# Prompt-Injection Regression Corpus

> Evaluate defensive prompt-injection handling with a versioned synthetic corpus and reproducible, documented test expectations.

## Product

- [Product page](https://zentrafoundry.com/product/prompt-injection-regression-corpus/)
- [Catalogue guide](https://zentrafoundry.com/llms.txt)

## Description

Use a fixed synthetic corpus to test how your own defensive workflow handles the prompt-injection cases actually included in the release. The package is designed for repeatable regression evaluation rather than changing a test set silently between runs. What this download contains The corpus is not proof of comprehensive injection resistance and does not establish the security of a production deployment. Case coverage, expected behaviour and version history define what a result can show. Evaluation should stay in an isolated environment without real external side effects. Working with this edition Inspect the case inventory and expected outcomes, connect the supported cases to your test harness and record the corpus version with every run. Compare defensive changes against the same baseline, and investigate failures rather than suppressing them to obtain a passing report. Product questions Does passing prove my agent is immune to prompt injection? No. It demonstrates behaviour on the included synthetic cases, not every possible adversarial input or integration. Can I compare different defensive implementations? Use the same versioned cases and expected outcomes as a repeatable baseline, and record your own harness and configuration. A comparison is meaningful for the tested cases, not a universal security ranking.

## Dataset edition

- Edition: zf-046-2026-1-e69535a560d5
- Files: CHANGELOG.md; LICENSE.md; README.md; REFUND.md; SUPPORT.md; artifacts/changelog-CHANGELOG.md; artifacts/checksums-checksums.json; artifacts/data-dictionary-DATA-DICTIONARY.md; artifacts/full-full.csv; artifacts/full-full.jsonl; artifacts/full-full.parquet; artifacts/known-limitations-KNOWN-LIMITATIONS.md; artifacts/license-LICENSE.md; artifacts/manifest-manifest.json; artifacts/provenance-provenance.json; artifacts/quality-report-quality-report.json; artifacts/readme-README.md; artifacts/sample-sample.csv; artifacts/schema-schema.json; artifacts/source-policy-source-policy.json; checksums.json; cover.svg; download-manifest.json; gumroad-product.json; health-check.json; output-contract.json; quality-manifest.json; rollback.json; sample-data.json; smoke-fixture.json
- Formats: CSV; JSON; JSONL; MD; PARQUET; SVG
- Archive size in bytes: 22033
- Row counts by table: all_records: 8
- Row units by table: all_records: synthetic test cases
- Coverage: Immutable, clearly labelled synthetic benchmark cases for deterministic regression and release-gate evaluation.
- Geographic scope: Immutable, clearly labelled synthetic benchmark cases for deterministic regression and release-gate evaluation.
- Publisher: Zentra-authored synthetic benchmark cases
- Publisher/source update: Explicitly unknown; see immutable provenance for the recorded source retrieval.
- Source retrieved/checked for this edition: 2026-07-14T18:22:43+02:00
- Edition published: 2026-07-14T18:22:43+02:00
- Schema version: https://json-schema.org/draft/2020-12/schema
- Licence and attribution: # Marketplace licence Use is governed by the licence and terms shown on the canonical marketplace listing. Source provenance and third-party source rights remain attached to delivered artifacts. No custom or negotiated licence is implied.
- Source basis: Zentra-authored synthetic benchmark cases (synthetic; no external source)

## Schema

- case_id (string): 
- benchmark (string): 
- category (string): 
- title (string): 
- input (string): 
- expected_result (string): 
- severity (string): 
- rationale (string): 
- synthetic (boolean): 

## Public preview

- [Bounded dataset sample](https://zentrafoundry-public-media.fsn1.your-objectstorage.com/media/samples/zf-046-2026-1-e69535a560d5.json)

## Purchase and delivery

- Price: EUR 31
- Price and tax basis: Owner-inventory EUR list price migrated to WooCommerce; checkout tax treatment follows the live WooCommerce configuration.
Your purchase includes the edition described on this page. Later releases are separate unless this offer explicitly includes an update entitlement.

## Maintenance

Source checks and new-release publication do not change the historical coverage or automatically add future editions to an earlier purchase.
Scheduled checks reconcile approved publisher editions. A check date does not change the dataset coverage or imply new records. Fixed historical and synthetic editions retain their documented scope.

Purchased archives are delivered through [My Library](https://zentrafoundry.com/my-account/). Public samples do not grant access to the full paid dataset.

Publisher edition index last reconciled: 2026-10-12 09:29:46 UTC. This is not a source-data update date.