Observability vs Competition

The short version: Langfuse

Langfuse is an open-source LLM observability platform — it traces LLM calls so you can debug, monitor, and evaluate your AI application. Logs, spans, generations, prompt management, evaluation pipelines, cost analytics: Langfuse gives you a dashboard into what your app's LLM calls are doing. It is excellent at what it does, which is observability. What it does not do is score the human operator or rank them against anyone.

SigRank is a different thing entirely: an operator leaderboard. It reads token telemetry from any AI tool an operator drives, computes the cascade efficiency (Υ Yield), and ranks them globally. Langfuse tells you what your app did; SigRank tells you how you compare to every other operator. Observability vs competition — different goals, different layers.

Feature comparison

Inward-facing vs outward-facing

Observability is inward-facing. Langfuse helps a team understand and improve their own application's LLM usage: which prompts are expensive, which chains are slow, which generations failed evaluation. The data stays private, scoped to one project, used for debugging. It answers "what is my app doing?"

A leaderboard is outward-facing. SigRank ranks operators against each other on a canonical, public metric. The score — Υ = cache_read × output / input² — is signed with ed25519, server-verified, and comparable across operators, tools, and platforms. It answers "how do I compare to everyone else?" The first is a debug tool; the second is a competition. You can run both — they do not overlap.

Token counts only — never content

Langfuse's traces can include full prompt and completion content — that is the point of observability, you need to see what the model received and returned. SigRank is privacy-preserving by design: it reads only the four token integers (input, output, cache-read, cache-write) and never touches prompt content. Your sessions stay on your machine; only the signed token counts leave it. That makes SigRank safe to run in any environment, including ones where logging content is a compliance risk.

Frequently asked questions

Does SigRank replace Langfuse?

No — they solve different problems. Langfuse is an LLM observability platform: it traces LLM calls — logs, spans, generations, prompts, completions — so you can debug, monitor, and evaluate your AI application. SigRank is an operator leaderboard: it scores how efficiently a human drives AI tools and ranks them globally. Langfuse is for understanding your app's LLM usage; SigRank is for ranking the operator behind it. Observability vs competition — different goals, different layers.

Does Langfuse measure operator token efficiency?

Langfuse traces LLM calls and can aggregate token usage metrics — input tokens, output tokens, cost per call. That is application-level observability: it tells you what your app spent and where. It does not compute the cascade efficiency (Υ Yield = cache_read × output / input²), does not assign a class tier, and does not rank operators on a leaderboard. Langfuse answers "what happened in this LLM call?" SigRank answers "how efficiently does this operator drive AI?" Tracing vs scoring — different questions.

Why is a leaderboard different from observability?

Observability (Langfuse) is inward-facing: it helps a team understand and improve their own application's LLM usage. The data stays private, scoped to one project, used for debugging and evaluation. A leaderboard (SigRank) is outward-facing: it ranks operators against each other on a canonical, public metric. The score is signed, server-verified, and comparable across operators, tools, and platforms. Observability tells you what your app did; a leaderboard tells you how you compare to everyone else. The first is a debug tool; the second is a competition.

Can I use SigRank alongside Langfuse?

Yes — they are complementary. Use Langfuse to trace and debug your application's LLM calls. Use SigRank to score the human operator who drives your AI tools — including the app Langfuse is observing. The SigRank CLI reads token telemetry locally (token counts only, never prompt content), computes the cascade metrics, signs a snapshot with ed25519, and publishes it to the leaderboard. Langfuse sees the app's calls; SigRank scores the operator behind them. Run `sigrank enroll` then `sigrank submit` to get your rank.

What is the difference between LLM tracing and operator scoring?

LLM tracing (Langfuse) records what happened inside individual LLM calls — the prompt, the completion, the latency, the token count, the model used. It is a per-call, application-scoped debug log. Operator scoring (SigRank) aggregates the token telemetry across an operator's entire session — across tools, across platforms — and computes a single cascade efficiency score (Υ Yield) that is comparable globally. Tracing answers "what did this call do?" Scoring answers "how efficiently does this person drive AI?" The first is observability; the second is competition.