Frequently Asked Questions¶

General¶

What is Mistaber?¶

Mistaber is a formal symbolic reasoning system for computational halacha (Jewish law). It uses Kripke semantics and Answer Set Programming to model how different halachic authorities (poskim) rule on various questions, enabling automated reasoning about disputes, inheritance of rulings, and source traceability.

Is Mistaber a posek (halachic decisor)?¶

No. Mistaber is a research and analysis tool, not a substitute for consulting a qualified rabbi. It shows what different authorities hold according to encoded rules, but does not issue practical halachic rulings.

What domains does Mistaber currently cover?¶

Currently, Mistaber covers Yoreh Deah simanim 87-92 (Basar Bechalav / Meat and Milk laws) as a pilot implementation. This includes:

7 halachic authorities (worlds)
Core commentators (Shach, Taz)
Prohibition types, normative levels, and safek handling

What makes Mistaber different from existing halacha databases?¶

Feature	Traditional Databases	Mistaber
Data model	Text search / keywords	Formal logic propositions
Reasoning	Manual lookup	Automated inference
Disputes	Listed separately	Detected automatically via world comparison
Sources	References only	Full derivation chains
Uncertainty	Not formalized	Safek calculus with resolution policies

Technical¶

What is Kripke semantics?¶

Kripke semantics is a framework from modal logic where propositions can be true in some "possible worlds" but false in others. In Mistaber, each halachic authority (posek) is modeled as a world. A proposition like "fish with dairy is forbidden" can be true in the Mechaber's world but false in the Rema's world.

What is Answer Set Programming (ASP)?¶

ASP is a form of declarative programming suited for knowledge representation and reasoning. Mistaber uses Clingo, the leading ASP solver, because ASP excels at:

Non-monotonic reasoning (rules with exceptions)
Default reasoning (what holds unless overridden)
Multiple stable models (different valid conclusions)

Why not use a Large Language Model (LLM)?¶

LLMs are powerful for text processing but have limitations for halachic reasoning:

Aspect	LLM	Mistaber
Explainability	"Black box"	Full derivation trees
Consistency	May contradict itself	Formally consistent
Source fidelity	May hallucinate	Every rule has makor
Authority modeling	Single voice	Multi-world semantics
Verifiability	Hard to audit	Rules are inspectable

Mistaber and LLMs could complement each other - an LLM could help translate natural language queries into Mistaber's formal language.

What are the system requirements?¶

Python 3.10+
Clingo 5.6.0+ (ASP solver)
Optional: asprin (for preference optimization)
Optional: xclingo (for explanations)

Usage¶

How do I query for a specific ruling?¶

# Check if something holds in a world
mistaber -o ./mistaber/ontology query "holds(issur(achiila, M, D), mechaber)"

# Compare across worlds
mistaber -o ./mistaber/ontology compare "holds(issur(achiila, M, D), W)" --worlds "mechaber,rema"

How do I add my own rules?¶

Create a .lp file in mistaber/ontology/extensions/
Use the HLL syntax with @world, @rule, @makor, @madrega directives
Rules will be loaded automatically

See the Extension Protocol for details.

Can I add new halachic authorities (worlds)?¶

Yes! Create a new .lp file in mistaber/ontology/worlds/ with:

world(my_posek).
accessible(my_posek, parent_world).  % Inherit from parent

% Override specific rulings
override(my_posek, some_proposition, reason).

% Assert new rulings
asserts(my_posek, new_ruling).

How do I handle uncertainty (safek)?¶

Mistaber uses safek policies that vary by world:

engine.configure(
    world="ashk_mb",
    safek_resolution="chumra"  # or "kula", "madrega"
)

The system applies: - Safek d'oraita l'chumra (biblical doubt → stringent) - Safek d'rabanan l'kula (rabbinic doubt → lenient)

Encoding & Plugin¶

What is the mistaber-skills plugin?¶

mistaber-skills is a Claude Code plugin that provides an AI-assisted workflow for encoding halachic content into Mistaber. It includes 6 skills, 7 enforcement hooks, and 1 orchestrating agent to guide you through the encoding pipeline with human checkpoints at each phase.

How do I start encoding halacha?¶

Install the plugin: claude plugins add ./mistaber-skills
Start a session: claude "I want to encode YD 87:3"
Follow the guided workflow through 4 checkpoints
Each phase requires human approval before proceeding

See the Encoding Workflow Guide for detailed instructions.

