risk_assessment_list

Python library for determining whether a chemical is on Japan's published risk-assessment obligation lists and for surfacing GHS-based hazard notices.

What it does

• downloads the requested JOHAS, MHLW, and NITE reference files into reference/
• converts the published MHLW obligation lists and NITE GHS classifications into a bundled SQLite database
• exposes exact lookup, synonym-aware fuzzy candidate search, and mixture evaluation APIs

Data sources

• https://cheminfo.johas.go.jp/step/list.html
• https://www.mhlw.go.jp/content/11300000/001168179.xlsx
• https://www.mhlw.go.jp/content/11300000/001474394.xlsx
• https://www.chem-info.nite.go.jp/chem/ghs/files/list_nite_all.xlsx

Install With uv

If you want to use the library directly from GitHub in another project:

uv add git+https://github.com/Ameyanagi/risk_assessment_list.git

Then import it normally:

from risk_assessment_list import evaluate_substance

result = evaluate_substance("50-00-0")
print(result.legal_ra_required)
print(result.nite_chrip_url)

Setup

uv sync --extra build --extra test
uv run python scripts/fetch_reference.py
uv run python scripts/build_db.py

fetch_reference.py is offline-first by default. If the files already exist in reference/ and their sha256 values match reference/manifest.json, the script uses them as cache hits and skips the network.

uv run python scripts/fetch_reference.py
uv run python scripts/fetch_reference.py --refresh
uv run python scripts/fetch_reference.py --refresh-key nite_list_nite_all

Build Workflow

scripts/build_db.py now builds the packaged database in stages:

1. stage raw MHLW and NITE workbook rows with file, sheet, and row provenance
2. normalize them into canonical substance, legal obligation, GHS classification, pictogram, and alias tables
3. validate row counts and URL cleanup rules
4. publish the SQLite snapshot atomically into src/risk_assessment_list/data/

The packaged SQLite snapshot is query-oriented rather than blob-oriented. GHS classifications and fuzzy-search aliases are stored in normalized tables, and an FTS5 index is built over aliases for runtime discovery.

The build also exports reference/generated_synonyms.sqlite3, a sidecar database containing synonym groups, identifiers, and aliases for the hazardous substances covered by the generated data set.

Internally, the repo is split into two private implementation areas:

• risk_assessment_list._runtime for lookup/search/evaluation logic
• risk_assessment_list._build for fetch/build/export orchestration

The checked-in scripts/*.py commands are thin wrappers over those internal modules, so the user-facing workflow stays the same while the package internals remain importable and testable.

Usage

from risk_assessment_list import MixtureComponent
from risk_assessment_list import evaluate_mixture
from risk_assessment_list import evaluate_substance
from risk_assessment_list import search_substances

candidates = search_substances("ホルムアルデヒド")
fuzzy_candidates = search_substances("maleik acid", mode="fuzzy")

result = evaluate_substance("50-00-0")
print(result.legal_ra_required)
print(result.ghs_notice_required)
print(result.ghs_pictograms)
print(result.nite_chrip_url)
print(result.nite_chrip_urls)
print(result.model_sds_url)

mixture = evaluate_mixture(
    [
        MixtureComponent(identifier="50-00-0", weight_percent=0.2),
        MixtureComponent(identifier="7732-18-5", weight_percent=99.8),
    ]
)
print(mixture.legal_ra_required)

Notes

• legal_ra_required follows the published union of the downloaded MHLW lists.
• ghs_notice_required is broader and fires whenever matched NITE data contains an assigned GHS classification, including environmental classes.
• nite_chrip_url is the first matched JOHAS NITE-CHRIP detail link returned with the legal/regulatory matches.
• nite_chrip_urls contains all matched CHRIP detail links for the result. Some regulatory rows map to multiple CHRIP pages because JOHAS splits one legal row into multiple CAS-specific detail pages.
• model_label_url and model_sds_url come from the NITE GHS workbook and point to the MHLW anzeninfo model label/SDS pages. They are separate from the JOHAS NITE-CHRIP link.
• search_substances(..., mode="balanced") is the default. It prefers exact aliases and strong prefix matches, and returns [] for obvious garbage input.
• search_substances(..., mode="fuzzy") is the broad finder mode. It keeps typo tolerance and can return wider near matches and derivatives.
• search returns scored candidates only; it does not perform the legal decision on its own.
• fuzzy search expands explicit aliases and conservative synonyms such as parenthetical aliases, Greek-letter variants, English/Japanese name pairs, and a small curated abbreviation set like PCB, DDT, TCE, and PCE.