feat(ebuy): extracted text on eBuy request attachments (API 5.9.1)

CHANGELOG.md

@@ -7,6 +7,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+
+- **Extracted text on eBuy request attachments** (Tango API 5.9.1). `attachments(extracted_text)` on eBuy requests returns the text extracted from an attachment's document. It is returned only when named: the default `get_ebuy_request()` shape and `attachments(*)` leave it out. An attachment with no text (an external link, an empty scan, or a document not yet extracted) is sent by the API without the key rather than with a null; the SDK's shaped result reads that as `None`.
+
+### Changed
+
+- Re-vendored `contracts/filter_shape_contract.json` (Tango API 5.9.1).
+
 ## [1.15.0] - 2026-10-06
 
 ### Added

contracts/filter_shape_contract.json

@@ -1,6 +1,6 @@
 {
   "meta": {
-    "api_version": "5.9.0",
+    "api_version": "5.9.1",
     "description": "Canonical API filter/shape contract. Downstream consumers (SDK, MCP) should validate their conformance against this manifest.",
     "generated_from": "scripts/filter_shape_conformance.py",
     "schema_version": 2
@@ -4339,6 +4339,7 @@
                 "doc_seq_num",
                 "doc_session_date",
                 "doc_type",
+                "extracted_text",
                 "is_link"
               ]
             },
@@ -4414,6 +4415,7 @@
           "attachments.doc_seq_num",
           "attachments.doc_session_date",
           "attachments.doc_type",
+          "attachments.extracted_text",
           "attachments.is_link",
           "award_method",
           "buyer_agency",

docs/API_REFERENCE.md

@@ -1758,6 +1758,24 @@ for attachment in request["attachments"]:
 
 The default shape returns every field plus two expands: `organization` (the buyer office, with the same seven keys as other resources' `organization` expand) and `attachments`. Each attachment carries `doc_seq_num`, `doc_name`, `doc_type`, `doc_path`, `is_link` and `doc_session_date`. `is_link=True` means `doc_path` is an outbound URL with no stored document behind it. `amendments`, `line_items` and `addresses` are lists of objects, served as eBuy publishes them.
 
+**Reading a document's text — `attachments(extracted_text)`** (Tango API 5.9.1):
+
+```python
+request = client.get_ebuy_request(
+    "RFQ1835158",
+    shape="rfq_id,attachments(doc_seq_num,doc_name,extracted_text)",
+)
+
+for attachment in request["attachments"]:
+    text = attachment.get("extracted_text")
+    if text:
+        print(attachment["doc_name"], len(text))
+```
+
+- `extracted_text` is the text extracted from the attachment's document.
+- **It is returned only when you name it.** The default shape does not include it, and neither does `attachments(*)`.
+- **The API omits the key, rather than sending null**, for an attachment with no text: an external link, an empty scan, or a document that has not been extracted yet. The SDK's shaped result reads an omitted key as `None`, so treat `None` as "no text", not as an error.
+
 ### get_ebuy_attachment_url()
 
 Get a short-lived download URL for one stored attachment.
@@ -2606,7 +2624,7 @@ entity = client.get_entity("UEI_KEY", shape=ShapeConfig.ENTITIES_COMPREHENSIVE)
 | `FEDERAL_REGISTER_MINIMAL` | `list_federal_register_documents` | uuid, document_number, publication_date, type, subtype, title, abstract, action, agencies, cfr_references, citation, significant, comments_close_on, effective_on, html_url, pdf_url |
 | `FEDERAL_REGISTER_COMPREHENSIVE` | `get_federal_register_document` | The list fields plus dates, signing_date, start_page, end_page, volume, docket_ids, dockets, regulation_id_numbers, topics, correction_of, corrections, executive_order_number, presidential_document_number, proclamation_number, comment_url, regulations_dot_gov_url, raw_text_url, body_html_url (omits `full_text`) |
 | `EBUY_REQUESTS_MINIMAL` | `list_ebuy_requests` | rfq_id, request_type, title, schedule, sin, status, buyer_name, buyer_agency, buyer_agency_code, reference_number, issue_date, close_date, attachment_count, link_count, last_seen |
-| `EBUY_REQUESTS_COMPREHENSIVE` | `get_ebuy_request` | Every field, plus the `organization` and `attachments` expands |
+| `EBUY_REQUESTS_COMPREHENSIVE` | `get_ebuy_request` | Every field, plus the `organization` and `attachments` expands (without `extracted_text`, which must be named) |
 | `BUDGET_ACCOUNTS_MINIMAL` | `list_budget_accounts`, `get_budget_account` | id, federal_account_symbol, fiscal_year, data_through_period, agency_code/name, bureau_name, account_title, bea_category, on_off_budget, subfunction_code, account_category, lifecycle (requested/enacted/apportioned/obligated/outlayed/unobligated), contract & assistance rollups, key ratios, next-year growth, source_anomalies |
 | `VEHICLE_ORDERS_MINIMAL` | `list_vehicle_orders` | key, piid, award_date, recipient(display_name,uei), total_contract_value, obligated |
 | `ITDASHBOARD_INVESTMENTS_MINIMAL` | `list_itdashboard_investments` | Minimal IT Dashboard investment fields |

tango/models.py

@@ -1365,6 +1365,10 @@ class EbuyAttachment:
     """Schema definition for one entry in an eBuy request's ``attachments`` (not used for instances).
 
