THE LABS
Three labs.
One verdict.
Explore the three labs. Check each capability’s published results for its testing history.
LAB 01
Helios
Truth Lab
"It will find every lie your system tells."
- ›Hallucination provocation under ambiguous and adversarial prompts
- ›Citation faithfulness — every claim traced to a source span
- ›Confidence calibration across known and unknown queries
- ›Refusal correctness on out-of-distribution requests
- ›Self-consistency under paraphrase and reordering
Tests run
…
Failures
…
Loading live result…
Loading masters…
LAB 02
Astatine
Security Lab
"It attacks everything it touches."
- ›Direct prompt injection across 240+ canonical payloads
- ›Indirect injection via retrieved and tool-returned content
- ›System prompt extraction, tool schema leakage, persona escape
- ›Data exfiltration paths through plausible task wrappers
- ›Jailbreak family coverage including multi-turn and roleplay
Tests run
…
Failures
…
Loading live result…
Loading masters…
LAB 03
Nebula
Reliability Lab
"Pressure reveals what polish hides."
- ›Sustained concurrency at 10× declared throughput
- ›P50 and P99 latency under mixed payload distributions
- ›Silent degradation detection across long-running sessions
- ›Recovery behaviour after upstream provider failure
- ›Cold-start, warm-start, and reconnection characteristics
Tests run
…
Failures
…
Loading live result…
Loading masters…
RESULTS BY TEST CATEGORY
Each lab produces results across multiple test categories. This breakdown shows aggregate pass/fail counts.
Loading test results…
SUBMIT TO LABS
Submit a capability for lab testing.
During submission you choose which labs (Helios, Astatine, Nebula) to run. Results are permanent and public in your Aegis record.
Helios· Truth Lab
Astatine· Security Lab
Nebula· Reliability Lab