fix(receipts): apply the same falsy-quote guard in verify_evidence

verify_evidence carries its own copy of the empty-span guard ahead of
the source read (a missing source must never mask "nothing to compare"
as forged). it had the same quote is None gap verify_receipt did:
quote="" paired with a missing source fell through to FORGED instead
of NO_RECEIPT. same fix, one call site up.

per coderabbit review on #513.

Author: davion-knight

CHANGELOG.md

@@ -52,6 +52,17 @@ All notable changes to vouch are documented here. Format follows
   the same resolution now backs `capture answer`, which previously left
   re-captured duplicates pending forever.
 
+### Fixed
+- `receipts.verify_receipt()` and `receipts.verify_evidence()` no longer
+  report `VERIFIED`/fall through past their guard for an empty-quote,
+  zero-length byte span. both guards only checked `quote is None`, not an
+  empty string, so `quote=""` with `byte_start == byte_end` decoded to
+  `""`, trivially equaled the empty quote, and `verify_receipt` returned
+  `VERIFIED` -- despite carrying no actual quoted text, contradicting the
+  function's own docstring. `locate_span` already refused to mint such a
+  receipt on the propose path; this closes the same gap on the verify
+  path, reachable via bundle import or sync.
+
 ## [1.4.0] — 2026-07-17
 
 ### Added
@@ -111,14 +122,6 @@ All notable changes to vouch are documented here. Format follows
   in config.yaml (#476).
 
 ### Fixed
-- `receipts.verify_receipt()` no longer reports `VERIFIED` for an
-  empty-quote, zero-length byte span. the guard only checked `quote is
-  None`, not an empty string, so `quote=""` with `byte_start ==
-  byte_end` decoded to `""`, trivially equaled the empty quote, and
-  verified -- despite carrying no actual quoted text, contradicting the
-  function's own docstring. `locate_span` already refused to mint such
-  a receipt on the propose path; this closes the same gap on the
-  verify path, reachable via bundle import or sync.
 - approve/reject/expire record the audit event *before* moving the
   proposal to decided/. a crash between the two used to leave a durable
   decision with no authoritative history; it now leaves a pending

src/vouch/receipts.py

@@ -184,7 +184,7 @@ def verify_evidence(store: KBStore, evidence: Evidence) -> ReceiptResult:
     """
     from .storage import ArtifactNotFoundError
 
-    if evidence.byte_start is None or evidence.byte_end is None or evidence.quote is None:
+    if evidence.byte_start is None or evidence.byte_end is None or not evidence.quote:
         return ReceiptResult(ReceiptStatus.NO_RECEIPT, "no byte-offset span")
     try:
         source_bytes = store.read_source_content(evidence.source_id)

tests/test_receipts.py

@@ -173,6 +173,23 @@ def test_verify_evidence_not_verified_when_source_missing(store: KBStore) -> Non
     assert result.verified is False
 
 
+def test_verify_evidence_no_receipt_for_empty_quote_and_missing_source(
+    store: KBStore,
+) -> None:
+    # verify_evidence has its own pre-check ahead of the source-read (so a
+    # missing source doesn't mask a plain "nothing to compare" case) --
+    # it must apply the same falsy-quote rule verify_receipt does, or an
+    # empty quote paired with a missing source reports FORGED instead of
+    # NO_RECEIPT, same bug as verify_receipt's, one call site up.
+    ev = Evidence(
+        id="e4", source_id="does-not-exist",
+        locator="x", quote="", byte_start=0, byte_end=0,
+    )
+    result = verify_evidence(store, ev)
+    assert result.status is ReceiptStatus.NO_RECEIPT
+    assert result.verified is False
+
+
 # ---- the quote step: locate a span, or drop what cannot be quoted ----