Corrections

2026-08-28 · /upprpb-constable-2025-07/ · wrong for about 2 hours

What we showed: The card showed 'Applications open 17 August – close 31 August 2026' for the UP Police Constable drive (Advt 07/2025, 32,679 posts).

What is true: Those dates are the document-verification and physical-standard-test window for candidates already shortlisted. Applications for this drive closed in 2025. The page now shows the DV/PST window as a timeline event and no application window.

Why it happened, and what changed: Our extractor read the phase window of a DV/PST notice (a Hindi document with a corrupted text layer) as the application window. The rule that phase windows are never application windows now exists in the extractor, and every drive within 7 days of a deadline gets its window independently re-read daily.

2026-08-30 · / · wrong for about 27 hours (29 August evening to 30 August 21:00 IST)

What we showed: Drives whose last date lay ahead but whose opening date we had not read were labelled 'closed' — 122 of them (IIT Kharagpur, IARI, UPSC, JNCASR, NABFINS, ESIC and others), and the home page counted 42 open drives.

What is true: A drive is open until its last date has passed, unless the opening date is known and lies ahead; where only the last date was read the page says so. Where only a last date more than 60 days out is known (UPSC's 2027 programme rows) the page says 'Closes …' and does not count it as open. 152 drives are open.

Why it happened, and what changed: Our status rule required both dates. Many recruiters print only a last date. Reported by a reader on 30 August.

2026-08-30 · / · wrong for up to 33 hours for the cap-failed pages (from 29 August morning), a few hours for the portal-page drives

What we showed: About 200 drive pages existed with a title and little else: 194 founded from documents our reader had filed as read when the model budget had stopped every question unanswered, and 3 founded from recruiters' portal index pages (one showed a 'Chairperson' drive closing in 2028 — the date of a defaulter-firm notice on the same page).

What is true: Those pages are withdrawn and answer 410 Gone. The documents behind the 194 are queued to be read properly and return under the same address where the post name matches.

Why it happened, and what changed: A budget stop was recorded as a read; a whole portal page was treated as one notification. Both rules are fixed.

2026-08-31 · /iari-junior-research-fellow-2026-2/ · wrong for about 18 hours (30 August afternoon to 31 August 07:30 IST)

What we showed: Three ICAR-IARI drives showed an application deadline up to a week later than the truth — one said applications closed 7 September when the advertisement said 'latest by 31st August 2026', with 7 September being the walk-in interview date.

What is true: The deadline now comes from the advertisement itself. Where IARI's own 'Last Date' column holds the interview date rather than a closing date, the page shows the advertisement's date, or none at all if the advertisement states none.

Why it happened, and what changed: IARI's jobs table is labelled 'Last Date', and for walk-in rows that column carries the interview date. Our loader was first-writer-wins, so the table's date stood against the document's. The table's date is now a hint only; the document decides. Found by the 31 August daily run, which re-reads every drive near its deadline and disputed these two.

2026-08-31 · /gail-important-instructions-before-you-apply-online-2018/ and eight others · wrong for some since 2026-08-27; all withdrawn 2026-08-31

What we showed: Nine pages were published as recruitment drives that were not recruitments. Seven were GAIL's "IMPORTANT INSTRUCTIONS BEFORE YOU APPLY ONLINE" — the instructions annex that sits beside a real GAIL advertisement — and one was Western Railway's notice about viewing your application status and changing an allotted exam group. Each showed a closing date and no post, no vacancy count and no eligibility, because there were none to read: they are not advertisements.

What is true: They have been withdrawn. A document whose own heading says it is instructions or a status notice, and which carries no read fact, no longer founds a drive. Their URLs return 410 Gone rather than quietly disappearing, and the documents themselves are still in the archive — what changed is that they are no longer presented as jobs. Records that carry a real fact are untouched: two ESIC advocate-empanelment notices state a qualification and a deadline and remain, because whether an empanelment is a recruitment is a question about the document, not about its title.

Why it happened, and what changed: The identity task read each annex as an advertisement — it is published on the recruitment page, beside the advertisement, and carries a date. Nothing downstream asked the obvious question: a recruitment with no post, no vacancy count and no eligibility is not a recruitment. That check now runs at load.

2026-09-01 · the home page, /open.json and every drive whose last date was 31 August 2026 · wrong for the small hours of every night since the site went live on 29 August 2026; fixed 01 September 2026, 02:50 IST

What we showed: Between midnight and 5:30 in the morning, Indian time, this site listed about twenty-three drives as still open whose last date had passed, and offered a calendar reminder for at least three deadlines that were already behind the reader. The three found by the audit were IIT Kharagpur research posts that closed on 31 August; the home page counted 229 open drives when 206 were open, and /open.json stamped the feed with yesterday's date.

What is true: A recruiter's last date is an Indian date, and the site now reads the clock in Indian Standard Time on every machine it runs on. The drives that closed on 31 August read "Applications closed" from midnight IST, and the calendar link is gone from them. Nothing about the dates themselves was wrong — every one was the recruiter's own; what was wrong was the day we compared them against.

Why it happened, and what changed: The site is served from a machine whose clock is set to UTC, and the code asked Python for "today" — which answers with the machine's day, not the reader's. For the five and a half hours between midnight in India and midnight in Greenwich, the server was still living in yesterday. It went unseen because the laptop that runs the audits is on Indian time, so the same corpus passed every check locally and only failed when the audit was pointed at the live site.

2026-09-01 · /iari-young-professional-ii-2026-2/ · wrong for since 29 August 2026, when the document was first read; corrected 1 September 2026

What we showed: This page said applications closed on 30 September 2026. They did not. The recruiter's notification names no last date at all: it is a walk-in interview, held on 3 September 2026 at 10:00 AM at the Division of Genetics, ICAR-IARI, New Delhi. The 30 September we printed is the end of the CONTRACT — "the post is purely temporary and will be filled on contractual basis initially for up to 30-09-2026, which may be extended periodically". A reader trusting this page would have believed they had four more weeks, and the interview would have been over.

What is true: The date now reads 3 September 2026, and the page says in its own words that the document calls this a walk-in interview: attend in person on the day, there is no application window to meet. Both facts are quoted from the notification.

Why it happened, and what changed: A model read the document and took the contract's end date for the last date to apply. Nothing downstream disagreed, because the date was real, the format was right and it was in the future — every check we had was about shape. What caught it was reading the document again without a model and asking what the SENTENCE around the date says: "initially for up to 30-09-2026" is a tenure, and the only actionable date in the notification is the interview.