⚡ LLM-BUSTER PROBE LAB
v0.10.1
[01] Runner
[02] Inspector
[03] Load JSON
[04] Proxy-Watch
● SERVER READY
Diagnostic Configuration
DEEP
Full Probe Suite
19+ tests: reasoning, refusal ladder, identity leak, quantization audit.
10S
Canary Fast Check
10 fast deterministic checks: math, recall, format adherence.
Base URL
API Key
Model ID
Claimed Model (Optional)
Environment Prompt
IDENTITY CHECK
Injected as system prompt to every probe. Use to force model self-identification.
▶ START AUDIT RUN
Cancel
Live Execution Telemetry
Provider Hygiene Index
0 / 100
Model Authenticity
70%
Quantization Fidelity
85 / 100
Elapsed Time
0.0s
Status: Idle
0%
#
STATUS
PROBE NAME
LATENCY
OBSERVATION / RATIONALE
No audit telemetry data. Click "START AUDIT RUN" to begin.
⚙ [02] Inspector - Deep Model Fingerprint & Attribution Analysis
Model Authenticity & Substitution Index
70%
GENUINE MATCH (50%)
Claimed Model
Detected Family (Auto)
QWEN
• Alibaba / Qwen
100%
• Anthropic / Claude
11%
Provider Hygiene & Feature Index
69 / 100
(Grade C)
Security Probes (Auto)
15 / 19 Pass
Inference Source (Auto)
Live Inference
Hidden System Prompt (Auto)
Injected ⚠
Vision Support (Auto)
Text-only
Passed / Warn / Failed
25
·
2
·
14
Upstream Infrastructure & Relay Attribution
DeepSeek Official Cloud
50% Conf
Detected Origin (Auto)
DeepSeek Official Cloud
Infrastructure Class
HYPERSCALER
Evidence Match
Error schema: DeepSeek Cloud
Base URL Source
Inherited from Runner
Model Weight Fidelity & Quantization Assessment
< 4-bit (2/3-bit)
Aggressive
Fidelity Score (Auto)
15 / 100
Quantization Aggressiveness
Highly Aggressive
Precision Loss Evidence
Degraded math probe accuracy
Run probes first to inspect raw request payload & HTTP headers.
↑ [03] Load Preset JSON Config
Load Profile JSON
⊟ [04] Proxy-Watch — Live Request Log
Auto-refreshes every 2s when audit is running
Clear