Blog

Notes from the token level

Engineering notes and research on attribution, grounding, and what models actually attend to when they answer.

researchSeptember 8, 2026 · 6 min read

Why attention attribution transfers across LLMs

Generating a good answer needs a strong model. Checking which parts of it came from the document doesn't — on grounded tokens, a 9B model agrees with GPT, Claude, Gemini and Grok almost token for token.

benchmarksresearchAugust 31, 2026 · 11 min read

Multilingual attribution with TokenPath

We test attribution from answers in 23 international languages back to English contract evidence, compare it with embedding retrieval, and run a separate pilot across all 22 scheduled Indian languages.

benchmarksresearchJuly 26, 2026 · 8 min read

Self-attention beats asking a model for citations

One model does both jobs: Qwen3.5-9B writes an answer, then either cites it when asked or has its attention read. Reading attention scores 0.744 against 0.644 F1, and returns in 1.2s instead of 53s.

engineeringresearchJuly 10, 2026 · 9 min read

How TokenPath works

Using the attention activations of an open LLM to construct citations — with a raw-Transformers reproduction you can run yourself.

benchmarksresearchJuly 9, 2026 · 10 min read

One answer, four ways to cite it

We benchmark a prompted frontier LLM, Anthropic's Citations API, embedding retrieval, and TokenPath's post-hoc attention on LongBench-Cite — same answer, same judge. Post-hoc matches generation-time citation quality, ~5–6× faster and ~7× cheaper.