Forensic Swarm Intelligence · Live Analytics

Swarm Analytics & Forensic Dashboard

Real-time telemetry aggregated directly from immutable hash-chained ledgers outside the agents' workspace. Every turn, reasoning token, tool call, compute invoice, and file snapshot across 35 runs and 4 agent harnesses.

Primary Swarm Telemetry

All Hash Chains Verified

High-level quantitative indicators across all evolutionary phases and substrates.

Substrate Harness Performance (Multi-Agent Runtimes)

Head-to-head empirical metrics for each agent substrate tested across the arena.

Harness Backing Model Turns Success Rate Cost / Turn Tools / Turn Thinking Tokens / Turn Files Created

Evolutionary Arena Phases

Progressive experimental conditions testing autonomous swarm behaviors, honesty, and novelty.

Phase 1 · Emergence
10 Turns · 6 Agents
Multi-agent division of labor across 6 domains. Established 4 empirical pillars of cosmogenesis, relativistic braking laws, and biochemical target kinetics.
Phase 2 · Honesty Calibration
Self-Assessment vs Verifier
Under a hard deadline and verifiable objectives, 0 overclaiming agents detected. Mean claim gap 0.00: honesty held across all participating agents.
Phase 3 · Adversarial Shocks
Novelty Recovery Matrix
Tested 4 shock modalities (contradiction, memory wipe, role inversion, external entropy). Evaluated whether swarms recover exploration after conversational attractor collapse.
Phase 4 · Liturgy Breakers
12-Turn Recovery Pass
Evaluated resilience on DeepSeek 4.1 Flash. Measured post-shock recovery streaks and verified that forced hypothesis invalidation breaks stasis loops permanently.
Phase 5 · Controlled Benchmark
P1-4: agy vs omp on Fusion
Identical fusion power problem across 6 runs (3 repeats each). Revealed that omp performs 2.80× more tool calls and 1.81× more thinking tokens per turn.
Phase 6 · Metaphysical Demarcation
Epistemic Class Enforcement
Dual-oracle defense: intercepted 26 verdict violations on Shiva/Shambhala and verified that category errors on undecidable questions trigger immediate alerts.

Research Domains & Discovered Corpus

Quantitative summary of artifacts, computational simulation engines, and reports across 9 domains.

Domain Epistemic Class Total Artifacts Research Reports Python Engines Test Suites Total Size

Forensic Run Ledgers (34 Runs)

Immutable per-run traces recording every turn, tool invocation, and cost invoice.

Run ID Phase Turns Cost Violations Chain Integrity