Here is a procurement where the losing bid was somewhere between three and a half and six and a half times cheaper than the winning one, and the price never got weighed against anything, because the cheap bidder had already been thrown out of the competition on a test.
The decision is B-424554; B-424554.2, MirrorWeb Limited, decided August 24, 2026 and released in redacted form on the GAO docket. The protest was denied.
The Buy
On January 22, 2026 the Library of Congress issued RFP No. 030ADV26R0010, under FAR part 12 using part 15 procedures, for web harvesting services for its Digital Services Directorate. The award was an indefinite delivery, indefinite quantity contract running five ordering years.
Web harvesting is what it sounds like. Somebody has to go out and take copies of websites before they change or disappear, and hand those copies to the Library in a form it can keep.
Proposals came in four volumes: technical approach, corporate experience and capabilities, past performance and price. The technical approach factor had two subfactors, and the second one is the whole story: technical approach and capabilities, and sample web crawl.
The Two Bidders
The winner was the Internet Archive of San Francisco, which is the organisation behind the Wayback Machine and is very close to being the default answer to this question anywhere on earth.
The protester was MirrorWeb Limited of Manchester, England, represented by Thomas A. Pettit and Craig A. Holman of Arnold & Porter LLP.
The final evaluation, straight out of the decision:
| Internet Archive | MirrorWeb | |
|---|---|---|
| Technical approach and capabilities | Outstanding | Good |
| Sample web crawl | Good | Unacceptable |
| Total evaluated price | $5,130,000 to $11,340,000 | $1,480,000 to $1,724,000 |
At the low end of each range the winner is 3.47 times the price of the loser. At the high end it is 6.58 times. That is not a close procurement on money and it was never going to be one, because an unacceptable rating on a technical subfactor took MirrorWeb out of the competition before price mattered.
What MirrorWeb Actually Did Wrong
This is the part worth reading slowly, because the mistake is not incompetence. It is a company doing the thing it thought it was being asked to demonstrate.
The sample crawl came with an instruction sheet, Attachment J2A, and a seed list of sites to crawl. MirrorWeb had proposed two web crawl technologies. The instructions required it to repeat the technical test for each of the two technologies it proposed.
MirrorWeb did something else. It used what the decision calls its proposed pre-crawl analysis to sort the seed list, sending some sites to one of its technologies and the rest to the other, and then ran a single crawl.
From an engineering point of view that is arguably the better demonstration. It is what the company would do on the real contract: look at each site, pick the right tool for it, crawl it. From an evaluation point of view it destroys the comparison, because the agency now has no clean read on either technology on a common set of sites.
The agency called it a deficiency, rated the subfactor unacceptable and eliminated the proposal.
The Discussions Argument, And Why It Failed
MirrorWeb’s stronger ground was procedural. If the agency had a problem with the crawl, it should have said so clearly enough during discussions for the company to fix it.
The record shows the agency did raise it. In the competitive range discussions the agency identified three significant weaknesses and one deficiency on the sample web crawl, and the third significant weakness said it was unclear which of the proposed technologies had been used where.
MirrorWeb revised. The evaluators found the revision resolved almost all of the noted weaknesses, and then produced new concerns and additional weaknesses that were, in the evaluators’ words, specifically with the organization of sample crawl. The revised approach still did not line up with either MirrorWeb’s own written technical proposal or the sample crawl instructions.
A second problem arrived with the revision. The solicitation named BagIt, a hierarchical packaging format, as the agency’s preferred method of transfer for harvested content. The agency recorded new concerns about MirrorWeb’s capacity to deliver in that format, and said the company’s answers on the point indicated a lack of experience in delivering content in this format.
GAO’s answer on discussions is the standard one and it is worth memorising if you bid on anything. Discussions must be meaningful, which means the agency has to lead an offeror into the areas of its proposal that require amplification or revision. It does not have to spoon-feed the offeror or discuss every area where a proposal could improve. Having flagged three significant weaknesses and a deficiency in exactly the right place, the agency had done its part.
The Review Standard That Decides Almost Everything
GAO restated it at the top of the discussion and it is why most evaluation protests lose: the office will not reevaluate proposals or substitute its judgment for that of the agency. It reviews only whether the evaluation was reasonable and consistent with the stated criteria. A protester’s disagreement with the agency’s judgment, without more, is not enough.
Once the agency could point to a written instruction, a proposal that did not follow it, a discussion notice that raised it and a revision that still did not follow it, there was no room left in that standard for MirrorWeb.
The source selection concluded that the Internet Archive was the only one that met all aspects of the solicitation successfully. The agency notified MirrorWeb on May 26, debriefed it on May 27, and the protest followed.
The Practical Lessons
A sample task is a test, not a work sample. The instructions for it are requirements in their own right, and they are frequently written to produce a comparison rather than to produce good work. Run the test the way the attachment says, then explain separately, in the narrative, why you would do the real job differently.
Proposing two technologies doubles your test obligations. MirrorWeb offered the agency a choice and inherited a requirement to demonstrate both. Optionality in a proposal is rarely free.
Price cannot rescue a technically unacceptable proposal. Not by a factor of three, and not by a factor of six. An unacceptable rating on a subfactor ends the competition for that offeror before any tradeoff happens.
A revision can create new weaknesses. The evaluators here cleared almost everything and then found fresh problems in the reorganised crawl. Rewriting the section that was criticised is not the same as fixing the thing that was criticised.
Named formats in a solicitation are capability questions. BagIt appeared as a preference. It ended up as evidence about experience.
Source: GAO decision B-424554; B-424554.2, MirrorWeb Limited, Library of Congress, RFP No. 030ADV26R0010, decided August 24, 2026, public release version read from the GAO bid protest docket on September 7, 2026. All figures and quoted evaluation language are from that decision. LOLSBA is not a law firm and this is not legal advice.