Where every number comes from, so a score can be argued with rather than trusted.

What it is for

The radar answers one question for one person: which Amsterdam companies are about to scale, and therefore about to need someone who builds the operating rhythm before the structure exists. It is not a job board. Roles that are already posted are the secondary output, listed under open matches. The primary output is the company list, ranked so that the top of it is worth a message before any role is posted.

Where the companies come from

The universe is every company on the Amsterdam startup map, which is powered by Dealroom, with headquarters in Amsterdam and 11 to 200 employees. That was 1609 companies at the last harvest. Per row the map gives a current headcount, a headcount history, funding rounds with dates and amounts, the number of open roles with their titles, the growth stage, the launch year and the sector. Headcounts come from LinkedIn employee counts sampled every few weeks, so they lag reality by a month or two and undercount companies whose staff do not list themselves. Funding rounds come from press and filings and are reliable. Open-role counts are scraped from job boards and can be stale for companies that update rarely. Every row shows the month its headcount was last sampled; when that is more than nine months ago the growth signal is not used, which is the case for roughly a third of the list.

The harvest needs a logged-in browser and is refreshed by hand every month or two. Everything else refreshes automatically on Sunday.

How accurate are the headcounts?

An independent check on the top 30 flagged companies: a web search for a recent employee count from news, the company's own site or a data aggregator, compared with Dealroom's figure. 7 could be compared; 7 were within 25% and 7 within 50%. 23 had no independent figure. Treat a Dealroom headcount as a bracket, not a number.

The fit score

Each company carries one number, fit, from 0 to 100. The measured part adds up: momentum, 5 points per level of each of the three signals below, up to 45; size, 25 for 30 to 100 people, 15 for 20 to 29 or 101 to 200, 5 under 20, 0 above 200; stage, 15 for breakout, 10 for early stage, 5 otherwise; sector, 10 for one she named, 6 neutral, 2 for one she cares less about, 0 and a cap of 20 for one she avoids; place, 5 for Amsterdam or the commute ring, −10 elsewhere. Companies that are not startups, or are closed, acquired or low-activity, are capped at 15.

For flagged companies Claude then reads the same facts against Sophie's own words, how her managers describe her, and what drains her, and returns a second 0 to 100 with a one-line note. The final fit is 0.6 × measured + 0.4 × read. Companies without a read show the measured part alone; the table says which is which.

The three signals

Each company gets three scores from 0 to 3. They are meant to be read separately; the total is a convenience for sorting.

Signal	3	2	1	0
F · raised funding	a round in the last 12 months	in the last 18 months	in the last 36 months	none, or only an acquisition, secondary or exit
G · grew the team quickly	headcount up 40% or more in 12 months	up 20% or more	up 10% or more	flat, shrinking, or the snapshot is older than nine months
H · hiring across the board	3+ open roles spanning 3+ functions	3+ roles spanning 2 functions, or roles equal to 15% of headcount	at least one open role	nothing open

Functions are read from the role titles: engineering, sales, marketing, product, operations, finance, people. Internships, working-student and junior roles are ignored, because a company hiring three interns is not scaling.

The flag and the fit adjustments

A company is flagged as about to scale when at least two of the three signals are present, or any one of them is at its maximum of 3, and the total score after adjustments is 4 or more. The adjustments encode the profile rather than the company:

• +1 for Dealroom's "breakout stage" label.
• +1 for 30 to 100 people, the sweet spot she named. Under 20 counts −1. Above 200, her stated maximum, counts −3.
• −3 for anything Dealroom does not label a startup: established businesses ("corporate"), coworking spaces, crowdfunding vehicles. −3 for companies that are closed, acquired or marked low-activity.
• −3 and never flagged for sectors she avoids, currently gambling, crypto, adtech and dating, read from the startup map's sector tags and the tagline. Other sectors do not change the score; the company list has a sector filter instead.
• −2 for a headquarters outside Amsterdam and its commuting ring.

So a 45-person company with a Series A eight months ago, a team up 60% and five roles across engineering, sales and operations scores 3 + 3 + 3 + 1 = 10. The same company at 250 people scores 6.

Open matches

For companies whose website is known, the radar finds the job board behind their careers page. Most scale-ups use one of a dozen applicant tracking systems with a public feed: Greenhouse, Lever, Ashby, Recruitee, Workable, SmartRecruiters, Personio, BambooHR, Teamtailor, Homerun, Join. Postings are stored with the date first seen and last seen, which is what turns a snapshot into a feed: new this week, closed this week, open for six weeks.

Two stages decide what she sees. A rule filter first: location in Amsterdam, the commuting ring, the Netherlands or remote; a title that matches one she searches for, an adjacent one, or a senior title containing words like strategy or operations; no dealbreaker phrase in the text. That removes about 98% of postings without any model. What survives is read by Claude against her own account of what she does and wants, taken from a voice note, her title lists, the company's facts and her earlier ratings, and gets a score from 0 to 100, a verdict of apply, look or skip, a one-line reason and up to three concerns.

What is deliberately not here

• No LinkedIn scraping. The only LinkedIn touch is a search-engine check of whether a matched role is also listed there, which yields the "not on LinkedIn" badge.
• No guess about whether a company already has a chief of staff. A search-based check was tried and returned unrelated people, so the radar only trusts "first chief of staff" language in a posting itself.
• No résumé parsing, no applications, no salary estimates.
• No data about anyone except Sophie.

Cadence and cost

A GitHub Action runs every Sunday at 17:00 UTC: it re-crawls the job boards, re-scores, regenerates these pages and sends the digest by email. The run costs a few euro a month, almost all of it the model reads of surviving postings. The expensive part is the two hours of calibration, rating 25 real postings, which is also the part that makes the matcher hers rather than generic.

Attribution

Company data comes from the Amsterdam startup map by Dealroom.co. Postings come from each company's own job board. Headcount and funding facts for the dream-list companies are read from public news and company statements found through web search.