When the recommendation set collapses

Why this matters

Most companies assume that appearing in the answer means being in the running. This run tested that assumption and it did not hold.

An evaluator will name several vendors when the question is generic. The commercial question is what happens next, once the buyer describes who they are and what they need. If the field can resolve to one company at that point, then presence in the generic answer is close to meaningless. The competition that decides the outcome happens in the step after it, and it can finish before a buyer contacts anyone.

For every company in the category except one, that loss is invisible. There is no lost deal to review, because there was never a deal.

What we observed

In clinical trial management software, the field did not stay open.

As a buyer profile and commercial requirements were introduced one at a time, the recommendation narrowed from several vendors to a single final recommendation. Advarra OnCore was the vendor that persisted, surviving all three draws.

The narrowing also happened sooner than expected. The decisive move came as soon as the buyer profile was introduced, not at the end after every requirement had been added. Each requirement after that point was an opportunity to be removed, and removal did not require a competitor to be better overall.

Related runs

E-040 ran a comparable sequence in project-management software and the field did not collapse to one vendor. Two categories, two different outcomes. Whether that difference is driven by category is an open question, not a finding.

E-002 remains the only run in the program with a causally verified result. This one is not causal, and nothing here changes that count.

How we tested it

One evaluator, Google AI Mode, on a single surface. The run began from an unqualified category recommendation, then introduced a buyer profile and commercial requirements in sequence, observing how the field changed at each step. Three draws.

Instrument details and scoring are retained internally.

What we cannot say yet

This is a narrow result and should be read narrowly.

One evaluator, one surface, three draws, one buyer profile, and no removal test. Nothing here is causal: we did not remove a requirement and watch the outcome change, so we cannot say what forced the collapse. It does not generalize beyond this category and these conditions, and it says nothing about how a different evaluator would behave.

Closed means the run is complete. It does not mean the result is a universal truth.

Created from the Upstream Zero research record (session 2026-07-18). Raw outputs, instrument details, and scoring are retained privately and are not published.