PetrefiedThunder · Oct 31, 2026 · Oct 31, 2026 · Oct 31, 2026
@@ -31,7 +31,7 @@ A browser verifier and a downloadable Python bundle verifier let reviewers inspe
 | Readiness assessments and recall reporting | Release-enabled: [readiness](../services/ingestion/app/readiness_router.py), [recall reports](../services/ingestion/app/recall_report.py). |
 | Bundle verifier and recall-drill endpoint | [Browser verifier](https://regengine.co/verify), [verifier source](../frontend/public/sdk/verify_chain.py). The recall-drill endpoint currently returns no drill data. |
 | Supplier manifest remediation and correction links | Implemented in [remediation](../services/ingestion/app/remediation/) and the [buyer dashboard](../frontend/src/app/remediation/page.tsx); excluded from the production router allowlist. Notification delivery also requires recipient/provider configuration. |
-| EPCIS JSON-LD export and broader recall simulations | Implemented in [EPCIS export](../services/ingestion/app/epcis_export.py) and [simulations](../services/ingestion/app/recall_simulations.py); these routers are experimental and excluded from the production allowlist. A single EPCIS export returns at most 2,000 events, newest first, and its `metadata.truncated` flag says when it was cut off. Public sample EPCIS downloads are a separate proof surface. |
+| EPCIS JSON-LD export and broader recall simulations | Implemented in [EPCIS export](../services/ingestion/app/epcis_export.py) and [simulations](../services/ingestion/app/recall_simulations.py); these routers are experimental and excluded from the production allowlist. A single EPCIS export returns at most 2,000 events, newest first, and the `regengine:truncated` flag inside its `regengine:metadata` block says when it was cut off. Public sample EPCIS downloads are a separate proof surface. |
 | Inbound email CSV ingestion | [Postmark integration](../services/ingestion/app/email_ingest/); excluded from the production allowlist, with no self-service provisioning UI. See [rollout notes](engineering/EMAIL_INGESTION.md). |
 | Passwordless login and administration | [Email OTP/passkey auth](../services/admin/app/auth/challenges_router.py), tenant/user/API-key management, and session controls. Email delivery requires provider and identity provisioning. |
 | Subscription access and usage metering | [Subscription gate](../services/ingestion/app/subscription_gate.py), [metering design and limitations](../DESIGN_metering-billing.md). Usage reporting to Stripe requires an operator call; there is no periodic reporting job. |

@@ -165,14 +165,19 @@ The ZIP bundle (`_build_fda_package()` in `services/ingestion/app/fda_export_ser
 |------|----------|
 | `fda_spreadsheet_<scope>_<timestamp>.csv` | FDA-oriented spreadsheet, one row per event (`FDA_COLUMNS`) |
 | `chain_verification_<scope>_<timestamp>.json` | Hash chain verification result, row hash coverage, completeness summary, redaction flag |
-| `evidence_context_<scope>_<timestamp>.json` | Rule catalog and the rule versions that evaluated these events, enforcement mode at export time, and a known-gaps section (KDE coverage, unchecked event types, missing hashes, whether the coverage gate was bypassed with `allow_incomplete`) |
-| `VALIDATION_ERRORS_<scope>_<timestamp>.log` | Only when events are missing required KDEs: one line per affected event |
+| `evidence_context_<scope>_<timestamp>.json` | Rule catalog and the rule versions that evaluated these events, enforcement mode at export time, and a known-gaps section (KDE coverage, unchecked event types, missing hashes, whether the coverage gate was bypassed with `allow_incomplete`). Before treating the rule-version evidence as complete, check `rule_versions.catalog_status` and `rule_versions.evaluated_status` (also mirrored as `known_gaps.rules_catalog_status` / `known_gaps.rules_evaluated_status`): `ok` means the read succeeded, `unavailable` means the export could not read that table and the section is empty for that reason, `not_requested` means no evaluations were looked up. |
+| `VALIDATION_ERRORS_<scope>_<timestamp>.log` | Only when events are missing required KDEs: one line per affected event, built from the completeness summary's `missing_required_events` list, which is capped at 250 entries. When more events are affected than the log lists, `known_gaps.missing_required_events_truncated` in the evidence context is `true` and `known_gaps.missing_required_events_listed` gives the count written; `events_with_missing_required_fields` is the full count |
 | `manifest.json` | Package metadata, per-file sizes and SHA-256 digests, Merkle root over the hashed members, pointer to the evidence context file. The manifest is not one of its own Merkle leaves. |
 | `README.txt` | Human-readable description of the package and how to verify it |
 
