Draft history.

Draft history

A draft history is the timestamped sequence of edits that produced the final essay. Google Docs stores it automatically; Word's track-changes feature stores it on request; Git stores it by design. In academic-integrity disputes, draft history is often the single strongest evidence that an essay was written, not generated.

Detection tools score the final text. Draft history shows the process. The two together are decisive in a way either alone is not.

What draft history actually proves

Version history shows the essay's growth, early messy passages, deletions, paragraph reordering, additions. AI-drafted essays usually arrive in one or two large paste events with little subsequent revision. Human-drafted essays show progression: small additions over hours or days, sentences crossed out and rewritten, paragraphs moved. The pattern of edits, more than any single edit, is the evidence.

Tooling that supports it by default

Google Docs version history is on by default and survives indefinitely. Microsoft Word's track-changes is opt-in but persistent once enabled. Notion captures revision metadata at block level. Git captures everything by design. The single most useful piece of pedagogical advice in 2026 is "draft your essays in a tool with version history enabled, every essay, from week one." The student-side how-to is on /for-students/how-to-appeal.

How draft history interacts with related metrics

Draft history functions as a temporal scaffold for other detection signals. Edit velocity, measured as the ratio of changed characters to elapsed seconds, derives meaning only when anchored to a timestamped sequence of drafts. A 400-word insertion appearing in version 3 after 90 seconds triggers different risk thresholds than the same text emerging gradually across versions 8 through 12 over 40 minutes. Revision depth, quantified by Levenshtein distance between consecutive snapshots, likewise depends on granular draft capture. Systems that sample every 15 seconds produce exponentially richer datasets than those polling at two-minute intervals, enabling differentiation between human pause patterns and the characteristic step-function output of language models.

Metadata integrity amplifies or undermines the value of revision logs. A draft history showing 47 versions across 18 minutes loses diagnostic power if client-side timestamps lack server validation, since motivated users can manipulate local clocks or inject fabricated edit events. Cross-referencing paste event listeners with sudden content deltas strengthens causal inference. When version 6 shows a 280-word block insertion coinciding with a recorded paste action at timestamp 14:32:18, the combined signal carries more weight than draft count alone. Session continuity markers, including IP consistency and browser fingerprint stability, contextualize whether a 12-hour gap between versions 3 and 4 reflects genuine overnight rest or a handoff between human and machine agents.

Practical implications for institutions and educators

Institutions implementing draft-tracking workflows must address consent frameworks and storage obligations. The European Union's General Data Protection Regulation classifies keystroke logs and incremental text snapshots as personal data requiring explicit opt-in and defined retention windows. A university mandating draft history collection for high-stakes assessments bears responsibility for encrypting at-rest logs, anonymizing datasets used in aggregate analysis, and purging records within documented timelines. Failure to separate identifying metadata from revision content exposes institutions to regulatory penalties. The California Consumer Privacy Act similarly grants students rights to access, export, and request deletion of their draft histories, obligating registrars to maintain retrieval systems that can isolate individual records from bulk archives within 45 days.

Pedagogical strategy shifts when revision data becomes visible to instructors. A composition course publishing anonymized draft metrics as formative feedback helps students recognize that typical essays undergo 8 to 15 substantial revisions, countering unrealistic expectations that strong writing emerges in two passes. Conversely, high-resolution tracking risks chilling authentic exploration if students fear that deleted paragraphs or abandoned thesis statements will be scrutinized. Transparency protocols that disclose which attributes trigger review, such as sub-threshold edit counts below 4 versions or mean inter-draft intervals exceeding 90 minutes, allow learners to calibrate effort while preserving academic honesty standards. Institutions report 23 percent reduction in detection flags when rubrics explicitly allocate completion credit for minimum draft milestones, incentivizing process documentation without penalizing non-linear workflows.