Apple's FY24 revenue is $391 billion. Pull it from SEC's free XBRL API and you might get back $182 billion. Both are technically correct.
Welcome to the actual state of public financial data in 2026.
The SEC publishes everything. None of it is usable.
The SEC runs the most comprehensive financial filing system on the planet. Form 4 insider trades, 10-K annual reports, 8-K material events, Schedule 13D activist stakes, 13F institutional holdings, XBRL-tagged financial statements. All free. All public. All updated within minutes of being filed.
So why do the big aggregators charge $$$$ a year and Bloomberg charges $24K per terminal? Because raw SEC data is a hot mess. Apple tags revenue under three different XBRL concepts depending on the quarter. Bunge reports two revenue numbers per filing - one with tax, one without - and a naive query returns whichever it sees first. Walmart's FY24 EPS comes back wrong by a factor of nine if you don't track which filing was filed after which split. Foreign issuers use a completely different taxonomy. Nothing is named consistently. The dimensional axes that hold segment revenue get collapsed before they reach you. You spend three weeks getting one ticker's numbers to match the cover page of its own 10-K.
The big aggregators sell you a way around the hot mess. That's the entire business. The data is free; the cleanup is paywalled.
What insiderdelta does
We ingest direct from SEC EDGAR and do the normalization ourselves. Two-pass XBRL alias resolution. Period-shape filtering that catches when issuers mislabel quarterly facts as annual. Split-adjusted historical EPS that doesn't double-count when a 10-K filing already restated post-split. IFRS taxonomy support so ASML and Toyota don't quietly disappear from your screen. Inline XBRL parsing for segment breakdowns - Apple's iPhone vs Services, Walmart's US vs International - the stuff the aggregators charge extra for.
We also surface things the big aggregators don't. Form 4 insider trades rolled up into cluster-buy signals (when 3+ insiders buy the same stock inside 30 days - documented academic alpha). 13D activist intent filings in a queryable feed. Smart-money 13F holdings for a curated set of activist and value managers. Dividend aristocrat streaks computed against real data, not stale spreadsheets. 8-K item-level parsing with the earnings press release attached as Exhibit 99.1 fetched and inlined under Item 2.02.
The receipts
Anyone can claim "we have accurate data." We publish a continuous validation suite against hand-curated values pulled from issuer 10-K cover pages and Yahoo Finance snapshots. Apple, Microsoft, Coca-Cola, J&J, Walmart, GameStop, Bunge, Alphabet. Tolerances are explicit (1% on headline financials, 5% on TTM dividends). When a check fails we surface the discrepancy publicly. The validation page is the credibility marker we want everyone to read before relying on anything else here.
Why this matters more in 2026
Every LLM agent built to answer financial questions hits this problem. Ask Claude or GPT what Apple's FY24 revenue was and you get a different number depending on which web search returns first. Hallucination on financials is a liability problem for the tool builders. The fix is a layer between the model and the SEC that's deterministic, fast, queryable, and honest about what it doesn't know.
That layer is what we're building. Right now it's a website. The MCP server is next.
Try it
Pick a ticker and walk through the surfaces. Start with AAPL or WMT. Compare what you see against the issuer's own 10-K cover. If something looks wrong, tell us - [email protected]. The validation suite means we can usually reproduce, isolate, and fix a real discrepancy fast.
If you're a buy-side analyst, a builder of LLM research agents, or you currently pay for one of the big aggregators and have an honest opinion on what we're missing - we want to talk to you. Same address.
Built independently. No affiliation with the SEC, Bloomberg, or any data vendor. Not investment advice; see Validation & accuracy for the full disclaimer.