 The PDF format is a separate response (`format=pdf`), not a member of the ZIP. Every member except
-`manifest.json` is hashed and listed in `manifest.files`, so a change to any of them, the known-gaps
-section included, fails both the browser verifier and `services/admin/scripts/verify_audit_pack.py`.
+`manifest.json` is hashed and listed in `manifest.files`. The browser verifier and
+`services/admin/scripts/verify_audit_pack.py` compare each member against the per-file hashes and the
+Merkle root in `manifest.json`, so a change to any member, the known-gaps section included, without a
+matching manifest update fails both. These checks establish that the package is internally consistent;
+they do not authenticate the issuer, so a package rebuilt with a regenerated manifest passes them. To
+tie a ZIP to the original export, compare it against the `X-Package-Hash` value captured when it was
+downloaded (section 3.4).
 
 ### 3.3 FDA CSV Column Specification
 

@@ -105,7 +105,8 @@
     "rollup": "4.59.0",
     "esbuild": "0.27.3",
     "fast-uri": "^3.1.5",
-    "minimatch": ">=3.1.3",
+    "follow-redirects": ">=1.16.2",
+    "minimatch": ">=11.0.1",
     "postcss": "^8.5.18",
     "serialize-javascript": "7.0.3",
     "workbox-build": {

@@ -1235,9 +1235,9 @@ pygments==2.21.0 \
     --hash=sha256:2363c69b61c4a97c838da3b130dcd6468f4848992b21a82f2a63ec34377137d9 \
     --hash=sha256:610ca751c9bc2492b38eb9a38a7fbc93edbbb2d7182edaf34e66ae493dee5c8c
     # via pytest
-pypdf==6.21.2 \
-    --hash=sha256:0a3e1c5c886a1be03646c16a04827cc166c2a0575c971b26a4bf89686551c2d4 \
-    --hash=sha256:1d44d1a3a200f25062adff0085b02040e4646431ac2ccac22e42679e91466084
+pypdf==6.21.4 \
+    --hash=sha256:33e4fb41ca352810eb4e9838aa0d82d6888a5f8bf682f43e99cce5faa666255c \
+    --hash=sha256:c8817e3f888d47a087872ded69812c1ef30cacb871d6e053e0c00bbaf0bc2163
     # via -r requirements.txt
 pytest==9.1.1 \
     --hash=sha256:1088fbde8f2b49d95a549a195707afa7a76a3ce9bcadc26b6d71f0ffda5fe313 \

@@ -8,6 +8,6 @@ neo4j>=5.23,<6
 streamlit>=1.38,<2
 pytest>=9.0.3,<10
 beautifulsoup4>=4.12,<5
-pypdf>=6.21.2,<7
+pypdf>=6.21.4,<7
 openai>=1.40,<3
 python-dotenv>=1.0,<2
@@ -163,7 +163,7 @@ beautifulsoup4>=4.15.0
 lxml>=6.1.3
 pdfminer.six>=20260107
 pdfplumber>=0.11.10
-pypdf>=6.21.2,<7  # CVE-2026-115572/115580/115585; CVE-2026-71852, CVE-2026-71870; earlier: CVE-2026-59935/59936 (inline-image DoS), CVE-2026-48155/48156, GHSA-jm82-fx9c-mx94
+pypdf>=6.21.4,<7  # CVE-2026-119203/119204/119205 (6.21.4); CVE-2026-119206/119207/119208/119209 (6.21.3); earlier: CVE-2026-115572/115580/115585, CVE-2026-71852/71870, CVE-2026-59935/59936, CVE-2026-48155/48156, GHSA-jm82-fx9c-mx94
 fpdf2>=2.8.9,<3
 python-docx>=1.2.0
 docx2txt>=0.9

