Public Release - 2026-07-18

When AI writes too smoothly, the writer can disappear first.

Artifact: identity_compression_check_v1. This note separates style from authorship: when the text becomes too smooth, the review record should still show which voice is native, which phrase is borrowed, and which dissent survives.

claim_id source_voice borrowed_pattern_ratio lived_experience_anchor dissenting_view reusable_phrase audience_memory evidence_timestamp boundary_update review_status
Identity Compression Check v1 public evidence visual.

Artifact

Smooth text can hide a shallow voice.

AI writing can look fluent, familiar, and correct while erasing the lived anchor that makes a piece distinct. The review object asks which voice is native, which parts are borrowed, and where the text still carries a real disagreement.

The risk is not only plagiarism. It is flattening.

Compression

The risk is not only plagiarism. It is flattening.

When every paragraph starts to sound the same, the audience sees pattern before judgment. The check names the borrowed ratio, the reusable phrase, and the memory it leaves behind so a smooth finish does not count as substance by default.

If audience memory is empty, the page has compressed too far.

Review

If audience memory is empty, the page has compressed too far.

A useful challenge should identify a missing field, an unsupported causal link, a stale evidence timestamp, or a condition under which the proposed control fails. If the piece only sounds like a content machine, the boundary update needs to be explicit.

1

claim_id

The writing, authorship, or content claim under review.

2

source_voice

The voice that the text claims as its own.

3

borrowed_pattern_ratio

The amount of phrasing, structure, or cadence copied from a familiar pattern.

4

lived_experience_anchor

The concrete event, practice, or observation that keeps the text from floating off.

5

dissenting_view

The sentence or section that survives disagreement instead of smoothing everything flat.

6

reusable_phrase

The line that could be reused anywhere and therefore needs scrutiny.

7

audience_memory

What the reader is likely to remember after the page is closed.

8

evidence_timestamp

The source date, publication date, or context window that anchors the claim.

9

boundary_update

The scope change required when style starts replacing authorship.

10

review_status

The recorded state: supported, narrowed, pending, or retired.