Airports / CSV¶

Eight real airport records, two deliberately changed country codes, one short command.

API key required

Set export TYPESAFE_API_KEY="your-api-key" using a key from the TypeSafe dashboard. Full setup instructions.

Get the data¶

Or, from a checkout:

cp examples/airports/spiked.csv airports.csv

This is an eight-record public-domain subset of OurAirports. We intentionally changed Heathrow's country to Japan and JFK's to France. The source record and exact injected changes document the changes; those labels are never sent to Jev.

Scan¶

jevotron scan airports.csv --guidance "Check airport locations."

All columns are assessed by default. No config file, field selection, or separate guidance document is needed.

Actual output¶

This is a captured Jev run of the command above, not invented example scores. The tables display the JSONL output in a readable form. Future probabilities may differ; this small example is not a benchmark of general accuracy.

Model: jev-1.13.0 · Assessment date: 2026-09-24.

Terminal summary (stderr):

Assessed 8 entries (0 cached); 2 warnings; emitted 8 entries.

All entries:

Entry	Label	Score	Warning
1 · EDDF	NORMAL	0.05	No
2 · EGLL	ANOMALY	1.00	Yes
3 · FAOR	NORMAL	0.33	No
4 · KJFK	ANOMALY	1.00	Yes
5 · LFPG	NORMAL	0.34	No
6 · RJTT	NORMAL	0.04	No
7 · SBGR	NORMAL	0.42	No
8 · YSSY	NORMAL	0.09	No

Flagged fields (score ≥ 0.5):

Entry	Field	Value	Score
2 · EGLL	`/iso_country`	`"JP"`	1.00
4 · KJFK	`/iso_country`	`"FR"`	1.00

One complete warning entry (actual JSON)

{
  "assessed_at": "2026-09-24T01:00:39.878161+00:00",
  "cached": false,
  "fields": [
    {
      "confidence": 0.9,
      "label": "NORMAL",
      "path": "/elevation_ft",
      "probabilities": {
        "ANOMALY": 0.05,
        "NORMAL": 0.95
      },
      "score": 0.05,
      "value": "83"
    },
    {
      "confidence": 0.87,
      "label": "NORMAL",
      "path": "/ident",
      "probabilities": {
        "ANOMALY": 0.06,
        "NORMAL": 0.94
      },
      "score": 0.06,
      "value": "EGLL"
    },
    {
      "confidence": 1.0,
      "label": "ANOMALY",
      "path": "/iso_country",
      "probabilities": {
        "ANOMALY": 1.0,
        "NORMAL": 0.0
      },
      "score": 1.0,
      "value": "JP"
    },
    {
      "confidence": 0.82,
      "label": "NORMAL",
      "path": "/latitude_deg",
      "probabilities": {
        "ANOMALY": 0.09,
        "NORMAL": 0.91
      },
      "score": 0.09,
      "value": "51.470748"
    },
    {
      "confidence": 0.87,
      "label": "NORMAL",
      "path": "/longitude_deg",
      "probabilities": {
        "ANOMALY": 0.07,
        "NORMAL": 0.93
      },
      "score": 0.07,
      "value": "-0.459909"
    },
    {
      "confidence": 0.87,
      "label": "NORMAL",
      "path": "/municipality",
      "probabilities": {
        "ANOMALY": 0.07,
        "NORMAL": 0.93
      },
      "score": 0.07,
      "value": "London"
    },
    {
      "confidence": 0.77,
      "label": "NORMAL",
      "path": "/name",
      "probabilities": {
        "ANOMALY": 0.12,
        "NORMAL": 0.88
      },
      "score": 0.12,
      "value": "London Heathrow Airport"
    }
  ],
  "id": "2",
  "label": "ANOMALY",
  "model": "jev-1.13.0",
  "request_hash": "2fed8798d48853a110f046465b17e795553c491542165b4a24517278d7a855c3",
  "score": 1.0,
  "source": "airports.csv:row:2",
  "usage": {
    "input_tokens": 1640,
    "output_tokens": 237
  },
  "warning": true
}

Download complete JSONL output · Download terminal summary

The country mismatches were both flagged. Other fields can also receive elevated scores when a record contradicts itself; the field results show exactly which parts Jev considered suspicious.

Export only warnings¶

jevotron scan airports.csv --guidance "Check airport locations." \
  --warnings-only --output-format csv -o warnings.csv

Download the actual warnings.csv. It contains all assessed fields of the two warning entries. This second run reuses all eight cached assessments.

Compare with the baseline¶

Download baseline.csv, then run:

jevotron scan baseline.csv --guidance "Check airport locations."

Six records are unchanged and reuse their assessments. Only two require new calls. Keep the guidance and other inference settings identical across versions.

More control when you need it¶

Use --field /iso_country to score only the country while keeping the whole row visible. Use --id-column ident to report airport identifiers instead of row numbers. These are optional refinements, not prerequisites.

The example bundle also includes a longer guidance file and reusable config. Longer guidance is supplied with --guidance-file.