Methodology
How tenders get here, how completeness is measured, and what we do not claim.
A crawl that finishes without error but accounts for fewer tenders than the source itself declares is recorded as a failed crawl — not a successful one.
Why this matters
Every tender portal misses things sometimes: a website changes, a certificate expires, a server goes down mid-crawl. The difference that matters is whether the platform knows it missed something. If it does not, neither do you — and a tender service you cannot audit is one you cannot rely on.
How the check works
Nearly every procurement portal publishes its own running total. Federal PPRA prints “Showing 50 of 1,408 tenders”. Punjab prints “785 items in 8 pages”. EPADS prints an active-listing count. On every crawl we parse that number and reconcile it against what we actually captured and stored.
- Fetch every page of every facet the portal exposes.
- Read the source's own declared total from the page.
- Count what was parsed and persisted.
- If captured < declared, the run is marked gap and an alert fires — even though nothing technically errored.
Why our count is often higher than the declared one
We crawl each portal along every filter it offers, not just its default list. On federal PPRA that means all four tender types and all three procurement categories. Records that never appear in the headline listing still surface under a secondary filter — currently 1,726 captured against a declared 1,408, a 22.6% difference a single-pass scraper would never see.
Other coverage guarantees
- Raw archival. Every page and document is stored before parsing, so a website redesign can be replayed offline instead of costing a day of tenders.
- Never delete. A tender that disappears from a source is marked closed, never removed. History is an audit trail.
- Sequential-ID sweeping. Portals with numeric detail IDs get a nightly sweep across the observed range, catching records their own search index omits.
- Change detection. Content hashes surface corrigenda — deadline extensions and cancellations — rather than silently overwriting the old values.
What we deliberately do not claim
No platform can honestly promise every tender in Pakistan. Some private opportunities are invitation-only. Some portals block automated access or go offline for days. Newspaper scans carry copyright constraints. Anyone promising 100% is either misunderstanding the problem or misrepresenting it.
What we commit to instead is measurable coverage of a published source list, with no silent failures — a claim you can audit on the coverage page and against the source list.
Data quality notes
- Deadlines are shown in Pakistan Standard Time, because that is the timezone they are enforced in.
- Publication dates carry no time. Portals publish a date only; printing 00:00 beside it would imply a precision the source never gave.
- "Bid security" is not contract value. Federal PPRA does not publish estimated contract value, so we label the earnest-money figure for exactly what it is rather than presenting it as a budget.
- Cities are canonicalised. Sources write “Karachi”, “Karachi.” and “karachi”; these are merged so a filter does not split one city across three entries. Genuinely ambiguous values are left unresolved rather than guessed.
Attribution
Government tender notices are public-interest information published expressly for wide circulation — the same basis on which the EU operates TED, the United States operates SAM.gov and the United Kingdom operates Find a Tender. Every tender here links back to its issuing source, and we crawl politely and identify our crawler honestly.
12,800 tenders from 54 actively crawled sources · 77 tenders unaccounted for · see the live reconciliation →