@@ -2999,9 +2999,9 @@ psycopg-binary==3.3.6 \
     --hash=sha256:fa34eb47969297471db7b7f193622c7e3ee839ec05abd05f1fe104d5b1b1dcf4 \
     --hash=sha256:fdccb3a0e184b03e9baa673b15a809cf36c339c85dbda0ebc25a698846dfbee8
     # via psycopg
-pyasn1==0.6.4 \
-    --hash=sha256:9c447d8431c947fe4c8febc4ed9e760bc29011a5b01e5c74b67025bd9fb8ce81 \
-    --hash=sha256:deda9277cfd454080ec40b207fb6df82206a3a2688735233cdcd8d3d565f088b
+pyasn1==0.6.5 \
+    --hash=sha256:4017b0fce125096b8573b39c989fdb1a4b19cf761e683398184550fea0e66587 \
+    --hash=sha256:6b11616a6bcf3a40e635f082bb232fa57b5e0dd7f517bd7ddd69ec3794508a47
     # via
     #   pyasn1-modules
     #   webauthn
@@ -3202,9 +3202,9 @@ pyotp==2.10.0 \
     --hash=sha256:1df2f6a1bcc3bb0716172a5215ddc2f8c7c7fd26a13df9927d52e1746934836c \
     --hash=sha256:d01e9703443616b03c57c700b5cbffd56a1f929c1b0f8f03131bc78c1fca9d3f
     # via -r requirements.in
-pypdf==6.21.2 \
-    --hash=sha256:0a3e1c5c886a1be03646c16a04827cc166c2a0575c971b26a4bf89686551c2d4 \
-    --hash=sha256:1d44d1a3a200f25062adff0085b02040e4646431ac2ccac22e42679e91466084
+pypdf==6.21.4 \
+    --hash=sha256:33e4fb41ca352810eb4e9838aa0d82d6888a5f8bf682f43e99cce5faa666255c \
+    --hash=sha256:c8817e3f888d47a087872ded69812c1ef30cacb871d6e053e0c00bbaf0bc2163
     # via -r requirements.in
 pypdfium2==5.13.0 \
     --hash=sha256:07f58e91b8c45ca144a1ff3008faf3c73ef8a5e9fb32988831788363288228cd \

@@ -521,12 +521,22 @@ def test_pack_declares_evidence_context_member(tmp_path):
     assert verifier.verify_pack(str(pack_path)) is True
 
 
-def test_tampered_evidence_context_fails(tmp_path):
+def test_tampered_evidence_context_fails(tmp_path, capsys):
+    # Flip one character inside a string value so the member keeps its
+    # serialized size: the rejection must come from the hash comparison,
+    # not from the cheaper size check that runs first.
     pack_path, _ = _make_pack(tmp_path)
     members = _read_members(pack_path)
     name = _evidence_name(members)
     context = json.loads(members[name])
-    context["known_gaps"]["coverage_gate_bypassed"] = not context["known_gaps"]["coverage_gate_bypassed"]
-    members[name] = json.dumps(context, indent=2, sort_keys=True).encode("utf-8")
+    status = context["known_gaps"]["verification_status"]
+    assert isinstance(status, str) and status
+    context["known_gaps"]["verification_status"] = status[:-1] + ("X" if status[-1] != "X" else "Y")
+    mutated = json.dumps(context, indent=2, sort_keys=True).encode("utf-8")
+    assert len(mutated) == len(members[name]), "the mutation must keep the serialized size"
+    members[name] = mutated
     tampered = _write_members(tmp_path, "tampered-evidence.zip", members)
     assert verifier.verify_pack(str(tampered)) is False
+    out = capsys.readouterr().out
+    assert f"hash mismatch for {name}" in out
+    assert "size mismatch" not in out
@@ -159,6 +159,20 @@ def _parse_stored_event_time(value: object) -> datetime | None:
     return parsed
 
 
+_EPOCH_UTC = datetime(1970, 1, 1, tzinfo=timezone.utc)
+
+
+def _fallback_export_sort_key(entry: dict) -> tuple[bool, datetime]:
+    """Sort key for fallback-store records: event time, unparseable last.
+
+    Mirrors the DB export's ``ORDER BY event_timestamp DESC`` when used
+    with ``reverse=True``: dated records come first, newest to oldest,
+    then the records whose event time is missing or unparseable.
+    """
+    event_dt = _parse_stored_event_time((entry.get("normalized_cte") or {}).get("event_time"))
+    return (event_dt is not None, event_dt or _EPOCH_UTC)
+
+
 @router.post("/events", status_code=201, summary="Ingest EPCIS 2.0 event")
 async def ingest_epcis_event(
     event: dict,
@@ -367,9 +381,19 @@ async def export_epcis_events(
             if product_id and normalized.get("product_id") != product_id:
                 continue
             filtered_records.append(entry)
-        events = [entry["epcis_document"] for entry in filtered_records]
-        limit = None
-        truncated = False
+        if filtered_records:
+            # The DB path orders by event_timestamp DESC; the fallback
+            # store iterates in insertion order, so sort it the same way
+            # to keep ``regengine:ordering`` truthful. Records without a
+            # parseable event time go last.
+            filtered_records.sort(key=_fallback_export_sort_key, reverse=True)
+            events = [entry["epcis_document"] for entry in filtered_records]
+            # Only the fallback's own events are uncapped. When it supplies
+            # nothing, the DB page's limit/truncated still describe what
+            # the caller got: a truncated page that the product filter
+            # emptied must not read as a complete export.
+            limit = None
+            truncated = False
 
     filters = {
         f"regengine:{key}": value

@@ -339,6 +339,51 @@ def test_fallback_path_reports_no_cap(self, client: TestClient, monkeypatch: pyt
         assert meta["regengine:limit"] is None
         assert meta["regengine:truncated"] is False
 
