How the largest crowdsourced compensation dataset turns base, equity, and bonus into one total comp number, where it is strong, where it is thin, and how everycity reconciles it with the public data.
Every salary figure on everycity rests on a question: where does the number come from. For technology roles, the most cited source on earth is levels.fyi, the crowdsourced compensation dataset that has reshaped how engineers read an offer. This piece explains how that dataset is built, what its total comp number actually measures, where it is strong, where it is thin, and how everycity reconciles it with the public sources to produce the city salary medians on the profiles and rankings. The methodology page carries the full weighting note.
Levels.fyi launched in 2017 as a tool to map job titles across technology companies and grew into the largest public ledger of technology pay, built from hundreds of thousands of self submitted compensation reports. A user enters their company, level, location, years of experience, and the breakdown of their pay, and the site aggregates those entries into medians and percentiles by role and level. The result is a living picture of what a given rung at a given company pays right now, refreshed as new entries land, rather than the lagging survey data that older sources rely on.
The core insight levels.fyi popularized is that a technology salary is not one number but three. Base salary is the cash paid through the year. Equity is the stock grant, usually restricted stock units that vest over four years, converted to an annual figure by dividing the grant by the vesting schedule. Bonus is the annual cash target, sometimes with a separate sign on bonus in the first year. Add the three and the result is total comp, the figure levels.fyi leads with, because at the senior levels the equity portion often rivals or exceeds the base.
Reading base salary alone badly understates senior technology pay. A staff engineer at a large public company might draw a 220,000 dollar base but a 250,000 dollar annual equity grant and a 40,000 dollar bonus, for a total comp near 510,000 dollars that the base figure hides entirely. This is why a recruiter quoting base and a candidate quoting total comp can describe the same job with numbers that differ by half. Everycity follows the total comp convention for technology roles so the cities for tech jobs and highest paying cities rankings compare like with like.
The genuinely hard problem levels.fyi solved is the ladder. A Google L4 is not a Meta E4 is not a Microsoft level 61 is not an Amazon SDE II, yet they describe broadly equivalent seniority. The site built a cross company mapping that lets a reader line up a level at one employer against a level at another, which is what makes the data comparable at all. Without that mapping the numbers are noise, since a company that simply uses higher level numbers would look like it pays more. The mapping is also where the most editorial judgment sits, and where the dataset is most often debated.
Entries are self reported, with an optional verification step where a user uploads an offer letter or a pay statement to mark a data point as confirmed. The verified entries carry more weight, but the bulk of the dataset is unverified, which is the first thing to hold in mind. The second is sample size: the cells are dense for a senior engineer at a large employer in a major hub and thin to empty for a niche role at a small company in a smaller city. A median built on six entries is a hint, not a fact, and everycity treats thin cells accordingly.
Four biases run through any crowdsourced pay dataset. Self selection: the people who report tend to be the engaged, often higher earning end of the distribution, which lifts the medians. Survivorship: the unhappy and the laid off report less. Equity valuation: a grant is valued at the share price on the day it was given, so a number from two years ago can be far above or below its worth today. And geographic and sector skew: the dataset leans heavily toward United States big technology, which is why it informs the technology lines on everycity but not the wider economy. The cities for startups and cities for finance rankings lean on other sources for the non technology roles.
Everycity does not copy levels.fyi figures onto a city page. For technology roles it uses the dataset as one input, weighted by sample size and verification rate, then blends it with Glassdoor, the OECD tax and wage database, Numbeo, and the national statistics offices to land a median that reflects the whole local market rather than the reporting tail. The blend is reconciled to the local currency at the May 2026 rate and stated in the city profile. So the technology salary on the San Francisco, New York, London, Zurich, and Singapore profiles draws on levels.fyi but is not equal to it.
A total comp figure is a gross number, and gross is not what lands in the account or buys the groceries. A 400,000 dollar total comp in San Francisco and a 250,000 dollar package in Zurich can leave similar money in hand once the tax band and the rent are taken out. This is the whole reason everycity pairs every salary with a tax and a cost view: the highest paying cities after tax ranking reweights the gross figures, the tax calculator runs the take home against a jurisdiction, and the cost of living calculator sets that take home against the local basket.
Read levels.fyi as a strong signal on technology pay, not a precise wage. It is the best public window into what the large employers pay at each rung, and it has done more for salary transparency than any survey, but the self reporting, the equity valuation, and the big technology skew mean the medians run a little high and a little narrow. Used with that caveat, and blended against the public data, it is a genuine asset, which is why it sits among everycity sources. The data is strongest for the named large employers in the major hubs and weakest for the small company, the niche role, and the city outside the United States and a handful of European and Asian capitals, where the public wage statistics carry more of the weight. For the wider picture, the 2026 cost of living report and the relocation score tool put a salary into the context of a place.
Levels.fyi turns base, equity, and bonus into a single total comp number and maps levels across companies so technology pay can be compared at all. Treat it as a strong signal rather than a precise wage: it is self reported, it values equity at the grant date, and it skews toward United States big technology. Everycity blends it with Glassdoor, the OECD, Numbeo, and national statistics, then pairs every gross figure with the tax and cost views, because the only number that matters is what lands after tax and after rent.
One measured email each month: the new city profiles, the refreshed cost baskets, and the rankings that moved. No tourism boards, no sponsored placement, just the numbers.
