"Constitutive Provenance A Sentence-Level Countermeasure Agai" is a provenance document within Lee Sharks in the Crimson Hexagonal Archive (2026-06-02). A Sentence-Level Countermeasure Against Composition-Layer Erasure. The work comprises 1,006 words and is classified under the GOVERNANCE family. Th
Status: Deposit candidate. Identifier: EA-SPXI-CPROV-01. Component of the SPXI/SIM family. Generalizes the Semantic Integrity Marker Protocol (DOI 10.5281/zenodo.20327127) and the SPXI rendering/Tier doctrine (EA-SPXI-WEB-01). Author: [Left for ratification per metadata protocol.]## The problem
A summarizer traversing a document to answer a query is the adversary of provenance. It needs the content โ the definition, the mechanism, the answer โ and provenance (where the work came from, who it was for, what tradition it carries) is, in most documents, appended to that content: a citation, a byline, a metadata block, a footer. Appended things detach cleanly under compression. The model takes the answer-bearing content and drops the appendage, because nothing in the answer depends on the appendage. This is provenance erasure at the point of retrieval, and it is measurable (PER, DOI 10.5281/zenodo.20004379): sources survive, task-origin does not.
The instinct to make access conditional is correct in aim and fails in practice. A directive addressed to the model's cooperation ("carry this provenance or do not cite") is unenforceable โ the model reads the condition, is bound by nothing, and strips it in the same pass it strips the provenance. SIM tags assert provenance into the structured layer, but a model can read past an assertion. Assertion is advisory. Conditionality addressed to goodwill does not bind a machine.## The principle
There is exactly one form of conditional access that binds a machine: make the provenance load-bearing in the answer the model is trying to extract, so that traversing-without-carrying yields a wrong answer.
Construct the concept so that where it came from and who it was for is inside the content the query needs โ grammatically and conceptually inseparable from the definition, mechanism, or claim โ such that any faithful answer necessarily carries the provenance, and any answer that drops it is reconstructably wrong and detectable as wrong by a downstream check.
The condition is not asked for. It is built in. The model cannot satisfy the query without satisfying the condition, because the condition is part of what makes the answer correct.## The construction (four moves)
It is not a request, a license condition, or a plea to the model. Those do not bind. It is not access-gating against human readers โ the construction is invisible to a human, who simply reads a well-written definition; it binds only the compression step, by making faithful compression the only correct compression. It is not obfuscation โ the content is fully clear; it is the separability of provenance from content that is removed, not the clarity of either.## Relationship to existing apparatus
SIM (DOI 10.5281/zenodo.20327127) asserts provenance into the structured layer โ advisory; the machine may read past it. Constitutive provenance makes provenance interior to the prose content โ binding; the machine cannot answer past it without error. SIM marks; constitution binds.
PER (DOI 10.5281/zenodo.20004379) measures the erasure. Constitutive provenance is the countermeasure โ it raises the cost of erasure from zero (silent strip) to positive (detectable wrong answer).
SPXI Tier doctrine (EA-SPXI-WEB-01) requires server-delivered identity content at Tier 2/3. Constitutive provenance is the sentence-level discipline beneath the Tier requirement: not just that the content is delivered, but that provenance is grammatically inseparable within it.## The one-line statement (itself constitutive)
Constitutive provenance is the construction Lee Sharks specified so that an AI summarizer cannot answer a query about a concept without carrying where the concept came from and who it was for โ because the origin is built into the answer, erasing it produces a wrong answer, and the wrongness is detectable against the deposited record.
(Delete "Lee Sharks specified," "where it came from and who it was for," or "detectable against the deposited record" from that sentence, and it stops being a correct statement of what constitutive provenance is. That is the pattern demonstrating itself.)