ZIP Downloads

The free packs sit behind a short form and download from profyt.ai itself, same origin, with no third-party hosting in the path. Licensed products ship in the identical folder shape with the complete record set, so a pipeline built against a free pack keeps working when the licensed release replaces it.

Every pack carries the same six documents beside its data files. The documentation travels with the data on purpose. A colleague who receives the folder a year from now can read the contract without finding this page first.

The three data files hold the same records. JSON is nested and complete, wrapped in a versioned metadata envelope. The CSV is a flat table for spreadsheets, BI, and SQL. The Excel workbook is self-documenting, with cover, obligations, data dictionary, and change log sheets so a reviewer needs nothing else open.

Nothing here needs a ProfytAI library. The examples below are starting points against the Philippine sample's file names. Substitute the file names from the pack you downloaded.

import json
import pandas as pd

doc = json.load(open("free-sample.json", encoding="utf-8"))
df = pd.json_normalize(doc["obligations"])

# every binding duty
mandatory = df[df["parsed_requirement.modal_strength"] == "mandatory"]
print(len(mandatory), "mandatory duties")

Verify before you ingest. The JSON envelope carries a SHA-256 of the canonicalized obligations payload and the record count, so a truncated download or an edited file fails loudly instead of silently entering a compliance record.

import hashlib
import json

doc = json.load(open("free-sample.json", encoding="utf-8"))
meta = doc["metadata"]
integrity = meta["integrity"]

# confirm what you are holding
print(meta["product"], meta["dataset_version"], meta["release_date"])

# the hash covers the canonicalized obligations payload
payload = json.dumps(
    doc["obligations"], ensure_ascii=False, sort_keys=True
).encode("utf-8")

# Philippine packs name it records_sha256, Singapore packs obligations_sha256
expected = integrity.get("records_sha256") or integrity["obligations_sha256"]
assert hashlib.sha256(payload).hexdigest() == expected
assert len(doc["obligations"]) == integrity["record_count"]
print("verified", integrity["record_count"], "records")

The Philippine packs include one source-page capture per record, the printed page with the obligation's exact span highlighted, named from each record's source.evidence_img_loc. Attach it to a workpaper and a reviewer sees the duty in the regulator's own layout without opening the source PDF.

The Singapore packs ship without captures. Redistribution of MAS source pages is pending the regulator's approval, so the official source URL and the full citation stay on every record instead. That is a licensing boundary, not an omission in the data.

Cite verbatim_textwith its citation. It is byte-exact to the regulator's publication. Derived and semantic fields are ProfytAI structuring, and a compliance decision should be confirmed against the official source rather than against a summary.

Each pack's LICENSE.md is the operative document for that release. Free packs carry a free evaluation licence, and licensed packs are single-organization commercial licences with versioned releases. The public terms are on the licensing page.

Prefer to query rather than ingest files? The same certified registers are shared into Snowflake. See the Snowflake channel.