Why do some companies consistently make the shortlist while others are never evaluated?

Making the shortlist is not a ranking outcome; it is the output of an evaluation. That process — surfacing, screening, comparison, validation — is commercial evaluation, the discipline this observatory studies. Visibility can get you surfaced; it does not carry you through the screening that decides whether you are evaluated at all.

Run enough of those evaluations and the same layer keeps appearing underneath them: requirements — the specific conditions a buyer needs satisfied. An RFP is a requirements list; a procurement workflow is a requirements filter; an AI screening a vendor is matching it against inferred requirements. The interfaces differ; the structure does not.

This is why the pattern is consistent rather than random. The companies that recur on shortlists are the ones whose fit against the operative requirements is easy to recognize and verify; the ones that are never evaluated are the ones whose fit cannot be reconstructed from what an evaluator can reach. Models and interfaces change; the requirements persist — which is what makes this worth studying rather than chasing.

Honestly, this answer is a founding position, not a demonstrated result. The observatory has published zero observations, and the claims beneath this answer sit at the lowest evidence tier:

• Portions of commercial evaluation have become observable through AI evaluators' behavior
  Narrated · 1 of 6
• Organizations will increasingly maintain a machine representation alongside their human representation
  Narrated · 1 of 6

Whether AI evaluation reaches human shortlists at all runs through the open bridge hypothesis H-1, and the first experiment against this framing is pre-registered in draft as EXP-0001. The full ledger is at Claims.

It does not establish which requirements dominate any specific category, that AI evaluation resembles or drives human committee evaluation (that is H-1, held as a hypothesis), or that changing what an evaluator can verify changes the outcome. No intervention effect has been measured. What would change the answer: published observations of evaluator screening that contradict the requirement framing, or stability results too noisy to attribute to evaluation at all (Q-3).

If this is your situation, the measurable starting point is observation, not optimization: capture how evaluators currently assess you, which requirements they appear to credit, and where the gaps are that keep you from being evaluated. That is what an engagement measures — it does not promise to change any evaluator's behavior, and the answer above is meant to be useful even if you never work with us.

Upstream Zero is the research company conducting this work — the institution behind the discipline, not the subject of it.

This is an editorial navigation page, not a research object. Its answer is informed by the claim C-0001 and relates to commercial evaluation, which observes requirements. The bridge hypothesis H-1 and experiment EXP-0001 carry the open evidence. The full machine graph is graph.json.

• Which requirements are eliminating us from consideration?
• Why are we difficult for AI systems to understand?
• What is the relationship between AI evaluation and human buying-committee evaluation?