Representative case

The current site uses the LK-99 news cycle to show multimodal input, cross-platform simulation, and report generation in one example.

Experiment Dashboard LK-99 Case

LK-99 case overview

This example combines article-style text, videos, structured outputs, and a clear progression from claim to debate to correction.

Multimodal input Multi-platform Report output Claim to correction
Input package long-form analysis + multiple videos
Story arc from excitement and speculation to verification and disillusion
Audience coverage researchers, science communicators, investors, and ordinary users
Narrative shift the run captures how public discussion changes once replication evidence arrives
Input Style Multi-modal

One run can combine explanatory text with several commentary videos.

Narrative Shape Escalation then correction

The example begins with excitement and gradually shifts toward verification.

Output View Report + details

Each run keeps the final report together with the simulation records.

Output Style Inspectable

Prompts, traces, reports, and graph artifacts remain available after the run.

Domain Science communication and public reaction Input package Long-form text + multiple videos Narrative Hype, replication, correction Audience Experts, media, and the general public
Opinion Trend Simulation Output

Discussion role evolution

Weighted activity by role across 24 simulated hours, showing how the claim, amplification, verification, and public interpretation evolved.

Twitter
Reddit
Weighted action score post ×3 · comment / quote ×2 · repost / reaction ×1
Signal Summary

Platform dynamics

The same event unfolds differently across platforms: excitement spreads quickly, evidence is contested, and the tone shifts again as technical discussion accumulates.

Twitter fast emotional swing

hype, reposting, and social amplification show up first

good for watching excitement spread
Reddit slower evidence sorting

technical questioning and source-checking become more visible

good for watching correction take shape
Influence Reading

Different roles in one event

The runtime separates amplifiers, interpreters, and correctors within the same event so role differences remain visible in the trace.

These inputs anchor the case in concrete source material rather than abstract prompts alone.

Multi-stage story claim, hype, replication, correction
Multiple voices labs, commentators, media, and ordinary users
Cross-platform reading fast reaction versus slower source-checking
Traceable outputs report and simulation details stay available after the run
A

Detailed process

The run keeps platform actions, memory updates, and topology traces instead of only a final summary.

B

Readable result

The LK-99 case is understandable to a public audience: a surprising claim spreads, evidence accumulates, and the overall mood shifts from excitement to caution.

C

Available outputs

The project keeps prompts, graphs, configs, summaries, and topology snapshots for review.

D

Case fit

One case links multimodal inputs, topology-aware interaction, and readable outputs without needing extra setup.

Input to graph

The LK-99 package is parsed into event text, media references, ontology terms, entity descriptions, and relation edges before simulation begins.

Current workflow

Runtime artifacts

The same run emits prompts, social-relation graphs, simulation configs, topology traces, and the final summary for inspection.