> ## Documentation Index > Fetch the complete documentation index at: https://www.datalegion.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # 1.5.0_2026-10-15 > Company office locations, fuller names, verified addresses, more accurate identity resolution, cleaner contact data, and more accurate job classification. ## New Features ### Company Locations Company records now include `locations`: the company's office addresses, merged across sources and kept current month to month. Each entry carries the street address, city, state, country, postal code, and coordinates, plus two flags: * `primary` marks the company's primary location (its headquarters). A company has at most one, and it is always an office that is still operating. * `current` tells you whether the office is still operating. Closed offices stay in the list with `current: false`, so you can see where a company used to be. Entries are ordered with the primary location first, then offices still operating, then closed ones, each group newest first. A company whose only known location is its country has a country-level entry (country filled, street, city, state, and postal code empty), so it can still be found by country. A country-level entry appears only when no fuller entry names that country. Addresses in London's postal districts (postal codes starting E, EC, N, NW, SE, SW, W, or WC) use *london* as the city. 64,487,480 companies (90%) carry at least one location, and 64,429,837 (90%) have a primary location. The address fields, `current`, and `primary` are Base tier. `geo`, `raw`, `num_sources`, and `last_seen` are Premium, matching person locations. See [Company Fields](/docs/company-data/fields#locations) and [Data Tiers](/docs/company-data/tiers). **Bulk deliveries:** `locations` is not added to existing delivery definitions automatically. Contact us to include it in your feed. ## Improvements ### Fuller, Cleaner Person Names * `full_name` now carries every name part, including the middle name. * Name parts are reconciled per person across all sources, so a stray initial no longer leaks into `first_name` or `last_name`, and a middle name no longer repeats the first or last name. * When sources disagree and the vote ties, the most recently seen name wins, so a LinkedIn name change shows up. * Short forms and spelling variants are matched to the full name (*Katie* and *Kaitlyn*, *Paco* and *Francisco*), and distinct names that only look alike (*Andrew* and *Andrea*, *Louis* and *Louise*) are no longer treated as one. * Names keep their accents (*Muñoz*, not *Muoz*). * Words that are never a given name (*Estate*, *Occupant*, *Manager*) no longer appear as `first_name`, and *Papa* is kept as the surname it usually is. * Names written in decorative Unicode letters (𝓙𝓸𝓱𝓷) now read as plain letters. * A middle name that repeats part of a hyphenated surname or first name (*Ana Lee Lee-Park*) is dropped, so it no longer prints twice in `full_name`. A stray initial at the end of `first_name` fell from 423,403 people in 1.4.1 to 1,484, at the end of `last_name` from 12,785 to 3,493, and at the start of `last_name` from 40,155 to 574. No middle name repeats the first or last name (1.4.1: 519,377). ### More Complete, More Accurate Locations * More people have a top-level `city`: when no location is flagged current, the top-level fields fall back to the best available one. 186,780,445 people now have a `city` (1.4.1: 147,087,992). * Region labels resolve to their city (*Chicago Area* becomes `chicago`), and county, parish, and metro-area labels (*Fargo-Moorhead*) are no longer reported as a city. * City names that include "City" keep it (*Johnson City*, *Bossier City*). * A city is checked against its state, and a US ZIP code corrects a mismatched state and fills a missing city from USPS. Addresses outside the United States that arrived without a country are no longer tagged as US. * Coordinates outside valid ranges or at 0,0 are removed. * Duplicate copies of one street address filed under another city are folded into the entry that carries the ZIP code. * A street address filed under two neighboring towns that share its ZIP code shows the town most sources report, and never a town that cannot be the ZIP code's. * LinkedIn locations written as *City, United States* now keep the city. * Street spellings that differ only by a USPS abbreviation or a word the source repeated (*Lake Park Dr* and *Lk Park Dr*) now fold into one entry. * A source that names only the city no longer counts toward a street address in that city, so some street addresses show a lower `num_sources` or an older `last_seen` than in 1.4.1. When that city sighting is newer than every street address in the city, it stays as its own entry, and the top-level location can follow it. * The same source records now always produce the same location entries and values, whatever order they arrive in. * Employer addresses are no longer reported as a person's home address. * A street address that arrives without a ZIP code is kept only when USPS confirms that street in or near the stated city. The city and state are kept either way. * An address USPS could not match keeps the source's own text instead of a best guess. * An office, campus, or store address that USPS identifies as a business is removed from every record that carries it. * A street that USPS confirmed for a person is no longer repeated under one of their other towns in another region. * When USPS reports that a person moved and the new address could not be validated, the old address stays with `current: false` instead of being removed with it. A USPS move or vacancy notice sets `current: false` whichever source supplied the address. * A street address that USPS reports is really a post office box (a street-style address assigned to a PO box) is removed. Addresses written as a PO box (*PO Box 975*) stay as written. Fewer people carry a street address as a result: 126,983,928 (1.4.1: 142,793,942), and 87,613,370 with a current one (1.4.1: 109,802,382). The