Employers Who Ask for Kubernetes Ask for Docker 17 Times More Often Than Average.

Change is how many percentage points the expectation rises once the first skill is present. Lift is the same movement as a multiple. Both are shown because either alone is misleading: lift favours rare skills, change favours common ones, and a pairing only counts as a shadow when it clears both.

Source: Seeker job corpus, 231,147 active job postings, snapshot 2026-07-22. seekerscore.com/insights/skill-shadow-2026

Why frequency is the wrong measure

Rank skills by how often they appear and you get the same list for everyone: cloud computing, communication, testing. Those are true and they are useless, because they describe the market rather than your position in it. Conditional probability asks a different question. Among the 8,880 postings that require Kubernetes, 82.6% also mention cloud computing. That sounds impressive until you notice cloud computing is common everywhere, so its lift is only 3.5. Docker appears in 59.1% of the same postings at a lift of 16.8, which means a Kubernetes posting is nearly seventeen times more likely to want Docker than a random posting is. The second number is the one that tells you something.

Some pairings are effectively one requirement

Docker and Kubernetes at 16.8x in both directions, and the symmetry matters. Docker postings want Kubernetes 66.2% of the time, Kubernetes postings want Docker 59.1% of the time, and each starts from a market rate under 4%. In practice these are not two skills an employer might want. They are one expectation written on two lines. React and TypeScript behave the same way at +43.7 points and 17.4x: if you write React professionally and do not write TypeScript, that is the clearest single gap in this dataset. Salesforce and CRM at +43.0 points is the same shape outside engineering.

Why one number could not have told us this

Ranked purely by lift, the highest pairing among the eight seeds we swept is React and Angular at 19.7x. That is real, and it is not the finding. Angular appears in about 1% of postings, and lift rewards rarity: a skill can post a huge multiple while barely moving the absolute expectation. Under change in percentage points, Angular does not reach the top eight for React, while TypeScript sits at +43.7. Ranked purely by change, the opposite failure appears: cloud computing tops several seeds at +59.3 points, but its lift is only 3.5 because it already appears in 23.3% of everything. It is not an expectation that Kubernetes creates, it is the weather. Requiring both is what leaves Docker and Kubernetes at the top of this set, at +62.3 points and 16.8x together, and it is why this page publishes two columns instead of one.

The shadow depends on the career, not the skill

SQL is the clearest case. In data and analytics postings it pairs with business intelligence at 6.6x and data engineering at 6.1x, pointing toward the pipeline and visualisation layer rather than deeper database work. Excel pairs with PowerPoint at 7.0x, Outlook at 6.6x and Word at 6.3x: the Office suite travels as a unit, and the lift is higher than most engineering pairings even though the shares are modest. Python is the weakest seed we measured. Its strongest partner is scripting at 6.3x, and nothing dominates, because Python is spread across five unrelated specialties and has no single natural next step.

How we measured it (and what we did not claim)

For each seed skill we took every active posting whose extracted skills include it, then counted how often each other skill appears alongside. Sample sizes run from 5,993 postings for TypeScript to 29,172 for Excel. Ranking is by lift, with a floor of 15% share so a pairing cannot top the list on ratio alone, and a lift floor of 1.8 to remove skills that are simply common everywhere. That filter is doing real work: without it the Kubernetes list is topped by cloud computing, teams, and benefits, the last two of which are boilerplate our extractor picks up from sections of the posting that are not requirements at all. Skills come from our extraction pipeline rather than a hand read of each posting, so anything an employer implied without naming is not counted and every share is a floor. This measures which skills appear together in postings. It does not measure which combinations get people hired, and co-occurrence is not causation: we can say employers ask for these together, not that learning one causes the other to be expected. One limit worth stating plainly: this is a sweep of eight seed skills, not an exhaustive search of the corpus. Pairs were ranked within that set, so a stronger relationship may well exist between two skills we did not use as a starting point. Read every ranking here as "highest among the eight we examined", never as "highest that exists".

Frequently asked questions

Related Data

Guides

• Skill Gap Analyzer →
• What Jobs Match My Skills? →
• Transferable Skills List →
• Resume Keywords That Matter →