A student sent you a link. Here is what you are looking at.
Two minutes, written for the person who has to make the call.
Receipts is a writing tool that records the process of writing as it happens: every keystroke, pause, deletion, and paste, with timing. The link your student sent replays their document being written, from the first character to the last. It was recorded while they wrote, not reconstructed afterwards.
How to read the replay
Press play and watch for the texture of composition. Real writing is uneven. It stalls while the writer thinks, sprints through familiar ground, doubles back to fix a sentence, and abandons wording mid stream. You will see hesitations before sentences, words deleted and replaced, and passages that were written and then cut entirely. That texture is difficult to manufacture and easy to recognise.
The signals, briefly
Process depth (0 to 100) How much of the writing process the record carries. It measures process, not prose quality, and it cannot be raised by pasting finished text.
Rhythm variance and cadence shape Human typing is uneven and long tailed. Scripts and transcription cluster tightly around one speed.
Revision ratio and non linear edits How much was deleted per character produced, and how often the writer returned to earlier text. Copying produces almost none of either.
Paste disclosure Pasted text is recorded and disclosed, never hidden. Moving one's own paragraph is distinguished from importing outside text.
Abandoned ideas The share of deleted words that never appear anywhere in the finished essay. Writing forward in time leaves dead ends, because the writer does not yet know where the argument lands. A history reconstructed backwards from a completed essay has very few, since the finished text is all its author had to work from. This counts in the student's favour when present and counts for nothing when absent, because plenty of people draft cleanly.
Key hold times How long each key was physically held down, which is a separate measurement from the gaps between keystrokes. Human hold times sit in a band with real spread. Injected or scripted keystrokes typically release instantly and hold every key for the same duration. Phone keyboards, dictation and assistive input produce no hold data at all, so a blank reading here means nothing either way and is never counted against a student.
Reworked spans Places the writer returned to three or more times, minutes apart. Revision in real documents is not one clean pass.
One caution worth stating plainly: several of these signals only ever count in the student's favour. That asymmetry is deliberate. A student who types on a phone, uses dictation, drafts on paper first, writes with two fingers, or simply never deletes anything will show blanks in places, and none of those blanks are evidence of anything at all.
What this record proves, and what it does not
It shows: that this text accumulated over time, with revision, hesitation, and non linear editing, on this student's device, in the rhythm shown. Pasted material, if any, is disclosed on the record rather than concealed.
It cannot prove: that the student did not read AI output from a second screen and retype it. No tool can prove that. Retyping does leave a visible signature, a flat rhythm with almost no revision and no abandoned wording, and the record makes that visible too, but the judgment belongs to you, not to the tool.
It is not unforgeable, and we do not claim it is: a determined person could drive a keyboard with software and manufacture a record. Doing it convincingly means reproducing hold times, finger transitions, pauses that land on the hard words, and ideas that get abandoned and never return, live and in real time across the whole essay. That is a great deal more work than writing the essay, which is the honest standard this record is built to meet. It is evidence for you to weigh, not a verdict, and no single signal on it should decide anything by itself.
Why this is stronger than a detector score
AI detectors estimate how statistically predictable a text is and output a percentage. They carry documented false positive rates, which run highest for non native English speakers, and several institutions have disabled them for exactly that reason. A detector hands you an opinion about the finished text. This record hands you evidence about how the text came to exist. You are not being asked to trust a number. You are being asked to watch the writing happen and judge it yourself.
Integrity of the record
The record carries a cryptographic hash computed over the full event log, shown on the exhibit page. It demonstrates the record has not been altered since it was produced. The record is self attested, meaning it was produced by the student's own browser; the independent cross check, where available, is the document's Google revision history, whose timestamps are written by Google rather than by the student or by us.
Receipts is free for students and requires nothing from you: no account, no download. Questions or feedback from educators are welcome. About Receipts