     ``is_link=True`` means the entry is an outbound URL in ``doc_path`` with no stored document behind it, so ``get_ebuy_attachment_url()`` raises ``TangoAttachmentLinkError`` for it.
+
+    ``extracted_text`` (Tango API 5.9.1) is the text extracted from the attachment's document.
+    It must be named, e.g. ``attachments(doc_seq_num,doc_name,extracted_text)``, because neither the default response nor ``attachments(*)`` includes it.
+    An attachment with no text (a link, an empty scan, or a document not yet extracted) omits the key from the API response rather than returning null; the SDK's shaped result reads that as ``None``.
     """
 
     doc_seq_num: int | None = None
@@ -1373,6 +1377,7 @@ class EbuyAttachment:
     doc_path: str | None = None
     is_link: bool | None = None
     doc_session_date: datetime | None = None
+    extracted_text: str | None = None
 
 
 class EbuyRequest:
@@ -1660,7 +1665,7 @@ class ShapeConfig:
         "link_count,last_seen"
     )
 
-    # Default for get_ebuy_request(). Mirrors the API's own retrieve default; the attachment fields are named rather than `attachments(*)` so their dates parse.
+    # Default for get_ebuy_request(). Mirrors the API's own retrieve default; the attachment fields are named rather than `attachments(*)` so their dates parse. `extracted_text` is left out because document bodies are large and the API serves them only when named.
     EBUY_REQUESTS_COMPREHENSIVE: Final = (
         "rfq_id,request_type,title,description,schedule,sin,status,buyer_name,"
         "buyer_agency,buyer_agency_code,buyer_email,buyer_user_id,reference_number,"

tango/shapes/explicit_schemas.py

@@ -886,6 +886,8 @@
     "doc_session_date": FieldSchema(
         name="doc_session_date", type=datetime, is_optional=True, is_list=False
     ),
+    # Only when named in the shape; the API omits the key for an attachment with no text.
+    "extracted_text": FieldSchema(name="extracted_text", type=str, is_optional=True, is_list=False),
 }
 
 

tests/test_ebuy.py

@@ -212,6 +212,26 @@ def test_detail_fields_and_expands_parse(self, mock_request):
         assert row["attachments"][1]["is_link"] is True
         assert isinstance(row["attachments"][0]["doc_session_date"], datetime)
 
+    @patch("tango.client.httpx.Client.request")
+    def test_extracted_text_is_read_when_named_and_none_without_text(self, mock_request):
+        shape = "rfq_id,attachments(doc_seq_num,is_link,extracted_text)"
+        _mock(
+            mock_request,
+            {
+                "rfq_id": RFQ_ID,
+                "attachments": [
+                    {"doc_seq_num": 1, "is_link": False, "extracted_text": "Statement of work"},
+                    {"doc_seq_num": 2, "is_link": True},
+                ],
+            },
+        )
+        row = TangoClient(api_key="k").get_ebuy_request(RFQ_ID, shape=shape)
+
+        assert _call_params(mock_request)["shape"] == shape
+        assert row["attachments"][0]["extracted_text"] == "Statement of work"
+        assert row["attachments"][1].get("extracted_text") is None
+        assert row["attachments"][1]["extracted_text"] is None
+
     @patch("tango.client.httpx.Client.request")
     def test_out_of_scope_request_is_not_found(self, mock_request):
         mock_request.return_value = _response(404, {"detail": "No EbuyRequest matches."})
@@ -310,12 +330,22 @@ class TestEbuyShapes:
             ShapeConfig.EBUY_REQUESTS_COMPREHENSIVE,
             "rfq_id,organization(agency_code,office_name),attachments(doc_seq_num,is_link)",
             "rfq_id,amendments,line_items,addresses,oco_aac,ocs_phone",
+            "rfq_id,attachments(extracted_text)",
+            "rfq_id,attachments(doc_seq_num,doc_name,extracted_text)",
         ],
     )
     def test_shape_validates(self, shape):
         parser = ShapeParser(cache_enabled=True)
         parser.validate(parser.parse(shape), EbuyRequest)
 
+    @pytest.mark.parametrize(
+        "shape",
+        [ShapeConfig.EBUY_REQUESTS_MINIMAL, ShapeConfig.EBUY_REQUESTS_COMPREHENSIVE],
+    )
+    def test_default_shapes_leave_extracted_text_opt_in(self, shape):
+        assert "extracted_text" not in shape
+        assert "attachments(*)" not in shape
+
     def test_contract_number_is_a_filter_not_a_response_field(self):
         parser = ShapeParser(cache_enabled=True)
         with pytest.raises(ShapeValidationError):