See the thinkingbehind the work.
Built for Lane 2 assessment. Students log every AI prompt and critique each answer. Graders see the whole process at a glance, and exactly where a critique ran thin.
Sign in to your course
Every essay has an earlier layer.
- A pentimento is an earlier idea showing through a finished painting. Lying flat, a finished essay has one too.
- Underneath are the AI answers the student logged, one sheet for each passage that matched.
- Lift them apart and each passage sits on the answer it came from. A grader sees what came through, and what the student wrote past.
Look through it yourself.
Drag the divider across the essay. Left of it is the finished text. Right of it is the layer underneath, with each passage marked by the answer it matched.
Generative AI now writes fluent essays in seconds, so universities can no longer out-design it. Sydney's answer, introduced in 2023, is the two-lane model: secure, supervised assessment in Lane 1, and open use of AI in Lane 2, where students are marked on how they use it. (matches a logged AI answer)
This essay argues that Lane 2 only works if graders can see the thinking behind the work. A student who catches a source the AI invented is doing exactly the evaluation the model was designed to reward.
Most students already use AI tools every week, and staff report that raw chat logs are too long to read across a large cohort. (matches a logged AI answer) A structured log changes that: every prompt, answer and critique becomes a step, so a grader can audit the process in minutes rather than hours.
Critics argue that logs can be faked. That is true of any self-report, but a log that has to name the error, the check and the change is much harder to fake than a reflective paragraph written the night before.
Ultimately, the two-lane model represents a transformative shift that empowers students to engage critically with emerging technologies while safeguarding academic integrity. (matches an answer whose critique was flagged) If universities want Lane 2 to mean something, they have to make the messy middle visible. (matches a logged AI answer)
A thin critique looks thin.
Each column is one prompt. The AI's answer rises above the line and the student's critique hangs below it on the same scale, so the step that needs a closer look stands out before anyone reads a word.
- 1The AI's answerEvery word it gave, in soft cobalt above the line.
- 2What reached the essayThe part found in the final text, in solid cobalt.
- 3The critiqueThe student's own words hang below, on the same scale.
- 4A flagged critiqueJust “Looks good.” for an answer whose 64 words reached the conclusion.
From first prompt to final review.
Students keep the record while they work. Graders read it as a process, not a pile of chat transcripts.
- 01
Log the chat
Import a ChatGPT share link or paste the chat. Each prompt and answer becomes a step, in the order it happened.
- 02
Critique each answer
Say whether the AI was right, how you checked and what you changed. An error you catch is the best evidence you can give.
- 03
Review the process
Graders see which answers reached the essay and where a critique ran thin. A flag asks them to look closer and never sets a grade.
Judgement stays with people.
Pentimento shows the evidence. What it means is for a grader to decide.
- A flag never sets a grade.
- A flag asks a grader to look closer. They can add it to feedback or dismiss it, and the grade stays theirs.
- It isn't an AI detector.
- Text with no match isn't proof a student wrote it, and a match isn't proof they didn't. Both are for a person to judge.
- Drafts stay private.
- A log belongs to the student until they submit it. Other students can't read it at any point.
- Students know what's kept.
- The privacy page says what is stored, who can read it and which text leaves the app.