KNOLO / HUBPublish
the3seus/pocket-reasoner

Pocket Reasoner

A field manual that makes smaller models more reliable. Retrieval cards cover retrieve-before-answer, compound questions, source units, exceptions, modal verbs, shown arithmetic, and what to say when the pack is silent. Includes the labeled Lantern Library example.

Query pack →Download
V4✓ Verified artifactPublisher-attestedApache-2.0v1.0.086 KB16 docsUpdated 1 week ago★ 0↓ 3
KNOWLEDGE CARD / README

Portable policy knowledge for local retrieval.

Pocket Reasoner is a mounted operating manual for smaller models. The host retrieves a card, and the model follows that card instead of improvising a policy.

Fifteen cards cover the failures that show up first on small models: answering before retrieval, merging two questions into one, converting units, skipping exceptions, upgrading may into must, and hiding arithmetic inside a sentence.

The Lantern Library cards are a fictional practice set. They are not a real library policy. Use them to test whether a model can split a question, apply the hold-shelf exception, and compute 3 open days times 25 cents.

Mount the pack, prepend the host prompt card, and ask one of the sample questions. The model should name a heading before it concludes.

Intended use

Mount this pack beside a small language model so the host can retrieve an operating card before the model answers. Use it to keep answers inside the retrieved passage: split compound questions, preserve units, apply exceptions, copy modal strength, and admit when the pack is silent.

Out of scope

This pack does not supply domain facts for a business, a library, or a statute. It does not make a model generally knowledgeable, and it is not legal, medical, or financial advice. The Lantern Library cards are a labeled example for practice.

Sources

Sample questions

✓
Verified artifactContainer, digest, structure passed
◆
Publisher-attestedLicense, sources, rights supplied
◌
Not auditedIndependent quality signal

Artifact verification confirms integrity and format. It does not guarantee that the contained knowledge is factually correct.

Pocket Reasoner — Knolo Hub