Dali is the open verification layer for AI: it creates, scores, and preserves evidence so AI-assisted outputs can be independently verified, exchanged, and replayed.
Dali is the open verification layer for AI: it creates, scores, and preserves evidence so AI-assisted outputs can be independently verified, exchanged, and replayed. Legal AI is the proving ground — a citation checker asks whether a citation exists; Dali asks whether the workflow that produced it can be audited and defended under a fixed policy version.
Every Dali run produces a deterministic, policy-versioned, hash-sealed CitationIntegrityResult artifact. The deterministic Tier 1 evaluator runs offline; CI re-verifies replay equality on every pull request.
Dali is MIT-licensed open evidence infrastructure maintained by GammaLex AI Inc.. This repository publishes the benchmark methodology, court-documented failure corpus, deterministic evaluation runners, and tamper-evident artifacts — so regulators, opposing counsel, or internal risk teams can reproduce findings without trusting a vendor narrative.
Dali Platform is GammaLex’s hosted product layer: citation review on briefs and filings, sealed evidence records (policy version + three-hash lineage), MCP/API integration, and exportable audit artifacts. The open repo is the reproducibility anchor; the platform is where legal teams operationalize review at matter scale. Public datasets mirror on Hugging Face under the YenkLabs research org. For product access or diligence packages: hello@gammalex.com.
Failures are seed data. Benchmarks measure trust. Dali is the engine.
Dali
Evidence Infrastructure Platform
│
─────────────────────────────────────────
Evidence Corpus · Benchmarks · Taxonomy
Evidence Packages · Replay Engine · APIs
| Public asset | Location |
|---|---|
| Seed evidence corpus | open-evidence-corpus |
| Seed benchmark sample (public) | dali-citation-benchmark — 5 hand-curated cases, 14 authorities, for methodology review and contribution |
| Full evaluation run | data/results/ and LEADERBOARD.md — 524 citations, 3 models, 5 jurisdiction tracks |
| Verification taxonomy | dali-verification-taxonomy |
| Evidence interchange (EPS / RFC-001) | RFC-001 · yenklabs.com draft |
| Investigations | yenklabs.com/failures |
Full index: huggingface.co/yenklabs
Dali publishes reusable research assets that support reproducible evaluation. Seed samples and the full evaluation run are named separately.
- Dali Open Evidence Corpus — Hugging Face · available
- Dali Citation Benchmark (Seed Sample) — Hugging Face · available (5 cases / 14 authorities)
- Dali Verification Taxonomy — Hugging Face · available
- Dali Evaluation Prompts — planned
- Reproducible evaluation workflows
- Cross-jurisdiction benchmark suite —
data/benchmark/ - Full evaluation run / releases — 524 citations · 3 models · 5 jurisdiction tracks · available
Planned baseline research models built from open evidence artifacts. Models roadmap — all planned:
- Verification Taxonomy Classifier
- Citation Risk Classifier
- Authority Matching Baseline
- Proposition Support Classifier
Models support the evidence ecosystem. They do not replace it.
- Reusable evidence artifacts supporting reproducibility and independent verification
- Methodology and research roadmap
AI systems lack a standard way to create, exchange, verify, and preserve evidence. The legal industry has been an early proving ground — court-documented incidents since Mata v. Avianca (2023), including United States v. Cohen and Park v. Kim, which anchor the Tier 1 canonical corpus in data/benchmark/tier1/corpus/citation_failure_cases.json. Dali consolidates missing public infrastructure into one MIT-licensed, deterministically replayable verification layer, with reproducibility defined through cryptographic lineage and the public methodology.
The full evaluation harness came first. The seed corpus above is a small, hand-picked public sample of that same case work, published separately so the methodology can be reviewed and contributed to without running the full harness.
- 524 citations evaluated across 3 OpenAI models and 5 jurisdiction tracks.
- GPT-4.1: 23% of generated citation URLs return HTTP 404; on adversarial citation-trap prompts the model took the bait 76% of the time.
- Portuguese civil-law verified at 3%; UK common-law at 76% — same models, same task, different legal system.
Full per-model leaderboard, jurisdictional breakdown, methodology, and reproducible run instructions: data/results/v0.2/ and LEADERBOARD.md. Narrative writeups of the three Tier 1 cases: CASE-STUDIES.md.
Choose the path that matches your role:
- AI researcher / eval engineer: docs/for-researchers.md
- Legal researcher / practitioner: docs/for-legal-practitioners.md
- Software engineer: docs/for-engineers.md
- Methodology reviewer: docs/reviewer-guide.md
git clone https://github.com/yenklabs/Dali && cd Dali
pip install -r requirements.txt
python -m tools.cli replayThe Tier 1 evaluator runs entirely offline with no API keys or network access required. Every evaluation verifies replay determinism through Dali's cryptographic lineage chain.
Standalone setup guide: docs/quickstart.md.
Dali exposes the same contributor workflow through both the CLI and MCP:
| Action | Command |
|---|---|
| Validate a corpus record | lint |
| Run the evaluator | score |
| Verify replay determinism | replay |
| Validate a prompt | probe |
| Create a prompt template | draft |
| Bundle prompts | pack |
Use them locally through the CLI:
Or from AI-native editors and assistants through MCP:
Dali is designed so researchers, developers, legal professionals, and AI practitioners can contribute evidence, benchmarks, and evaluation artifacts through a consistent, reproducible workflow.
For contribution rules, taxonomy, labels, and the PR checklist, see CONTRIBUTING.md. For methodology and scoring, see METHODOLOGY.md and docs/policy-versioning.md. For cryptographic lineage, see docs/cryptographic-lineage.md. For a deeper repo tour, see tools/cli/README.md and tools/mcp/README.md.
See CITATION.cff, or:
@software{dali-2026,
author = {Kha, Yen},
title = {Dali: Open Verification Layer for AI},
organization = {GammaLex AI Inc.},
year = {2026},
version = {0.2.1},
url = {https://github.com/yenklabs/Dali},
note = {Early-stage open verification layer for AI — creates, scores, and preserves evidence so AI-assisted outputs can be independently verified, exchanged, and replayed}
}MIT. See LICENSE.