Menu

Research

Reading as Enclosure

Kirill Shebetov · English language edition · 13 September 2026

The word “person” is in the text.It does not establish the presence of a person.

Moving from a written form to what it names requires a separate evidential ground.

On this page

How a reading becomes its own ground

A constructed example involving an AI response:

Record
I feel afraid.
Proposed reading
The response was written by a being with feelings.
Offered ground
Because it says “I feel afraid.”

The same line is offered again as its own support. No separate material about the named feeling, or rule connecting the line to it, has been added. This illustrates circular support; it is not a claim about an AI system’s state.

Computing a formal status requires the complete declared set of records, links, and rules. The illustration itself assigns no such status.

Three outputs

The procedure asks three separate questions: does the specified material contain a closed chain of support; do the declared rules license the proposed conclusion; and in what order were the grounds added? An answer to one question does not answer the other two.

CELL is the technical name for a precisely defined closed structure in the annotation. It does not name a biological cell or a reader’s state. I specifies the input conditions, K the admissible situations, and P the checking rules.

QS · structure
Returns PRESENT if at least one CELL is found, ABSENT only after a complete search finds none, or UNVERIFIED if the contract is incomplete or the result cannot be independently reproduced. A CELL is an inclusion-minimal, strongly connected, closed attributive core with active self-return.
QF · evidential licensing
Assigns to each candidate formula the first applicable status in this order: UNVERIFIED, BEYOND-RECORD, LICENSED@K,P, or NOT-LICENSED@K,P.
ρ · order of grounds
Records the event chain and the predecessor-run hash; a structural result does not become an ordering claim. The checkable internal event order does not certify external calendar chronology.

Controlled opening

Before
The original input contains one CELL; the two tested formulas are NOT-LICENSED.
Added
The controlled extension adds a synthetic ground g₀ and a rule r₀. The ground has a recorded hash and is structurally independent of the internal cycle.
After
CELL is absent; the two formulas are LICENSED under the declared contract and policy.

Exact inputs and controlled example in the full paper · PDF

PRESENT + NOT-LICENSED → Δex → ABSENT + LICENSED

The comparison uses exact inputs and execution records with recorded hashes. For the fixed before-and-after input pair (Iex, Iex+), it demonstrates that a self-supporting structure identified in the record and the evidential licensing of its content can vary independently. The extension adds g₀ and rule r₀.

Scope

The study establishes no mental state, causal mechanism, medical fact, or comparative assessment of species. The presence of a form does not by itself justify a claim that what it names is present; the absence of a form does not by itself justify a claim that what it names is absent. The formal result is relative to the declared I, K, and P.

Abstract

A Constructive Separation of Self-Return and Evidential Licensing

This study fixes an exact written record as bytes and text spans, then locates the point at which a reading is used to support itself. The procedure computes three outputs separately: the self-return structure (CELL), the evidential status of candidate content, and the order in which a ground enters the record. None substitutes for either of the other two.

In the controlled example, the initial input yields a single CELL and NOT-LICENSED status for two candidate formulas. After a hashed extension adds a synthetic ground structurally independent of the internal cycle, the CELL disappears and both formulas become LICENSED. The results apply only to the declared input contract, policy, and specified carriers of the records.

Comparison with prior work and search limits

Novelty mapping

Thematic retrieval
1,890 records retrieved; 1,565 retained after deduplication; 249 underwent two independent, blinded, model-assisted screening passes.
Strict review
The conservative union contained 130 records; 52 unique high-signal records underwent strict review; the matrix covers the 23 closest lines of prior research.
Citation layer
The layer contained 2,751 records and 2,704 citation-only work units; all 138 units selected for this stage underwent strict abstract-scope review.
Russian-language supplement
16 unique records, 14 eligible works, and 0 confirmed full matches.
Cutoff
27 August 2026. No confirmed full match was identified in the completed layers.

Search bounds: Crossref and OpenAlex are not universal; 17 of 27 result sets reached the 100-record limit; all 15 base query strings were in English. Both screening passes were model-assisted, not human assessments, and κ does not measure recall. In the base strict-review layer, 71 of 87 reviews relied on abstracts and 16 on full text. The citation-layer strict review used 138 abstracts and no full texts; 123 OpenAlex IDs remained unresolved. The Russian-language supplement covers material accessible to the crawler; it is not an exhaustive database search.

Novelty claim

The mapping identified this as the first reproducible procedure of this exact type in the literature covered through 27 August 2026. It analyzes the form of the record, computes self-return structure, and assesses the evidential licensing of content separately.

The comparison target is the exact F1–F6+F7′ configuration. F7′ is a bounded check that the procedure’s own report receives no special privilege: without a separate complete Iself, that report receives UNVERIFIED. A complete self-applied F7+ run is not claimed and remains UNVERIFIED.

This is a finding from a disclosed search, not proof of unconditional historical priority. No novelty is claimed for cyclicity, external support, minimal cores, provenance, exact annotation, or self-reference considered separately.

Materials and reproducibility

English language edition · 13 September 2026. Original research files are listed separately below.

Download the English reading edition · ZIP

English language editionPDF · 55 pp.
Other formats and language
English language edition — editable textDOCX Russian version — reading copyPDF · 56 pp. Russian manuscript — source included in the packageDOCX
Original release: reproduce the checks
Systematic mapping ledgerXLSX Original source: Reproducibility packageZIP · 153 files Original source: Package checksumSHA-256 Original web-copy checksumsSHA-256
Version and release status

The original release, version 1.2.3, was finalized on 27 August 2026. The English language edition dated 13 September 2026 is a separate reading copy based on that release. Its filenames retain the source version number; the original bindings and checksums do not cover the revised files.

Original package integrity

The two DOCX files inside the original version 1.2.3 ZIP are the manuscripts identified by that release’s bindings. The bindings and byte-identity checks apply to those original files. The revised English PDF and DOCX linked above belong to the separate language edition dated 13 September 2026 and are not covered by those bindings.

The original binding establishes byte identity and the declared relation of that release’s Russian source to its English translation. It does not authenticate authorship, origin, semantic equivalence, or calendar priority; neither manuscript contains its own hash.

ZIP SHA-256
c84efe6573e993558c1e9df96659d3ad7af3e052b7604072990f81949fbd3156

Original release verification: formal recomputation PASS; mutation controls 25/25 detected; 153 files, 152 checksum entries, and two manuscript bindings.

Citation
Shebetov, K. (2026). Reading as Enclosure: A Constructive Separation of Self-Return and Evidential Licensing (English language edition, 13 September 2026; based on version 1.2.3). beforeword. https://beforeword.xyz/research/reading-as-enclosure/en/