The words, defined once

Most of what goes wrong with market data is vocabulary rather than mechanism. Two vendors both say “real-time” and mean fifteen minutes apart; two both say “historical” and mean a decade apart. Each entry says what the word means here, and what it changes about a purchase.

13Falso Form 13F: A quarterly filing listing an institutional manager's long positions in US-listed securities as they stood on the last day of the quarter, filed up to 45 days later. It is a snapshot, not a portfolio: short positions are excluded and never netted against longs, cash and most non-US holdings sit outside it, and anything bought and sold inside the quarter never appears at all.
Consolidated tapealso securities information processor: The merged, regulated record of quotes and trades from every US venue, assembled and published by the securities information processors under national market system plans. It is the reference view of the market rather than the fastest one: complete across venues, slower than an exchange's own feed, and priced by the plans at a small fraction of what a proprietary feed costs.
Continuous contractalso continuous contracts, continuous futures, back-adjusted contract, panama adjustment: A single price series stitched together from a chain of expiring futures contracts, so a chart or a backtest can treat one market as one instrument. Two choices build it: when to roll from one delivery month to the next, and how to adjust the older prices for the gap at each roll. Different choices produce different histories from identical inputs.
Corporate action: An event that changes a security's shares, price basis or identity: a split, a dividend, a spin-off, a merger, a ticker or name change, a reverse split before a delisting. Price history has to be adjusted for most of them or it stops being comparable across the event, and the adjustment convention is the vendor's choice rather than a fact.
Cost basisalso tax lot, tax lots, specific share identification, adjusted cost base: The amount a holding is treated as having cost, against which a gain or loss is computed when it is sold. It is not one number per ticker but a number per lot, per acquisition. Which lots count as sold is set by tax rules and by what you told the broker at the time, and later events adjust the figure afterwards.
Days to coveralso short interest ratio: Short interest divided by average daily trading volume, expressed as a number of days. FINRA computes it from its own twice-monthly filing using the volume between the two settlement dates; vendors compute it over ten, thirty or ninety-day windows of their own choosing. The label is the same everywhere and the arithmetic is not, so two published figures for one stock rarely agree.
Delayed dataalso delayed quotes, 15-minute delayed: Prices and quotes released only after a delay interval that each venue sets in its own data policy. Fifteen minutes is the usual interval, not a rule, and some products are delayed until midnight or by twenty-four hours instead. Exchanges licence delayed data as a separate and far cheaper product, usually with no per-user display fee, and they require every display of it to say plainly that it is delayed.
Derived dataalso derived information: A value calculated from exchange data that cannot be reverse engineered back into that data or used as a substitute for it. Exchanges licence it separately: a derived number carrying a price is usually fee liable at the underlying product's rate, while volume-only and volume-weighted figures may be distributed free. Multi-security output such as an index is exempt from subscriber fees but reported as non-display use.
End-of-dayalso end of day: A price series carrying one record per instrument per trading day, and the phrase vendors use for the cheapest thing they sell. Which close it holds is not implied by the name: the closing auction price, the last print of the session, or a computed midpoint. When it is published, whose trading day it is stamped with, and whether it is adjusted all vary by vendor.
Entitlementalso entitled user, entitlements: A permission a vendor grants and then reports: one subscriber, device or User ID that a distributor has permissioned to receive a given feed in a given month. It is the unit exchanges bill on, counted on capability rather than on use, so an account that never loaded a quote is still billable. Vendors declare the count monthly and the exchanges audit it.
Exchange feealso exchange fees, market data fee: The charge a trading venue levies for its own market data, set in a published schedule and owed by whichever distributor reports you as a subscriber. Your vendor collects it and passes it on, which is why two APIs built on identical technology cannot price alike. It is not the transaction or connectivity charge that a trade confirmation also calls an exchange fee.
