feat(ttp): implement evidence-shape validation and confidence range constraint

- TolerantTagger.tag validates evidence keys against EVIDENCE_SCHEMA TypedDicts;
  TypeError (programmer error) propagates instead of being swallowed
- IntelEvidence and EmailEvidence expanded from stubs to full per-provider
  key sets (total=False); IntelEvidence old stub fields replaced wholesale
- EVIDENCE_SCHEMA map added to models/ttp.py and imported by base.py
- TTPTag __table_args__ gains confidence [0,1] CheckConstraint (DB-enforced)
- xfail removed from test_confidence_outside_range_rejected_at_insert and
  test_evidence_shape_violation_propagates_as_typeerror — both now pass
- TypeError removed from _SWALLOWED_EXCS fuzz list; test_intel_evidence_keys
  updated to assert the real provider key set

decnet/ttp/base.py

@@ -19,7 +19,7 @@ import logging
 from abc import ABC, abstractmethod
 from typing import Any, Final, NamedTuple, Protocol, runtime_checkable
 
-from decnet.web.db.models.ttp import TTPTag
+from decnet.web.db.models.ttp import EVIDENCE_SCHEMA, TTPTag
 
 _log = logging.getLogger(__name__)
 
@@ -109,7 +109,32 @@ class TolerantTagger(Tagger):
 
     async def tag(self, event: TaggerEvent) -> list[TTPTag]:
         try:
-            return await self._tag_impl(event)
+            results = await self._tag_impl(event)
+            # Validate evidence shape: unknown keys are a programmer error,
+            # not a runtime absence. Raise TypeError so the caller sees the
+            # bug rather than silently dropping the tag.
+            td = EVIDENCE_SCHEMA.get(event.source_kind)
+            if td is not None:
+                declared = (
+                    getattr(td, "__required_keys__", frozenset())
+                    | getattr(td, "__optional_keys__", frozenset())
+                )
+                for tag in results:
+                    ev = getattr(tag, "evidence", None)
+                    if ev is None:
+                        continue
+                    unknown = set(ev) - declared
+                    if unknown:
+                        raise TypeError(
+                            f"lifter {self.name!r} emitted evidence keys "
+                            f"{unknown!r} not declared in "
+                            f"{td.__name__} for source_kind={event.source_kind!r}"
+                        )
+            return results
+        except TypeError:
+            # Programmer error — bad evidence shape or type mismatch.
+            # Propagate; do NOT swallow.
+            raise
         except Exception:
             # ``Exception`` deliberately, not ``BaseException``:
             # ``KeyboardInterrupt`` / ``SystemExit`` /

decnet/web/db/models/ttp.py

@@ -69,19 +69,46 @@ class CommandEvidence(TypedDict):
     rule_pattern: str  # regex source string, never user input
 
 
-class IntelEvidence(TypedDict):
-    intel_uuid: str
-    provider: Literal["abuseipdb", "greynoise", "feodo", "threatfox"]
-    category: Optional[int]
-    score: float  # already normalized to [0.0, 1.0]
+class IntelEvidence(TypedDict, total=False):
+    # AbuseIPDB
+    abuseipdb_categories: list[int]
+    abuseipdb_score: float
+    abuse_confidence_score: int
+    # GreyNoise
+    greynoise_classification: str
+    greynoise_tags: list[str]
+    greynoise_name: str
+    # Feodo
+    feodo_listed: bool
+    feodo_malware_family: str
+    first_seen_feodo: str
+    malware_family: str
+    # ThreatFox
+    threatfox_threat_types: list[str]
+    threatfox_ioc_types: list[str]
+    threatfox_malware_families: list[str]
+    threat_types: list[str]
+    malware_families: list[str]
+    ioc_types: list[str]
+    # Aggregate meta-rule
+    aggregate_verdict: str
+    bumped_rule_ids: list[str]
 
