Distributed Discovery

Collective search under dispersed information

Distributed Discovery

How groups turn evidence into portfolios of action.

Better shared information can produce worse collective discovery.

A group can improve its best guess and still waste its attempts by sending everyone to the same place. This project studies how evidence, roles, incentives, and timing shape a portfolio of search actions.

Start with three resultsExplore the researchSee the program map Browse Labs Read papers

How the system works

From evidence to discovery

A discovery architecture determines more than what a group knows. It shapes who sees evidence, which actions are taken, and what the group learns next.

  1. 01Acquire evidenceChoose sources and measurements.
  2. 02Share evidenceDecide who can see which signals.
  3. 03Choose actionsSpread attempts across the search space.
  4. 04Learn and adaptUse outcomes to update the next move.
  5. 05Reward discoveryAlign reporting, attention, and effort.
  6. 06Measure outcomesTrack coverage, quality, and discovery.

Key findings

Where architecture changes the result

Shared Discovery Paradox

One better answer can produce fewer discoveries.

0.383468709731 discovery under one shared answer in the canonical fixture.

See the interactive guide

Common-Source Trap

Many reports can still come from one source.

p(1−p)/N is the exact all-common trap width in the frozen model.

Read the paper

Truthful discovery mechanisms

Rewards can support truthful, differentiated action.

16 strict rows in the registered exact mechanism class.

Open the study

Architecture Atlas

Coherent designs reveal real tradeoffs.

12 nondominated cells in the bounded synthetic Atlas.

Explore the Atlas

Key finding · The Incentive to Ignore

One reader can beat a broadcast.

In the frozen shared-versus-private action class, the first reader can improve discovery while duplicate shared-signal use reduces it.

Read the paper

Key finding · Threshold Discovery

Form the smallest viable teams, then diversify the teams.

When candidates open only after a minimum team forms, too little overlap strands effort and too much overlap crowds the portfolio.

Read the paper

Explore the work

Choose your depth

Research

Browse every completed, active, and planned study.

Labs

Change inputs and inspect precomputed model behavior.

Papers

Read the long-form arguments and validated artifacts.

Share the evidence. Diversify the actions.

When actions require teams, form the smallest viable teams and diversify those teams.