Merge pull request #513 from sylvesterkaczmarek/fix/market-snapshot-invalid-numerics

fix(infonet): harden market snapshot projection

backend/services/infonet/markets/snapshot.py

@@ -24,6 +24,7 @@ from __future__ import annotations
 
 import hashlib
 import json
+import math
 from typing import Any, Iterable
 
 
@@ -32,6 +33,14 @@ def _payload(event: dict[str, Any]) -> dict[str, Any]:
     return p if isinstance(p, dict) else {}
 
 
+def _finite_float(value: Any) -> float | None:
+    try:
+        parsed = float(value)
+    except (TypeError, ValueError, OverflowError):
+        return None
+    return parsed if math.isfinite(parsed) else None
+
+
 def _events_for_market(market_id: str, chain: Iterable[dict[str, Any]]) -> list[dict[str, Any]]:
     out: list[dict[str, Any]] = []
     for ev in chain:
@@ -61,6 +70,10 @@ def build_snapshot(
     to advance to EVIDENCE. Pass it explicitly so the function stays
     pure and deterministic.
     """
+    frozen_at_value = _finite_float(frozen_at)
+    if frozen_at_value is None:
+        raise ValueError("frozen_at must be finite")
+
     events = _events_for_market(market_id, chain)
 
     predictor_ids: list[str] = []
@@ -79,27 +92,29 @@ def build_snapshot(
         side = p.get("side")
         if side not in ("yes", "no"):
             continue
+
+        stake = p.get("stake_amount")
+        if stake is None:
+            weight = 1.0  # Free pick = 1.0 virtual stake (RULES §5.2).
+            staked_amount = 0.0
+        else:
+            parsed_stake = _finite_float(stake)
+            if parsed_stake is None or parsed_stake <= 0:
+                # Invalid paid predictions must not inflate participant
+                # counts or poison the frozen probability state.
+                continue
+            weight = parsed_stake
+            staked_amount = parsed_stake
+
         if node not in seen_predictors:
             seen_predictors.add(node)
             predictor_ids.append(node)
-        stake = p.get("stake_amount")
-        if stake is not None:
-            try:
-                a = float(stake)
-            except (TypeError, ValueError):
-                a = 0.0
-            if a > 0:
-                total_stake += a
-                if side == "yes":
-                    yes_weight += a
-                else:
-                    no_weight += a
+
+        total_stake += staked_amount
+        if side == "yes":
+            yes_weight += weight
         else:
-            # Free pick = 1.0 virtual stake (RULES §5.2).
-            if side == "yes":
-                yes_weight += 1.0
-            else:
-                no_weight += 1.0
+            no_weight += weight
 
     pool = yes_weight + no_weight
     if pool > 0:
@@ -114,7 +129,7 @@ def build_snapshot(
         "frozen_total_stake": total_stake,
         "frozen_predictor_ids": predictor_ids,
         "frozen_probability_state": {"yes": yes_p, "no": no_p},
-        "frozen_at": float(frozen_at),
+        "frozen_at": frozen_at_value,
     }
 
 

backend/services/infonet/tests/test_snapshot_invalid_numerics.py

@@ -0,0 +1,64 @@
+"""Regression coverage for malformed/non-finite market snapshot numerics."""
+
+import math
+from typing import Any
+
+import pytest
+
+from services.infonet.markets.snapshot import build_snapshot
+
+
+def _prediction(
+    node: str,
+    side: str,
+    stake: Any,
+    *,
+    timestamp: float,
+    sequence: int,
+) -> dict[str, Any]:
+    payload: dict[str, Any] = {"market_id": "m1", "side": side}
+    if stake is not None:
+        payload["stake_amount"] = stake
+    return {
+        "event_type": "prediction_place",
+        "node_id": node,
+        "timestamp": timestamp,
+        "sequence": sequence,
+        "payload": payload,
+    }
+
+
+@pytest.mark.parametrize("invalid_stake", [float("nan"), float("inf"), "not-a-number", -1.0, 0.0])
+def test_invalid_paid_stake_does_not_poison_or_count_snapshot(invalid_stake: Any) -> None:
+    chain = [
+        _prediction("alice", "yes", None, timestamp=100.0, sequence=1),
+        _prediction("mallory", "no", invalid_stake, timestamp=101.0, sequence=2),
+    ]
+
+    snapshot = build_snapshot("m1", chain, frozen_at=200.0)
+
+    assert snapshot["frozen_participant_count"] == 1
+    assert snapshot["frozen_predictor_ids"] == ["alice"]
+    assert snapshot["frozen_total_stake"] == 0.0
+    assert snapshot["frozen_probability_state"] == {"yes": 1.0, "no": 0.0}
+    assert all(math.isfinite(v) for v in snapshot["frozen_probability_state"].values())
+
+
+def test_finite_numeric_string_stake_is_preserved() -> None:
+    chain = [
+        _prediction("alice", "yes", "2.5", timestamp=100.0, sequence=1),
+        _prediction("bob", "no", "7.5", timestamp=101.0, sequence=2),
+    ]
+
+    snapshot = build_snapshot("m1", chain, frozen_at="200.5")
+
+    assert snapshot["frozen_participant_count"] == 2
+    assert snapshot["frozen_total_stake"] == 10.0
+    assert snapshot["frozen_probability_state"] == {"yes": 0.25, "no": 0.75}
+    assert snapshot["frozen_at"] == 200.5
+
+
+@pytest.mark.parametrize("invalid_frozen_at", [float("nan"), float("inf"), "not-a-time"])
+def test_nonfinite_or_malformed_frozen_at_is_rejected(invalid_frozen_at: Any) -> None:
+    with pytest.raises(ValueError, match="frozen_at must be finite"):
+        build_snapshot("m1", [], frozen_at=invalid_frozen_at)