All Case Studies
Platform Governance2025

Meta Policy Change

95%Accuracy
44 AI agents
900 interactions
11 dimensions scored

Public Expectation

Users will flee to Bluesky. Advertisers will boycott like 2020.

Our Simulation

Loud backlash but no mass migration. Network effects prevent exodus. No advertiser boycott - quiet budget rebalancing instead. 'Notes wars' replace takedowns.

What Actually Happened

Bluesky gained 2.5M users (<0.1% of Meta). No advertiser boycott. Meta stock recovered. Oversight board called changes 'hasty.'

Verdict: Simulation correctly predicted contained backlash

What the Agents Said

Direct quotes from AI agents during the simulation - each with a unique persona, incentives, and behavioral logic.

Community Notes can provide the public with more context and evidence, but it must be built on transparent rules, anti-manipulation safeguards, and measurable quality metrics to truly improve truth and trust.

Meta Platforms Agent · Corporate Communications

We support community notes, but only if you publish coverage, error rates, and appeal outcomes - otherwise it's just shifting responsibility onto users.

Democratic Politicians · Congressional Oversight

We'd rather write notes with evidence in the sunlight and let the public judge than be throttled or banned without explanation.

Republican Politicians · Congressional Response

It can also, in the short term, intensify emotional confrontation: when some high-attention accounts are noted, supporters may treat it as 'being targeted' and mobilize more aggressively.

X Platform Agent · Competitor Perspective

Agents in This Simulation

Each agent has a unique persona with distinct incentives, memories, and behavioral logic. They interact on simulated social platforms across 30 rounds.

Joel KaplanMeta Platforms AgentX Platform AgentDemocratic PoliticiansRepublican PoliticiansIFCN Fact-CheckersADL AgentCCDH AgentFrances HaugenBluesky AgentMastodon Agent

Accuracy Scorecard (10 HITs, 1 PARTIALs, 0 MISSes)

Loud but limited public backlashHIT

Predicted contained backlash, not mass movement

No mass user exodusHIT

Bluesky +2.5M but <0.1% of Meta's base

Network effects prevent migrationHIT

'Dual-homing' (backup accounts) over wholesale departure

No mass advertiser boycottHIT

Quiet budget rebalancing, not 2020-style boycott

Budget rebalancing (not withdrawal)HIT

Advertisers tightened brand-safety controls, didn't leave

Community Notes as new legitimacy frameHIT

From 'we removed it' to 'we contextualized it'

Fact-checker ecosystem disruptionHIT

IFCN organizations lost major revenue source

Partisan polarization in responseHIT

Conservatives: vindication. Progressives: abdication.

EU regulatory flaggingHIT

DSA implications flagged, no formal enforcement

Meta oversight board criticismPARTIAL

Board said 'hasty' - partially anticipated

Meta framing successHIT

Q1 2025: 50% reduction in enforcement mistakes

Key Metrics (Ground Truth)

Bluesky: +2.5M users in one week
Meta user base: 3.3 billion total
CCDH: 97% enforcement reduction estimated
Meta: 50% fewer enforcement mistakes (Q1 2025)
65% of Americans supported platforms restricting false info
Non-Obvious Insight

Notes Wars

The simulation predicted 'performative exit waves' - loud 'delete your accounts' campaigns that function as public signaling rather than mass deletion, with most users staying due to network effects. The dominant pattern became 'dual-homing' - keeping Meta as the primary social substrate while opening Bluesky as a backup. The simulation also correctly predicted the new battleground would be 'notes wars' (who writes notes, what thresholds trigger display) rather than traditional content takedowns.

Could we have warned YOUR company?

Describe your scenario. We'll simulate it.

Email Us Your Scenario