streets removed were mostly employer offices and streets USPS could not place. ### Fewer Duplicate Skills and History Entries Skills that differ only in spacing, punctuation, or plural form (*Project Management* and *project-management*) now appear once. Experience, education, and certification entries that differ only in punctuation or spacing are merged, while *C*, *C#*, and *C++* stay distinct. A social profile URL appears once per person. ### Cleaner Emails and Phones * Emails on a misspelled webmail domain (*gamil.com*, *yaho.com*, *sbcgobal.net*) are no longer delivered. * Emails whose domain ends in a top-level domain that does not exist (*gmail.c*, the retired *.yu*) are no longer delivered. * People who work at a company whose domain also offers consumer email, such as a telecom provider, now get their `work_email` on the company's own domain. * A `.edu` address is `professional` when the person worked at that school, and `personal` otherwise (for example, an alumni address). * Fewer valid work emails are removed by the check for addresses that belong to someone else. * Phone numbers no North American line can have (an area code or exchange starting with 0 or 1, or an N11 area code such as 911) and numbers that can only be dialed locally are no longer delivered. * Facebook and X links that point to a site page instead of a profile (*facebook.com/home.php*, *facebook.com/settings*, *x.com/intent*) are no longer reported in `socials`, nor are links that cannot name an account (a username cut off by the source, or a website typed where the username goes). ### More Real People About 726,000 people whose job history listed employers without titles are now included, as are about 12,700 people whose LinkedIn account appeared under two profile links (after a name change, or with the same name split differently). Test and placeholder profiles are no longer delivered. About 3.2 million people whose only sign of being in the United States was an employer's office address are no longer included: with that address gone, nothing shows they are in the United States. The release has 188,866,145 people (1.4.1: 188,095,903). ### Birth Dates `birth_date` now comes only from sources whose dates agree with other sources, so the dates are more accurate. 64,968,832 people carry one (1.4.1: 62,121,898). ### Job Classification Seniority, job function, and degree are more accurate. Staff roles, "head of" titles, buyers, physician assistants, graduate students, foremen, board seats, and others now land in the right `seniority_level` or `job_function`. Degree abbreviations such as *B.Com*, *B.Tech*, *MSN*, *MPP*, and *LL.M* are recognized, and trade licenses (*Master Electrician*) are no longer counted as master's degrees. More gig and marketplace workers are flagged `is_platform_worker`. Degrees written with their field (*JD, Law*), spaced or dotted (*Ph. D.*), and international qualifications (*SSC*, *HSC*, *A.B.*, *Licenciatura*) are recognized too, and titles such as *Busser*, *Sonographer*, and *Ramp Agent* now get a `job_function` instead of `other`. Answers and place names that a profile listed as its job title (*No*, *Mr*, *Greater Boston Area*) no longer appear as `job_title`. Ranks outside the corporate ladder are read for what they are. Associate and full professors are `senior`, deans `director`, school principals `director`, and assistant or vice principals `manager`. Assistant, adjunct, and visiting professors, fellows in training, and student-body officers get no `seniority_level`, and an assistant no longer takes the rank of the person they support (a *CEO's Executive Assistant* is not `c_level`). Principals at firms (*Managing Principal*, *Principal Associate*) are no longer classified under `job_function` *education*. Rules for titles written with an apostrophe or hyphen now apply, so *Buyer's Agent* is no longer filed under operations and *President's Club* awards no longer read as `c_level`. Company workforce distributions (`legion_seniority_distribution`, `legion_job_function_distribution`, `legion_education_distribution`) shift with these fixes. ### Company Websites on Shared Hosts A company whose website is a page on a shared host (a hotel on its chain's site, an agent on an insurer's site) now carries that page as its website, such as `marriott.com/`, instead of the host alone, and so does the matching job in a person's `experience`. The host's own domain resolves to the company that owns it. A path on a social network, marketplace, link shortener, or listing site is no longer used as a company website. In the API, a full website URL finds these companies. A company's website is read by the name before its public suffix, so a domain under a two-part ending (*cope-technology.co.uk*) is matched to its company, and a department's subdomain that the company's name spells out (*civil.iitm.ac.in*) is kept. A group's page on an organizing site (`meetup.com/`) is its website. ### Companies with Generic Names Real companies whose name is a generic word are now included. Placeholder pages with names such as *Self Employed* or *Confidential* stay out. ### Industries Company and person industries use LinkedIn's current industry names throughout. Older LinkedIn names that some sources still carry (*Information Technology & Services*) map to their current name (*IT Services and IT Consulting*), and mappings that disagreed with LinkedIn's own industry codes are corrected: real estate companies now show *Real Estate* instead of *Real Estate and Equipment Rental Services*. *Artificial Intelligence*, which LinkedIn shows on company pages, is a new value. A source's own label that is not a LinkedIn industry is no longer reported; the company takes another source's industry instead. See [Enums](/docs/enums#company-industry). 