Insights page: /insights

One page for “the market and my progress”. The top right switches between 7 / 30 / 90 days (?range= in the URL, so refreshes and shared links keep it). “Today” and each day boundary follow timezone in schedule.yaml, and the range includes today.

Configuration: insights.yaml

Defaults live in data/config/insights.yaml; your own copy lives in data/config/insights.yaml (see configuration.md).

Key	Meaning
ranges_days / default_range_days	Which time ranges are offered (short to long) and the default
cities	Cities listed separately in the city card, written like the /jobs city filter (München, not Munich); empty (the default) = the locations of your preferences
top_filter_reasons	How many rows the “Filter reasons” card lists
score_bucket_width / score_bucket_cap	Bar width of the rank-value histogram, and the value from which everything goes into the last bar
salary_bucket / salary_bucket_cap	Bar width of the annual-salary histogram (in the market’s base currency), and where the last bar starts

The web app re-reads this file each time the page opens, so no restart is needed.

How each card is computed

Every number is a SQL aggregate over the database, counted by “jobs first seen within the range” (first_seen_at, the same reading as the “first seen” date filter on /jobs). “Passed filter” = the job’s latest score record is not filtered_out (jobs with no score record yet are also shown by default on /jobs, so they count as passed here too).

Card	Definition	Click goes to
Key numbers	New (all), passed filter, share with salary disclosed, average rank value, applied	New -> /jobs (including filtered); passed filter -> /jobs; salary disclosed -> salary=true; applied -> /pipeline
Trend lines of the key numbers	Days that have data (days with new > 0, the pass rate of days that have jobs, the rank value of days that have scored jobs) are joined into one continuous step line; days without data are skipped instead of drawn as scattered points. With fewer than 2 days of data no line is drawn and “Not enough data (< 2 days)” is shown; the line width shrinks with the card and never draws into a neighbouring card (lib/insights/trend.ts)	n/a
New per day	Jobs per day that passed / were filtered out; days without jobs still get an empty bar	Passed bar -> /jobs for that day; filtered bar -> /jobs for that day, filtered-out jobs only (filteredOnly=true)
Source breakdown	By source family (channels.family); a job that appears in several sources is counted once per source; shows the share that passed the filter	Total -> source= + include filtered; pass share -> source=
City breakdown	Passed jobs only. A job counts if any of its locations matches a configured city (one job can be in several cities); remote is by workplace; other = neither remote nor in a listed city	City -> city=; remote -> workplace=remote; other has no matching filter and is not clickable
Salary disclosure	Passed jobs only: the share that states a salary. The annual-salary histogram only uses “comparable EUR annual salaries” (the same definition as the detail page and /me: qualityStatus is not review and not a salary from another region; ranges use the midpoint, one-sided figures use that end), bucketed with a median (the label states the range and count, e.g. “median · last 30 days · 8”; hidden when there are fewer than ui.yaml: salary_stats.min_sample)	Share -> salary=true; histogram bars are not clickable
Signals	Passed jobs only: for each signal category (except tag categories, in the order of /interests), the number of jobs with at least one signal of that category and their average rank value; and the number of jobs with none of them	The “high-value signals” count (it names the high-value categories) -> high=true (same definition as the quick filter on /jobs, including the salary floor rule of the amount category); hidden when no category is high value
Rank-value distribution	Passed and scored jobs only, bucketed by rank value; jobs still waiting for AI get their own count	Not clickable
Application funnel	New jobs = jobs that appeared in the range and passed the filter; each later stage = applications that entered that stage (or a later one) through a status change within the range; rejected / withdrawn / ignored are listed separately. Note that “interested” may be a job that appeared earlier, so new -> interested is not a conversion rate; the share relative to the previous stage is only given from “interested” onward	New -> /jobs; the rest -> /pipeline
Filter reasons	Jobs hard-filtered out within the range, counted by reason code (without the detail after the colon); a job can have several reasons	“Open filtered jobs” -> /jobs, filtered-out jobs only (filteredOnly=true)

When there are no jobs and no application actions at all in a range, only an empty state is shown; when only a single card has no data, that card shows its own empty state.