 
-class EmailEvidence(TypedDict):
-    body_sha256: str  # hash, never raw body
+class EmailEvidence(TypedDict, total=False):
+    body_sha256: str          # hash, never raw body
     matched_headers: list[str]  # header NAMES, not values
     rcpt_domain_set: list[str]  # domains, not addresses
     attachment_sha256s: list[str]
     rcpt_count: int
+    # PII-safe match discriminators (subset of _EMAIL_EVIDENCE_ALLOWED_KEYS)
+    matched_kit: str
+    matched_trigger: str
+    matched_url_host: str
+    matched_signals: list[str]
+    matched_subject_kw: list[str]
+    matched_body_kw: list[str]
+    encoded_byte_count: int
 
 
 class CanaryFingerprintEvidence(TypedDict):
@@ -98,6 +125,18 @@ class HttpFingerprintEvidence(TypedDict):
     raw: Optional[dict]  # raw settings dict for h2_settings / h3_settings
 
 
+# Maps source_kind → its evidence TypedDict. Used by TolerantTagger to
+# validate that lifters do not emit undeclared keys (programmer error →
+# TypeError, not the swallowed absence-of-data case).
+EVIDENCE_SCHEMA: dict[str, type] = {
+    "command": CommandEvidence,
+    "intel": IntelEvidence,
+    "email": EmailEvidence,
+    "canary_fingerprint": CanaryFingerprintEvidence,
+    "http_fingerprint": HttpFingerprintEvidence,
+}
+
+
 # ── Tables ──────────────────────────────────────────────────────────
 
 
@@ -176,6 +215,10 @@ class TTPTag(SQLModel, table=True):
             "attacker_uuid IS NOT NULL OR identity_uuid IS NOT NULL",
             name="ttp_tag_has_anchor",
         ),
+        CheckConstraint(
+            "confidence >= 0.0 AND confidence <= 1.0",
+            name="ttp_tag_confidence_range",
+        ),
         Index(
             "ix_ttp_tag_identity_technique",
             "identity_uuid",

tests/ttp/test_base.py

@@ -140,7 +140,8 @@ _SWALLOWED_EXCS: tuple[type[Exception], ...] = (
     ValueError,
     RuntimeError,
     KeyError,
-    TypeError,
+    # TypeError is intentionally NOT swallowed — it propagates as a
+    # programmer-error signal (bad evidence shape). See TolerantTagger.tag.
     AttributeError,
     LookupError,
     OSError,

tests/ttp/test_evidence_shape.py

@@ -67,7 +67,19 @@ def test_intel_evidence_keys() -> None:
     keys = (
         IntelEvidence.__required_keys__ | IntelEvidence.__optional_keys__
     )
-    assert keys == {"intel_uuid", "provider", "category", "score"}
+    assert keys == {
+        # AbuseIPDB
+        "abuseipdb_categories", "abuseipdb_score", "abuse_confidence_score",
+        # GreyNoise
+        "greynoise_classification", "greynoise_tags", "greynoise_name",
+        # Feodo
+        "feodo_listed", "feodo_malware_family", "first_seen_feodo", "malware_family",
+        # ThreatFox
+        "threatfox_threat_types", "threatfox_ioc_types", "threatfox_malware_families",
+        "threat_types", "malware_families", "ioc_types",
+        # Aggregate meta-rule
+        "aggregate_verdict", "bumped_rule_ids",
+    }
 
 
 def test_canary_fingerprint_evidence_keys() -> None:
@@ -140,10 +152,6 @@ def test_lifter_emits_evidence_matching_typeddict(
 # ── Negative case: shape violation propagates (impl phase) ──────────
 
 
-@pytest.mark.xfail(
-    strict=True,
-    reason="impl phase: TolerantTagger currently swallows TypeError",
-)
 def test_evidence_shape_violation_propagates_as_typeerror() -> None:
     """A lifter that emits an evidence dict with a key not in its
     ``TypedDict`` is a programmer error — it MUST propagate past the

tests/ttp/test_schema.py

@@ -169,7 +169,6 @@ def test_guard_runs_before_super_init() -> None:
 # ── confidence range guard (impl phase) ─────────────────────────────
 
 
-@pytest.mark.xfail(strict=True, reason="impl phase: confidence range guard not yet enforced")
 async def test_confidence_outside_range_rejected_at_insert(session: AsyncSession) -> None:
     """``confidence`` outside [0.0, 1.0] must be rejected. The contract
     schema currently types it as bare ``float`` without a range