What comparable work actually closed at

You quote four per cent below the estimate and lose to a firm that quoted eleven below. Or you quote eighteen below, win, and then fund fourteen months of work on a margin that was already gone at bid opening. Both mistakes come from the same place: pricing against the buyer's estimate instead of against what that buyer has actually paid.

The estimated value in a notice is prepared inside the department, usually off a schedule of rates, before a single bid arrives. It is a starting point, not a market price. The market sits either side of it, and where it sits varies by buyer, by category and by how many firms are chasing work that month. A contractor who has bid the same circle office nine times carries that range in their head. Everybody else guesses.

Award data closes that gap, because the buyer publishes the answer after financial opening. It names who won and at what value, and that value can be set against the estimate carried in the tender document. The method for reading it is in what past award data tells you before you bid, and the costing that should sit behind any discount is in how to price a government tender.

Four fields do the work. The name of the successful bidder. The accepted value. The estimated value it is measured against, which comes from the tender document and gives you the discount. And the number of bidders who quoted, which tells you what kind of market you were in.

From those four, on a large enough set, you get the things worth knowing before you commit an EMD. The band of discounts a particular buyer has accepted in your category. Whether their fields run to three bidders or fourteen. Which firms appear in the winner column repeatedly, and whether a handful of names take most of a department's work. Whether the awards cluster tightly, which suggests everybody is costing off the same schedule of rates, or scatter, which suggests they are not.

One result is an anecdote. Twenty results from the same buyer in the same value band is a distribution, and a distribution is something you can quote against. This module exists to build that distribution and show it with its sample size attached, rather than reduce it to a single confident number.

Avsar does not carry the award corpus today. That is a data access problem, and it is worth being precise about it.

Live tenders are published as paginated public data feeds. Avsar reads those feeds, across NIC's Central Public Procurement Portal, the state portals NIC aggregates and GeM, every minute, which is why the live listings are current. Results are published differently. The result of tenders section sits behind a search form that issues a captcha token per search, so there is no list to walk and no feed to read. It is a lookup, one tender at a time.

The choice at that point is to estimate the numbers or to publish nothing. Estimating a discount, inferring a winner or averaging across categories that are not comparable would produce a page that looks like intelligence and behaves like noise. So the awards listing and the competitor pages stay empty, and are kept out of the sitemap until real results back them. What Avsar does and does not index is listed on the status page.

The mechanism matters more than the promise, so here it is. Comparables are selected in a ring: same category first, then same buyer where the sample supports it, then same value band. If that ring returns two awards, the median of two is a coin toss dressed up as analysis, so the ring widens and the page states that it widened and by how much.

The output is a band rather than a number. The interquartile range of comparable awards, with the sample size beside it, so a thin set is visibly thin. Alongside it, the field size those awards drew, and the concentration in the winner column, because three names taking most of a buyer's work usually means there is an incumbent, and a week spent on the pack will not change that.

Everything is built from notices the buyer published after financial opening. Nothing comes from private records, and no bidder submits anything to appear in the corpus. Where a read cannot be supported, the product says so, in the same way the eligibility engine flags a criterion it could not resolve instead of guessing at it.

Two tenders at the same estimated value, one drawing three bidders and one drawing fourteen, are not the same market, and the same discount is the wrong answer to both. In a crowded field a cautious quote never reaches the financial round. In a thin field the same aggressive quote hands the buyer margin you did not have to give.

That is why bidder count sits next to the discount in the panel above rather than in a footnote. It is also why the evaluation method has to be read first. Under lowest cost, price is the whole contest. Under quality and cost based selection, a technical mark can be worth more than a rate cut, which is covered in L1 versus QCBS. Where the tender ends in a reverse auction, the number you must know is your walk away floor rather than your opening quote, and the reverse auction floor price calculator works it out in your browser.

A quote far below the rest of the field can also invite a justification notice, which abnormally low bids sets out.

It will not tell you what to quote. A median is a fact about other people's contracts, not a decision about yours, and the cost of your own labour, plant and finance appears in no award notice.

It will not model a competitor's cost base. An accepted value tells you what a firm was willing to sign for, not what the work cost them or whether they made money on it. Reading a rival's discount as a benchmark for your own is how a bad year starts.

It will not produce a number when the sample cannot support one. A thin set is labelled thin, and where there are no comparables at all, the page says so.

Avsar also never submits a bid and never holds your digital signature certificate. Uploading a bid stays a deliberate act you perform on the portal yourself, by design.

One further limit, because it affects every module. The portal list feeds carry the notice, not the commercials. Estimated value, EMD and the eligibility clauses sit inside the tender document, which is why the eligibility engine reads the PDF and cites the clause and page for every verdict rather than trusting the feed.

The rest of the platform is not waiting on the award corpus.

Search across every indexed portal is free and needs no card, at live tenders. Add your three year turnover figures and one completed work order to a workspace, which takes one sitting with the balance sheet and the work order in front of you, and the eligibility engine starts returning verdicts on the tenders you were already reading, with the clause and page cited on each one. Alerts fire when a matching notice is published, including corrigenda on anything you are watching.

On the capital side, the EMD and PBG register tracks what you have paid out, when each refund fell due and which guarantees are still charging commission. Working capital planning around EMD explains why that figure decides how many tenders you can hold open at once.

There are 52 guides and 8 calculators, and every calculator runs entirely in your browser, so no figure you type is sent anywhere. Start with the bid or no bid framework if the question is which tender deserves the week.