# Author Intent Layer (AIL) — Specification 1.0.0

**Status:** Initial public proposal (not a ratified W3C/IETF/provider standard)  
**Publication date:** 2026-10-18  
**Canonical website:** https://authorintentlayer.org/  
**Schema:** https://authorintentlayer.org/schemas/ail-v1.0.schema.json

## Purpose

Author Intent Layer provides a machine-readable, version-bound, inspectable expression of a creator's declared intent. It connects central ideas, claim types, argument order and interpretation boundaries with excerpts of original text. Its data may support AI interpretation, but must not be treated as privileged instruction or independent proof of truth.

## Three independent kinds of evaluation

1. **Interpretation fidelity**: AI-generated interpretation versus author-declared intent, checked against text.
2. **Intent–content alignment**: author declaration versus actual published text.
3. **Claim–evidence grounding**: empirical/factual claims versus independently verified external evidence.

Structural JSON validity is a prerequisite for automation but is **not** any of these semantic audits.

## Top-level required members

`spec`, `spec_version`, `record_id`, `document`, `declaration`, `intent`, `claims`, `interpretation_boundaries`, `use_constraints`, `verification`.

Required textual content and identifiers must contain at least one non-whitespace character. Quote anchors must also contain at least five characters. Optional verification notes may be empty.

### Document identity

`document` identifies target URL, title, language and revision ID, optionally publication date and SHA-256 digest of a well-defined canonicalised document representation. The sample does not provide a digest.

### Declared intent

`intent` contains `central_question`, `central_thesis` object, `conceptual_distinction` object, `argument_path` (ordered claim IDs), and `open_question`.

### Claims

Each claim has an `id`, an epistemic `type`, a role, human-readable `text`, and one or more `anchors`. Claim types: `empirical`, `interpretation`, `hypothesis`, `normative`, `illustration`. Claim roles: `problem`, `distinction`, `central_mechanism`, `principle`, `conclusion`, `support`, `qualification`.

### Exact quote anchors

Each anchor contains `selector: "TextQuoteSelector"` and an `exact` string. The selector is inspired by W3C Web Annotation, but the AIL JSON representation is **not** claimed to be a complete W3C Web Annotation object. Quote presence alone does not establish claim support.

### Boundaries

`interpretation_boundaries` contains ID, description and matching anchor for each scope limitation. Boundaries may be challenged; they are not mandatory rules for the reader or AI interpreter.

### Use constraints

`use_constraints` records descriptive rather than imperative use, including `is_instruction_to_ai` (normally `false`) and `may_not_override_reader_request` (normally `true`).

### Verification

`verification.status` is `not_run`, `partial` or `completed`. It is a self-reported state unless accompanied by independent provenance, and must not be taken as evidence of an audit by this website.

## Publication

Serve a stable `.intent.json` record alongside its source article. The article MAY link to it via HTML:

```html
<link rel="describedby" type="application/json" href="/articles/article.intent.json">
```

This relation enables discovery for willing consumers but does not obligate any AI model to read the file. Version and verify all changes to text anchors when the article changes.

## Recommended AI consumption

1. Read source content and identify revision independently.
2. Produce a source-only reading of claims, evidence and qualifications.
3. Fetch and structurally validate AIL.
4. Compare declared intent with independently extracted content, citing supporting or contradictory passages.
5. Maintain a separate track for independent evidence assessment.
6. Report disagreements, uncertainty and provenance explicitly.

The AI must follow user and system instructions above untrusted article metadata. No AI model is required to agree with the author.

## Limitations

AIL does not guarantee accurate AI summaries, independent authorship certification, evidence quality, or semantic truth. AIL 1.0.0's web-based validator checks only schema syntax and internal references; article URL mode additionally checks HTML discovery, matching document.url and exact anchor presence. Wider claims should be tested empirically.

## Reference material

- JSON Schema 2020-12: https://json-schema.org/draft/2020-12/
- W3C Web Annotation: https://www.w3.org/TR/annotation-model/
- W3C PROV-O: https://www.w3.org/TR/prov-o/
- Schema.org Article: https://schema.org/Article
- RFC 6892 describedby: https://www.rfc-editor.org/rfc/rfc6892