Edit history signals

What edit history is

Word-processor documents — DOCX, ODT, RTF — record more than just the final text. They keep a trail of:

• Tracked changes that were inserted or deleted at some point and may or may not have been accepted.
• Hidden text runs marked with a formatting flag that the reader doesn't see by default.
• Revision marks on every paragraph showing when (and by whom) it was last edited.
• Comments and their replies, including resolved threads.
• The embedded baseline — the previous version Word keeps cached for "compare to last save" features.

Most of this exists to support collaborative editing. None of it is exposed in the rendered document a reader sees. Autotend Forensics surfaces it because it's where copy-paste-then-clean workflows tend to leave residue.

What it tells you

The most useful edit-history signals are:

• Accept-all residue. If a student pastes a draft from an AI tool and the AI's diff machinery left tracked changes behind, "Accept all changes" hides them visually but leaves the acceptance metadata in the file. Autotend Forensics surfaces the count of accepted-but-recorded changes.
• Hidden-text blocks. Sometimes used legitimately for printer instructions; almost never used legitimately in student essays. A 500-word hidden block tucked at the end is loud.
• Comment residue from a different person. If a student's document carries comments authored under a name that isn't the student's, that's a strong signal something came from elsewhere.
• Revision-mark divergence from edit time. Documents with a high revision-mark count but very low recorded edit time tend to be paste-edits, not real authoring.

What it cannot tell you

Edit history is asymmetric: presence of an anomaly is a strong signal, absence is weak. A student who pastes a clean block of AI text directly (no diff trail) leaves no edit-history residue at all. The metadata channel and the linguistic channel are better suited for that case.

Edit history also has known false-positive paths:

• Students editing across multiple machines may show divergent Creator and Last modified by — that's just a laptop + lab PC workflow.
• Some IT environments rewrite document metadata on save (campus-wide Word configurations with custom templates).
• Template-baked hidden text from old assignment templates can trip the hidden-text detector if you don't recognize the source.

What we surface

Autotend Forensics flags:

• Accept-all-tracked-changes residue with a count and the recorded session timestamps.
• Hidden text blocks with their content visible in the inspector (so you can decide whether it's an instruction comment or a paste-buffer).
• Comments whose author doesn't match Creator / Last modified by.
• High revision-mark count with low total edit time.

As with every signal, none of these prove anything on their own. They're prompts to ask the student where the original draft came from.