Zeno Divergent

Next experiment · not started

Eight say 80%. Is that eight sources, or one?

Agreement can't vouch for itself. Zeno's next test: withhold one input at a time, see who moves, and pick the one observation that would settle the decision.

80%80%80%80%80%80%80%80%one faulty input breaks all eight
Illustration, not data. Same eight numbers in both pictures — the numbers alone cannot tell them apart.

the test

Four ways to spend the same budget.

  1. Zeno picks

    The observation with the highest expected decision benefit after its cost, or none.

  2. Random

    Any affordable observation.

  3. Fixed rule

    Always the nearest station.

  4. More research

    Plain extra reading, same budget.

One weather alert decision, the same events for every arm, at least 200 independent site-days. Costs, delays and false alarms count. If Zeno's pick doesn't beat all three rivals, the idea is rejected here, in public.

Related work and limits
  • Close prior work exists: shared-information elicitation, CastFlow (adaptive evidence for ensembles, 2026), and budgeted answer/abstain/acquire studies (2026). We claim only the narrow question above.
  • Withholding a source is possible only where we control the inputs. Outside forecasters (markets, labs) get an observed-dependence estimate, labelled weaker.
  • Forecasts made after seeing Zeno never count as independent proof of Zeno.
  • "Several missed on the same side" is an outcome, not a shared cause. It counts as a blind spot only if it beats a rival that knows event difficulty and each forecaster's past accuracy.

All research →