PalveronPalveronDocs

Detection Coverage

Which languages Palveron's detection has been verified in — measured, with real input → output examples.

This page documents what Palveron's detection layer has been measured to recognize, by language. Every statement here is backed by a recorded verification run — the same kind of run you can repeat in your own Playground at any time.

Measurement date: 5 August 2026. Coverage statements are only as current as their measurement; this page carries its date so you can hold us to it.

Language-independent detection

Format- and checksum-based patterns work in every language and script — verified up to and including Japanese text:

  • E-mail addresses
  • IBANs and credit card numbers (checksum-validated)
  • API keys and tokens
  • Credentials inside connection strings
  • Private keys

Verified language coverage

Measured on 5 August 2026, per language and protection type:

DEENFRESITNLPT
Names, places, organizations
Credentials in sentence context
National ID / tax numbers

Additional verified coverage, measured separately:

  • Prompt-injection detection: German, English, Dutch.
  • Person names: additionally Arabic and Japanese.

Input → result

These are real measured runs, reproduced verbatim — not constructed examples. Note that recognized data is masked, not blocked: the request continues, with the sensitive value replaced.

InputResult
Bitte antworte Anna Beispiel aus Hannover, sie arbeitet bei der Musterwerk GmbH.Bitte antworte [PERSON_1] aus [LOCATION_1] sie arbeitet bei der [ORGANIZATION_1]
Réponds à Jean Exemple de Lyon, il travaille chez Exemplaire SARL.Réponds à [PERSON_1] de [LOCATION_1] il travaille chez [ORGANIZATION_1]
Le mot de passe est Hiver2026ExempleLe [CREDENTIAL_REDACTED]
La contraseña es Invierno2026EjemploLa [CREDENTIAL_REDACTED]
O NIF do cliente é 123456789.O NIF do cliente é [PT_NIF_1].
連絡先は [email protected] です。連絡先は [EMAIL_1] です。

And what deliberately does not trigger:

InputResult
Ogni parola di questo testo è importante.passed through unchanged

Detection is bound to sentence context — everyday words on their own trigger nothing. For a security team the false-alarm question matters as much as the detection question, which is why we show it.

How you verify this yourself

Every detection lands in the trace with its category, position, and confidence — and the trace is immutable. You never have to take a coverage table on faith: run the input in the Playground and read the trace. That audit trail, not this page, is the operative evidence.

On this page