Recruiting radar

This is an internal sourcing screen for whoever at Vercel is hiring senior frontend and product engineers. You pick a city and a role; it shows companies in that area that have had something go wrong recently, ranked by how heavy and how fresh the signal is, with the people who do that job listed underneath each one and the triggering article shown as proof. The copy is deliberately careful: these are people having a bad quarter, not leads, so the page opens with a note saying so and the only action on a person is a shortlist button. Nobody in this session named a city or a role, so I chose them: the San Francisco Bay Area, and senior frontend / product engineer (React, Next.js, TypeScript), which is what Vercel actually hires for.

Everything on the screen came from Trayo. trayo_find_companies returned 100 Bay Area software and internet companies of 150-10,000 people; after dropping rows with no website and obvious non-companies, 78 were imported as accounts. I wrote four new signals for this job - layoffs, acquisition, restructuring or leadership exit, and financial or operational trouble - and ran one discovery over all 78 accounts with a 90-day lookback. It settled after 15 minutes with 12 events at 11 companies and no blocked signals. The layoff signal, the heaviest one, fired at zero companies: the window was quiet for cuts, and the page says so rather than hiding the empty filter. trayo_search_stakeholders_batch, run against all 78 companies under a hiring brief rather than a buying committee, returned 253 matching engineers at 68 of them; 22 of those sit at the companies with a signal and are shown, four per company. Tenure is not a field the stakeholder search returns, so trayo_research_person_batch supplied each person’s dated role history and the app derives the tenure from it - all 22 resolved.

Two of the twelve events are left off the radar and listed on the Method page with the reason: one matched the wrong company (the article is about Lattice Semiconductor, the account is lattice.com) and one was labelled an acquisition when the article describes Lime losing a city scooter contract. Showing a wrong trigger as proof seemed like the worst failure for a screen about people’s jobs. Shortlist is the only stubbed thing in the app - no ATS, CRM or mail is connected, so it raises a preview toast describing what would have been sent and writes nothing. On the palette: Vercel’s brand gave one colour, #171717, so brandFromWorkspace made it the primary and applyBrand reported three adjustments - no background colour was supplied so a white page was assumed before the dark theme was applied, #171717 has no hue so dark mode fills accents with near-white #e8e8e8 instead of lifting it, and the top bar takes a deep #171717 step. Those notes are repeated on the Method page.