Six new cuts of the full 3.95M-tender, 3.17M-award corpus (last 15 years): which sectors account for the money, when tenders get published, how often they're amended, the minimum-bid boundary, bunching near the global-tender threshold, and signs of contract splitting. Built to answer a researcher's data request with public, reproducible analysis rather than a private hand-off.
Award value by sector, last 15 years. Sectors are a keyword classification over buyer names, not literal ministry names - a broad-strokes view, not an org chart.
State Govt (dept. unspecified) tops the list by both value and count - not a finding, a data gap: a large share of state-level award records capture only the state name, not the department. No amount of text-matching recovers a ministry that was never scraped. Other/Unclassified is a smaller, genuinely unresolved remainder (garbage/placeholder buyer names in the source). Finance/Banking (*) is dominated by periodic Food Corporation of India consortium cash-credit tenders - Rs 25,000-75,000cr lending ceilings for grain-procurement working capital, not one-time spend - so it is not comparable to the other rows.
Month-of-year distribution of tender publication dates - a different thing from the site's March-rush finding, which is about when contracts are awarded/paid. Publication timing turns out to be close to flat.
Neutral baseline if publication had no seasonality at all: 25.0% (3 of 12 months). The actual share sits close to that baseline - unlike award/spending timing, tender issuance doesn't show a strong year-end rush.
Share of tenders that had at least one corrigendum (amendment) issued after publication.
The source only links to CPPP's corrigendum page, not structured change data - this counts whether a tender was amended, not what changed or why. 2026 is a partial year and excluded from the trend above.
A valid tender process requires a minimum of 3 bids. Awards that land on exactly 3 are the boundary case worth watching.
Since 15-May-2020 (GFR Rule 161(iv), Dept. of Expenditure), no Global Tender Enquiry is required for contracts up to Rs 200 crore. If buyers structure awards to just avoid a GTE, you'd expect a spike just below Rs 200cr. This uses awarded value as a proxy for the tender's bracket - the real test is the pre-bid estimated value, which isn't captured anywhere in the source data.
Read honestly: this isn't a clean signal. Award counts in this band are small (dozens per Rs-5cr bin), and the data shows a small bump at Rs 200cr rather than a clear ramp just below it - more consistent with round-number pricing than deliberate threshold-gaming. Flagging as an open question worth more data, not a finding.
A cluster = 2-6 awards to the same vendor from the same buyer, all within 30 days of each other, where each award individually sits under one approval-threshold tier (Rs 10/25/50/100L - the same bins as the Integrity page's threshold heuristic) but the combined total crosses into a higher tier.
Heavy caveat, read before citing any row: an early version of this heuristic had no cap on cluster size and its top "finding" was a state oil PSU awarding the same vendor 482 separate Rs 1.1L contracts over years - a routine standing rate-contract relationship, not splitting. That pattern is now excluded by the 6-award cap. What remains below is a real pattern (same buyer, same vendor, tight window, suspiciously threshold-adjacent values) but is not verified against whether the awards are actually one project cut into pieces or genuinely separate small jobs. Buyer names are exact strings, not CIN-resolved. Treat every row as a lead, not a finding. Contract packaging (clubbing small work into one large tender to lock out small bidders) is out of scope - this data has no principled baseline for "unusually large for this category of work."