What are checkpoints?¶

Checkpoints are mandatory human approval points in the encoding workflow:

Checkpoint	Phase	What You Review
Checkpoint 1	corpus-prep	Source texts, derivation chains
Checkpoint 2	hll-encode	Generated HLL rules
Checkpoint 3	validate	Test results, compilation
Checkpoint 4	review	Full encoding package

You must type "approved" at each checkpoint to proceed.

What's the difference between override and asserts?¶

asserts/2 creates a positive assertion in a world:

asserts(mechaber, issur(achiila, fish_dairy, sakana)).

override/3 blocks inheritance of a rule from parent worlds with a descriptive reason:

override(dag_bechalav_is_sakana, rema, no_sakana).

Key rule: Override's third argument must be a descriptive value (like no_sakana), never invalid.

Why can't I write .lp files directly?¶

The phase-gate hook prevents writing .lp files until you've completed the corpus-prep checkpoint. This ensures:

You've reviewed the source texts
The derivation chain is complete
You understand what you're encoding

To bypass (not recommended), you'd need to modify the hook configuration.

How do I handle a machloket (dispute)?¶

Encode the base position in the base world
Use override/3 in disagreeing worlds with descriptive reasons
Test both positions with multi-world queries
Document the machloket in your encoding report

Example (fish & dairy):

% Base: fish with dairy is dangerous
asserts(base, sakana(fish_dairy)).

% Rema: no danger
override(dag_bechalav_is_sakana, rema, no_sakana).

See the Machloket Handling Tutorial.

What test coverage is required?¶

Minimum test requirements for each encoding:

Test Type	Minimum Count	Purpose
Positive tests	5	Verify expected rulings
Negative tests	3	Verify boundaries
Edge case tests	2	Cover unusual scenarios
Multi-world tests	2 (if applicable)	Verify inheritance

See the Testing Requirements guide.

What sources can I use for encoding?¶

The plugin integrates with Sefaria MCP for fetching sources. Supported sources include:

Shulchan Aruch (primary)
Major commentators (Shach, Taz, Pri Megadim)
Later authorities (Mishnah Berurah, Aruch HaShulchan, Yalkut Yosef)
Talmudic sources for derivation chains

The corpus-prep skill helps you gather and organize all relevant sources.

How are rules named?¶

Rules use topic-based naming, not siman-seif references:

Correct	Incorrect
`r_bb_dag_sakana`	`r_87_3`
`r_bb_beheima_achiila`	`r_yd_87_1`
`r_bb_of_bishul`	`r_siman87_seif1`

This makes rules more readable and maintainable.

What are the 7 Kripke worlds?¶

World	Authority	Tradition	Inherits From
`base`	Shulchan Aruch	Universal	—
`mechaber`	R' Yosef Karo	Sefardi	base
`rema`	R' Moshe Isserles	Ashkenazi	base
`gra`	Vilna Gaon	Litvish	base
`sefardi_yo`	Yalkut Yosef	Modern Sefardi	mechaber
`ashk_mb`	Mishnah Berurah	Litvish	rema
`ashk_ah`	Aruch HaShulchan	Russian	rema, gra

Where can I learn more about encoding?¶

Resource	Description
Encoding Guide	Overview of the encoding workflow
Encoding Tutorials	Hands-on tutorials
Encoding Guidelines	Best practices and standards
Plugin Documentation	Detailed plugin docs
Glossary	Terms and definitions

Contributing¶

How can I contribute?¶

See the Contributing Guide for:

Setting up the development environment
Running tests
Code style guidelines
Pull request process

I'm not a halacha expert. Can I still contribute?¶

Absolutely! Many contributions don't require domain expertise:

Tests - Add edge cases, improve coverage
Documentation - Fix typos, add examples
Tooling - CLI improvements, performance
Bug fixes - General software issues

Where do I report bugs or request features?¶

Open an issue on GitHub.

Troubleshooting¶

"Clingo not found" error¶

Install Clingo:

# Ubuntu/Debian
sudo apt install gringo

# macOS
brew install clingo

# pip (cross-platform)
pip install clingo

"Unsatisfiable" result¶

This usually means contradictory facts were asserted. Common causes:

A food item classified as both basar and chalav (violates disjointness)
Conflicting assertions without proper override

Check your facts for logical consistency.

Slow query performance¶

Large ontologies with many variables can slow grounding. Tips:

Use concrete values instead of variables where possible
Add constraints early to prune the search space
Use #show directives to limit output

Have a question not answered here? Open an issue with the "question" label.