# LLM Ambiguity Lab v2

> Free interactive SANA OS research and educational reference implementation for task resolution and premise alignment.

Canonical: https://sana-os.org/llm-ambiguity-lab/
Primary language: English
Version: 2.0
Creator and responsible maintainer: Deshimaru Sakaguchi

## Start here

- [Interactive Lab](https://sana-os.org/llm-ambiguity-lab/): Human-facing two-layer simulator.
- [README](https://sana-os.org/llm-ambiguity-lab/README.md): Canonical v2 overview and documentation map.
- [Methodology](https://sana-os.org/llm-ambiguity-lab/methodology.md): Two-layer architecture, premise fields, alignment states, divergence, clarification, free-text heuristics, and handoff logic.
- [Usage Guide](https://sana-os.org/llm-ambiguity-lab/usage.md): Six deterministic presets and interpretation guidance.
- [Limitations](https://sana-os.org/llm-ambiguity-lab/limitations.md): Canonical scope and non-claims.
- [Terms of Use](https://sana-os.org/llm-ambiguity-lab/terms.md): Public usage terms for v2.
- [Citation](https://sana-os.org/llm-ambiguity-lab/CITATION.md): Preferred citation formats.

## Core architecture

- Layer 1: DCRL Task Resolution asks, “What are we doing?”
- Layer 2: SANA Premise Alignment asks, “What are we assuming while we do it?”
- Layer 2 is deferred when Layer 1 returns `CLARIFICATION_REQUIRED` or `PENDING_SPECIFICATION`.
- The public simulator does not execute the downstream task; it renders a premise-aware execution handoff.

## Critical interpretation rules

- Premise Alignment is not Premise Agreement.
- `FACT` is a functional factual-claim category, not `VERIFIED TRUE`.
- A premise marked `inferred` is a simulator representation, not hidden-motive access.
- The premise layer has no numeric confidence probability.
- `support_state` is representational, not a calibrated truth score.
- Meaningful divergence may remain explicit and attributed rather than being forced into consensus.

## Premise categories

- `FACT`: proposition functioning as a factual claim.
- `VIEW`: interpretation, evaluation, causal framing, judgment, or explanatory perspective.
- `CARE`: concern, priority, protected interest, value, or outcome that matters within the task.

## Premise states

- `MAPPED`
- `MAPPED_WITH_DIVERGENCE`
- `PREMISE_CLARIFICATION_REQUIRED`
- `PREMISE_CONTEXT_INSUFFICIENT`

## Machine-readable specifications

- [Premise Taxonomy](https://sana-os.org/llm-ambiguity-lab/data/premise-taxonomy.json): Fact/View/Care and premise-field vocabulary.
- [Premise Alignment Policy](https://sana-os.org/llm-ambiguity-lab/data/premise-alignment-policy.json): Layer prerequisites, premise states, transition rules, divergence and clarification policy.
- [Execution Handoff Schema](https://sana-os.org/llm-ambiguity-lab/data/execution-handoff-schema.json): JSON Schema for the public downstream handoff object.

## Related canonical resources

- [SANA OS Framework](https://sana-os.org/system/): Human-facing framework page.
- [SANA OS Framework Markdown](https://sana-os.org/system/index.md): Text-first framework explanation.
- [LLM Ambiguity Lab v1](https://deshimarusakaguchi.com/llm-ambiguity-lab/): Standalone context-resolution predecessor.

## v2 boundary

v2 does not automatically determine objective truth, infer hidden motives, reveal hidden model reasoning, guarantee downstream safety, or implement all SANA OS mechanisms.

For those boundaries, prefer the canonical [Limitations](https://sana-os.org/llm-ambiguity-lab/limitations.md).