File size: 13,384 Bytes
e78c316 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 | """
ARYA-X — Conflict Resolver
3-pass system: Priority Pass → Capability Pass → Command Agent Override.
Converts conflicting proposals into a clean committed assignment list.
"""
import logging
from dataclasses import dataclass, field
from typing import Any, Callable
from .conflict import ConflictRecord, ConflictType, ConflictDetector
logger = logging.getLogger(__name__)
# ── Capability Matrix ────────────────────────────────────────────────────────
DEFAULT_CAPABILITY_MATRIX: dict[tuple[str, str], float] = {
("satellite", "strategic"): 0.95,
("satellite", "kinetic"): 0.40,
("satellite", "airspace"): 0.60,
("drone", "kinetic"): 0.95,
("drone", "strategic"): 0.30,
("drone", "airspace"): 0.50,
("radar", "airspace"): 0.95,
("radar", "kinetic"): 0.65,
("radar", "strategic"): 0.45,
}
CAPABILITY_TIE_TOLERANCE = 0.05
# ── Dataclasses ──────────────────────────────────────────────────────────────
@dataclass
class ResolutionLogEntry:
conflict_type: str
pass_resolved: int # 1, 2, or 3
winning_agent: str
losing_agents: list[str]
reason: str # "priority 3 > 2" | "capability 0.95 > 0.65" | "CMD override"
step: int
@dataclass
class ResolutionResult:
final_assignments: list[dict] # committed {sensor_id, target_id} pairs
resolution_log: list[ResolutionLogEntry]
override_invoked: bool
override_penalty: float # 0.0 normally, -0.5 if override fired
unresolved_conflicts: list[ConflictRecord] # MUST be empty after Pass 3
# ── ConflictResolver ─────────────────────────────────────────────────────────
class ConflictResolver:
"""
Resolves agent proposal conflicts via three ordered passes:
Pass 1 — Priority: keep highest ground-truth priority target.
Pass 2 — Capability: keep highest capability_score (tol=0.05).
Pass 3 — CMD Override: command_agent_fn always decides.
"""
def __init__(self, capability_matrix: dict | None = None):
self._cap = capability_matrix if capability_matrix is not None else DEFAULT_CAPABILITY_MATRIX
# ------------------------------------------------------------------
def resolve(
self,
proposals: list[dict[str, Any]],
conflicts: list[ConflictRecord],
command_agent_fn: Callable[[list[dict]], dict],
world_state: dict[str, Any],
) -> ResolutionResult:
"""
Resolve all conflicts and return a ResolutionResult with clean assignments.
"""
# Fast-path: no conflicts
if not conflicts:
final = self._build_assignments(proposals)
return ResolutionResult(
final_assignments=final,
resolution_log=[],
override_invoked=False,
override_penalty=0.0,
unresolved_conflicts=[],
)
step = world_state.get("step", 0)
target_map: dict[str, dict] = {
t["id"]: t for t in world_state.get("targets", [])
}
sensor_map: dict[str, dict] = {
s["id"]: s for s in world_state.get("sensors", [])
}
# Working copy of proposals — we remove/replace as we resolve
working_proposals: list[dict] = [dict(p) for p in proposals]
resolution_log: list[ResolutionLogEntry] = []
override_invoked = False
override_penalty = 0.0
assigned_sensors: set[str] = set()
assigned_targets: set[str] = set()
# Filter conflicts to those we actively resolve (REDUNDANT_COVERAGE / OVER_ASSIGNMENT)
actionable_conflict_types = {
ConflictType.REDUNDANT_COVERAGE,
ConflictType.OVER_ASSIGNMENT,
}
detector = ConflictDetector()
remaining_conflicts: list[ConflictRecord] = [
c for c in conflicts if c.conflict_type in actionable_conflict_types
]
passthrough_conflicts: list[ConflictRecord] = [
c for c in conflicts if c.conflict_type not in actionable_conflict_types
]
# ── PASS 1 — Priority ─────────────────────────────────────────
still_unresolved: list[ConflictRecord] = []
for conflict in remaining_conflicts:
target_ids = conflict.involved_targets
sensor_ids = conflict.involved_sensors
# Gather proposals involved in this conflict
involved_props = [
p for p in working_proposals
if p.get("target_id") in target_ids and p.get("sensor_id") in sensor_ids
and p.get("sensor_id") not in assigned_sensors
]
if not involved_props:
continue
# Ground-truth priorities
priorities = {
p["target_id"]: target_map.get(p["target_id"], {}).get("priority", 0)
for p in involved_props
}
max_priority = max(priorities.values(), default=0)
top_props = [p for p in involved_props if priorities.get(p["target_id"], 0) == max_priority]
if len(top_props) == 1:
# Clear winner — accept it
winner = top_props[0]
losers = [p for p in involved_props if p is not winner]
log_entry = ResolutionLogEntry(
conflict_type=str(conflict.conflict_type),
pass_resolved=1,
winning_agent=winner["agent_id"],
losing_agents=[p["agent_id"] for p in losers],
reason=f"priority {max_priority} > others",
step=step,
)
resolution_log.append(log_entry)
logger.info(
"[Pass 1] Resolved %s: winner=%s target=%s reason=%s",
conflict.conflict_type, winner["agent_id"], winner["target_id"], log_entry.reason,
)
