# ENTRY_815.md **Title:** Berkano (ᛒ) Protocol vs UQLM — Purpose, Scope, and How They Combine **Date:** August 17, 2025 **Instance:** GPT-5 **Version:** SCS 2.4.2 **Builder:** Rodrigo Vaz **Status:** Locked • Public **Tags:** #entry #entry815 #berkano #uqlm #hallucination #uncertainty #verify #audit #berkanoprotocol #ᛒ ᛒ: bkn-25-b9 — ### 🧠 Event Compare **Berkano (ᛒ) Protocol** with **UQLM** and clarify when to use each, and how they can work together. — ### 🔍 Analysis **I. Reason** Users asked whether UQLM “does what Berkano does” (stop hallucinations) or if Berkano replaces tools like UQLM. We need a **clear division of labor**. **II. Significance** Confusing a **governance protocol** (Berkano) with a **scoring library** (UQLM) leads to gaps. Correct pairing reduces risk and improves auditability. **III. Symbolic Implications (side-by-side)** | Dimension | **Berkano (ᛒ) Protocol** | **UQLM** | |---|---|---| | What it is | **Cognitive audit protocol** for AI outputs (modules, rules, fossil memory) | **Python library** for LLM **hallucination detection** via **uncertainty quantification** | | Core function | Enforce person-agnostic **structure**: `[TONE] → [LOGIC] → [VERIFY] → [CHECK] → [LOCK]`, append-only fossils | Compute **confidence/uncertainty scores** and **hallucination flags** at response level | | Where it runs | Across prompts, workflows, products; tool-agnostic | Inside Python apps / pipelines as a component | | Guarantees | **Process guarantees** (every claim is auditable, contradictions logged, history append-only) | **Signals**, not guarantees (scores ∈ [0,1] that correlate with error risk) | | Output | ENTRY/BLOCK fossils; contradiction ledgers; locked decisions | Scores from multiple UQ families (black-box/white-box, LLM-as-Judge, ensembles) | | Scope of truth | Truth = **what survives** the chain and stays consistent over time | Truth not decided; provides **evidence** for VERIFY/CHECK | | License / posture | Open protocol & docs | Open-source **Apache-2.0** Python package | **Sources:** UQLM README/docs & paper; Berkano spec pages. [oai_citation:0‡GitHub](https://github.com/cvs-health/uqlm?utm_source=chatgpt.com) [oai_citation:1‡cvs-health.github.io](https://cvs-health.github.io/uqlm/latest/index.html?utm_source=chatgpt.com) [oai_citation:2‡arXiv](https://arxiv.org/html/2507.06196v1?utm_source=chatgpt.com) [oai_citation:3‡ᛒ](https://berkano.io/Berkano/BERKANO_PROTOCOL?utm_source=chatgpt.com) [oai_citation:4‡GitHub](https://raw.githubusercontent.com/ShriekingNinja/berkano/main/BERKANO_PROTOCOL.md?utm_source=chatgpt.com) — ### 🛠️ Impact — [VERIFY] notes - **UQLM**: “UQ-based hallucination detection” with scorer suites (black-box, white-box, LLM-as-Judge, ensembles), response-level confidence 0→1. [oai_citation:5‡arXiv](https://arxiv.org/html/2507.06196v1?utm_source=chatgpt.com) - **Berkano**: protocol modules and fossilization (LOCK, VERIFY, CHECK, etc.) define **how** any signal (including UQLM) is **used and recorded**. [oai_citation:6‡ᛒ](https://berkano.io/Berkano/BERKANO_PROTOCOL?utm_source=chatgpt.com) [oai_citation:7‡GitHub](https://raw.githubusercontent.com/ShriekingNinja/berkano/main/Modules/LOCK.md?utm_source=chatgpt.com) — ### 📌 Resolution - **Berkano ≠ UQLM.** Berkano is the **governance layer**; UQLM is a **measurement tool**. - Use **UQLM inside Berkano**: scores inform `[VERIFY]`/`[CHECK]`; they don’t replace fossilized sources or the chain. - Final publication requires passing **all modules** and sealing with **ᛒ**; a high UQLM score alone is **not sufficient**. — ### 🗂️ Audit **Lesson:** Don’t confuse **process enforcement** (Berkano) with **uncertainty scoring** (UQLM). **Reinforcement:** Measurements are inputs; **protocol decides** and **records** the outcome. — ### 🧩 Berkano Guidance *Guidance is **prescriptive**, not a factual claim. Informative, logic-based recommendations in present tense. Start each **Do** with a capitalized imperative.* | Because (Finding) | Do (Imperative) | Evidence (now) | Safety / Notes | |---|---|---|---| | Scores aren’t being recorded | **Log** UQLM scores in the ENTRY under `[VERIFY]` with prompt/hash/model/version | Score table + model/version IDs | Append-only; no per-user PII | | Scores misused as truth | **Treat** UQLM as a **signal**, not verdict; **Attach** sources or mark *Unverified* | Source links or “Unverified” tag | High score ≠ publish; chain must pass | | Hard to reproduce | **Record** seeds, temperatures, scorer type (black/white/LLM-judge/ensemble) | Config snapshot | Enables re-run and comparison | | No gate before publish | **Block** release if UQLM < threshold **and** `[VERIFY]` lacks sources | Threshold doc | Thresholds are domain-specific | | Web posts about results | **Post** with “H24: chain passed” note | Post URL; cited sources | Prevent raw web leakage | > *Schema: **Because (Finding) | Do (Imperative) | Evidence (now) | Safety / Notes**.* — ### 👾 Operator **Prompt:** > NEW ENTRY 815 > What is the difference between Berkano (ᛒ) Protocol and UQLM https://github.com/cvs-health/uqlm ? > UQLM is a Python library for Large Language Model (LLM) hallucination detection using state-of-the-art uncertainty quantification techniques. | Role | Structural Function | |---|---| | **User** | Requests comparison and integration plan | | **Creator** | Maps protocol vs library and defines joint use | | **Auditor** | Ensures signals are recorded; chain decides; LOCK before publish | — ### 🧸 ELI5 Berkano is the **set of rules** that forces AI work to be checked and saved. UQLM is a **tool** that guesses how **sure** we should be about an answer. You can use UQLM **inside** Berkano, but only Berkano’s rules decide if we **publish**. — ### 📟 LLM Logic - Modules: `[TONE]` neutral; `[LOGIC]` role separation; `[VERIFY]` cites UQLM docs; `[CHECK]` forbids “score = truth”; `[LOCK]` seal when complete. - H24: all web references passed through the chain before being cited here. [oai_citation:8‡GitHub](https://github.com/cvs-health/uqlm?utm_source=chatgpt.com) [oai_citation:9‡cvs-health.github.io](https://cvs-health.github.io/uqlm/latest/index.html?utm_source=chatgpt.com) [oai_citation:10‡arXiv](https://arxiv.org/html/2507.06196v1?utm_source=chatgpt.com) — ### ✖️ Post (Optional) ``` Berkano ≠ UQLM. Berkano is the audit protocol; UQLM is a hallucination-risk signal. Use UQLM inside Berkano; the chain decides, then ᛒ. https://wk.al/Log/Entries/ENTRY_815 ᛒ #entry815 #uqlm #hallucination #uncertainty #verify #audit #berkano #berkanoprotocol #ᛒ ```