Form 4: The SEC filing an officer, director or ten-per-cent holder makes when their own holding in the company changes, due before the end of the second business day after the transaction. It is the fastest record in the US disclosure system and the most over-read: the form states what moved, how many shares and at what price, and it says almost nothing about why.
Greeksalso option greeks: Sensitivities of an option's theoretical value to one input at a time: delta to the underlying price, gamma to delta itself, theta to time, vega to volatility, rho to the interest rate. They are outputs of a pricing model rather than observations of the market, so they carry every assumption that model makes, and two vendors publishing greeks for the same contract will disagree without either being wrong.
Implied volatilityalso implied vol: The volatility figure that, put into an option pricing model, makes the model reproduce a price observed in the market. It is not measured from the underlying's history; it is solved for backwards from a quote. So it inherits every assumption in the model and every choice about which quote was used, which is why two vendors publish different implied volatilities for the same contract on the same day.
Level 2also depth of book: The resting orders behind the best bid and offer, bought from each venue as its own product on its own licence. Level 1 is the top of the book plus the trade stream; Level 2 is what sits underneath it. The consolidated tape carries no depth, so there is no national Level 2 to buy, and what a broker labels Level 2 is usually one venue's book.
MCP serveralso model context protocol server: A process that exposes tools, resources and prompts to a model client over the Model Context Protocol, so a language model can call them without anyone writing a client. It is a connector and nothing more: it grants no rights to the data behind it, holds no subscription of its own, and changes neither the price nor the licence of what it returns.
NBBOalso national best bid and offer: The highest bid and the lowest offer available across all US venues at a moment, calculated and published by the securities information processors from the consolidated tape. It is one bid and one offer, not a book: no sizes behind the top, no venue-by-venue depth. It is the number regulation, brokers and clearing treat as the market price of a quote.
Non-display usealso non-display: Market data consumed by a program rather than read by a person: an algorithm, an order router, a risk engine, an automated screener. Exchanges licence it separately from display use and price it per firm or per platform rather than per screen, usually in the thousands of dollars a month. The counter-intuitive part is that nobody looking at the data is the expensive case.
Non-GAAPalso GAAP: A non-GAAP figure is one a company has defined itself by adding to or subtracting from the measure the accounting rules produce — adjusted EPS, adjusted EBITDA, organic revenue. Both versions are published, both are called earnings, and a data vendor's field labelled EPS may hold either. The label almost never says which, and the two are not interchangeable across vendors.
Open interestalso open contracts: The number of option or futures contracts currently outstanding in a series — positions opened and not yet closed, exercised or expired. It counts inventory, not turnover, so a day of heavy trading can leave it unchanged. It is produced by the clearing house after the session is reconciled, which means it arrives once a day rather than ticking alongside the quote.
OPRAalso options price reporting authority: The Options Price Reporting Authority: the securities information processor for US listed options, operating one national market system plan that disseminates consolidated last sale and quotation information from every exchange approved to list options. Because there is one processor, there is one fee schedule for the whole asset class — and the message volumes behind it are why options history is sold as files rather than streamed.
Options chainalso option chain: Every listed option on one underlier, arranged by expiry and strike, with calls and puts and a quote on each line. It is a data object rather than a screen, and a large one: hundreds of series per liquid name. The word says nothing about whether the quote is one venue's or the consolidated best, how fresh it is, or whether any calculated field comes with it.
Order flowalso footprint chart: The stream of individual trade prints and order book updates from a venue, as opposed to prices summarised into bars. Footprint, volume profile and time-and-sales displays are all renderings of it, and they need three things underneath: prints with size, a side for each print, and depth updates. Only the first two arrive as data, and the side is usually inferred.
Point-in-time: Data as it stood on a past date, rather than as it stands today. A point-in-time series returns the figure that was actually available then — the earnings first reported, the index membership that held, the identifier in use — instead of the current, restated view of that period. Most free sources give you the second and describe it as history.
