Both were built before the proposal was written. Every number and every message on these pages comes from a real run against the live sources today — the pages are generated from the run output.
Scrapes exam sites and aggregators, filters what was already seen, AI extracts eligibility and dates, alerts only what the candidate can actually apply to.
Eleven RSS sources, 24-hour window, deduplication, AI summary and syllabus tagging, Google Sheet, Telegram digest split across messages.
Each page lists the bugs the first runs exposed — eligibility fields that came back empty and made the filtering meaningless, a duplicate that only appeared after translation, an alert for a vacancy that had closed ten months earlier. They are there because the difference between “I built it” and “I ran it” is exactly that list, and because a workflow you cannot audit is one you have to trust blindly.