ClearDraft LogoClearDraft
GENERAL
DashboardTo-do ListImportsLive MailboxIMAP
MY SPACES
EvaluationRL AuditRLHFSettings

Benchmark Evaluation

Audited verification parity across shipping documents, operator dwell telemetry, and scoring isolation.

98.8%
GROUND-TRUTH BENCHMARK ACCURACY98.8% Verified

514 of 520 Shipments Verified with Exact Ground-Truth Agreement

Dual-tier model verification scored against organizer criteria: 100% category accuracy, 0 false clears.

Run:e430879f-173c…
Model: DeepSeek-V3
Dev (144): 97.9%
Blind (376): 99.2%
Dwell Time
60s
vs. 240s baseline
Time Saved
96.3%
14.2h saved
False Clears
0
100% fail-safe
Auto-Cleared Parity
10842.5% straight-through (108/254)
Operator Action Queue
118118 field discrepancies & missing values
Counterparty Chases
2828 missing B/L or wrong document format
Arbitration Insurance
00 unresolved gateway conflicts

Inbound Email Classification Distribution

520 records evaluated

Every inbound email is routed via ClearDraft's dual-tier classification gateway (deterministic regex rules + DeepSeek-V3 LLM).

BL COMPARISONCore Comparison Pipeline254 (48.8%)
SI REQUEST133 (25.6%)
INVOICE QUERY133 (25.6%)

Core Operational Invariants

Zero False-Clear Circuit Breaker

ClearDraft never silently guesses on borderline shipping documents. Borderline cases are automatically protected by an arbitration queue with 0 observed false clears.

7-Field Cross-Document Parity

Every draft Bill of Lading is verified against customer Shipping Instructions across Shipper, Consignee, Container, Seal, POL, POD, and Cargo Weight.

Deterministic Value Normalization

Weights normalize automatically to kilograms (KG), container codes check ISO 6346 check-digits, and port locations canonicalize against international UN/LOCODE registers.