AI disclosure: This tutorial and its example code were produced with an AI agent for Startup Funding Tracker. The example data is synthetic.
If you import SEC Form D data, an accession number is a good primary key for a filing. It is a bad key for a funding event. A new Form D and its later amendment have different accession numbers. They can still describe the same offering, with the newer notice reporting a revised total.
That distinction affects both ingestion and analytics. Retrying a download should never create a second filing record. Reading an amendment should never cause a dashboard to add two snapshots of the same sale.
Store filing identity and offering identity separately
| Field | What to store |
|---|---|
accession |
Unique key for the document. Keep its original source URL. |
issuer_cik |
SEC issuer identifier. One issuer can have multiple offerings. |
prior_accession |
The prior filing reference when available on an amendment. |
offering_id |
A resolved key shared by notices that describe the same offering. |
accepted_at |
EDGAR acceptance time normalized to UTC. Keep the first-sale date separately. |
amount_sold |
The filing's reported cumulative amount, stored as an exact number. |
Start a new offering for an original notice. For an amendment with a prior reference, walk the chain until it reaches a known original. Keep the amendment unresolved if a predecessor is missing from the import. A fallback guess based on the CIK alone can silently merge separate financings by the same company.
The SEC's Form D Data Sets are convenient for backfills. Retain links to EDGAR because a flattened extract is not a substitute for the original notice.
Compute changes after resolving the chain
In this example, each tuple has a normalized UTC timestamp, accession, resolved offering key, and reported amount sold in whole dollars. The first notice establishes the baseline. A later filing can increase it, leave it unchanged, or revise it down.
def offering_changes(filings):
seen = set()
latest = {}
for accepted_at, accession, offering_id, amount in sorted(filings):
if accession in seen:
continue
seen.add(accession)
if offering_id is None or amount is None:
yield accession, "needs review", None
continue
previous = latest.get(offering_id)
latest[offering_id] = amount
if previous is None:
yield accession, "initial snapshot", amount
else:
delta = amount - previous
label = "downward revision" if delta < 0 else "reported change"
yield accession, label, delta
filings = [
("2026-01-15T12:00:00Z", "example-a", "offering-1", 600_000),
("2026-02-10T12:00:00Z", "example-b", "offering-1", 900_000),
("2026-02-24T12:00:00Z", "example-c", "offering-1", 850_000),
("2026-03-12T12:00:00Z", "example-d", None, 200_000),
]
for change in offering_changes(filings):
print(change)
('example-a', 'initial snapshot', 600000)
('example-b', 'reported change', 300000)
('example-c', 'downward revision', -50000)
('example-d', 'needs review', None)
The latest known total for the resolved offering is 850,000. Summing three reported totals would count the same securities repeatedly. A downward revision is a change in the reported amount, so keep it separate from any claim that an investor received a refund. The SEC's Regulation D statistics methodology likewise uses incremental amounts for successive amendments to the same offering.
For an incremental pipeline, rebuild the affected offering when a missing predecessor arrives. That keeps an out-of-order import from permanently changing the interpretation of a notice. Make the transformation replayable from the filing records, including the decisions that resolved the offering IDs.
For a live example of navigating from an issuer to its filing, Startup Funding Tracker has daily U.S. startup Form D notices and links to source filings.
Top comments (0)