v1.4.0

[makegov-mark]

Release cut of everything merged since 1.3.0 — #52, plus the two API-side fixes it depends on (makegov/tango#2944 and #2948, both merged and deployed).

Why minor, not patch

list_protests() gained naics_code and list_psc() gained has_awards. Both are additive public-surface additions, so SemVer says minor. Nothing was renamed, removed, or narrowed.

What ships

13 fields the SDK used to reject client-side now validate. ShapeParser was rejecting Entity.display_name, entity_type, entity_structure, organization_structure, profit_structure, purpose_of_registration, country_of_incorporation, past_performance, registered, sba_business_types, and Protest.decisions, resolved_agency, resolved_protester — against an API that returns 200 for every one. list_entities(shape="display_name") raised ShapeValidationError and never issued the request. Two of those (registered, sba_business_types) had regressed in 1.3.0.

The reverse shape-coverage gate no longer goes blind. It skipped any resource with a falsy contract shape tree, which was exactly the seven resources it most needed to check. Re-vendoring the corrected contract surfaced 69 real gaps (36 fields, 20 expands, 13 flattened), all closed by a regenerated overlay. A resource declaring shape_supported: true with no tree is now a hard finding attributed upstream rather than a silent skip.

Seven resources were never conformance-checked. RESOURCE_TO_METHOD omitted protests, psc, mas_sins, departments, business_types, and assistance_listings, and mapped offices to None despite list_offices existing. Wiring them up found the two new params above.

CONTRACT_OMITTED_PARAMS is empty, which is the goal state. Both entries it ever held were retired by the self-cleaning warning added alongside it, once the upstream generator learned to publish them.

Verification

• 366 unit tests, 124 integration (cassette) tests pass.
• mypy tango/ strict: clean. ruff format --check: clean.
• Conformance: exit 0, no errors. Shape coverage: 0 gaps, 0 baselined.
• Version synced across pyproject.toml and tango/__init__.py; import tango reports 1.4.0.
• ruff check reports 23 errors, all pre-existing on main in files this release doesn't touch.

⚠️ Unrelated production issue found while verifying

tests/production/test_production_smoke.py::test_search_filters fails, and it is not flaky and not caused by this release — I reproduced it on unmodified main before #52 landed.

/api/contracts/?awarding_agency= times out with a 504 after ~20.6s, consistently, across multiple agency codes. It is isolated to that one filter on that one endpoint:

request	result
contracts?awarding_agency=4700	504, 20.6s
contracts?awarding_agency=9700	504, 20.5s
contracts?fiscal_year=2024	200, 0.1s
contracts?naics_code=541519	200, 0.1s
contracts?search=cloud	200, 1.7s
idvs?awarding_agency=4700	200, 2.1s
grants?awarding_agency=4700	200, 0.2s

So the filter works fine on other resources, and other filters work fine on contracts — it is specifically awarding_agency on the contracts endpoint. That is a common filter on the busiest endpoint, so it likely affects real traffic. Needs a tango-side issue; nothing to fix in the SDK.

After this releases

tango-mcp #64 unblocks: its entity regression clears, and its hand-maintained shape allowlists (79 fields behind the API) can finally be derived from a contract with no blind spots.

🤖 Generated with Claude Code