in token for k in medium):
    return 0.4
if any(k in token for k in high):
    return 0.75
return 0.5  # default fallback

Silent Failure Cases

Position Label	Expected Friction	Actual	Why
keep_subscription	0.2 (low)	0.2	✅ "keep" matches
hold	0.2 (low)	0.5	❌ Not in keywords
retain	0.2 (low)	0.5	❌ Not in keywords
stop	0.75 (high)	0.5	❌ Not in keywords
quit	0.75 (high)	0.5	❌ Not in keywords
sign_up	0.75 (high)	0.5	❌ Not in keywords
join	0.75 (high)	0.5	❌ Not in keywords
leave	0.75 (high)	0.5	❌ Not in keywords
drop	0.4 (medium)	0.5	❌ Not in keywords

The LLM-generated scenario outcomes can use any phrasing. When they don't match the hardcoded keywords, friction defaults to 0.5, which affects:

• _infer_private_anchor_position() - picks wrong default
• Flip resistance calculations

Suggested Fix

Add explicit option_friction to outcome definitions in scenario spec:

# scenario.v1.yaml
outcomes:
  - name: position
    type: categorical
    options:
      - value: keep_current_plan
        friction: 0.2
      - value: switch_to_competitor  
        friction: 0.75
      - value: reduce_usage
        friction: 0.4

Then in engine:

def _position_action_friction(self, position: str | None) -> float:
    if not position:
        return 0.5
    # First check explicit friction from scenario spec
    if position in self._primary_option_friction:
        return self._primary_option_friction[position]
    # Fallback to heuristic with warning
    logger.warning(f"No explicit friction for '{position}', using heuristic")
    return self._heuristic_friction(position)

Impact

Medium - affects behavioral realism but has a reasonable default fallback (0.5).

References

• extropy/simulation/engine.py:1350-1385
• Used by _infer_private_anchor_position() at line 1387