SEC financials — as-reported, free, no key

The tool skill under intrinsic-value. It exists so a valuation is never built on a number someone remembered.

A summary site gives you a number with no period, no basis and no filing behind it. Restatements, "adjusted" figures and stale comparatives all look identical once they are prose — and in a valuation every downstream multiple inherits the error silently.

python3 sec_financials.py CTVA --years 7
python3 sec_financials.py CTVA --ttm       # trailing twelve months
python3 sec_financials.py CTVA --json
python3 sec_financials.py CTVA --concept AmortizationOfIntangibleAssets

Source is data.sec.gov/api/xbrl/companyfacts. No key, no account. The only requirement is a User-Agent carrying a real contact address — SEC blocks anonymous clients outright rather than rate-limiting them. Set it once: export SEC_USER_AGENT="Your Name you@example.com". Responses are cached for a day under ~/.cache/sec-financials because a 30-name screen otherwise pulls a couple of hundred MB.

The five traps, each of which returns a plausible wrong number

1. fy is the FILING's year, not the data's period. The big one. A 10-K carries two prior years as comparatives and stamps them all with the filing's fy. Corteva's FY2019 D&A ($1,599M) appears under fy=2019, fy=2020 and fy=2021 — key on fy and you get the same stale year three times, which reads as a beautifully flat trend rather than as an error. Key on the period end date.

2. Issuers switch tags mid-history. Corteva reported capex under PaymentsToAcquirePropertyPlantAndEquipment through FY2021, then PaymentsToAcquireProductiveAssets. First-tag-wins across the whole history silently truncates the series to blanks. Candidates are tried per period in priority order — they are alternates, never components, so never sum them.

3. Annual and Q4 facts share a tag. Flows are filtered to a 300–400 day start→end span, or you book a quarter as a year.

4. Not every line is tagged. Corteva never tags GrossProfit or OperatingIncomeLoss. Derive (revenue − COGS), label it DERIVED, or go read the statement — never leave a silent blank where a number is expected.

5. companyfacts can miss a whole filing for months. One issuer's companyfacts stopped at its Q1 10-Q while the Q2 10-Q had been filed for months, so --ttm and "latest quarter" were one quarter stale with no error. The same quarter WAS in the frames API (/api/xbrl/frames/us-gaap/GrossProfit/USD/CY2026Q2.json) and in the filing's own XBRL. Before quoting "latest", compare the newest accn in companyfacts against the submissions list (/submissions/CIK##########.json).

Rules

• Pre-spin history is a different company. Corteva's FY2018 D&A is $2,790M against $1,599M in FY2019 — that is DowDuPont, not Corteva. Same CIK, different entity. Check the separation date before charting a long series. The same applies to any carve-out: GE Vernova and Constellation both carry parent-era years that belong to the former parent, not to the business you are valuing.
• Split d_and_a before using it. Depreciation and AmortizationOfIntangibleAssets are separate tags meaning opposite things for valuation — see intrinsic-value, where the whole adjustment turns on it.
• --ttm needs a prior-year year-to-date figure to subtract and returns n/a with the reason rather than guessing. A TTM that double-counts a seasonal half-year is worse than no TTM: some businesses earn the whole year in H1.
• Foreign private issuers file 20-F/40-F with ifrs-full tags, not us-gaap. Handled, but the line names differ and coverage is thinner.
• Segment data is NOT in companyfacts. Segments are dimensional XBRL and this endpoint is flat — take segment splits from the earnings 8-K or the 10-K.

Related

Feeds intrinsic-value and buffett-checklist. For live quotes and where a name sits in its own cycle, that is price-history.