Generous and terrifying: the best late homework policy of all time

I want all of my interactions with students to be about the transmission of wondrous ideas. All the other bullshit should be defined out of my life as an educator.

But life happens, and students can flake on you and on their classmates, and if you don’t discourage it, it gets worse. So now the transmission of wonder is being crowded out by discussion about your late policy. And late policies are a trap.

For a softy like me, any policy that is strong enough to actually discourage tardy work is too harsh to be credible. To say NO LATE WORK WILL BE ACCEPTED is all well and good until you hit the exceptions: personal tragedies you don’t want to know about, the student who thoughtfully gave you three weeks advance notice by email, your own possible mistakes. Suddenly you’re penalizing thoughtfulness, incentivizing students to dishonestly inflate their excuse into an unspeakable tragedy, and setting yourself up to be the stern looker-past-of-quivering-chins. And what’s the alternative? 10% off for each day late? I don’t want to be rooting through month-past late-night emails from stressed students, looking up old deadlines, counting hours since submission, or calculating 10% decrements for this person and 30% for that one, especially not when such soft alternatives actually incentivize students to do the math and decide that 10% is worth another 24 hours. Plus, with all of these schemes, you’re pretending you care about a 10:02 submission on a 10:00 deadline—or even worse, you’re forgetting reality and convincing yourself that you actually do care.

My late policy should be flagrantly generous and utterly fearsome. It should be easy to compute and clear and reasonable. It should most certainly not increase the amount of late work, especially because that increases the work on me. It should be so fair that no one who challenges it has a leg to stand on, and so tough that all students are very strongly incentivized to get their work in on time. It should softly encourage students to be good to themselves, while allowing students flexibility in their lives, while not being so arbitrarily flexible that you’re always being challenged and prodded for more flexibility.

What I wanted was a low effort, utterly fair policy that nevertheless had my students in constant anxiety for every unexcused minute that they were late.


Is that even possible? Meet the Gamble Protocol. It’s based around one idea: because humans are risk averse, you can define systems that students simultaneously experience as rationally generous and emotionally terrifying. All you have to do is create a very friendly policy with small, steadily increasing probabilities of awful outcomes.

The Gamble Protocol is a lot like the well-known “10% off for every day late.” In fact, in the limit of infinite assignments, they’re statistically indistinguishable. Under the Protocol, a student who gets an assignment in before the deadline has a 100% chance of fair assessment of their work. After the deadline, they have a steadily increasing chance of getting 0% credit for all of their hard work. No partial credit: either a fair grade or nothing at all. On average, a student who submits 100 perfect assignments at 90% probability gets an A-, not because all submissions got 90%, but because ten got 0%. A bonus, for my purposes, is that I teach a lot of statistical reasoning, so the Protocol has extra legitimacy as an exercise in experiential learning.

After experimenting a bit, and feeling out my own feelings, I settled on the following: for each assignment, I draw a single number that applies to everyone (rather than recalculating for every late student). I draw it whenever I like, and I always tell students what number got drawn, and how many students got caught. The full details go in the syllabus:

Deadline. If the schedule says something is due for a class, it is due the night before that class at 10:00PM. There is no partial credit for unexcused lateness; late assignments are worth 0%. However, assignments submitted after the deadline will get a backup deadline subject to the Gamble Protocol.
The Gamble Protocol. I will randomly generate a backup deadline between 0 and 36 hours after the main deadline, following a specific pattern. Under this scheme:

  • an assignment that is less than 2 hours late (before midnight), has a 99% chance of earning credit,
  • an assignment turned in before 2:00AM has a 98% chance of earning credit,
  • an assignment turned in 12 hours late, by 10AM, has a 90% chance of earning credit,
  • that jumps suddenly down to 80% between 12–14 hours, getting worse faster,
  • an assignment turned in 24 hours late, before the next 10:00PM, has a 60% chance of earning credit,
  • and an assignment turned in more than 36 hours late is guaranteed to earn zero credit.

I will not calculate the backup deadline until well after its assignment was due.

Calculating is easy. For each assignment,

  • you can put the following numbers in a hat and draw:

    0 2 4 5 6 7 8 9 10 11 12 12 12 12 12 12
    12 12 12 12 14 15 16 17 18 19 20 21 22 23 14 15
    16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
    32 33 34 35 24 25 26 27 28 29 30 31 32 33 34 35
    24 25 26 27 28 29 30 31 32 33 34 35 24 25 26 27
    28 29 30 31 32 33 34 35 24 25 26 27 28 29 30 31
    32 33 34 35
  • or you can open any online R console and paste this code:
    deadline <- c( 0,2, c(4,5,6,7,8,9,10,11), rep(12, 10), rep(14:23,2), rep(24:35,5) ) sample(deadline)[1]

I'm keeping data from classes that did and did not use this policy to see if it reduces late work. I still haven't chugged any of it, but I will if requested. For future classes, I was thinking of extending from 36 hours to a few days, so that it really is directly equivalent to 10% for a day's tardiness.