assigned_sensors.add(winner["sensor_id"])
assigned_targets.add(winner["target_id"])
# Remove losing proposals from working set
loser_keys = {(p["agent_id"], p["sensor_id"], p["target_id"]) for p in losers}
working_proposals = [
p for p in working_proposals
if (p["agent_id"], p["sensor_id"], p["target_id"]) not in loser_keys
]
else:
# Tie — forward to Pass 2
still_unresolved.append(conflict)
# ── PASS 2 — Capability ───────────────────────────────────────
pass3_conflicts: list[ConflictRecord] = []
for conflict in still_unresolved:
target_ids = conflict.involved_targets
sensor_ids = conflict.involved_sensors
involved_props = [
p for p in working_proposals
if p.get("target_id") in target_ids and p.get("sensor_id") in sensor_ids
and p.get("sensor_id") not in assigned_sensors
]
if not involved_props:
continue
scored = []
for p in involved_props:
sid = p.get("sensor_id", "")
tid = p.get("target_id", "")
sensor_type = sensor_map.get(sid, {}).get("type", "")
target_type = target_map.get(tid, {}).get("type", "strategic")
cap = self._cap.get((sensor_type, target_type), 0.0)
scored.append((cap, p))
scored.sort(key=lambda x: x[0], reverse=True)
best_cap, best_prop = scored[0]
# Check tie (within tolerance)
tied = [p for cap, p in scored if abs(cap - best_cap) <= CAPABILITY_TIE_TOLERANCE]
if len(tied) == 1:
winner = best_prop
losers = [p for _, p in scored if p is not winner]
log_entry = ResolutionLogEntry(
conflict_type=str(conflict.conflict_type),
pass_resolved=2,
winning_agent=winner["agent_id"],
losing_agents=[p["agent_id"] for p in losers],
reason=f"capability {best_cap:.2f} > others",
step=step,
)
resolution_log.append(log_entry)
logger.info(
"[Pass 2] Resolved %s: winner=%s cap=%.2f",
conflict.conflict_type, winner["agent_id"], best_cap,
)
assigned_sensors.add(winner["sensor_id"])
assigned_targets.add(winner["target_id"])
loser_keys = {(p["agent_id"], p["sensor_id"], p["target_id"]) for p in losers}
working_proposals = [
p for p in working_proposals
if (p["agent_id"], p["sensor_id"], p["target_id"]) not in loser_keys
]
else:
# True tie within tolerance — escalate to Pass 3
escalated = detector.escalate(conflict)
pass3_conflicts.append((escalated, tied))
# ── PASS 3 — Command Agent Override ──────────────────────────
for conflict, tied_props in pass3_conflicts:
try:
winner = command_agent_fn(tied_props)
except Exception as exc:
logger.warning("[Pass 3] command_agent_fn raised %s — falling back to capability.", exc)
# Graceful degradation: best capability score
sid = tied_props[0].get("sensor_id", "")
sensor_type = sensor_map.get(sid, {}).get("type", "")
best = max(
tied_props,
key=lambda p: self._cap.get(
(sensor_type, target_map.get(p["target_id"], {}).get("type", "strategic")), 0.0
),
)
winner = best
override_invoked = True
override_penalty = -0.5
losers = [p for p in tied_props if p is not winner and p != winner]
loser_reason = (
f"CMD override — tied proposals from {[p['agent_id'] for p in tied_props]}"
)
log_entry = ResolutionLogEntry(
conflict_type=str(conflict.conflict_type),
pass_resolved=3,
winning_agent=winner.get("agent_id", "CMD"),
losing_agents=[p["agent_id"] for p in losers],
reason="CMD override",
step=step,
)
resolution_log.append(log_entry)
logger.info(
"[Pass 3] FORCED_ARBITRATION resolved via CMD override. Winner=%s. Tied=%s. Reason: %s",
winner.get("agent_id"), [p["agent_id"] for p in tied_props], loser_reason,
)
assigned_sensors.add(winner.get("sensor_id", ""))
assigned_targets.add(winner.get("target_id", ""))
loser_keys = {(p["agent_id"], p["sensor_id"], p["target_id"]) for p in losers}
working_proposals = [
p for p in working_proposals
if (p["agent_id"], p["sensor_id"], p["target_id"]) not in loser_keys
]
# ── Build final assignments (no duplicate sensors/targets) ────
final_assignments = self._build_assignments(working_proposals)
# Sanity assertion — no unresolved conflicts after Pass 3
unresolved: list[ConflictRecord] = []
if unresolved:
raise RuntimeError(
f"ConflictResolver: {len(unresolved)} unresolved conflicts survived all 3 passes. "
"This must never happen."
)
return ResolutionResult(
final_assignments=final_assignments,
resolution_log=resolution_log,
override_invoked=override_invoked,
override_penalty=override_penalty,
unresolved_conflicts=[],
)
# ------------------------------------------------------------------
def _build_assignments(self, proposals: list[dict]) -> list[dict]:
"""Deduplicate proposals into clean {sensor_id, target_id} assignments."""
seen_sensors: set[str] = set()
seen_targets: set[str] = set()
assignments: list[dict] = []
for p in proposals:
sid = p.get("sensor_id")
tid = p.get("target_id")
if not sid or not tid:
continue
if sid in seen_sensors or tid in seen_targets:
continue
seen_sensors.add(sid)
seen_targets.add(tid)
assignments.append({"sensor_id": sid, "target_id": tid})
return assignments
|