> ## Documentation Index
> Fetch the complete documentation index at: https://docs.letsfind.work/llms.txt
> Use this file to discover all available pages before exploring further.

# Fields from the job text

> Pay, workplace type, benefits and visa sponsorship, read from each job's own text by fixed rules, with measured accuracy.

Some fields come from the job's own text, not from the employer's board fields. Fixed rules read them when we store the job: word lists and patterns in English, German, French, Spanish, Dutch, Portuguese, Italian and Polish. No language model is used. Treat these fields as a strong hint, not as the employer's word.

## The fields

| Field | What the rules read |
| - | - |
| `salary*` | When the board gives no pay fields and no pay line, an amount or range with its currency in the text, next to a pay word ("salary", "Gehalt", "salaire"). Bonuses, perks and company figures are left out. |
| `workplaceType` | `hybrid` or `remote` when the title, place or text says so ("hybrid role", "2 Tage Homeoffice", "teletravail partiel", "100% remoto"). A board's own remote or hybrid field always wins; its `onsite` gives way to the text, since many boards set `onsite` by default. |
| `benefits` | Any of: `health` (health, medical, dental or vision insurance), `retirement` (pension, 401(k), retirement plan), `paid_time_off` (vacation days, paid holidays, paid sick leave), `wellness` (gym, fitness, wellbeing or mental health programs). Empty when the text names none of these. A bare "great benefits" names none. |
| `visaSponsorship` | `true`: the text says it sponsors visas. `false`: it says it does not, or that you must already have the right to work there. `null`: it says nothing. |

Filters: `benefit` (any of the values above) and `visaSponsorship` (`true` or `false`; jobs whose text says nothing are left out). `GET /v1/facets` counts both. Pay from the text works with `salaryMinEur` like any other pay, and the workplace type with `workplaceType` and `location=Remote`. See [Search and filters](https://docs.letsfind.work/search.md).

## How often they are right

We measured the rules on 100 open jobs, picked at random across 26 applicant tracking systems, 18 countries and 8 languages, in any field. We labelled every job by hand first, without seeing the rules' answers, on 15 November 2026. We tuned the rules on half of the jobs; the other half was never used for tuning. The table shows the half the rules never saw, then all 100.

| Field | Right: unseen half / all 100 | Measured as |
| - | - | - |
| Pay, all jobs | 86% / 91% | Same amounts, currency and period, or both empty |
| Pay, jobs whose text states pay | 74% / 81% | Same rule. Without reading the text: 32% |
| `workplaceType` | 84% / 91% | Same value (remote, hybrid, onsite). Before these rules: 82% / 88% |
| `benefits` | health 89% found, 100% right; retirement 88%, 100%; paid\_time\_off 93%, 76%; wellness 83%, 71% (all 100: 92% / 100%, 94% / 100%, 97% / 86%, 80% / 86%) | Per value: share of the jobs that offer it where we find it, and share of our values that are right |
| `visaSponsorship` | 96% / 98% | Same value, or both empty. When we give a value: 5 of 5 right |

100 jobs is a small set: a row can move by a few points on another sample. Fields we tried and dropped because the rules were not right often enough: years of experience (57%), education (83%), hours a week, office days, and other benefits (bonus, equity, learning, meals, transport and more).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.