Skip to content

Import inspection

Meat, poultry, and egg imports to inspect first

Not every imported lot of meat, poultry, or eggs can be opened. This model reads every record in the Food Safety and Inspection Service's public import data since fiscal 2014 and ranks them twice: first by the chance of refusal for public health, and separately by the chance of refusal for the product's condition.

Lots and outcomes from FSIS Import and Export Data, fiscal years 2014 to 2026. Every figure on this page is measured on lots the model was not trained on. Model fsis-import-risk 0.2.0.

Open a share of the arrivals. How many refused lots are inside it?

Fiscal years 2025 and 2026, the years the model never saw: 70 refused lots. Drag the handle. The model's order holds this many of them inside the share of arrivals you open.

Open 7.5% · catch 36 of 70

0% 25% 50% 75% 100% 7.5% · the share we compare at 0% 10% 20% 30% share of a year's arrivals opened
Of 70 refused lots, the share inside the first 7.5%
This model 51% Plant history only 37% Random draw 8%

Review workflow prototype

Seven cases. Different reasons to ask for a human look.

This case library demonstrates how a future uncertainty queue could route more than new establishments. Examples come from the February 2026 point-in-time snapshot; outcomes were not used to select or explain them. They use a separate exploratory documentation/damage ranking, not the capture-curve model above.

Research-only demo. Rank percentiles are not probabilities. The 207-row formal support-review lane is still homogeneous; the other profiles below are candidate routing signals, not validated uncertainty bands or inspection directives. No live scorer or Haiku call is connected.

Selected case · February 2026

Reactivated establishment

Beef · raw intact cuts · 99 prior plant lots; return after a 180+ day gap.

Documentation rank percentile
99.71
Damage rank percentile
74.85
Severity assessment
INDETERMINATE · no direct defect evidence in this payload

Offline LLM-style reasoning

A long gap can make old operating context less current, but inactivity is not evidence of failure.

Evidence to request / verify

Confirm current establishment identity and records; keep all required inspection checks in force.

This is an authored dry-run, not model-generated advice. Required FSIS checks remain unchanged. An unflagged lot is not cleared or established as safe.

Public health is a failed laboratory test, a pathogen, or a recall: 70 refused lots in fiscal 2025 and 2026. Condition is what an inspector finds on the product, such as pathology, extraneous material, a failed product exam, or compromised packaging: 984 refused lots. Plant history only ranks a lot by its plant's refusals in the previous year. A random draw catches the share it opens. 7.5% is a share we chose so the methods compare on equal terms.

See what others miss

Understand the industrial networks shaping foreign military capabilities and vulnerabilities.

Talk with us