Agrofit datasets¶
Four datasets expose current Agrofit/MAPA tables through the semantic layer. They preserve the filters, values, multiplicities, and contracts of the Defensivos source APIs, using the single defensivos source. No generic product argument is required.
Function in agrobr.datasets |
Content and identity | Validated contract | Columns |
|---|---|---|---|
defensivos_formulados |
One formulated product per textual registration | agrofit_formulados 1.1 |
12 |
defensivos_tecnicos |
One technical product per textual registration | agrofit_tecnicos 1.1 |
10 |
autorizacoes_defensivos |
Published product, crop, and pest relationships; repetitions retained | agrofit_autorizacoes 1.1 |
10 |
composicao_defensivos |
One component per family, registration, and position | agrofit_composicao 1.0 |
9 |
These are the existing source contracts, available under their agrofit_* names; the wrappers do not create duplicate schemas named after datasets. Validation also runs without metadata requested. See the current dataset and contract index.
Queries and filters¶
from agrobr import datasets
technical, meta = await datasets.defensivos_tecnicos(
nr_registro="00301", return_meta=True,
)
components = await datasets.composicao_defensivos(
tipo="tecnicos", nr_registro="00301",
)
formulated = await datasets.defensivos_formulados(nr_registro="08725")
authorizations = await datasets.autorizacoes_defensivos(nr_registro="08725")
| Dataset | Available filters, all keyword-only |
|---|---|
defensivos_formulados |
ingrediente_ativo, classe_toxicologica, classe_ambiental, titular, organicos, marca, formulacao, classe, nr_registro, situacao |
defensivos_tecnicos |
ingrediente_ativo, titular, classe, marca, nr_registro |
autorizacoes_defensivos |
nr_registro, cultura, ingrediente_ativo, classe, situacao |
composicao_defensivos |
tipo="formulados" or "tecnicos", nr_registro, ingrediente_ativo |
Optional filters accept str | None, defaulting to None. The source strips surrounding whitespace from filter values. Registration and organicos use exact equality; situacao ignores case and surrounding whitespace; other filters search literal substrings without regular expressions. Registration remains textual, preserving leading zeros and Unicode. The technical dataset does not accept situacao.
All functions are asynchronous and accept boolean flags use_cache=True, as_polars=False, and return_meta=False. Non-boolean flags, invalid filters, and unknown composition types raise InvalidParameterError before cache/network access. Unknown parameters and positional arguments raise TypeError from the public signature, also before I/O.
Cache and provenance¶
The first query downloads the entire family CSV, even with filters. The September 2026 capture was approximately 391 MB for formulated products and 0.7 MB for technical products; this is not a limit or a guaranteed future size. The formulated bundle serves products, authorizations, and composition; the technical bundle serves technical products and their composition. Subsequent queries can reuse the same cache for 24 hours from acquisition. use_cache=False skips cache reads and writes, as in the source API.
return_meta=True returns (DataFrame, MetaInfo). The selected/attempted source is defensivos; source identifies datasets.<name>/defensivos, and dataset identifies the function. Schema/contract versions match the table, while parser version remains 3. fetched_at retains the original acquisition with a UTC offset; fetch_timestamp is the same acquisition. Reading cache does not claim a fresh acquisition.
Raw hash/size, cache key/expiry, and acquisition/parsing durations propagate through the dataset layer. They still describe the resource and source work, rather than the filtered DataFrame, a separate dataset cache, or total wrapper duration. from_cache reflects the actual result. Resource details, layout, counts, filters, ignored fields, and diagnostics in source_details are preserved in an independent copy.
datasets.deterministic(...) is rejected before any I/O: the CSV is a current export, and its cache retains one acquisition identified by content. Neither can reconstruct an arbitrary historical register. These datasets return snapshot=None.
Identity and types¶
Text columns, including registrations and original composition, come in the installed pandas default dtype (str on pandas 3, object on 2). ordem_componente uses Int64; concentracao_valor uses float64. Empty results retain every column and its type, including optional contract additions. See the complete column description.
Repeated ingredients remain at separate positions. Authorizations are not assigned an artificial key or deduplicated after projection. Components and authorizations are not joined automatically. Literal units, ambiguous text, and null values retain source semantics; TRUE is not converted into a validity classification or an application recommendation.
Sync, Polars, and discovery¶
from agrobr.sync import datasets
components = datasets.composicao_defensivos(
tipo="tecnicos", nr_registro="00301", as_polars=True,
)
print(datasets.list_datasets())
print(datasets.info("composicao_defensivos"))
Polars requires its optional dependency and is applied after pandas validation. The single source has no fallback to another register. Companies, labels, cancellation history, and the official interpretation of situation remain outside these tables; see coverage and limits and the source license.
Export of 2026-09-18¶
The two CSVs of the 2026-09-18 export contain 4,403 formulated products, 279,707 authorization occurrences and 2,992 technical products. Two ambiguous expressions in that export, 1.9 10*10 UFC/g and 200 1x10E10 UFC/g, retain their text and come out with null value and unit and a diagnostic. Numeric interpretation is not guaranteed for the whole component population.
The cache retains non-null values, dtypes, component position and UTC provenance; None and pd.NA are equivalent only in nullable fields. Composition and situation text retain the literal; other fields retain the documented cleanup. Authorizations are not deduplicated.
A suffix published in the concentration fields can include an ambiguous expression and does not by itself certify a measurement unit or numeric interpretation. Catalogue, CSV and cache share an origin: none of them confirms the historical population.