test-corpus

An index of PDF-centric corpora

A Qiqqa Test Library / Test Corpus which contains various PDF document samples, etc. collected from live Qiqqa libraries to showcase issues and check regressions in the software.

Collection of certificates and keys with different encodings and encryptions

An open, reproducible PDF corpus for benchmarking and regression testing across real-world, synthetic, malformed, accessibility, archival, rendering, parsing, repair, and conformance cases.

The test corpus and harness that check line-of-code counters against their own declared rules

PineForge TV validation corpus — 312 reference strategies: 311 excellent + 1 declared anomaly on the engine's parity gate (engine main) (~431k TradingView trades). Public and reproducible; write access maintainers only.

the gauntlet corpus

A false-positive corpus for secret scanners: 109 formats of ordinary log and build output containing zero credentials, plus a recall half of 42 formats with 81 planted synthetic credentials and an answer key. Ships as a GitHub Action that gates the build.

An adversarial conformance corpus for IFC parsers — 52 trap files, and a scoreboard of how IfcOpenShell, web-ifc and friends survive them.

Shared test corpus and harness for YINI parser implementations — validating clear, indentation-insensitive configuration with nested sections and predictable parsing.

Language-agnostic JSON test vectors for NIP behavioral interpretation

539 regex constructs run on 16 real engines. 264 behave differently somewhere and only 83 tell you. MIT-licensed, reproducible.

CSV parser conformance test corpus

An LLM-as-a-judge evaluation harness built as a reference project: a hand-authored voice-agent corpus with seeded defects, a deterministic tier and a judged tier, and every claim held by a test, a verifier or a control.

BigCorp: a synthetic enterprise Java system for exploring code analysis and agent tooling across messaging, batch jobs, shared databases, and business rules.

GitGalaxy cross-language control corpus: one program, every language (gitgalaxy#1096)

A synthetic codebase built to be hard to index — 800 files, 72 file types, comments in 8 writing systems. Measure a context tool against something difficult without pointing it at your own code. MIT.

A repo of 50+ languages to test the detection capabilities of repo-scanners. Built to stress-test any multi-language scanner, prove regression resistance, and validate structural signature extraction at scale.

Reproducible benchmark and test corpus for prompt-injection defenses — adapters for common sanitization libraries, scored leaderboard, CI-ready. Security-adjacent, evergreen.