Do not describe an unwatched video as an observed scene

A written description and an observed scene are different evidence

A title naming a performance or version can support describing that label. It does not establish an expression, movement or camera view. Rewriting description metadata as a personally seen scene changes the kind of evidence implied by the sentence.

The same distinction applies to other sites: a supplied document, program output and directly used interface each support different scopes. A substantial amount of reading does not replace a new observation needed by an additional claim. Identify that change before drafting more detail.

Link each vivid statement to the source sentence or actual observation supporting it. A missing link can call for more evidence. Reducing certainty in the wording does not create the observation that the original claim required.

Inspect a family of claims instead of one phrase

An August 7 edit groups statements such as we can see, a specific video showing something and an opening scene into one claim type. Different wording can carry the same unsupported assumption. Historical results also show conditions being broadened and previous drafts re-evaluated.

These are boundaries for English manuscripts at the time, not guaranteed interpretation across every language and sentence. Read the surrounding prose and the field’s responsibility after a match. A regex condition should not inherit the complete responsibility for semantic judgment.

A general explanation could trigger a real rejection too

The cycle recorded early on September 30 in Korea stops at an on-screen phrase in an international-reader context field. That field describes a format generally rather than claiming observation of one scene. The later context change and passing output are connected. A rejected phrase does not independently establish a fabricated visual claim.

I distinguish explaining a concept from describing a specific upload as watched. Both need accuracy, but their evidence requirements differ. After rejection, inspect the purpose and required evidence instead of merely hiding the matched words. That keeps a checker’s useful alert from becoming an unquestioned verdict.

Do not replace unavailable scene evidence with other facts

The operating document also retains a hold condition for track names visible only inside a video when they cannot be read. Other metadata does not authorize guessing that on-screen text. A story missing its essential evidence differs from a story adequately supported within a narrower scope.

The new review prompt below restates the boundary for another task. Changing a phrase does not create observation. Decide whether to seek evidence, hold the story or remove an unnecessary claim while still answering the reader’s question.

Evidence available: title, description, observed metrics
Claim to review: a specific expression, movement or camera view
Needed next evidence: actual observation of that scene
Decision if unavailable: remove the unsupported claim or hold the story

This is a new review prompt, not a proof that every regex match is false.

Read the same boundary in drafting and publication

The drafting request states the unwatched condition, while publication also checks claim types. Clear instructions did not ensure every result stayed within scope. I want the writing and review stages to read the same evidence boundary, beyond retaining a longer forbidden-word list.

A newly invented scene example should not be presented as an actual historical draft. This account connects code edits, rejection outputs and the context field’s before-and-after input. It does not establish present viewing of all original videos or verification of every automatic verdict. Apply the boundary again to each new source and claim.