3,306,587 companies show *Real Estate*, and 3,375 show *Artificial Intelligence*. ### Workforce Trends End at the As-of Month Hiring, attrition, growth, and turnover now end at the release's as-of month: the latest month most employee profiles have been updated since, and the last entry of `legion_employee_count_by_month`. Job changes reach profiles months after they happen, so ending at the as-of month keeps the most recent windows complete. The windows are the 1, 3, 6, and 12 months ending at the as-of month. See [Data Freshness](/docs/data-quality/data-freshness#workforce-trends). In this release the as-of month is August 2025 (1.4.1: July 2025). Hires and departures are also recounted across the whole series. A promotion is no longer counted as a departure, people who were hired and later left now count as hires, and headcount change now equals hires minus departures in every month and window. Expect monthly hires to rise and departures to fall throughout the history, not only in recent months. ### Fresher Profile Data Headlines now come from the most recent profile data that carries one, so a person whose latest record lacks a headline keeps the previous one, and a tied name vote goes to the most recent name. A placeholder headline or summary such as *--* counts as missing, so it no longer hides a real one. ### More Precise International States and Provinces International addresses now resolve to an ISO 3166-2 `state_code` far more often. Names written without accents (*Tamil Nadu*, *Quebec*), English and native names (*Bavaria*, *North Holland*, *Community of Madrid*), and values carrying a trailing country code (*Ontario CA*) now match. 76% of company addresses outside the United States that name a city or street carry a `state_code`. State codes for names shared by two subdivisions (Dhaka is both a division and a district) now always resolve to the top-level subdivision, so the same address gets the same code in every release. ## Location Coordinates Coordinates (`locations.geo`) are no longer included for people associated with Connecticut. Other location fields are unaffected. ## Phone Coverage Phone numbers of unknown type associated with Colorado are no longer included, as mobile numbers have not been since 1.3.2. Landline and VoIP numbers are unaffected. ## Bug Fixes ### Fewer Duplicate Person Records Records of one person now resolve to one record more often, including after a name change or when sources write the name differently. Records of different people that had been combined, such as relatives who share contact details or people with a common name, are now separate. **If you store `legion_id`, you will need to reconcile.** Merged records retire all but one `legion_id`, and a separated record receives a new one. Re-resolve stored ids against the current data. This release records 1,252,304 merges, retiring 1,283,355 `legion_id`s from 1.4.1, and 4,888,619 splits. LinkedIn IDs on more than one person record fell from 88,869 to 57,336. ### Old-Format LinkedIn Profile Links LinkedIn profile links in the old `linkedin.com/pub/...` format, and LinkedIn's `/pub/dir/` name search pages, are no longer reported or used to match records. ### Non-Person Records Removed Records whose name is a business, a household ("John & Jane Smith"), a job headline, or a placeholder no longer appear as people. ### Foreign State Abbreviations No Longer Read as US States A two-letter state written under another country (`SC` for Santa Catarina in Brazil, `WA` for Western Australia, `MI` for Milan) was read as a US state, and the address was reported as being in the United States. These now resolve to the local subdivision, and the country stays as written. ### Mexico City Addresses in Mexico City now keep `city: "mexico city"` with `state_code: "MX-CMX"`. Previously the city was shortened to `mexico`. ### Company Details from Another Job When a person's work email pointed to a company, that company's domain, industry, and size could be applied to a different current job. They now fill only the matching current job, and only when the company names agree. ### Jobs Linked to Their Company A job whose work email names its company now links to that company: `company_legion_id` and `experience[].organization.legion_id` are filled beside the website and LinkedIn link the email already supplied. This covers jobs read from a LinkedIn headline and employers whose domain writes their name run together or as initials. Jobs whose LinkedIn link is a school page (`linkedin.com/school/...`) or LinkedIn's numeric company link (`linkedin.com/company/2532`) now link to that company as well. 85,927,193 people now have a `company_legion_id` (1.4.1: 81,274,866). ### LinkedIn Links Keep Their Accents LinkedIn links now keep the accented characters LinkedIn uses in the address (*linkedin.com/company/dâme-en-poésie*), for people and companies. Company `linkedin_url` changes for about 1.6 million companies (1,610,482), so update any stored join on it. ### Garbled Characters Repaired Text that a source had decoded with the wrong character set (*•* for a bullet, *Rosa María* for *Rosa María*) is repaired in names, headlines, summaries, job and education descriptions, and addresses. ### Company Names from Renamed LinkedIn Pages A LinkedIn page that was renamed to resemble another company, or redirected to an acquirer, no longer merges into that company, so its old name and website no longer appear in that company's `name.raw`, `domains`, and `socials`. A company with several LinkedIn pages is described by its main page, usually the one with the most followers: `linkedin_url` points to it, and the name, website, industry, type, size, founding year, headline, and description come from that page first. Companies no longer merge through a placeholder name. ### HTML Escapes in Names and Text Names and text that a source delivered with HTML escapes now read as written: `Art & Barrique` is *Art & Barrique* and `O'Leary` is *O'Leary*. This covers company names, headlines, and descriptions, employer and school names, job titles, and person names. ### Stray Null Markers A source's `\N` null marker no longer appears as a street, city, or state value. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.