TRACEGov
Technical Deep-Dive

RAG vs TAMR+

Vector similarity treats "shall ensure compliance" and "may consider compliance" as identical. In regulatory AI, that's an audit failure. Here's what we built instead.

0.680
TAMR+ Score
vs 0.385 RAG
250
Benchmark Qs
4 domains
207ms
Avg Latency
End-to-end
$0.03
Per Workspace
2,300x cheaper
0 LLM
Retrieval Calls
Fully deterministic

Architecture Comparison

Why Vector Similarity Fails for Regulation

Standard RAG retrieves by similarity. TAMR+ retrieves by regulatory structure — entity relationships, causal chains, and multi-hop traversal.

5 Things RAG Gets Wrong

Standard retrieval augmented generation was not designed for regulatory compliance.

1

"shall ensure" ≈ "may consider"

RAG Problem

In embedding space, mandatory obligations and optional guidance are nearly identical vectors.

TAMR+ Solution

TAMR+ uses causal density signals to separate legal obligations from discretionary guidance.

2

Single-Hop Retrieval

RAG Problem

Regulations reference other regulations. Article 9 requires Article 13 transparency. RAG finds one chunk.

TAMR+ Solution

TAMR+ follows the chain — up to 3 hops with decay weighting [1.0, 0.5, 0.25].

3

Opaque Scores

RAG Problem

RAG gives you an answer with no confidence signal. How much should you trust it?

TAMR+ Solution

TAMR+ provides 5-dimension TRACE scores mapped to specific EU AI Act articles.

4

No Audit Trail

RAG Problem

Article 51 requires technical documentation. RAG pipelines are stateless — no lineage.

TAMR+ Solution

TAMR+ produces SHA-256 chained audit trails for every retrieval decision.

5

Expensive at Scale

RAG Problem

Most RAG systems cost $0.50-12.00 per complex query with LLM calls at every stage.

TAMR+ Solution

TAMR+ runs at $0.03/workspace. Zero LLM calls during retrieval. 207ms latency.

Benchmark Results

250 regulatory questions across 4 domains. Open-source. Apache 2.0.

TAMR+ v2.3 (3-hop)
EU-RegQA
0.74
MedRegQA
0.69
FinRegQA
0.66
CrimNet
0.63
Average
0.680
TAMR+ v2.3 (1-hop)
EU-RegQA
0.67
MedRegQA
0.63
FinRegQA
0.61
CrimNet
0.59
Average
0.625
GraphCompliance
EU-RegQA
0.554
MedRegQA
FinRegQA
CrimNet
Average
0.554
Vector-only RAG
EU-RegQA
0.41
MedRegQA
0.38
FinRegQA
0.39
CrimNet
0.36
Average
0.385

Ablation: removing any single component degrades performance by 6-27%. Vector-only scores 38.8% below full pipeline (p<0.001).

The Gap Is the Product

A 67% with full gap attribution tells a compliance officer exactly what to fix. A 95% with no explanation tells them nothing.

42%
SCG
Source Coverage Gap

Small document workspace. Fix: add more regulatory sources.

28%
PKC
Parametric Knowledge Cost

LLM fills gaps from training data. Fix: domain-specific documents.

12%
ADG
Attribution Density Gap

Formatting over evidence. Fix: improve citation density.

8%
DLT
Domain Language Tax

Regulatory vocabulary precision. Fix: glossary expansion.

10%
FSC
Structural Ceiling

Irreducible system floor (3% system-wide). Disclosed per Art. 13.

Framework Comparison

How TAMR+ compares to existing evaluation and retrieval frameworks.

Gap Attribution
TAMR+5 categories
RAGASNo
DeepEvalNo
COMPL-AINo
GraphComp.No
Predictive Gaps
TAMR+Yes
RAGASNo
DeepEvalNo
COMPL-AINo
GraphComp.No
Formula-Based (No ML)
TAMR+Yes
RAGASNo
DeepEvalNo
COMPL-AIPartial
GraphComp.Partial
EU AI Act Mapping
TAMR+8/8 articles
RAGAS0/8
DeepEval0/8
COMPL-AI3/8
GraphComp.0/8
Cross-Domain
TAMR+4 domains
RAGASN/A
DeepEvalN/A
COMPL-AI1
GraphComp.1
Audit Trail
TAMR+Yes (Art. 51)
RAGASNo
DeepEvalNo
COMPL-AINo
GraphComp.No
Production Deployed
TAMR+Yes
RAGASN/A
DeepEvalN/A
COMPL-AINo
GraphComp.No
Open Benchmark
TAMR+250 Qs
RAGASNo
DeepEvalNo
COMPL-AINo
GraphComp.No

Publication Status

European Patent
EP26162901.8 — Filed March 2026
Filed
Zenodo (DOI)
10.5281/zenodo.18929634 — Open Access
Live
OSF Preprints — Law Archive
Submitted — Under moderation
Review
SSRN Preprint
ID 6359818 — Processing
Pending
GitHub Repository
quantamixsol/tamr-plus — Apache 2.0
Public
arXiv Preprint
Pending endorsement for cs.IR
Pending
Conference Submission
JURIX 2026 / ICAIL 2027 targets
Planned

136days until enforcement

Run Your System Against Our Benchmark

250 regulatory questions, 4 domains, Apache 2.0. Download the benchmark, run your RAG system, and compare.

No credit card requiredEU-hostedGDPR compliant