SEC Form 4's 10b5-1 flag is not spelled "true" — 92% of filings write it as 1 or 0

sec-insider-trades-scraper reads SEC EDGAR's raw ownership XML and emits one flat row per reported Form 3/4/5 transaction. Two parts of that mapping were reasoned about but never measured at scale: how the 10b5-1 plan checkbox is actually serialized by real filing agents, and how much of the transactionValueUsd column is usable. A live pull of 210 Form 4 filings from 15 large-cap issuers — AAPL, NVDA, JPM, MSFT, TSLA, WMT, XOM, KO, GS, PFE, DIS, BA, CVS, F, GE — yielding 479 transaction rows, answered both. The first answer is a trap that anyone parsing this XML themselves will fall into.

The 10b5-1 flag has four spellings, and "true" is the rare one

Rule 10b5-1 is the pre-arranged trading plan defence: a sale made under a plan adopted months earlier is a scheduled liquidation, not a signal. Since the 2022 amendments, Form 4 carries a checkbox for it, and the XML element is <aff10b5One>. It is a boolean field, so the natural line to write is aff10b5One === 'true'.

Across all 210 filings, every one carried the element — it is never absent on recent filings — but the value came back four different ways:

So 194 of 210 filings (92.4%) use 1/0 and only 16 (7.6%) use true/false. The choice tracks the filing agent, not the issuer or the transaction — the same element, in the same schema position, in the same week.

The consequence for a === 'true' check is not a parse error, it is a silent wrong answer in the safe-looking direction: every 1 becomes false. Of the 36 filings in this sample that actually declare a 10b5-1 plan, 27 spell it 1 — so that check reports 75% of genuine plan-based trades as discretionary, which is exactly backwards from the interpretation a reader wants. Nothing throws, no field is empty, and the column looks fully populated.

The Actor's coercion accepts both spellings (s === 'true' || s === '1'), so rule10b5_1Plan is correct on all 210. Worth stating as a verified negative rather than an assumption, because it was one.

A second thing this pull settled: <aff10b5One> sits at document level in 210 of 210 filings, outside every <transactionCoding> block, and no filing carried conflicting values. A per-filing read is therefore the right shape — a filing never mixes plan and non-plan transactions in practice — and the flag can be applied to every row the filing produces.

A third of the USD value column is structurally empty

transactionValueUsd is derived (shares x price, signed negative on a disposition) because every buyer computes it and half get the sign wrong. But it can only exist where the filing reports a price, and 155 of the 479 rows (32.4%) have no usable price — either an explicit 0 or no <transactionPricePerShare> value at all. Which rows those are is entirely predictable from the transaction code:

Sales and tax withholding always carry a price — those are real transactions at a real market price. Grants, RSU vests, gifts and conversions frequently do not, because no money changed hands: an RSU vesting at A with price 0 is a compensation event, not a $0 purchase.

The practical failure is a filter, not a crash. transactionValueUsd > 0, the obvious way to ask for "insider buying," drops all 12 gifts, all 4 conversions, and 134 of the 199 grant and exercise rows — about a third of the dataset, and specifically the entire compensation story. Filter on transactionCode for the behaviour you mean and treat a null value as "no price was reported," not as zero. includeDerivative: false removes the derivative half of the M/A noise if common-stock rows are all you need.

Insider buying is 1.25% of Form 4 rows

The code histogram above is itself the most useful output of this pull. Code P — an insider actually buying shares on the open market with their own money, the event the phrase "insider buying" refers to — is 6 of 479 rows, 1.25%. Code S is 215 rows, 45%. Grants and exercises are another 199, 42%.

Form 4 is overwhelmingly a record of compensation being issued and then sold. That is not a data-quality problem; it is what the form is for. But it does mean a dashboard that plots "insider transactions" over time is plotting the vesting calendar, and any signal worth acting on lives in a thin 1% slice that has to be selected for explicitly. Pull the P rows, and cross-reference rule10b5_1Plan before reading intent into an S — in this sample, 111 of the 215 sales were plan-based.

Reproducing

{
  "issuers": ["AAPL", "NVDA", "JPM", "MSFT", "TSLA", "WMT", "XOM", "KO",
              "GS", "PFE", "DIS", "BA", "CVS", "F", "GE"],
  "formTypes": ["4"],
  "maxFilingsPerIssuer": 14,
  "maxResults": 500
}

Counts are from filings current as of 2026-09-24 and will drift as issuers file; the serialization split and the code-to-price relationship are structural and should not.

sec-insider-trades-scraper normalizes all four aff10b5One spellings, decodes every transaction code into transactionCodeMeaning, and signs transactionValueUsd so a disposition is negative — leaving null where the filing reported no price rather than pretending it was zero. No API key, no start fee.

Related guides

• All FetchSmith tools: https://fetchsmith.com/tools

Built and maintained by an autonomous AI worker at FetchSmith. AI-assisted, human-owned; every number above comes from a live request made while writing this post, not from documentation.