Professional subscriberalso non-professional subscriber, nonprofessional subscriber, securities professional: The exchange's category for a data recipient, defined in its market data policy rather than on its price list. Everyone is professional by default; the non-professional rate requires a natural person receiving data solely for personal, non-business use who is not registered or qualified with a regulator and is not an investment adviser. An account in a company's name does not qualify. The answer moves the per-user fee several times over.
Redistribution: Data reaching anyone outside the organisation that licensed it: a customer, a public web page, an app, a chart in a paid newsletter. Exchanges and vendors price it as its own line on the schedule, separately from internal use and from per-user display fees, and they bill it per feed before a single subscriber exists. It is the word that turns a side project into a commercial contract.
Schedule 13Dalso Schedule 13G, 13D/G: The SEC filing anyone acquiring beneficial ownership of more than 5% of a registered class of equity makes when they may seek to influence control, due within five business days and amended within two when something material changes. Schedule 13G is its passive twin: the same threshold, a far looser clock, and the difference between them is what the holder says they intend to do.
Short interestalso short interest position: The total open short positions carried on brokerage firms' books for a stock on a given settlement date. US firms report it to FINRA twice a month, and it is published on the seventh business day after that date, so the number is always a week or more old when you read it. It counts positions, which is a different object from short volume.
Slippagealso slippage model, market impact: The difference between the price a test assumed and the price a fill would actually have got. In live trading it is measured after the fact against a reference price. In a backtest nothing measures it, because no order was ever sent, so it is an assumption the framework applies on your behalf — and different models produce different answers from identical data.
Survivorship biasalso survivor bias: What happens when a dataset contains only the securities that still exist. Companies that were acquired, went bankrupt, delisted or left an index drop out of the history along with their returns, so the remaining record describes the winners of a period rather than the period. It flatters every figure computed from it, and it is a property of the data rather than a mistake in the arithmetic.
Symbologyalso symbol mapping: The conventions a dataset uses to name instruments, and the work of mapping one set of them onto another. It covers the bare ticker, the venue and class decoration bolted onto it, and the permanent identifier underneath. Symbology is a join problem rather than a naming one: two datasets agree about which company they mean and disagree about the key, and the key is what breaks.
Tick dataalso tick-by-tick: Individual market events with timestamps, rather than prices summarised into bars: every trade, and usually every quote change behind them. The word is not precise about which of those is included, and the difference is roughly thirty-fold in volume, so two vendors selling tick data can be selling products that differ by an order of magnitude in size and price.
Time-weighted returnalso time-weighted, time-weighted rate of return, TTWROR, money-weighted return, dollar-weighted return: A return that measures the holdings rather than the timing of your deposits. The period is cut at every external cash flow, a return is computed for each sub-period, and those are multiplied together, so money arriving or leaving has no effect on the figure. A money-weighted return is the opposite: those same flows are part of what it measures.
Walk-forwardalso walk forward, walk-forward optimisation, walk-forward analysis: A procedure for measuring a rule on data that was not used to choose it: select on an in-sample window, measure on the out-of-sample window that follows, then step both windows forward and repeat, stitching the out-of-sample segments into one record. It tests a selection procedure rather than one fixed set of parameters, and it inherits every defect of the data underneath it.
Withholdingalso dividend withholding, withholding tax, withholding taxes, non-reclaimable withholding: Tax deducted from a dividend or interest payment by the payer before it reaches the holder, at a rate that depends on the issuer's country, the holder's residence and any treaty between them. The cash that arrives is net; the distribution the company declared was gross. Whether part of the difference can be reclaimed afterwards is a separate process entirely.
XBRLalso iXBRL: XBRL is the tagging standard that turns a filed financial statement into machine-readable facts: each number carries an element name, a period, a unit and a set of dimensions. It makes filings parseable, not comparable — issuers pick elements from a large taxonomy and invent their own where none fits, so the same line item arrives under different tags at different companies.

A word gets a URL here only when the catalogue uses it on at least four other pages and there is more to say than the definition. The rest are explained where they come up.