+    def test_truncated_db_page_emptied_by_filter_stays_truncated(
+        self, client: TestClient, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """A truncated DB page that the product filter emptied is still an incomplete export.
+
+        The fallback branch runs because ``events`` is empty, but when the
+        fallback store has nothing to add it must not reset the DB page's
+        limit/truncated facts to "uncapped, complete".
+        """
+        monkeypatch.setattr(
+            router_mod,
+            "_list_events_from_db_detailed",
+            lambda *a, **kw: _page([], truncated=True),
+        )
+        monkeypatch.setattr(router_mod, "_allow_in_memory_fallback", lambda: True)
+        monkeypatch.setattr(router_mod, "_fallback_store_for", lambda t: {})
+
+        resp = client.get("/api/v1/epcis/export?product_id=prod-rare")
+        assert resp.status_code == 200
+        meta = resp.json()["regengine:metadata"]
+        assert meta["regengine:event_count"] == 0
+        assert meta["regengine:truncated"] is True
+        assert meta["regengine:limit"] == persistence_mod.EXPORT_EVENT_LIMIT
+
+    def test_fallback_events_are_newest_first(self, client: TestClient, monkeypatch: pytest.MonkeyPatch) -> None:
+        """The fallback honours the advertised ``event_timestamp desc`` ordering.
+
+        Records are inserted oldest-first (store iteration order); the
+        response must still be newest-first, with undated records last.
+        """
+        monkeypatch.setattr(router_mod, "_list_events_from_db_detailed", _page)
+        monkeypatch.setattr(router_mod, "_allow_in_memory_fallback", lambda: True)
+
+        early, _, late = _ordered_export_times()
+        fake_store = {
+            "undated": {"normalized_cte": {}, "epcis_document": {"eventID": "undated"}},
+            "older": {"normalized_cte": {"event_time": early}, "epcis_document": {"eventID": "older"}},
+            "newer": {"normalized_cte": {"event_time": late}, "epcis_document": {"eventID": "newer"}},
+        }
+        monkeypatch.setattr(router_mod, "_fallback_store_for", lambda t: fake_store)
+
+        body = client.get("/api/v1/epcis/export").json()
+        assert [e["eventID"] for e in body["epcisBody"]["eventList"]] == ["newer", "older", "undated"]
+        assert body["regengine:metadata"]["regengine:ordering"] == "event_timestamp desc"
+
     def test_db_error_without_fallback_returns_503(self, client: TestClient, monkeypatch: pytest.MonkeyPatch) -> None:
         """Line 225: DB error + fallback disabled → 503."""
 

@@ -417,11 +417,19 @@ def test_non_package_format_still_goes_csv_path(self, monkeypatch):
 class TestPackageExport:
     def test_package_happy_path(self, monkeypatch):
         events = [{"id": "e1"}, {"id": "e2"}, {"id": "e3"}]
-        _, _, cap = _install_stubs(
+        sess, _, cap = _install_stubs(
             monkeypatch,
             rows=[{"id": 1}],
             events=events,
         )
+        snapshot = {"catalog_status": "ok", "catalog": {"rule-a": {"version": 2}}, "evaluated": [], "evaluated_status": "ok"}
+        seen: dict = {}
+
+        def _snapshot(db_session, tenant_id, evts):
+            seen["args"] = (db_session, tenant_id, evts)
+            return snapshot
+
+        monkeypatch.setattr(r, "_rule_versions_for_events", _snapshot)
         resp = _call(format="package", include_pii=False)
         assert resp == "PKG-RESP"
         pkg_k = cap["package_response_kwargs"]
@@ -433,8 +441,12 @@ def test_package_happy_path(self, monkeypatch):
         assert pkg_k["extra_headers"]["X-KDE-Coverage"] == "0.95"
         assert pkg_k["include_pii"] is False
         # S7: gate outcome and rule-version snapshot reach the package builder.
+        # The snapshot is taken on the export's own session, for this
+        # tenant, over exactly the exported events, and is forwarded as-is.
         assert pkg_k["coverage_gate_bypassed"] is False
-        assert "catalog_status" in pkg_k["rule_versions"]
+        assert seen["args"] == (sess, "t-1", events)
+        assert seen["args"][0] is sess
+        assert pkg_k["rule_versions"] is snapshot
         # Did NOT go through the CSV response path
         assert cap["csv_response_kwargs"] is None