Selective Peer Disclosure and Reputational Forecasting

Organizations often collect forecasts independently but control peer feedback before final reports are submitted.

I study two career-concerned experts who lodge locked binary forecasts; one expert then receives a second signal and may observe the other's lodged forecast before revising.

A platform commits to a report-contingent reveal-or-silence rule.

Every policy admits a completely mixed babbling equilibrium, so the unrestricted equilibrium correspondence yields no strict policy ranking.

The design analysis therefore conditions on regular-monotone sequential equilibrium, a maintained selection rather than a refinement or implementation result.

Within this class, randomized silence has a likelihood ratio that governs low-type forecast inertia.

A concealment-ray theorem shows that the candidate value of every finite downstream decision problem and the locked-report incentive slacks are affine along rays, reducing the policy search to two boundaries and full disclosure.

At an exact rational profile, a computer-assisted certificate identifies a unique classification-optimal policy within the reveal-or-silence class: never reveal disagreement and reveal approximately 71.08 percent of agreements, just enough for silence to eliminate low-type inertia.

It raises accuracy by approximately 1.37 percentage points over sealing and 0.30 percentage points over full disclosure.

After recalibrating the threshold, strict feasibility and unique optimality persist on an open set.

The polar experiments are Blackwell incomparable, so the policy conclusion is objective-specific.