Equibles is built for decisions, so every number has to be traceable and mean exactly what it says. Here's how the data is sourced and handled.
Sources
Regulatory, fundamental, and trading data comes straight from the source of record — not a downstream aggregator:
- SEC EDGAR — filings (10-K/10-Q/8-K/20-F/6-K/40-F and amendments, DEF 14A proxies, S-1/F-1 IPO registrations), XBRL company facts, 13F holdings, Forms 3/4/5 insider activity, Form D, NPORT-P / N-CEN funds, Form ADV advisers.
- FINRA — short volume, short interest, and off-exchange (dark-pool) volume.
- FRED, CFTC, CBOE — economic series, futures positioning, and VIX / put-call ratios.
- Congressional disclosures — member trades and net-worth bands.
- USAspending — federal contract awards.
- FDA — the advisory-committee meeting calendar.
- Company webcasts — earnings-call audio, transcribed and speaker-attributed.
Each figure a tool returns carries its provenance — the form, filing date, and period — so you can trace it back to the document it came from.
Prices come from a market-data feed on two separate lanes, and it matters which one you're reading:
- End-of-day bars — the daily OHLCV history, written once a session has settled. Customer-facing history, latest-price ranges, and technical indicators exclude zero-volume carry-forward candles because they do not establish a traded market price.
- Live intraday quotes — the latest available market reading for an active US listing: last trade price, its UTC timestamp, and bid/ask when the feed carries them. Each reading states whether it is real-time or 15-minute delayed, names both its market session and the session expected now, and marks an older-session reading stale. A ticker with no live reading is reported as having none rather than being filled with a placeholder price.
The two never silently substitute for each other: a current-session quote may enrich a stored close, while a stale quote is labelled historical and customer summaries fall back to the most recent traded close. Intraday readings are not written into daily history. Reach them through GetLiveQuote, /v1/stocks/{ticker}/quote, /v1/quotes, or /v1/stream.
No heuristics
We never classify financial data by pattern-matching. A company is a REIT because its SIC code says so; a security is common stock because the filing's security type says so — never because of how a ticker or name looks. When the authoritative field is missing, the value is omitted, never guessed.
Fundamentals are as-reported
Fundamentals come from tagged XBRL — the company's own numbers. A metric with a missing input is excluded from a calculation rather than filled in. Non-GAAP measures (adjusted EPS, FFO/AFFO) are read from the company's own reconciliation and labelled with its verbatim measure name; they're never compared against GAAP actuals.
As-reported vs derived
Most of what a tool returns is as-reported — the exact figure from a filing, with its provenance. Some outputs are derived on top of that data, and we say so:
- Composite scores — the short-squeeze and insider-sentiment scores are peer-relative 0–100 ranks computed from the underlying figures (short interest, days to cover, net insider buying), not values a company reports.
- Short-interest estimates — the nowcast predicts the immediate next official FINRA settlement from eligible daily data available through its stated as-of date. It is always labelled as a model estimate, names the settlement and expected FINRA publication date, and stays separate from the official short-interest history. A missing estimate stays missing rather than being filled with a fallback.
- Index outlooks — current constituents and observed changes come from disclosed tracking-fund holdings. Outlooks apply supported published rules company by company and disclose unsupported rules: S&P results are screens because its committee decides membership, while Russell and Nasdaq-100 results apply their published rankings as projections. They are not index-provider announcements.
- Narrative extractions — KPIs, non-GAAP bridges, and going-concern flags come from written filings; customer concentration uses supported dimensioned XBRL directly when available and otherwise uses the filing narrative. Forward guidance can also come from calls, while earnings briefs and insights come from transcripts. Each narrative extraction is checked by an independent verifier before it's published and carries the verbatim source quote, so you can confirm it against the document.
Split adjustment
Price surfaces use the stored daily close; raw price rows carry no split-basis metadata because a full-history Yahoo reconciliation can be served on either basis. Share counts and 13F positions are restated across captured splits so they're comparable across time. Same-series ratios can use raw filed counts, but totals or percentages spanning sibling listings restate each exact class first because their split histories can differ.
Fiscal periods
Periods are handled precisely: a trailing-twelve-month (TTM) figure sums the four most recent discrete quarters (not a year-to-date total), and every figure is tagged with its fiscal year and period. The Claude Skill captures the conventions we recommend an agent follow.
Freshness
Data is ingested continuously as it's published — new filings, prices, and disclosures are picked up as they land. A quarter's 13F view stays "combined" (funds that haven't filed yet carry their prior-quarter positions) for the 45 days filers have to report.